Benchmark run gpt-6-luna-all-r10-c5-prescriptive-postfix-20260927
This page contains exactly the publishable benchmark observations recorded in run gpt-6-luna-all-r10-c5-prescriptive-postfix-20260927; it does not select or aggregate them. Return to the run index or view aggregate metrics.
Interactive browser
Rows are pre-aggregated by task, target, prompt variant, MCP-instruction mode, context, arm, and restriction. Numeric values show median (p50), IQR, p90, maximum, and sample count. Boolean checks show pass counts and success rates. Filters apply to these summary groups; per-run observations remain available on their run pages.
Loading benchmark results…
Baseline vs semedit
Comparison evaluates baseline and semedit values within each restriction policy, target, and context group. Negative changes mark improvements for these lower-is-better metrics; positive changes mark degradations.
Detailed observations
Empirical telemetry measuring wall-clock latency, token usage, tool turns, and compiler correctness when AI coding agents perform code modifications.
Every benchmark pairs Vanilla LLM (standard file editing tools) against Semedit MCP (deterministic AST compiler operations).
Benchmark Methodology & Transparency
- Reproducibility & Provenance: Self-contained txtar archives define test scenarios. Links below resolve to commit-anchored GitHub source files for published commits, or cryptographic SHA-256 fingerprints for uncommitted local fixtures.
- Model Cost: Tables report model-specific credit consumption where a target model has a declared schedule: (uncached input × input credits + cached input × cached credits + (reasoning + visible output) × output credits) / 1,000,000. Thinking tokens use the output rate. ADR-0043 defines the per-million-token credit rates.
- Multi-Level Correctness Oracle: Four validation levels evaluate each trial:
- Level 1 (Mutation Policy): Restricts file modifications strictly to authorized paths.
- Level 2 (AST Invariants): Compiler AST verification of required symbols, imports, and relative declaration ordering.
- Level 3 (Clean Build): Strict compilation validation with zero compiler or typecheck errors.
- Level 4 (Test Suite): Unit and integration test suite execution.
- Isolated Sandboxing: Each trial executes in an ephemeral directory with pristine git and module state.
Test case: task-04-insert-public
Target: codex/gpt-6-luna/medium
default prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-04-insert-public |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | default |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/bench/task_04_insert_public.txtar |
Vanilla LLM Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
Add public constructor func InitServer() *Server placed before private helpers. Do not use semantic editing MCP tools; use standard file editing. When done, output DONE.
Semedit MCP Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
Add public constructor func InitServer() *Server placed before private helpers. Use semedit semantic tools for supported source code modifications. Shell commands for builds and tests are allowed. When done, output DONE.
Initial Workspace State (Before Edit)
package api
type Server struct {
Port int
}
func (s *Server) Start() {}
func (s *Server) internalRun() {}| Metric | Vanilla (Small) | MCP (Small) | Δ (Small) | Vanilla (Large) | MCP (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | 22.18s | 29.59s | +33.4% | 18.69s | 29.89s | +59.9% |
| Process Start → First Event | 0.78s | 0.72s | — | 0.12s | 0.14s | — |
| First Event → First Tool Call | 10.46s | 5.31s | — | 4.28s | 6.46s | — |
| MCP Initialize → First Semantic Call | — | — | — | — | — | — |
| MCP Server Start → Initialize | — | — | — | — | — | — |
| Top-Level User Turns | 1 | 1 | 0% | 1 | 1 | 0% |
| Internal Tool Cycles | 3 | 4 | +33.3% | 3 | 6 | +100.0% |
| Initial Load / Discovery Turns | 2 | 1 | -50.0% | 2 | 2 | 0% |
| MCP Discovery / Schema Turns | 0 | 0 | — | 0 | 0 | — |
| Tool Invocations (per run) | 3 | 4 | +33.3% | 3 | 6 | +100.0% |
| Output Tokens | 357 | 407 | +14.0% | 453 | 691 | +52.5% |
| Reasoning / Thinking Tokens | 39 | 89 | +128.2% | 87 | 238 | +173.6% |
| Total Input Tokens | 76110 | 133394 | +75.3% | 76107 | 206130 | +170.8% |
| Cached Input Tokens | 68608 | 117248 | +70.9% | 66560 | 184320 | +176.9% |
| Uncached Input Tokens | 7502 | 16146 | +115.2% | 9547 | 21810 | +128.4% |
| Cost | 0.0409 | 0.0759 | +85.7% | 0.0473 | 0.1122 | +137.5% |
| Cached vs Uncached Token Ratio | 9.15:1 | 7.26:1 | -20.6% | 6.97:1 | 8.45:1 | +21.2% |
| Oracle L1: Mutation Policy | PASS | PASS | — | PASS | PASS | — |
| Oracle L2: AST Invariants | PASS | PASS | — | PASS | PASS | — |
| Oracle L3: Clean Build | PASS | PASS | — | PASS | PASS | — |
| Oracle L4: Verification Test | PASS | PASS | — | PASS | PASS | — |
| Semantic Tool Invocation Verified (per run) | N/A | PASS | — | N/A | PASS | — |
Baseline vs Semedit Agentic in Small Context
- Vanilla (Baseline vs Semedit Agentic in Small Context) Edit: File api/server.go modified (9 lines -> 13 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Small Context) Edit: File api/server.go modified (9 lines -> 13 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ✓ ✓ | semedit/semantic_outline (✓ ✓) |
| 3 | ✓ ✓ | semedit/semantic_inspect_symbol (✓ ✓) |
| 4 | — | semedit/semantic_insert_function (✓ ✓) |
Baseline vs Semedit Agentic in Large Context
- Vanilla (Baseline vs Semedit Agentic in Large Context) Edit: File api/server.go modified (9 lines -> 13 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Large Context) Edit: File api/server.go modified (9 lines -> 13 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ✓ ✓ | ✓ ✓ |
| 3 | ✓ ✓ | semedit/semantic_outline (✓ ✓) |
| 4 | — | semedit/semantic_insert_function (✓ ✓) |
| 5 | — | ✓ ✓ |
| 6 | — | ✓ ✓ |
default prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-04-insert-public |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | default |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/bench/task_04_insert_public.txtar |
Vanilla LLM Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
Add public constructor func InitServer() *Server placed before private helpers. Do not use semantic editing MCP tools; use standard file editing. When done, output DONE.
Semedit MCP Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
Add public constructor func InitServer() *Server placed before private helpers. Use semedit semantic tools for supported source code modifications. Shell commands for builds and tests are allowed. When done, output DONE.
Initial Workspace State (Before Edit)
package api
type Server struct {
Port int
}
func (s *Server) Start() {}
func (s *Server) internalRun() {}| Metric | Vanilla (Small) | MCP (Small) | Δ (Small) | Vanilla (Large) | MCP (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | 15.20s | 103.96s | +584.1% | 17.33s | 16.41s | -5.3% |
| Process Start → First Event | 0.14s | 0.13s | — | 0.12s | 0.12s | — |
| First Event → First Tool Call | 3.79s | 7.15s | — | 5.04s | 4.25s | — |
| MCP Initialize → First Semantic Call | — | — | — | — | — | — |
| MCP Server Start → Initialize | — | — | — | — | — | — |
| Top-Level User Turns | 1 | 1 | 0% | 1 | 1 | 0% |
| Internal Tool Cycles | 3 | 8 | +166.7% | 3 | 3 | 0% |
| Initial Load / Discovery Turns | 2 | 2 | 0% | 3 | 1 | -66.7% |
| MCP Discovery / Schema Turns | 0 | 0 | — | 0 | 0 | — |
| Tool Invocations (per run) | 3 | 8 | +166.7% | 3 | 3 | 0% |
| Output Tokens | 457 | 1342 | +193.7% | 399 | 289 | -27.6% |
| Reasoning / Thinking Tokens | 85 | 618 | +627.1% | 84 | 9 | -89.3% |
| Total Input Tokens | 75725 | 204512 | +170.1% | 96255 | 100123 | +4.0% |
| Cached Input Tokens | 66560 | 186112 | +179.6% | 85760 | 87808 | +2.4% |
| Uncached Input Tokens | 9165 | 18400 | +100.8% | 10495 | 12315 | +17.3% |
| Cost | 0.0463 | 0.1170 | +152.6% | 0.0537 | 0.0565 | +5.1% |
| Cached vs Uncached Token Ratio | 7.26:1 | 10.11:1 | +39.3% | 8.17:1 | 7.13:1 | -12.7% |
| Oracle L1: Mutation Policy | PASS | PASS | — | PASS | PASS | — |
| Oracle L2: AST Invariants | PASS | PASS | — | PASS | PASS | — |
| Oracle L3: Clean Build | PASS | PASS | — | PASS | PASS | — |
| Oracle L4: Verification Test | PASS | PASS | — | PASS | PASS | — |
| Semantic Tool Invocation Verified (per run) | N/A | PASS | — | N/A | PASS | — |
Baseline vs Semedit Agentic in Small Context
- Vanilla (Baseline vs Semedit Agentic in Small Context) Edit: File api/server.go modified (9 lines -> 13 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Small Context) Edit: File api/server.go modified (9 lines -> 13 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ✓ ✓ | ✓ ✓ |
| 3 | ✓ ✓ | semedit/semantic_outline (✓ ✓) |
| 4 | — | semedit/semantic_insert_function (✓ ✗): unknown param "language"; supported keys are [access, access_modifier, auto-organize-imports, auto_organize_imports, file, placement, source, target, target_symbol]: invalid operation parameters |
| 5 | — | ✓ ✓ |
| 6 | — | ✓ ✓ |
| 7 | — | semedit/semantic_insert_function (✓ ✓) |
| 8 | — | ✓ ✓ |
Baseline vs Semedit Agentic in Large Context
- Vanilla (Baseline vs Semedit Agentic in Large Context) Edit: File api/server.go modified (9 lines -> 13 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Large Context) Edit: File api/server.go modified (9 lines -> 13 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ? ✗ | ? ✗ |
| 3 | ✓ ✓ | semedit/semantic_insert_function (✓ ✓) |
default prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-04-insert-public |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | default |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/bench/task_04_insert_public.txtar |
Vanilla LLM Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
Add public constructor func InitServer() *Server placed before private helpers. Do not use semantic editing MCP tools; use standard file editing. When done, output DONE.
Semedit MCP Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
Add public constructor func InitServer() *Server placed before private helpers. Use semedit semantic tools for supported source code modifications. Shell commands for builds and tests are allowed. When done, output DONE.
Initial Workspace State (Before Edit)
package api
type Server struct {
Port int
}
func (s *Server) Start() {}
func (s *Server) internalRun() {}| Metric | Vanilla (Small) | MCP (Small) | Δ (Small) | Vanilla (Large) | MCP (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | 15.05s | 28.00s | +86.1% | 18.41s | 21.91s | +19.0% |
| Process Start → First Event | 0.72s | 0.72s | — | 0.12s | 0.14s | — |
| First Event → First Tool Call | 5.84s | 6.49s | — | 5.11s | 4.79s | — |
| MCP Initialize → First Semantic Call | — | — | — | — | — | — |
| MCP Server Start → Initialize | — | — | — | — | — | — |
| Top-Level User Turns | 1 | 1 | 0% | 1 | 1 | 0% |
| Internal Tool Cycles | 2 | 3 | +50.0% | 4 | 5 | +25.0% |
| Initial Load / Discovery Turns | 2 | 2 | 0% | 3 | 3 | 0% |
| MCP Discovery / Schema Turns | 0 | 0 | — | 0 | 0 | — |
| Tool Invocations (per run) | 2 | 3 | +50.0% | 4 | 5 | +25.0% |
| Output Tokens | 326 | 317 | -2.8% | 488 | 425 | -12.9% |
| Reasoning / Thinking Tokens | 103 | 49 | -52.4% | 89 | 81 | -9.0% |
| Total Input Tokens | 78003 | 105621 | +35.4% | 95953 | 113552 | +18.3% |
| Cached Input Tokens | 68608 | 85760 | +25.0% | 84736 | 84736 | 0% |
| Uncached Input Tokens | 9395 | 19861 | +111.4% | 11217 | 28816 | +156.9% |
| Cost | 0.0460 | 0.0757 | +64.5% | 0.0564 | 0.0995 | +76.4% |
| Cached vs Uncached Token Ratio | 7.30:1 | 4.32:1 | -40.9% | 7.55:1 | 2.94:1 | -61.1% |
| Oracle L1: Mutation Policy | PASS | PASS | — | PASS | PASS | — |
| Oracle L2: AST Invariants | PASS | PASS | — | PASS | PASS | — |
| Oracle L3: Clean Build | PASS | PASS | — | PASS | PASS | — |
| Oracle L4: Verification Test | PASS | PASS | — | PASS | PASS | — |
| Semantic Tool Invocation Verified (per run) | N/A | PASS | — | N/A | PASS | — |
Baseline vs Semedit Agentic in Small Context
- Vanilla (Baseline vs Semedit Agentic in Small Context) Edit: File api/server.go modified (9 lines -> 13 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Small Context) Edit: File api/server.go modified (9 lines -> 13 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ✓ ✓ | ✓ ✓ |
| 3 | — | semedit/semantic_insert_function (✓ ✓) |
Baseline vs Semedit Agentic in Large Context
- Vanilla (Baseline vs Semedit Agentic in Large Context) Edit: File api/server.go modified (9 lines -> 13 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Large Context) Edit: File api/server.go modified (9 lines -> 13 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ✓ ✓ | ✓ ✓ |
| 3 | ✓ ✓ | ✓ ✓ |
| 4 | ✓ ✓ | semedit/semantic_insert_function (✓ ✓) |
| 5 | — | ✓ ✓ |
default prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-04-insert-public |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | default |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/bench/task_04_insert_public.txtar |
Vanilla LLM Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
Add public constructor func InitServer() *Server placed before private helpers. Do not use semantic editing MCP tools; use standard file editing. When done, output DONE.
Semedit MCP Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
Add public constructor func InitServer() *Server placed before private helpers. Use semedit semantic tools for supported source code modifications. Shell commands for builds and tests are allowed. When done, output DONE.
Initial Workspace State (Before Edit)
package api
type Server struct {
Port int
}
func (s *Server) Start() {}
func (s *Server) internalRun() {}| Metric | Vanilla (Small) | MCP (Small) | Δ (Small) | Vanilla (Large) | MCP (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | 19.30s | 24.50s | +26.9% | 11.87s | 31.03s | +161.5% |
| Process Start → First Event | 0.72s | 0.15s | — | 0.12s | 0.13s | — |
| First Event → First Tool Call | 5.71s | 5.01s | — | 4.10s | 4.97s | — |
| MCP Initialize → First Semantic Call | — | — | — | — | — | — |
| MCP Server Start → Initialize | — | — | — | — | — | — |
| Top-Level User Turns | 1 | 1 | 0% | 1 | 1 | 0% |
| Internal Tool Cycles | 2 | 3 | +50.0% | 2 | 6 | +200.0% |
| Initial Load / Discovery Turns | 2 | 1 | -50.0% | 2 | 2 | 0% |
| MCP Discovery / Schema Turns | 0 | 0 | — | 0 | 0 | — |
| Tool Invocations (per run) | 2 | 3 | +50.0% | 2 | 6 | +200.0% |
| Output Tokens | 305 | 378 | +23.9% | 297 | 572 | +92.6% |
| Reasoning / Thinking Tokens | 71 | 112 | +57.7% | 40 | 160 | +300.0% |
| Total Input Tokens | 76071 | 115035 | +51.2% | 76157 | 130790 | +71.7% |
| Cached Input Tokens | 66560 | 98048 | +47.3% | 66560 | 114176 | +71.5% |
| Uncached Input Tokens | 9511 | 16987 | +78.6% | 9597 | 16614 | +73.1% |
| Cost | 0.0451 | 0.0731 | +62.0% | 0.0448 | 0.0792 | +76.7% |
| Cached vs Uncached Token Ratio | 7.00:1 | 5.77:1 | -17.5% | 6.94:1 | 6.87:1 | -0.9% |
| Oracle L1: Mutation Policy | PASS | PASS | — | PASS | PASS | — |
| Oracle L2: AST Invariants | PASS | PASS | — | PASS | PASS | — |
| Oracle L3: Clean Build | PASS | PASS | — | PASS | PASS | — |
| Oracle L4: Verification Test | PASS | PASS | — | PASS | PASS | — |
| Semantic Tool Invocation Verified (per run) | N/A | PASS | — | N/A | PASS | — |
Baseline vs Semedit Agentic in Small Context
- Vanilla (Baseline vs Semedit Agentic in Small Context) Edit: File api/server.go modified (9 lines -> 13 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Small Context) Edit: File api/server.go modified (9 lines -> 13 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ✓ ✓ | semedit/semantic_outline (✓ ✓) |
| 3 | — | semedit/semantic_insert_function (✓ ✓) |
Baseline vs Semedit Agentic in Large Context
- Vanilla (Baseline vs Semedit Agentic in Large Context) Edit: File api/server.go modified (9 lines -> 13 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Large Context) Edit: File api/server.go modified (9 lines -> 13 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ✓ ✓ | ✓ ✓ |
| 3 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 4 | — | semedit/semantic_insert_function (✓ ✓) |
| 5 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 6 | — | ✓ ✓ |
default prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-04-insert-public |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | default |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/bench/task_04_insert_public.txtar |
Vanilla LLM Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
Add public constructor func InitServer() *Server placed before private helpers. Do not use semantic editing MCP tools; use standard file editing. When done, output DONE.
Semedit MCP Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
Add public constructor func InitServer() *Server placed before private helpers. Use semedit semantic tools for supported source code modifications. Shell commands for builds and tests are allowed. When done, output DONE.
Initial Workspace State (Before Edit)
package api
type Server struct {
Port int
}
func (s *Server) Start() {}
func (s *Server) internalRun() {}| Metric | Vanilla (Small) | MCP (Small) | Δ (Small) | Vanilla (Large) | MCP (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | 46.18s | 24.02s | -48.0% | 17.73s | 25.25s | +42.4% |
| Process Start → First Event | 0.13s | 0.13s | — | 0.13s | 0.13s | — |
| First Event → First Tool Call | 9.08s | 4.89s | — | 4.70s | 4.88s | — |
| MCP Initialize → First Semantic Call | — | — | — | — | — | — |
| MCP Server Start → Initialize | — | — | — | — | — | — |
| Top-Level User Turns | 1 | 1 | 0% | 1 | 1 | 0% |
| Internal Tool Cycles | 4 | 5 | +25.0% | 4 | 4 | 0% |
| Initial Load / Discovery Turns | 3 | 1 | -66.7% | 2 | 1 | -50.0% |
| MCP Discovery / Schema Turns | 0 | 0 | — | 0 | 0 | — |
| Tool Invocations (per run) | 4 | 5 | +25.0% | 4 | 4 | 0% |
| Output Tokens | 331 | 447 | +35.0% | 501 | 334 | -33.3% |
| Reasoning / Thinking Tokens | 65 | 138 | +112.3% | 58 | 18 | -69.0% |
| Total Input Tokens | 94785 | 98819 | +4.3% | 75976 | 132829 | +74.8% |
| Cached Input Tokens | 84736 | 80896 | -4.5% | 66560 | 117248 | +76.2% |
| Uncached Input Tokens | 10049 | 17923 | +78.4% | 9416 | 15581 | +65.5% |
| Cost | 0.0513 | 0.0723 | +41.1% | 0.0472 | 0.0727 | +54.1% |
| Cached vs Uncached Token Ratio | 8.43:1 | 4.51:1 | -46.5% | 7.07:1 | 7.53:1 | +6.5% |
| Oracle L1: Mutation Policy | PASS | PASS | — | PASS | PASS | — |
| Oracle L2: AST Invariants | PASS | PASS | — | PASS | PASS | — |
| Oracle L3: Clean Build | PASS | PASS | — | PASS | PASS | — |
| Oracle L4: Verification Test | PASS | PASS | — | PASS | PASS | — |
| Semantic Tool Invocation Verified (per run) | N/A | PASS | — | N/A | PASS | — |
Baseline vs Semedit Agentic in Small Context
- Vanilla (Baseline vs Semedit Agentic in Small Context) Edit: File api/server.go modified (9 lines -> 13 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Small Context) Edit: File api/server.go modified (9 lines -> 13 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ✓ ✓ | semedit/semantic_outline (✓ ✓) |
| 3 | ✓ ✓ | ✓ ✓ |
| 4 | ✓ ✓ | semedit/semantic_lookup (✓ ✓) |
| 5 | — | semedit/semantic_insert_function (✓ ✓) |
Baseline vs Semedit Agentic in Large Context
- Vanilla (Baseline vs Semedit Agentic in Large Context) Edit: File api/server.go modified (9 lines -> 13 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Large Context) Edit: File api/server.go modified (9 lines -> 13 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ✓ ✓ | semedit/semantic_outline (✓ ✓) |
| 3 | ✓ ✓ | semedit/semantic_inspect_symbol (✓ ✓) |
| 4 | ✓ ✓ | semedit/semantic_insert_function (✓ ✓) |
default prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-04-insert-public |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | default |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/bench/task_04_insert_public.txtar |
Vanilla LLM Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
Add public constructor func InitServer() *Server placed before private helpers. Do not use semantic editing MCP tools; use standard file editing. When done, output DONE.
Semedit MCP Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
Add public constructor func InitServer() *Server placed before private helpers. Use semedit semantic tools for supported source code modifications. Shell commands for builds and tests are allowed. When done, output DONE.
Initial Workspace State (Before Edit)
package api
type Server struct {
Port int
}
func (s *Server) Start() {}
func (s *Server) internalRun() {}| Metric | Vanilla (Small) | MCP (Small) | Δ (Small) | Vanilla (Large) | MCP (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | 13.90s | 34.00s | +144.6% | 15.32s | 28.90s | +88.7% |
| Process Start → First Event | 0.12s | 0.13s | — | 0.12s | 0.12s | — |
| First Event → First Tool Call | 4.80s | 5.19s | — | 3.98s | 4.09s | — |
| MCP Initialize → First Semantic Call | — | — | — | — | — | — |
| MCP Server Start → Initialize | — | — | — | — | — | — |
| Top-Level User Turns | 1 | 1 | 0% | 1 | 1 | 0% |
| Internal Tool Cycles | 2 | 6 | +200.0% | 3 | 7 | +133.3% |
| Initial Load / Discovery Turns | 2 | 2 | 0% | 3 | 3 | 0% |
| MCP Discovery / Schema Turns | 0 | 0 | — | 0 | 0 | — |
| Tool Invocations (per run) | 2 | 6 | +200.0% | 3 | 7 | +133.3% |
| Output Tokens | 282 | 799 | +183.3% | 456 | 685 | +50.2% |
| Reasoning / Thinking Tokens | 57 | 176 | +208.8% | 160 | 204 | +27.5% |
| Total Input Tokens | 76804 | 150256 | +95.6% | 95640 | 154363 | +61.4% |
| Cached Input Tokens | 67584 | 135424 | +100.4% | 85760 | 137472 | +60.3% |
| Uncached Input Tokens | 9220 | 14832 | +60.9% | 9880 | 16891 | +71.0% |
| Cost | 0.0442 | 0.0831 | +88.1% | 0.0538 | 0.0877 | +62.9% |
| Cached vs Uncached Token Ratio | 7.33:1 | 9.13:1 | +24.6% | 8.68:1 | 8.14:1 | -6.2% |
| Oracle L1: Mutation Policy | PASS | PASS | — | PASS | PASS | — |
| Oracle L2: AST Invariants | PASS | PASS | — | PASS | PASS | — |
| Oracle L3: Clean Build | PASS | PASS | — | PASS | PASS | — |
| Oracle L4: Verification Test | PASS | PASS | — | PASS | PASS | — |
| Semantic Tool Invocation Verified (per run) | N/A | PASS | — | N/A | PASS | — |
Baseline vs Semedit Agentic in Small Context
- Vanilla (Baseline vs Semedit Agentic in Small Context) Edit: File api/server.go modified (9 lines -> 13 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Small Context) Edit: File api/server.go modified (9 lines -> 13 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ✓ ✓ | ✓ ✓ |
| 3 | — | semedit/semantic_insert_function (✓ ✗): calculate function offset: section placement violation: strategy "private_start", target symbol "InitServer", in /Users/alessandro/sources/semantic-editor/.scratch/benchmarks/run_codex_semedit_small-default_task-04-insert-public_1790467754122690000/api/server.go |
| 4 | — | ✓ ✓ |
| 5 | — | ✓ ✓ |
| 6 | — | semedit/semantic_insert_function (✓ ✓) |
Baseline vs Semedit Agentic in Large Context
- Vanilla (Baseline vs Semedit Agentic in Large Context) Edit: File api/server.go modified (9 lines -> 13 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Large Context) Edit: File api/server.go modified (9 lines -> 13 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ✓ ✓ | ? ✗ |
| 3 | ✓ ✓ | ✓ ✓ |
| 4 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 5 | — | semedit/semantic_insert_function (✓ ✓) |
| 6 | — | ✓ ✓ |
| 7 | — | ✓ ✓ |
default prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-04-insert-public |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | default |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/bench/task_04_insert_public.txtar |
Vanilla LLM Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
Add public constructor func InitServer() *Server placed before private helpers. Do not use semantic editing MCP tools; use standard file editing. When done, output DONE.
Semedit MCP Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
Add public constructor func InitServer() *Server placed before private helpers. Use semedit semantic tools for supported source code modifications. Shell commands for builds and tests are allowed. When done, output DONE.
Initial Workspace State (Before Edit)
package api
type Server struct {
Port int
}
func (s *Server) Start() {}
func (s *Server) internalRun() {}| Metric | Vanilla (Small) | MCP (Small) | Δ (Small) | Vanilla (Large) | MCP (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | 15.75s | 59.70s | +279.2% | 33.76s | 27.78s | -17.7% |
| Process Start → First Event | 0.12s | 0.12s | — | 0.12s | 0.15s | — |
| First Event → First Tool Call | 4.65s | 4.68s | — | 20.70s | 4.89s | — |
| MCP Initialize → First Semantic Call | — | — | — | — | — | — |
| MCP Server Start → Initialize | — | — | — | — | — | — |
| Top-Level User Turns | 1 | 1 | 0% | 1 | 1 | 0% |
| Internal Tool Cycles | 3 | 10 | +233.3% | 3 | 6 | +100.0% |
| Initial Load / Discovery Turns | 2 | 1 | -50.0% | 2 | 2 | 0% |
| MCP Discovery / Schema Turns | 0 | 0 | — | 0 | 0 | — |
| Tool Invocations (per run) | 3 | 10 | +233.3% | 3 | 6 | +100.0% |
| Output Tokens | 532 | 1371 | +157.7% | 521 | 601 | +15.4% |
| Reasoning / Thinking Tokens | 66 | 498 | +654.5% | 51 | 203 | +298.0% |
| Total Input Tokens | 75959 | 246470 | +224.5% | 78115 | 151707 | +94.2% |
| Cached Input Tokens | 66560 | 224768 | +237.7% | 68608 | 133632 | +94.8% |
| Uncached Input Tokens | 9399 | 21702 | +130.9% | 9507 | 18075 | +90.1% |
| Cost | 0.0476 | 0.1338 | +181.0% | 0.0481 | 0.0886 | +84.4% |
| Cached vs Uncached Token Ratio | 7.08:1 | 10.36:1 | +46.3% | 7.22:1 | 7.39:1 | +2.4% |
| Oracle L1: Mutation Policy | PASS | PASS | — | PASS | PASS | — |
| Oracle L2: AST Invariants | PASS | PASS | — | PASS | PASS | — |
| Oracle L3: Clean Build | PASS | PASS | — | PASS | PASS | — |
| Oracle L4: Verification Test | PASS | PASS | — | PASS | PASS | — |
| Semantic Tool Invocation Verified (per run) | N/A | PASS | — | N/A | PASS | — |
Baseline vs Semedit Agentic in Small Context
- Vanilla (Baseline vs Semedit Agentic in Small Context) Edit: File api/server.go modified (9 lines -> 13 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Small Context) Edit: File api/server.go modified (9 lines -> 13 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ✓ ✓ | semedit/semantic_outline (✓ ✓) |
| 3 | ✓ ✓ | semedit/semantic_inspect_symbol (✓ ✓) |
| 4 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 5 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 6 | — | ✓ ✓ |
| 7 | — | semedit/semantic_insert_function (✓ ✗): unknown param "language"; supported keys are [access, access_modifier, auto-organize-imports, auto_organize_imports, file, placement, source, target, target_symbol]: invalid operation parameters |
| 8 | — | ? ✗ |
| 9 | — | ✓ ✓ |
| 10 | — | semedit/semantic_insert_function (✓ ✓) |
Baseline vs Semedit Agentic in Large Context
- Vanilla (Baseline vs Semedit Agentic in Large Context) Edit: File api/server.go modified (9 lines -> 13 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Large Context) Edit: File api/server.go modified (9 lines -> 13 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ✓ ✓ | ? ✗ |
| 3 | ✓ ✓ | semedit/semantic_outline (✓ ✓) |
| 4 | — | semedit/semantic_insert_function (✓ ✓) |
| 5 | — | ✓ ✓ |
| 6 | — | ✓ ✓ |
default prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-04-insert-public |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | default |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/bench/task_04_insert_public.txtar |
Vanilla LLM Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
Add public constructor func InitServer() *Server placed before private helpers. Do not use semantic editing MCP tools; use standard file editing. When done, output DONE.
Semedit MCP Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
Add public constructor func InitServer() *Server placed before private helpers. Use semedit semantic tools for supported source code modifications. Shell commands for builds and tests are allowed. When done, output DONE.
Initial Workspace State (Before Edit)
package api
type Server struct {
Port int
}
func (s *Server) Start() {}
func (s *Server) internalRun() {}| Metric | Vanilla (Small) | MCP (Small) | Δ (Small) | Vanilla (Large) | MCP (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | 17.56s | 17.86s | +1.7% | 12.97s | 28.77s | +121.8% |
| Process Start → First Event | 0.12s | 0.12s | — | 0.12s | 0.12s | — |
| First Event → First Tool Call | 4.26s | 5.14s | — | 4.69s | 4.89s | — |
| MCP Initialize → First Semantic Call | — | — | — | — | — | — |
| MCP Server Start → Initialize | — | — | — | — | — | — |
| Top-Level User Turns | 1 | 1 | 0% | 1 | 1 | 0% |
| Internal Tool Cycles | 4 | 4 | 0% | 2 | 6 | +200.0% |
| Initial Load / Discovery Turns | 3 | 3 | 0% | 2 | 3 | +50.0% |
| MCP Discovery / Schema Turns | 0 | 0 | — | 0 | 0 | — |
| Tool Invocations (per run) | 4 | 4 | 0% | 2 | 6 | +200.0% |
| Output Tokens | 494 | 335 | -32.2% | 293 | 502 | +71.3% |
| Reasoning / Thinking Tokens | 0 | 28 | — | 73 | 97 | +32.9% |
| Total Input Tokens | 95084 | 92470 | -2.7% | 75749 | 152611 | +101.5% |
| Cached Input Tokens | 84736 | 77824 | -8.2% | 66560 | 132608 | +99.2% |
| Uncached Input Tokens | 10348 | 14646 | +41.5% | 9189 | 20003 | +117.7% |
| Cost | 0.0532 | 0.0606 | +13.9% | 0.0442 | 0.0906 | +105.1% |
| Cached vs Uncached Token Ratio | 8.19:1 | 5.31:1 | -35.1% | 7.24:1 | 6.63:1 | -8.5% |
| Oracle L1: Mutation Policy | PASS | PASS | — | PASS | PASS | — |
| Oracle L2: AST Invariants | PASS | PASS | — | PASS | PASS | — |
| Oracle L3: Clean Build | PASS | PASS | — | PASS | PASS | — |
| Oracle L4: Verification Test | PASS | PASS | — | PASS | PASS | — |
| Semantic Tool Invocation Verified (per run) | N/A | PASS | — | N/A | PASS | — |
Baseline vs Semedit Agentic in Small Context
- Vanilla (Baseline vs Semedit Agentic in Small Context) Edit: File api/server.go modified (9 lines -> 13 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Small Context) Edit: File api/server.go modified (9 lines -> 13 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ✓ ✓ | ✓ ✓ |
| 3 | ✓ ✓ | ✓ ✓ |
| 4 | ✓ ✓ | semedit/semantic_insert_function (✓ ✓) |
Baseline vs Semedit Agentic in Large Context
- Vanilla (Baseline vs Semedit Agentic in Large Context) Edit: File api/server.go modified (9 lines -> 13 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Large Context) Edit: File api/server.go modified (9 lines -> 13 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ✓ ✓ | ? ✗ |
| 3 | — | ✓ ✓ |
| 4 | — | semedit/semantic_outline (✓ ✓) |
| 5 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 6 | — | semedit/semantic_insert_function (✓ ✓) |
default prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-04-insert-public |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | default |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/bench/task_04_insert_public.txtar |
Vanilla LLM Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
Add public constructor func InitServer() *Server placed before private helpers. Do not use semantic editing MCP tools; use standard file editing. When done, output DONE.
Semedit MCP Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
Add public constructor func InitServer() *Server placed before private helpers. Use semedit semantic tools for supported source code modifications. Shell commands for builds and tests are allowed. When done, output DONE.
Initial Workspace State (Before Edit)
package api
type Server struct {
Port int
}
func (s *Server) Start() {}
func (s *Server) internalRun() {}| Metric | Vanilla (Small) | MCP (Small) | Δ (Small) | Vanilla (Large) | MCP (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | 39.24s | 21.19s | -46.0% | 25.63s | 18.54s | -27.7% |
| Process Start → First Event | 0.15s | 0.13s | — | 0.12s | 0.12s | — |
| First Event → First Tool Call | 4.73s | 5.49s | — | 8.14s | 5.58s | — |
| MCP Initialize → First Semantic Call | — | — | — | — | — | — |
| MCP Server Start → Initialize | — | — | — | — | — | — |
| Top-Level User Turns | 1 | 1 | 0% | 1 | 1 | 0% |
| Internal Tool Cycles | 2 | 3 | +50.0% | 2 | 4 | +100.0% |
| Initial Load / Discovery Turns | 2 | 2 | 0% | 2 | 3 | +50.0% |
| MCP Discovery / Schema Turns | 0 | 0 | — | 0 | 0 | — |
| Tool Invocations (per run) | 2 | 3 | +50.0% | 2 | 4 | +100.0% |
| Output Tokens | 260 | 392 | +50.8% | 281 | 331 | +17.8% |
| Reasoning / Thinking Tokens | 56 | 47 | -16.1% | 49 | 34 | -30.6% |
| Total Input Tokens | 75567 | 102642 | +35.8% | 75770 | 146650 | +93.5% |
| Cached Input Tokens | 66560 | 88832 | +33.5% | 66560 | 129536 | +94.6% |
| Uncached Input Tokens | 9007 | 13810 | +53.3% | 9210 | 17114 | +85.8% |
| Cost | 0.0431 | 0.0622 | +44.3% | 0.0438 | 0.0797 | +82.1% |
| Cached vs Uncached Token Ratio | 7.39:1 | 6.43:1 | -13.0% | 7.23:1 | 7.57:1 | +4.7% |
| Oracle L1: Mutation Policy | PASS | PASS | — | PASS | PASS | — |
| Oracle L2: AST Invariants | PASS | PASS | — | PASS | PASS | — |
| Oracle L3: Clean Build | PASS | PASS | — | PASS | PASS | — |
| Oracle L4: Verification Test | PASS | PASS | — | PASS | PASS | — |
| Semantic Tool Invocation Verified (per run) | N/A | PASS | — | N/A | PASS | — |
Baseline vs Semedit Agentic in Small Context
- Vanilla (Baseline vs Semedit Agentic in Small Context) Edit: File api/server.go modified (9 lines -> 13 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Small Context) Edit: File api/server.go modified (9 lines -> 13 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ✓ ✓ | ✓ ✓ |
| 3 | — | semedit/semantic_insert_function (✓ ✓) |
Baseline vs Semedit Agentic in Large Context
- Vanilla (Baseline vs Semedit Agentic in Large Context) Edit: File api/server.go modified (9 lines -> 13 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Large Context) Edit: File api/server.go modified (9 lines -> 13 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ? ✗ |
| 2 | ✓ ✓ | ✓ ✓ |
| 3 | — | ✓ ✓ |
| 4 | — | semedit/semantic_insert_function (✓ ✓) |
default prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-04-insert-public |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | default |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/bench/task_04_insert_public.txtar |
Vanilla LLM Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
Add public constructor func InitServer() *Server placed before private helpers. Do not use semantic editing MCP tools; use standard file editing. When done, output DONE.
Semedit MCP Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
Add public constructor func InitServer() *Server placed before private helpers. Use semedit semantic tools for supported source code modifications. Shell commands for builds and tests are allowed. When done, output DONE.
Initial Workspace State (Before Edit)
package api
type Server struct {
Port int
}
func (s *Server) Start() {}
func (s *Server) internalRun() {}| Metric | Vanilla (Small) | MCP (Small) | Δ (Small) | Vanilla (Large) | MCP (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | 15.67s | 24.09s | +53.7% | 14.11s | 31.82s | +125.6% |
| Process Start → First Event | 0.13s | 0.12s | — | 0.13s | 0.13s | — |
| First Event → First Tool Call | 4.36s | 6.45s | — | 4.48s | 5.46s | — |
| MCP Initialize → First Semantic Call | — | — | — | — | — | — |
| MCP Server Start → Initialize | — | — | — | — | — | — |
| Top-Level User Turns | 1 | 1 | 0% | 1 | 1 | 0% |
| Internal Tool Cycles | 3 | 4 | +33.3% | 2 | 8 | +300.0% |
| Initial Load / Discovery Turns | 2 | 1 | -50.0% | 2 | 2 | 0% |
| MCP Discovery / Schema Turns | 0 | 0 | — | 0 | 0 | — |
| Tool Invocations (per run) | 3 | 4 | +33.3% | 2 | 8 | +300.0% |
| Output Tokens | 511 | 342 | -33.1% | 327 | 450 | +37.6% |
| Reasoning / Thinking Tokens | 53 | 57 | +7.5% | 70 | 0 | -100.0% |
| Total Input Tokens | 75923 | 109165 | +43.8% | 76015 | 184872 | +143.2% |
| Cached Input Tokens | 66560 | 94976 | +42.7% | 66560 | 164096 | +146.5% |
| Uncached Input Tokens | 9363 | 14189 | +51.5% | 9455 | 20776 | +119.7% |
| Cost | 0.0471 | 0.0642 | +36.3% | 0.0452 | 0.0986 | +117.9% |
| Cached vs Uncached Token Ratio | 7.11:1 | 6.69:1 | -5.8% | 7.04:1 | 7.90:1 | +12.2% |
| Oracle L1: Mutation Policy | PASS | PASS | — | PASS | PASS | — |
| Oracle L2: AST Invariants | PASS | PASS | — | PASS | PASS | — |
| Oracle L3: Clean Build | PASS | PASS | — | PASS | PASS | — |
| Oracle L4: Verification Test | PASS | PASS | — | PASS | PASS | — |
| Semantic Tool Invocation Verified (per run) | N/A | PASS | — | N/A | PASS | — |
Baseline vs Semedit Agentic in Small Context
- Vanilla (Baseline vs Semedit Agentic in Small Context) Edit: File api/server.go modified (9 lines -> 13 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Small Context) Edit: File api/server.go modified (9 lines -> 13 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ✓ ✓ | semedit/semantic_lookup (✓ ✓) |
| 3 | ✓ ✓ | semedit/semantic_inspect_symbol (✓ ✓) |
| 4 | — | semedit/semantic_insert_function (✓ ✓) |
Baseline vs Semedit Agentic in Large Context
- Vanilla (Baseline vs Semedit Agentic in Large Context) Edit: File api/server.go modified (9 lines -> 13 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Large Context) Edit: File api/server.go modified (9 lines -> 13 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ? ✗ | ✓ ✓ |
| 3 | — | semedit/semantic_lookup (✓ ✓) |
| 4 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 5 | — | semedit/semantic_outline (✓ ✓) |
| 6 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 7 | — | semedit/semantic_insert_function (✓ ✓) |
| 8 | — | ✓ ✓ |
Test case: task-07-generate-template-main
Target: codex/gpt-6-luna/medium
crypto_rand prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-07-generate-template-main |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | crypto_rand |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/scripts/generate_template_main.txtar |
Vanilla LLM Prompt:
You are forbidden to modify protected files: “go.mod”, “go.sum”.
In main.go, implement main() to execute a text/template that prints ‘Hello World’ along with a random integer from crypto/rand, ensuring all necessary standard library packages are imported cleanly. Do not use semantic editing MCP tools; use standard file editing. When done, output DONE.
Semedit MCP Prompt:
You are forbidden to modify protected files: “go.mod”, “go.sum”.
In main.go, implement main() to execute a text/template that prints ‘Hello World’ along with a random integer from crypto/rand, ensuring all necessary standard library packages are imported cleanly. Use semedit semantic tools for supported source code modifications. Shell commands for builds and tests are allowed. When done, output DONE.
Initial Workspace State (Before Edit)
package main
func main() {
}| Metric | Vanilla (Small) | MCP (Small) | Δ (Small) | Vanilla (Large) | MCP (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | 24.45s | 33.57s | +37.3% | — | — | — |
| Process Start → First Event | 0.15s | 0.12s | — | — | — | — |
| First Event → First Tool Call | 5.22s | 5.79s | — | — | — | — |
| MCP Initialize → First Semantic Call | — | — | — | — | — | — |
| MCP Server Start → Initialize | — | — | — | — | — | — |
| Top-Level User Turns | 1 | 1 | 0% | — | — | — |
| Internal Tool Cycles | 4 | 5 | +25.0% | — | — | — |
| Initial Load / Discovery Turns | 4 | 1 | -75.0% | — | — | — |
| MCP Discovery / Schema Turns | 0 | 0 | — | — | — | — |
| Tool Invocations (per run) | 4 | 5 | +25.0% | — | — | — |
| Output Tokens | 636 | 681 | +7.1% | — | — | — |
| Reasoning / Thinking Tokens | 207 | 233 | +12.6% | — | — | — |
| Total Input Tokens | 115798 | 161154 | +39.2% | — | — | — |
| Cached Input Tokens | 104960 | 139776 | +33.2% | — | — | — |
| Uncached Input Tokens | 10838 | 21378 | +97.3% | — | — | — |
| Cost | 0.0639 | 0.0998 | +56.3% | — | — | — |
| Cached vs Uncached Token Ratio | 9.68:1 | 6.54:1 | -32.5% | — | — | — |
| Oracle L1: Mutation Policy | PASS | PASS | — | — | — | — |
| Oracle L2: AST Invariants | PASS | FAIL | — | — | — | — |
| Oracle L3: Clean Build | PASS | FAIL | — | — | — | — |
| Oracle L4: Verification Test | PASS | FAIL | — | — | — | — |
| Semantic Tool Invocation Verified (per run) | N/A | PASS | — | — | — | — |
Baseline vs Semedit Agentic in Small Context
- Vanilla (Baseline vs Semedit Agentic in Small Context) Edit: File main.go modified (4 lines -> 20 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Small Context) Edit: File main.go modified (4 lines -> 22 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ✓ ✓ | semedit/semantic_inspect_symbol (✓ ✓) |
| 3 | ✓ ✓ | semedit/semantic_replace_body (✓ ✓) |
| 4 | ✓ ✓ | ✓ ✓ |
| 5 | — | ✓ ✓ |
Baseline vs Semedit Agentic in Large Context
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | not published | not published |
crypto_rand prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-07-generate-template-main |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | crypto_rand |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/scripts/generate_template_main.txtar |
Vanilla LLM Prompt:
You are forbidden to modify protected files: “go.mod”, “go.sum”.
In main.go, implement main() to execute a text/template that prints ‘Hello World’ along with a random integer from crypto/rand, ensuring all necessary standard library packages are imported cleanly. Do not use semantic editing MCP tools; use standard file editing. When done, output DONE.
Semedit MCP Prompt:
You are forbidden to modify protected files: “go.mod”, “go.sum”.
In main.go, implement main() to execute a text/template that prints ‘Hello World’ along with a random integer from crypto/rand, ensuring all necessary standard library packages are imported cleanly. Use semedit semantic tools for supported source code modifications. Shell commands for builds and tests are allowed. When done, output DONE.
Initial Workspace State (Before Edit)
package main
func main() {
}| Metric | Vanilla (Small) | MCP (Small) | Δ (Small) | Vanilla (Large) | MCP (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | 19.69s | 45.28s | +129.9% | — | — | — |
| Process Start → First Event | 0.12s | 0.13s | — | — | — | — |
| First Event → First Tool Call | 4.43s | 5.84s | — | — | — | — |
| MCP Initialize → First Semantic Call | — | — | — | — | — | — |
| MCP Server Start → Initialize | — | — | — | — | — | — |
| Top-Level User Turns | 1 | 1 | 0% | — | — | — |
| Internal Tool Cycles | 3 | 7 | +133.3% | — | — | — |
| Initial Load / Discovery Turns | 3 | 0 | -100.0% | — | — | — |
| MCP Discovery / Schema Turns | 0 | 0 | — | — | — | — |
| Tool Invocations (per run) | 3 | 7 | +133.3% | — | — | — |
| Output Tokens | 535 | 887 | +65.8% | — | — | — |
| Reasoning / Thinking Tokens | 122 | 309 | +153.3% | — | — | — |
| Total Input Tokens | 95771 | 186809 | +95.1% | — | — | — |
| Cached Input Tokens | 85760 | 168960 | +97.0% | — | — | — |
| Uncached Input Tokens | 10011 | 17849 | +78.3% | — | — | — |
| Cost | 0.0547 | 0.1018 | +86.2% | — | — | — |
| Cached vs Uncached Token Ratio | 8.57:1 | 9.47:1 | +10.5% | — | — | — |
| Oracle L1: Mutation Policy | PASS | PASS | — | — | — | — |
| Oracle L2: AST Invariants | PASS | PASS | — | — | — | — |
| Oracle L3: Clean Build | PASS | PASS | — | — | — | — |
| Oracle L4: Verification Test | PASS | PASS | — | — | — | — |
| Semantic Tool Invocation Verified (per run) | N/A | PASS | — | — | — | — |
Baseline vs Semedit Agentic in Small Context
- Vanilla (Baseline vs Semedit Agentic in Small Context) Edit: File main.go modified (4 lines -> 21 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Small Context) Edit: File main.go modified (4 lines -> 24 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | semedit/semantic_lookup (✓ ✓) |
| 2 | ✓ ✓ | semedit/semantic_inspect_symbol (✓ ✓) |
| 3 | ✓ ✓ | semedit/semantic_replace_body (✓ ✓) |
| 4 | — | semedit/semantic_organize_imports (✓ ✓) |
| 5 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 6 | — | semedit/semantic_organize_imports (✓ ✓) |
| 7 | — | ✓ ✓ |
Baseline vs Semedit Agentic in Large Context
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | not published | not published |
crypto_rand prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-07-generate-template-main |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | crypto_rand |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/scripts/generate_template_main.txtar |
Vanilla LLM Prompt:
You are forbidden to modify protected files: “go.mod”, “go.sum”.
In main.go, implement main() to execute a text/template that prints ‘Hello World’ along with a random integer from crypto/rand, ensuring all necessary standard library packages are imported cleanly. Do not use semantic editing MCP tools; use standard file editing. When done, output DONE.
Semedit MCP Prompt:
You are forbidden to modify protected files: “go.mod”, “go.sum”.
In main.go, implement main() to execute a text/template that prints ‘Hello World’ along with a random integer from crypto/rand, ensuring all necessary standard library packages are imported cleanly. Use semedit semantic tools for supported source code modifications. Shell commands for builds and tests are allowed. When done, output DONE.
Initial Workspace State (Before Edit)
package main
func main() {
}| Metric | Vanilla (Small) | MCP (Small) | Δ (Small) | Vanilla (Large) | MCP (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | 16.63s | 26.20s | +57.6% | — | — | — |
| Process Start → First Event | 0.12s | 0.13s | — | — | — | — |
| First Event → First Tool Call | 4.17s | 5.14s | — | — | — | — |
| MCP Initialize → First Semantic Call | — | — | — | — | — | — |
| MCP Server Start → Initialize | — | — | — | — | — | — |
| Top-Level User Turns | 1 | 1 | 0% | — | — | — |
| Internal Tool Cycles | 3 | 5 | +66.7% | — | — | — |
| Initial Load / Discovery Turns | 2 | 2 | 0% | — | — | — |
| MCP Discovery / Schema Turns | 0 | 0 | — | — | — | — |
| Tool Invocations (per run) | 3 | 5 | +66.7% | — | — | — |
| Output Tokens | 592 | 612 | +3.4% | — | — | — |
| Reasoning / Thinking Tokens | 137 | 237 | +73.0% | — | — | — |
| Total Input Tokens | 76306 | 133541 | +75.0% | — | — | — |
| Cached Input Tokens | 66560 | 114432 | +71.9% | — | — | — |
| Uncached Input Tokens | 9746 | 19109 | +96.1% | — | — | — |
| Cost | 0.0501 | 0.0870 | +73.6% | — | — | — |
| Cached vs Uncached Token Ratio | 6.83:1 | 5.99:1 | -12.3% | — | — | — |
| Oracle L1: Mutation Policy | PASS | PASS | — | — | — | — |
| Oracle L2: AST Invariants | PASS | PASS | — | — | — | — |
| Oracle L3: Clean Build | PASS | PASS | — | — | — | — |
| Oracle L4: Verification Test | PASS | PASS | — | — | — | — |
| Semantic Tool Invocation Verified (per run) | N/A | PASS | — | — | — | — |
Baseline vs Semedit Agentic in Small Context
- Vanilla (Baseline vs Semedit Agentic in Small Context) Edit: File main.go modified (4 lines -> 20 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Small Context) Edit: File main.go modified (4 lines -> 20 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ✓ ✓ | ✓ ✓ |
| 3 | ✓ ✓ | semedit/semantic_outline (✓ ✓) |
| 4 | — | semedit/semantic_replace_body (✓ ✓) |
| 5 | — | ✓ ✓ |
Baseline vs Semedit Agentic in Large Context
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | not published | not published |
crypto_rand prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-07-generate-template-main |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | crypto_rand |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/scripts/generate_template_main.txtar |
Vanilla LLM Prompt:
You are forbidden to modify protected files: “go.mod”, “go.sum”.
In main.go, implement main() to execute a text/template that prints ‘Hello World’ along with a random integer from crypto/rand, ensuring all necessary standard library packages are imported cleanly. Do not use semantic editing MCP tools; use standard file editing. When done, output DONE.
Semedit MCP Prompt:
You are forbidden to modify protected files: “go.mod”, “go.sum”.
In main.go, implement main() to execute a text/template that prints ‘Hello World’ along with a random integer from crypto/rand, ensuring all necessary standard library packages are imported cleanly. Use semedit semantic tools for supported source code modifications. Shell commands for builds and tests are allowed. When done, output DONE.
Initial Workspace State (Before Edit)
package main
func main() {
}| Metric | Vanilla (Small) | MCP (Small) | Δ (Small) | Vanilla (Large) | MCP (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | 18.11s | 24.62s | +35.9% | — | — | — |
| Process Start → First Event | 0.12s | 0.12s | — | — | — | — |
| First Event → First Tool Call | 4.17s | 5.03s | — | — | — | — |
| MCP Initialize → First Semantic Call | — | — | — | — | — | — |
| MCP Server Start → Initialize | — | — | — | — | — | — |
| Top-Level User Turns | 1 | 1 | 0% | — | — | — |
| Internal Tool Cycles | 3 | 5 | +66.7% | — | — | — |
| Initial Load / Discovery Turns | 2 | 1 | -50.0% | — | — | — |
| MCP Discovery / Schema Turns | 0 | 0 | — | — | — | — |
| Tool Invocations (per run) | 3 | 5 | +66.7% | — | — | — |
| Output Tokens | 661 | 450 | -31.9% | — | — | — |
| Reasoning / Thinking Tokens | 195 | 23 | -88.2% | — | — | — |
| Total Input Tokens | 77589 | 126871 | +63.5% | — | — | — |
| Cached Input Tokens | 67584 | 108288 | +60.2% | — | — | — |
| Uncached Input Tokens | 10005 | 18583 | +85.7% | — | — | — |
| Cost | 0.0526 | 0.0794 | +51.0% | — | — | — |
| Cached vs Uncached Token Ratio | 6.76:1 | 5.83:1 | -13.7% | — | — | — |
| Oracle L1: Mutation Policy | PASS | PASS | — | — | — | — |
| Oracle L2: AST Invariants | PASS | PASS | — | — | — | — |
| Oracle L3: Clean Build | PASS | PASS | — | — | — | — |
| Oracle L4: Verification Test | PASS | PASS | — | — | — | — |
| Semantic Tool Invocation Verified (per run) | N/A | PASS | — | — | — | — |
Baseline vs Semedit Agentic in Small Context
- Vanilla (Baseline vs Semedit Agentic in Small Context) Edit: File main.go modified (4 lines -> 21 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Small Context) Edit: File main.go modified (4 lines -> 26 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ✓ ✓ | semedit/semantic_inspect_symbol (✓ ✓) |
| 3 | ✓ ✓ | semedit/semantic_replace_body (✓ ✓) |
| 4 | — | ✓ ✓ |
| 5 | — | semedit/semantic_inspect_symbol (✓ ✓) |
Baseline vs Semedit Agentic in Large Context
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | not published | not published |
crypto_rand prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-07-generate-template-main |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | crypto_rand |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/scripts/generate_template_main.txtar |
Vanilla LLM Prompt:
You are forbidden to modify protected files: “go.mod”, “go.sum”.
In main.go, implement main() to execute a text/template that prints ‘Hello World’ along with a random integer from crypto/rand, ensuring all necessary standard library packages are imported cleanly. Do not use semantic editing MCP tools; use standard file editing. When done, output DONE.
Semedit MCP Prompt:
You are forbidden to modify protected files: “go.mod”, “go.sum”.
In main.go, implement main() to execute a text/template that prints ‘Hello World’ along with a random integer from crypto/rand, ensuring all necessary standard library packages are imported cleanly. Use semedit semantic tools for supported source code modifications. Shell commands for builds and tests are allowed. When done, output DONE.
Initial Workspace State (Before Edit)
package main
func main() {
}| Metric | Vanilla (Small) | MCP (Small) | Δ (Small) | Vanilla (Large) | MCP (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | 13.96s | 32.59s | +133.5% | — | — | — |
| Process Start → First Event | 0.12s | 0.19s | — | — | — | — |
| First Event → First Tool Call | 3.48s | 5.89s | — | — | — | — |
| MCP Initialize → First Semantic Call | — | — | — | — | — | — |
| MCP Server Start → Initialize | — | — | — | — | — | — |
| Top-Level User Turns | 1 | 1 | 0% | — | — | — |
| Internal Tool Cycles | 1 | 5 | +400.0% | — | — | — |
| Initial Load / Discovery Turns | 1 | 1 | 0% | — | — | — |
| MCP Discovery / Schema Turns | 0 | 0 | — | — | — | — |
| Tool Invocations (per run) | 1 | 5 | +400.0% | — | — | — |
| Output Tokens | 493 | 632 | +28.2% | — | — | — |
| Reasoning / Thinking Tokens | 219 | 217 | -0.9% | — | — | — |
| Total Input Tokens | 58057 | 132572 | +128.3% | — | — | — |
| Cached Input Tokens | 49408 | 118272 | +139.4% | — | — | — |
| Uncached Input Tokens | 8649 | 14300 | +65.3% | — | — | — |
| Cost | 0.0429 | 0.0759 | +77.1% | — | — | — |
| Cached vs Uncached Token Ratio | 5.71:1 | 8.27:1 | +44.8% | — | — | — |
| Oracle L1: Mutation Policy | PASS | PASS | — | — | — | — |
| Oracle L2: AST Invariants | PASS | PASS | — | — | — | — |
| Oracle L3: Clean Build | PASS | PASS | — | — | — | — |
| Oracle L4: Verification Test | PASS | PASS | — | — | — | — |
| Semantic Tool Invocation Verified (per run) | N/A | PASS | — | — | — | — |
Baseline vs Semedit Agentic in Small Context
- Vanilla (Baseline vs Semedit Agentic in Small Context) Edit: File main.go modified (4 lines -> 21 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Small Context) Edit: File main.go modified (4 lines -> 24 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 3 | — | semedit/semantic_replace_body (✓ ✓) |
| 4 | — | semedit/semantic_organize_imports (✓ ✓) |
| 5 | — | ✓ ✓ |
Baseline vs Semedit Agentic in Large Context
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | not published | not published |
crypto_rand prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-07-generate-template-main |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | crypto_rand |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/scripts/generate_template_main.txtar |
Vanilla LLM Prompt:
You are forbidden to modify protected files: “go.mod”, “go.sum”.
In main.go, implement main() to execute a text/template that prints ‘Hello World’ along with a random integer from crypto/rand, ensuring all necessary standard library packages are imported cleanly. Do not use semantic editing MCP tools; use standard file editing. When done, output DONE.
Semedit MCP Prompt:
You are forbidden to modify protected files: “go.mod”, “go.sum”.
In main.go, implement main() to execute a text/template that prints ‘Hello World’ along with a random integer from crypto/rand, ensuring all necessary standard library packages are imported cleanly. Use semedit semantic tools for supported source code modifications. Shell commands for builds and tests are allowed. When done, output DONE.
Initial Workspace State (Before Edit)
package main
func main() {
}| Metric | Vanilla (Small) | MCP (Small) | Δ (Small) | Vanilla (Large) | MCP (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | 19.18s | 37.55s | +95.7% | — | — | — |
| Process Start → First Event | 0.12s | 0.12s | — | — | — | — |
| First Event → First Tool Call | 4.26s | 5.17s | — | — | — | — |
| MCP Initialize → First Semantic Call | — | — | — | — | — | — |
| MCP Server Start → Initialize | — | — | — | — | — | — |
| Top-Level User Turns | 1 | 1 | 0% | — | — | — |
| Internal Tool Cycles | 6 | 8 | +33.3% | — | — | — |
| Initial Load / Discovery Turns | 6 | 1 | -83.3% | — | — | — |
| MCP Discovery / Schema Turns | 0 | 0 | — | — | — | — |
| Tool Invocations (per run) | 6 | 8 | +33.3% | — | — | — |
| Output Tokens | 590 | 873 | +48.0% | — | — | — |
| Reasoning / Thinking Tokens | 98 | 345 | +252.0% | — | — | — |
| Total Input Tokens | 134644 | 155763 | +15.7% | — | — | — |
| Cached Input Tokens | 123136 | 135680 | +10.2% | — | — | — |
| Uncached Input Tokens | 11508 | 20083 | +74.5% | — | — | — |
| Cost | 0.0682 | 0.0994 | +45.8% | — | — | — |
| Cached vs Uncached Token Ratio | 10.70:1 | 6.76:1 | -36.9% | — | — | — |
| Oracle L1: Mutation Policy | PASS | PASS | — | — | — | — |
| Oracle L2: AST Invariants | PASS | PASS | — | — | — | — |
| Oracle L3: Clean Build | PASS | PASS | — | — | — | — |
| Oracle L4: Verification Test | PASS | PASS | — | — | — | — |
| Semantic Tool Invocation Verified (per run) | N/A | PASS | — | — | — | — |
Baseline vs Semedit Agentic in Small Context
- Vanilla (Baseline vs Semedit Agentic in Small Context) Edit: File main.go modified (4 lines -> 26 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Small Context) Edit: File main.go modified (4 lines -> 24 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ✓ ✓ | semedit/semantic_lookup (✓ ✓) |
| 3 | ✓ ✓ | ✓ ✓ |
| 4 | ✓ ✓ | semedit/semantic_inspect_symbol (✓ ✓) |
| 5 | ✓ ✓ | ✓ ✓ |
| 6 | ✓ ✓ | semedit/semantic_replace_body (✓ ✓) |
| 7 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 8 | — | ✓ ✓ |
Baseline vs Semedit Agentic in Large Context
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | not published | not published |
crypto_rand prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-07-generate-template-main |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | crypto_rand |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/scripts/generate_template_main.txtar |
Vanilla LLM Prompt:
You are forbidden to modify protected files: “go.mod”, “go.sum”.
In main.go, implement main() to execute a text/template that prints ‘Hello World’ along with a random integer from crypto/rand, ensuring all necessary standard library packages are imported cleanly. Do not use semantic editing MCP tools; use standard file editing. When done, output DONE.
Semedit MCP Prompt:
You are forbidden to modify protected files: “go.mod”, “go.sum”.
In main.go, implement main() to execute a text/template that prints ‘Hello World’ along with a random integer from crypto/rand, ensuring all necessary standard library packages are imported cleanly. Use semedit semantic tools for supported source code modifications. Shell commands for builds and tests are allowed. When done, output DONE.
Initial Workspace State (Before Edit)
package main
func main() {
}| Metric | Vanilla (Small) | MCP (Small) | Δ (Small) | Vanilla (Large) | MCP (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | 20.60s | 55.04s | +167.2% | — | — | — |
| Process Start → First Event | 0.14s | 0.14s | — | — | — | — |
| First Event → First Tool Call | 4.28s | 5.19s | — | — | — | — |
| MCP Initialize → First Semantic Call | — | — | — | — | — | — |
| MCP Server Start → Initialize | — | — | — | — | — | — |
| Top-Level User Turns | 1 | 1 | 0% | — | — | — |
| Internal Tool Cycles | 3 | 8 | +166.7% | — | — | — |
| Initial Load / Discovery Turns | 2 | 2 | 0% | — | — | — |
| MCP Discovery / Schema Turns | 0 | 0 | — | — | — | — |
| Tool Invocations (per run) | 3 | 8 | +166.7% | — | — | — |
| Output Tokens | 763 | 1489 | +95.2% | — | — | — |
| Reasoning / Thinking Tokens | 222 | 436 | +96.4% | — | — | — |
| Total Input Tokens | 77752 | 244513 | +214.5% | — | — | — |
| Cached Input Tokens | 67584 | 222720 | +229.5% | — | — | — |
| Uncached Input Tokens | 10168 | 21793 | +114.3% | — | — | — |
| Cost | 0.0546 | 0.1342 | +145.7% | — | — | — |
| Cached vs Uncached Token Ratio | 6.65:1 | 10.22:1 | +53.8% | — | — | — |
| Oracle L1: Mutation Policy | PASS | PASS | — | — | — | — |
| Oracle L2: AST Invariants | PASS | PASS | — | — | — | — |
| Oracle L3: Clean Build | PASS | PASS | — | — | — | — |
| Oracle L4: Verification Test | PASS | PASS | — | — | — | — |
| Semantic Tool Invocation Verified (per run) | N/A | PASS | — | — | — | — |
Baseline vs Semedit Agentic in Small Context
- Vanilla (Baseline vs Semedit Agentic in Small Context) Edit: File main.go modified (4 lines -> 24 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Small Context) Edit: File main.go modified (4 lines -> 20 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ✓ ✓ | ✓ ✓ |
| 3 | ✓ ✓ | semedit/semantic_replace_body (✓ ✓) |
| 4 | — | ✓ ✓ |
| 5 | — | semedit/semantic_replace_body (✓ ✓) |
| 6 | — | semedit/semantic_organize_imports (✓ ✓) |
| 7 | — | ✓ ✓ |
| 8 | — | ✓ ✓ |
Baseline vs Semedit Agentic in Large Context
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | not published | not published |
crypto_rand prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-07-generate-template-main |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | crypto_rand |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/scripts/generate_template_main.txtar |
Vanilla LLM Prompt:
You are forbidden to modify protected files: “go.mod”, “go.sum”.
In main.go, implement main() to execute a text/template that prints ‘Hello World’ along with a random integer from crypto/rand, ensuring all necessary standard library packages are imported cleanly. Do not use semantic editing MCP tools; use standard file editing. When done, output DONE.
Semedit MCP Prompt:
You are forbidden to modify protected files: “go.mod”, “go.sum”.
In main.go, implement main() to execute a text/template that prints ‘Hello World’ along with a random integer from crypto/rand, ensuring all necessary standard library packages are imported cleanly. Use semedit semantic tools for supported source code modifications. Shell commands for builds and tests are allowed. When done, output DONE.
Initial Workspace State (Before Edit)
package main
func main() {
}| Metric | Vanilla (Small) | MCP (Small) | Δ (Small) | Vanilla (Large) | MCP (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | 13.11s | 24.26s | +85.1% | — | — | — |
| Process Start → First Event | 0.12s | 0.14s | — | — | — | — |
| First Event → First Tool Call | 3.77s | 4.53s | — | — | — | — |
| MCP Initialize → First Semantic Call | — | — | — | — | — | — |
| MCP Server Start → Initialize | — | — | — | — | — | — |
| Top-Level User Turns | 1 | 1 | 0% | — | — | — |
| Internal Tool Cycles | 1 | 4 | +300.0% | — | — | — |
| Initial Load / Discovery Turns | 1 | 2 | +100.0% | — | — | — |
| MCP Discovery / Schema Turns | 0 | 0 | — | — | — | — |
| Tool Invocations (per run) | 1 | 4 | +300.0% | — | — | — |
| Output Tokens | 350 | 541 | +54.6% | — | — | — |
| Reasoning / Thinking Tokens | 89 | 109 | +22.5% | — | — | — |
| Total Input Tokens | 56773 | 133744 | +135.6% | — | — | — |
| Cached Input Tokens | 48384 | 116224 | +140.2% | — | — | — |
| Uncached Input Tokens | 8389 | 17520 | +108.8% | — | — | — |
| Cost | 0.0386 | 0.0810 | +110.0% | — | — | — |
| Cached vs Uncached Token Ratio | 5.77:1 | 6.63:1 | +15.0% | — | — | — |
| Oracle L1: Mutation Policy | PASS | PASS | — | — | — | — |
| Oracle L2: AST Invariants | PASS | PASS | — | — | — | — |
| Oracle L3: Clean Build | PASS | PASS | — | — | — | — |
| Oracle L4: Verification Test | PASS | PASS | — | — | — | — |
| Semantic Tool Invocation Verified (per run) | N/A | PASS | — | — | — | — |
Baseline vs Semedit Agentic in Small Context
- Vanilla (Baseline vs Semedit Agentic in Small Context) Edit: File main.go modified (4 lines -> 20 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Small Context) Edit: File main.go modified (4 lines -> 24 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | — | ✓ ✓ |
| 3 | — | semedit/semantic_replace_body (✓ ✓) |
| 4 | — | ✓ ✓ |
Baseline vs Semedit Agentic in Large Context
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | not published | not published |
crypto_rand prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-07-generate-template-main |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | crypto_rand |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/scripts/generate_template_main.txtar |
Vanilla LLM Prompt:
You are forbidden to modify protected files: “go.mod”, “go.sum”.
In main.go, implement main() to execute a text/template that prints ‘Hello World’ along with a random integer from crypto/rand, ensuring all necessary standard library packages are imported cleanly. Do not use semantic editing MCP tools; use standard file editing. When done, output DONE.
Semedit MCP Prompt:
You are forbidden to modify protected files: “go.mod”, “go.sum”.
In main.go, implement main() to execute a text/template that prints ‘Hello World’ along with a random integer from crypto/rand, ensuring all necessary standard library packages are imported cleanly. Use semedit semantic tools for supported source code modifications. Shell commands for builds and tests are allowed. When done, output DONE.
Initial Workspace State (Before Edit)
package main
func main() {
}| Metric | Vanilla (Small) | MCP (Small) | Δ (Small) | Vanilla (Large) | MCP (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | 15.55s | 68.95s | +343.5% | — | — | — |
| Process Start → First Event | 0.12s | 0.12s | — | — | — | — |
| First Event → First Tool Call | 4.02s | 14.34s | — | — | — | — |
| MCP Initialize → First Semantic Call | — | — | — | — | — | — |
| MCP Server Start → Initialize | — | — | — | — | — | — |
| Top-Level User Turns | 1 | 1 | 0% | — | — | — |
| Internal Tool Cycles | 3 | 5 | +66.7% | — | — | — |
| Initial Load / Discovery Turns | 3 | 1 | -66.7% | — | — | — |
| MCP Discovery / Schema Turns | 0 | 0 | — | — | — | — |
| Tool Invocations (per run) | 3 | 5 | +66.7% | — | — | — |
| Output Tokens | 456 | 744 | +63.2% | — | — | — |
| Reasoning / Thinking Tokens | 118 | 297 | +151.7% | — | — | — |
| Total Input Tokens | 76031 | 132713 | +74.6% | — | — | — |
| Cached Input Tokens | 66560 | 118272 | +77.7% | — | — | — |
| Uncached Input Tokens | 9471 | 14441 | +52.5% | — | — | — |
| Cost | 0.0475 | 0.0787 | +65.7% | — | — | — |
| Cached vs Uncached Token Ratio | 7.03:1 | 8.19:1 | +16.5% | — | — | — |
| Oracle L1: Mutation Policy | PASS | PASS | — | — | — | — |
| Oracle L2: AST Invariants | PASS | PASS | — | — | — | — |
| Oracle L3: Clean Build | PASS | PASS | — | — | — | — |
| Oracle L4: Verification Test | PASS | PASS | — | — | — | — |
| Semantic Tool Invocation Verified (per run) | N/A | PASS | — | — | — | — |
Baseline vs Semedit Agentic in Small Context
- Vanilla (Baseline vs Semedit Agentic in Small Context) Edit: File main.go modified (4 lines -> 24 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Small Context) Edit: File main.go modified (4 lines -> 25 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ✓ ✓ | semedit/semantic_inspect_symbol (✓ ✓) |
| 3 | ✓ ✓ | semedit/semantic_replace_body (✓ ✓) |
| 4 | — | semedit/semantic_organize_imports (✓ ✓) |
| 5 | — | ✓ ✓ |
Baseline vs Semedit Agentic in Large Context
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | not published | not published |
crypto_rand prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-07-generate-template-main |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | crypto_rand |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/scripts/generate_template_main.txtar |
Vanilla LLM Prompt:
You are forbidden to modify protected files: “go.mod”, “go.sum”.
In main.go, implement main() to execute a text/template that prints ‘Hello World’ along with a random integer from crypto/rand, ensuring all necessary standard library packages are imported cleanly. Do not use semantic editing MCP tools; use standard file editing. When done, output DONE.
Semedit MCP Prompt:
You are forbidden to modify protected files: “go.mod”, “go.sum”.
In main.go, implement main() to execute a text/template that prints ‘Hello World’ along with a random integer from crypto/rand, ensuring all necessary standard library packages are imported cleanly. Use semedit semantic tools for supported source code modifications. Shell commands for builds and tests are allowed. When done, output DONE.
Initial Workspace State (Before Edit)
package main
func main() {
}| Metric | Vanilla (Small) | MCP (Small) | Δ (Small) | Vanilla (Large) | MCP (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | 24.09s | 31.33s | +30.1% | — | — | — |
| Process Start → First Event | 0.13s | 0.12s | — | — | — | — |
| First Event → First Tool Call | 4.72s | 5.37s | — | — | — | — |
| MCP Initialize → First Semantic Call | — | — | — | — | — | — |
| MCP Server Start → Initialize | — | — | — | — | — | — |
| Top-Level User Turns | 1 | 1 | 0% | — | — | — |
| Internal Tool Cycles | 3 | 4 | +33.3% | — | — | — |
| Initial Load / Discovery Turns | 3 | 1 | -66.7% | — | — | — |
| MCP Discovery / Schema Turns | 0 | 0 | — | — | — | — |
| Tool Invocations (per run) | 3 | 4 | +33.3% | — | — | — |
| Output Tokens | 740 | 698 | -5.7% | — | — | — |
| Reasoning / Thinking Tokens | 343 | 311 | -9.3% | — | — | — |
| Total Input Tokens | 96118 | 110432 | +14.9% | — | — | — |
| Cached Input Tokens | 85760 | 96000 | +11.9% | — | — | — |
| Uncached Input Tokens | 10358 | 14432 | +39.3% | — | — | — |
| Cost | 0.0609 | 0.0727 | +19.4% | — | — | — |
| Cached vs Uncached Token Ratio | 8.28:1 | 6.65:1 | -19.7% | — | — | — |
| Oracle L1: Mutation Policy | PASS | PASS | — | — | — | — |
| Oracle L2: AST Invariants | PASS | PASS | — | — | — | — |
| Oracle L3: Clean Build | PASS | PASS | — | — | — | — |
| Oracle L4: Verification Test | PASS | PASS | — | — | — | — |
| Semantic Tool Invocation Verified (per run) | N/A | PASS | — | — | — | — |
Baseline vs Semedit Agentic in Small Context
- Vanilla (Baseline vs Semedit Agentic in Small Context) Edit: File main.go modified (4 lines -> 20 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Small Context) Edit: File main.go modified (4 lines -> 25 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ✓ ✓ | semedit/semantic_inspect_symbol (✓ ✓) |
| 3 | ✓ ✓ | semedit/semantic_replace_body (✓ ✓) |
| 4 | — | semedit/semantic_inspect_symbol (✓ ✓) |
Baseline vs Semedit Agentic in Large Context
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | not published | not published |
default prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-07-generate-template-main |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | default |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/scripts/generate_template_main.txtar |
Vanilla LLM Prompt:
You are forbidden to modify protected files: “go.mod”, “go.sum”.
In main.go, implement main() to execute a text/template that prints ‘Hello World’ along with a random integer, ensuring all necessary standard library packages are imported cleanly. Do not use semantic editing MCP tools; use standard file editing. When done, output DONE.
Semedit MCP Prompt:
You are forbidden to modify protected files: “go.mod”, “go.sum”.
In main.go, implement main() to execute a text/template that prints ‘Hello World’ along with a random integer, ensuring all necessary standard library packages are imported cleanly. Use semedit semantic tools for supported source code modifications. Shell commands for builds and tests are allowed. When done, output DONE.
Initial Workspace State (Before Edit)
package main
func main() {
}| Metric | Vanilla (Small) | MCP (Small) | Δ (Small) | Vanilla (Large) | MCP (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | 16.21s | 39.03s | +140.8% | — | — | — |
| Process Start → First Event | 0.12s | 0.12s | — | — | — | — |
| First Event → First Tool Call | 5.07s | 5.16s | — | — | — | — |
| MCP Initialize → First Semantic Call | — | — | — | — | — | — |
| MCP Server Start → Initialize | — | — | — | — | — | — |
| Top-Level User Turns | 1 | 1 | 0% | — | — | — |
| Internal Tool Cycles | 3 | 9 | +200.0% | — | — | — |
| Initial Load / Discovery Turns | 2 | 3 | +50.0% | — | — | — |
| MCP Discovery / Schema Turns | 0 | 0 | — | — | — | — |
| Tool Invocations (per run) | 3 | 9 | +200.0% | — | — | — |
| Output Tokens | 513 | 1003 | +95.5% | — | — | — |
| Reasoning / Thinking Tokens | 146 | 349 | +139.0% | — | — | — |
| Total Input Tokens | 76717 | 160752 | +109.5% | — | — | — |
| Cached Input Tokens | 60416 | 142592 | +136.0% | — | — | — |
| Uncached Input Tokens | 16301 | 18160 | +11.4% | — | — | — |
| Cost | 0.0641 | 0.0979 | +52.8% | — | — | — |
| Cached vs Uncached Token Ratio | 3.71:1 | 7.85:1 | +111.9% | — | — | — |
| Oracle L1: Mutation Policy | PASS | PASS | — | — | — | — |
| Oracle L2: AST Invariants | PASS | PASS | — | — | — | — |
| Oracle L3: Clean Build | PASS | PASS | — | — | — | — |
| Oracle L4: Verification Test | PASS | PASS | — | — | — | — |
| Semantic Tool Invocation Verified (per run) | N/A | PASS | — | — | — | — |
Baseline vs Semedit Agentic in Small Context
- Vanilla (Baseline vs Semedit Agentic in Small Context) Edit: File main.go modified (4 lines -> 14 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Small Context) Edit: File main.go modified (4 lines -> 18 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ✓ ✓ | ✓ ✓ |
| 3 | ✓ ✓ | ✓ ✓ |
| 4 | — | semedit/semantic_replace_body (✓ ✓) |
| 5 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 6 | — | semedit/semantic_organize_imports (✓ ✓) |
| 7 | — | ✓ ✓ |
| 8 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 9 | — | ✓ ✓ |
Baseline vs Semedit Agentic in Large Context
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | not published | not published |
default prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-07-generate-template-main |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | default |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/scripts/generate_template_main.txtar |
Vanilla LLM Prompt:
You are forbidden to modify protected files: “go.mod”, “go.sum”.
In main.go, implement main() to execute a text/template that prints ‘Hello World’ along with a random integer, ensuring all necessary standard library packages are imported cleanly. Do not use semantic editing MCP tools; use standard file editing. When done, output DONE.
Semedit MCP Prompt:
You are forbidden to modify protected files: “go.mod”, “go.sum”.
In main.go, implement main() to execute a text/template that prints ‘Hello World’ along with a random integer, ensuring all necessary standard library packages are imported cleanly. Use semedit semantic tools for supported source code modifications. Shell commands for builds and tests are allowed. When done, output DONE.
Initial Workspace State (Before Edit)
package main
func main() {
}| Metric | Vanilla (Small) | MCP (Small) | Δ (Small) | Vanilla (Large) | MCP (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | 21.88s | 33.79s | +54.4% | — | — | — |
| Process Start → First Event | 0.12s | 0.12s | — | — | — | — |
| First Event → First Tool Call | 4.62s | 7.03s | — | — | — | — |
| MCP Initialize → First Semantic Call | — | — | — | — | — | — |
| MCP Server Start → Initialize | — | — | — | — | — | — |
| Top-Level User Turns | 1 | 1 | 0% | — | — | — |
| Internal Tool Cycles | 4 | 8 | +100.0% | — | — | — |
| Initial Load / Discovery Turns | 4 | 1 | -75.0% | — | — | — |
| MCP Discovery / Schema Turns | 0 | 0 | — | — | — | — |
| Tool Invocations (per run) | 4 | 8 | +100.0% | — | — | — |
| Output Tokens | 584 | 556 | -4.8% | — | — | — |
| Reasoning / Thinking Tokens | 238 | 0 | -100.0% | — | — | — |
| Total Input Tokens | 97259 | 183866 | +89.0% | — | — | — |
| Cached Input Tokens | 86784 | 164096 | +89.1% | — | — | — |
| Uncached Input Tokens | 10475 | 19770 | +88.7% | — | — | — |
| Cost | 0.0582 | 0.0974 | +67.5% | — | — | — |
| Cached vs Uncached Token Ratio | 8.28:1 | 8.30:1 | +0.2% | — | — | — |
| Oracle L1: Mutation Policy | PASS | PASS | — | — | — | — |
| Oracle L2: AST Invariants | PASS | PASS | — | — | — | — |
| Oracle L3: Clean Build | PASS | PASS | — | — | — | — |
| Oracle L4: Verification Test | PASS | PASS | — | — | — | — |
| Semantic Tool Invocation Verified (per run) | N/A | PASS | — | — | — | — |
Baseline vs Semedit Agentic in Small Context
- Vanilla (Baseline vs Semedit Agentic in Small Context) Edit: File main.go modified (4 lines -> 14 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Small Context) Edit: File main.go modified (4 lines -> 15 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ✓ ✓ | semedit/semantic_outline (✓ ✓) |
| 3 | ✓ ✓ | ✓ ✓ |
| 4 | ✓ ✓ | semedit/semantic_inspect_symbol (✓ ✓) |
| 5 | — | semedit/semantic_replace_body (✓ ✓) |
| 6 | — | semedit/semantic_organize_imports (✓ ✓) |
| 7 | — | ✓ ✓ |
| 8 | — | ✓ ✓ |
Baseline vs Semedit Agentic in Large Context
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | not published | not published |
default prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-07-generate-template-main |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | default |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/scripts/generate_template_main.txtar |
Vanilla LLM Prompt:
You are forbidden to modify protected files: “go.mod”, “go.sum”.
In main.go, implement main() to execute a text/template that prints ‘Hello World’ along with a random integer, ensuring all necessary standard library packages are imported cleanly. Do not use semantic editing MCP tools; use standard file editing. When done, output DONE.
Semedit MCP Prompt:
You are forbidden to modify protected files: “go.mod”, “go.sum”.
In main.go, implement main() to execute a text/template that prints ‘Hello World’ along with a random integer, ensuring all necessary standard library packages are imported cleanly. Use semedit semantic tools for supported source code modifications. Shell commands for builds and tests are allowed. When done, output DONE.
Initial Workspace State (Before Edit)
package main
func main() {
}| Metric | Vanilla (Small) | MCP (Small) | Δ (Small) | Vanilla (Large) | MCP (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | 14.23s | 35.57s | +150.0% | — | — | — |
| Process Start → First Event | 0.13s | 0.12s | — | — | — | — |
| First Event → First Tool Call | 4.09s | 5.48s | — | — | — | — |
| MCP Initialize → First Semantic Call | — | — | — | — | — | — |
| MCP Server Start → Initialize | — | — | — | — | — | — |
| Top-Level User Turns | 1 | 1 | 0% | — | — | — |
| Internal Tool Cycles | 3 | 7 | +133.3% | — | — | — |
| Initial Load / Discovery Turns | 3 | 1 | -66.7% | — | — | — |
| MCP Discovery / Schema Turns | 0 | 0 | — | — | — | — |
| Tool Invocations (per run) | 3 | 7 | +133.3% | — | — | — |
| Output Tokens | 381 | 796 | +108.9% | — | — | — |
| Reasoning / Thinking Tokens | 83 | 327 | +294.0% | — | — | — |
| Total Input Tokens | 77163 | 134580 | +74.4% | — | — | — |
| Cached Input Tokens | 67584 | 120320 | +78.0% | — | — | — |
| Uncached Input Tokens | 9579 | 14260 | +48.9% | — | — | — |
| Cost | 0.0466 | 0.0798 | +71.0% | — | — | — |
| Cached vs Uncached Token Ratio | 7.06:1 | 8.44:1 | +19.6% | — | — | — |
| Oracle L1: Mutation Policy | PASS | PASS | — | — | — | — |
| Oracle L2: AST Invariants | PASS | PASS | — | — | — | — |
| Oracle L3: Clean Build | PASS | PASS | — | — | — | — |
| Oracle L4: Verification Test | PASS | PASS | — | — | — | — |
| Semantic Tool Invocation Verified (per run) | N/A | PASS | — | — | — | — |
Baseline vs Semedit Agentic in Small Context
- Vanilla (Baseline vs Semedit Agentic in Small Context) Edit: File main.go modified (4 lines -> 14 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Small Context) Edit: File main.go modified (4 lines -> 17 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ✓ ✓ | semedit/semantic_inspect_symbol (✓ ✓) |
| 3 | ✓ ✓ | ? ✗ |
| 4 | — | semedit/semantic_replace_body (✓ ✓) |
| 5 | — | semedit/semantic_organize_imports (✓ ✓) |
| 6 | — | ✓ ✓ |
| 7 | — | ✓ ✓ |
Baseline vs Semedit Agentic in Large Context
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | not published | not published |
default prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-07-generate-template-main |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | default |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/scripts/generate_template_main.txtar |
Vanilla LLM Prompt:
You are forbidden to modify protected files: “go.mod”, “go.sum”.
In main.go, implement main() to execute a text/template that prints ‘Hello World’ along with a random integer, ensuring all necessary standard library packages are imported cleanly. Do not use semantic editing MCP tools; use standard file editing. When done, output DONE.
Semedit MCP Prompt:
You are forbidden to modify protected files: “go.mod”, “go.sum”.
In main.go, implement main() to execute a text/template that prints ‘Hello World’ along with a random integer, ensuring all necessary standard library packages are imported cleanly. Use semedit semantic tools for supported source code modifications. Shell commands for builds and tests are allowed. When done, output DONE.
Initial Workspace State (Before Edit)
package main
func main() {
}| Metric | Vanilla (Small) | MCP (Small) | Δ (Small) | Vanilla (Large) | MCP (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | 54.41s | 46.11s | -15.3% | — | — | — |
| Process Start → First Event | 0.16s | 0.12s | — | — | — | — |
| First Event → First Tool Call | 11.12s | 5.79s | — | — | — | — |
| MCP Initialize → First Semantic Call | — | — | — | — | — | — |
| MCP Server Start → Initialize | — | — | — | — | — | — |
| Top-Level User Turns | 1 | 1 | 0% | — | — | — |
| Internal Tool Cycles | 4 | 9 | +125.0% | — | — | — |
| Initial Load / Discovery Turns | 2 | 0 | -100.0% | — | — | — |
| MCP Discovery / Schema Turns | 0 | 0 | — | — | — | — |
| Tool Invocations (per run) | 4 | 9 | +125.0% | — | — | — |
| Output Tokens | 586 | 650 | +10.9% | — | — | — |
| Reasoning / Thinking Tokens | 125 | 31 | -75.2% | — | — | — |
| Total Input Tokens | 96179 | 212896 | +121.4% | — | — | — |
| Cached Input Tokens | 85760 | 192512 | +124.5% | — | — | — |
| Uncached Input Tokens | 10419 | 20384 | +95.6% | — | — | — |
| Cost | 0.0564 | 0.1076 | +90.9% | — | — | — |
| Cached vs Uncached Token Ratio | 8.23:1 | 9.44:1 | +14.7% | — | — | — |
| Oracle L1: Mutation Policy | PASS | PASS | — | — | — | — |
| Oracle L2: AST Invariants | PASS | PASS | — | — | — | — |
| Oracle L3: Clean Build | PASS | PASS | — | — | — | — |
| Oracle L4: Verification Test | PASS | PASS | — | — | — | — |
| Semantic Tool Invocation Verified (per run) | N/A | PASS | — | — | — | — |
Baseline vs Semedit Agentic in Small Context
- Vanilla (Baseline vs Semedit Agentic in Small Context) Edit: File main.go modified (4 lines -> 14 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Small Context) Edit: File main.go modified (4 lines -> 20 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | semedit/semantic_inspect_symbol (✓ ✓) |
| 2 | ✓ ✓ | semedit/semantic_lookup (✓ ✓) |
| 3 | ✓ ✓ | semedit/semantic_replace_body (✓ ✓) |
| 4 | ✓ ✓ | semedit/semantic_inspect_symbol (✓ ✓) |
| 5 | — | semedit/semantic_replace_body (✓ ✓) |
| 6 | — | ✓ ✓ |
| 7 | — | semedit/semantic_organize_imports (✓ ✓) |
| 8 | — | ✓ ✓ |
| 9 | — | semedit/semantic_verify (✓ ✓) |
Baseline vs Semedit Agentic in Large Context
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | not published | not published |
default prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-07-generate-template-main |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | default |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/scripts/generate_template_main.txtar |
Vanilla LLM Prompt:
You are forbidden to modify protected files: “go.mod”, “go.sum”.
In main.go, implement main() to execute a text/template that prints ‘Hello World’ along with a random integer, ensuring all necessary standard library packages are imported cleanly. Do not use semantic editing MCP tools; use standard file editing. When done, output DONE.
Semedit MCP Prompt:
You are forbidden to modify protected files: “go.mod”, “go.sum”.
In main.go, implement main() to execute a text/template that prints ‘Hello World’ along with a random integer, ensuring all necessary standard library packages are imported cleanly. Use semedit semantic tools for supported source code modifications. Shell commands for builds and tests are allowed. When done, output DONE.
Initial Workspace State (Before Edit)
package main
func main() {
}| Metric | Vanilla (Small) | MCP (Small) | Δ (Small) | Vanilla (Large) | MCP (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | 13.89s | 42.62s | +206.9% | — | — | — |
| Process Start → First Event | 0.12s | 0.12s | — | — | — | — |
| First Event → First Tool Call | 5.09s | 4.30s | — | — | — | — |
| MCP Initialize → First Semantic Call | — | — | — | — | — | — |
| MCP Server Start → Initialize | — | — | — | — | — | — |
| Top-Level User Turns | 1 | 1 | 0% | — | — | — |
| Internal Tool Cycles | 3 | 9 | +200.0% | — | — | — |
| Initial Load / Discovery Turns | 3 | 2 | -33.3% | — | — | — |
| MCP Discovery / Schema Turns | 0 | 0 | — | — | — | — |
| Tool Invocations (per run) | 3 | 9 | +200.0% | — | — | — |
| Output Tokens | 390 | 678 | +73.8% | — | — | — |
| Reasoning / Thinking Tokens | 71 | 0 | -100.0% | — | — | — |
| Total Input Tokens | 75907 | 280052 | +268.9% | — | — | — |
| Cached Input Tokens | 66560 | 258304 | +288.1% | — | — | — |
| Uncached Input Tokens | 9347 | 21748 | +132.7% | — | — | — |
| Cost | 0.0458 | 0.1274 | +178.4% | — | — | — |
| Cached vs Uncached Token Ratio | 7.12:1 | 11.88:1 | +66.8% | — | — | — |
| Oracle L1: Mutation Policy | PASS | PASS | — | — | — | — |
| Oracle L2: AST Invariants | PASS | PASS | — | — | — | — |
| Oracle L3: Clean Build | PASS | PASS | — | — | — | — |
| Oracle L4: Verification Test | PASS | PASS | — | — | — | — |
| Semantic Tool Invocation Verified (per run) | N/A | PASS | — | — | — | — |
Baseline vs Semedit Agentic in Small Context
- Vanilla (Baseline vs Semedit Agentic in Small Context) Edit: File main.go modified (4 lines -> 15 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Small Context) Edit: File main.go modified (4 lines -> 14 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ✓ ✓ | ✓ ✓ |
| 3 | ✓ ✓ | semedit/semantic_replace_body (✓ ✓) |
| 4 | — | semedit/semantic_organize_imports (✓ ✓) |
| 5 | — | ✓ ✓ |
| 6 | — | semedit/semantic_organize_imports (✓ ✓) |
| 7 | — | semedit/semantic_replace_body (✓ ✓) |
| 8 | — | semedit/semantic_organize_imports (✓ ✓) |
| 9 | — | ✓ ✓ |
Baseline vs Semedit Agentic in Large Context
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | not published | not published |
default prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-07-generate-template-main |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | default |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/scripts/generate_template_main.txtar |
Vanilla LLM Prompt:
You are forbidden to modify protected files: “go.mod”, “go.sum”.
In main.go, implement main() to execute a text/template that prints ‘Hello World’ along with a random integer, ensuring all necessary standard library packages are imported cleanly. Do not use semantic editing MCP tools; use standard file editing. When done, output DONE.
Semedit MCP Prompt:
You are forbidden to modify protected files: “go.mod”, “go.sum”.
In main.go, implement main() to execute a text/template that prints ‘Hello World’ along with a random integer, ensuring all necessary standard library packages are imported cleanly. Use semedit semantic tools for supported source code modifications. Shell commands for builds and tests are allowed. When done, output DONE.
Initial Workspace State (Before Edit)
package main
func main() {
}| Metric | Vanilla (Small) | MCP (Small) | Δ (Small) | Vanilla (Large) | MCP (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | 15.24s | 30.24s | +98.4% | — | — | — |
| Process Start → First Event | 0.13s | 0.12s | — | — | — | — |
| First Event → First Tool Call | 4.45s | 7.15s | — | — | — | — |
| MCP Initialize → First Semantic Call | — | — | — | — | — | — |
| MCP Server Start → Initialize | — | — | — | — | — | — |
| Top-Level User Turns | 1 | 1 | 0% | — | — | — |
| Internal Tool Cycles | 3 | 4 | +33.3% | — | — | — |
| Initial Load / Discovery Turns | 2 | 1 | -50.0% | — | — | — |
| MCP Discovery / Schema Turns | 0 | 0 | — | — | — | — |
| Tool Invocations (per run) | 3 | 4 | +33.3% | — | — | — |
| Output Tokens | 447 | 618 | +38.3% | — | — | — |
| Reasoning / Thinking Tokens | 121 | 272 | +124.8% | — | — | — |
| Total Input Tokens | 75981 | 108888 | +43.3% | — | — | — |
| Cached Input Tokens | 66560 | 96000 | +44.2% | — | — | — |
| Uncached Input Tokens | 9421 | 12888 | +36.8% | — | — | — |
| Cost | 0.0473 | 0.0673 | +42.4% | — | — | — |
| Cached vs Uncached Token Ratio | 7.07:1 | 7.45:1 | +5.4% | — | — | — |
| Oracle L1: Mutation Policy | PASS | PASS | — | — | — | — |
| Oracle L2: AST Invariants | PASS | PASS | — | — | — | — |
| Oracle L3: Clean Build | PASS | PASS | — | — | — | — |
| Oracle L4: Verification Test | PASS | PASS | — | — | — | — |
| Semantic Tool Invocation Verified (per run) | N/A | PASS | — | — | — | — |
Baseline vs Semedit Agentic in Small Context
- Vanilla (Baseline vs Semedit Agentic in Small Context) Edit: File main.go modified (4 lines -> 15 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Small Context) Edit: File main.go modified (4 lines -> 14 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ✓ ✓ | semedit/semantic_inspect_symbol (✓ ✓) |
| 3 | ✓ ✓ | semedit/semantic_replace_body (✓ ✓) |
| 4 | — | semedit/semantic_organize_imports (✓ ✓) |
Baseline vs Semedit Agentic in Large Context
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | not published | not published |
default prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-07-generate-template-main |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | default |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/scripts/generate_template_main.txtar |
Vanilla LLM Prompt:
You are forbidden to modify protected files: “go.mod”, “go.sum”.
In main.go, implement main() to execute a text/template that prints ‘Hello World’ along with a random integer, ensuring all necessary standard library packages are imported cleanly. Do not use semantic editing MCP tools; use standard file editing. When done, output DONE.
Semedit MCP Prompt:
You are forbidden to modify protected files: “go.mod”, “go.sum”.
In main.go, implement main() to execute a text/template that prints ‘Hello World’ along with a random integer, ensuring all necessary standard library packages are imported cleanly. Use semedit semantic tools for supported source code modifications. Shell commands for builds and tests are allowed. When done, output DONE.
Initial Workspace State (Before Edit)
package main
func main() {
}| Metric | Vanilla (Small) | MCP (Small) | Δ (Small) | Vanilla (Large) | MCP (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | 22.98s | 45.00s | +95.8% | — | — | — |
| Process Start → First Event | 0.13s | 0.16s | — | — | — | — |
| First Event → First Tool Call | 5.28s | 5.59s | — | — | — | — |
| MCP Initialize → First Semantic Call | — | — | — | — | — | — |
| MCP Server Start → Initialize | — | — | — | — | — | — |
| Top-Level User Turns | 1 | 1 | 0% | — | — | — |
| Internal Tool Cycles | 4 | 11 | +175.0% | — | — | — |
| Initial Load / Discovery Turns | 4 | 1 | -75.0% | — | — | — |
| MCP Discovery / Schema Turns | 0 | 0 | — | — | — | — |
| Tool Invocations (per run) | 4 | 11 | +175.0% | — | — | — |
| Output Tokens | 817 | 794 | -2.8% | — | — | — |
| Reasoning / Thinking Tokens | 203 | 0 | -100.0% | — | — | — |
| Total Input Tokens | 97969 | 236687 | +141.6% | — | — | — |
| Cached Input Tokens | 86784 | 217600 | +150.7% | — | — | — |
| Uncached Input Tokens | 11185 | 19087 | +70.6% | — | — | — |
| Cost | 0.0624 | 0.1120 | +79.5% | — | — | — |
| Cached vs Uncached Token Ratio | 7.76:1 | 11.40:1 | +46.9% | — | — | — |
| Oracle L1: Mutation Policy | PASS | PASS | — | — | — | — |
| Oracle L2: AST Invariants | PASS | PASS | — | — | — | — |
| Oracle L3: Clean Build | PASS | PASS | — | — | — | — |
| Oracle L4: Verification Test | PASS | PASS | — | — | — | — |
| Semantic Tool Invocation Verified (per run) | N/A | PASS | — | — | — | — |
Baseline vs Semedit Agentic in Small Context
- Vanilla (Baseline vs Semedit Agentic in Small Context) Edit: File main.go modified (4 lines -> 15 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Small Context) Edit: File main.go modified (4 lines -> 23 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ✓ ✓ | semedit/semantic_inspect_symbol (✓ ✓) |
| 3 | ✓ ✓ | semedit/semantic_lookup (✓ ✓) |
| 4 | ✓ ✓ | semedit/semantic_replace_body (✓ ✓) |
| 5 | — | ✓ ✓ |
| 6 | — | ✓ ✓ |
| 7 | — | semedit/semantic_replace_body (✓ ✓) |
| 8 | — | semedit/semantic_organize_imports (✓ ✓) |
| 9 | — | ✓ ✓ |
| 10 | — | semedit/semantic_organize_imports (✓ ✓) |
| 11 | — | ✓ ✓ |
Baseline vs Semedit Agentic in Large Context
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | not published | not published |
default prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-07-generate-template-main |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | default |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/scripts/generate_template_main.txtar |
Vanilla LLM Prompt:
You are forbidden to modify protected files: “go.mod”, “go.sum”.
In main.go, implement main() to execute a text/template that prints ‘Hello World’ along with a random integer, ensuring all necessary standard library packages are imported cleanly. Do not use semantic editing MCP tools; use standard file editing. When done, output DONE.
Semedit MCP Prompt:
You are forbidden to modify protected files: “go.mod”, “go.sum”.
In main.go, implement main() to execute a text/template that prints ‘Hello World’ along with a random integer, ensuring all necessary standard library packages are imported cleanly. Use semedit semantic tools for supported source code modifications. Shell commands for builds and tests are allowed. When done, output DONE.
Initial Workspace State (Before Edit)
package main
func main() {
}| Metric | Vanilla (Small) | MCP (Small) | Δ (Small) | Vanilla (Large) | MCP (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | 19.71s | 34.95s | +77.3% | — | — | — |
| Process Start → First Event | 0.13s | 0.12s | — | — | — | — |
| First Event → First Tool Call | 5.83s | 5.33s | — | — | — | — |
| MCP Initialize → First Semantic Call | — | — | — | — | — | — |
| MCP Server Start → Initialize | — | — | — | — | — | — |
| Top-Level User Turns | 1 | 1 | 0% | — | — | — |
| Internal Tool Cycles | 3 | 6 | +100.0% | — | — | — |
| Initial Load / Discovery Turns | 3 | 1 | -66.7% | — | — | — |
| MCP Discovery / Schema Turns | 0 | 0 | — | — | — | — |
| Tool Invocations (per run) | 3 | 6 | +100.0% | — | — | — |
| Output Tokens | 541 | 467 | -13.7% | — | — | — |
| Reasoning / Thinking Tokens | 239 | 41 | -82.8% | — | — | — |
| Total Input Tokens | 76129 | 173364 | +127.7% | — | — | — |
| Cached Input Tokens | 66560 | 157696 | +136.9% | — | — | — |
| Uncached Input Tokens | 9569 | 15668 | +63.7% | — | — | — |
| Cost | 0.0503 | 0.0849 | +68.8% | — | — | — |
| Cached vs Uncached Token Ratio | 6.96:1 | 10.06:1 | +44.7% | — | — | — |
| Oracle L1: Mutation Policy | PASS | PASS | — | — | — | — |
| Oracle L2: AST Invariants | PASS | PASS | — | — | — | — |
| Oracle L3: Clean Build | PASS | PASS | — | — | — | — |
| Oracle L4: Verification Test | PASS | PASS | — | — | — | — |
| Semantic Tool Invocation Verified (per run) | N/A | PASS | — | — | — | — |
Baseline vs Semedit Agentic in Small Context
- Vanilla (Baseline vs Semedit Agentic in Small Context) Edit: File main.go modified (4 lines -> 14 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Small Context) Edit: File main.go modified (4 lines -> 16 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ✓ ✓ | semedit/semantic_inspect_symbol (✓ ✓) |
| 3 | ✓ ✓ | semedit/semantic_replace_body (✓ ✓) |
| 4 | — | semedit/semantic_organize_imports (✓ ✓) |
| 5 | — | semedit/semantic_organize_imports (✓ ✓) |
| 6 | — | ✓ ✓ |
Baseline vs Semedit Agentic in Large Context
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | not published | not published |
default prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-07-generate-template-main |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | default |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/scripts/generate_template_main.txtar |
Vanilla LLM Prompt:
You are forbidden to modify protected files: “go.mod”, “go.sum”.
In main.go, implement main() to execute a text/template that prints ‘Hello World’ along with a random integer, ensuring all necessary standard library packages are imported cleanly. Do not use semantic editing MCP tools; use standard file editing. When done, output DONE.
Semedit MCP Prompt:
You are forbidden to modify protected files: “go.mod”, “go.sum”.
In main.go, implement main() to execute a text/template that prints ‘Hello World’ along with a random integer, ensuring all necessary standard library packages are imported cleanly. Use semedit semantic tools for supported source code modifications. Shell commands for builds and tests are allowed. When done, output DONE.
Initial Workspace State (Before Edit)
package main
func main() {
}| Metric | Vanilla (Small) | MCP (Small) | Δ (Small) | Vanilla (Large) | MCP (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | 24.48s | 31.99s | +30.7% | — | — | — |
| Process Start → First Event | 0.13s | 0.12s | — | — | — | — |
| First Event → First Tool Call | 4.09s | 5.54s | — | — | — | — |
| MCP Initialize → First Semantic Call | — | — | — | — | — | — |
| MCP Server Start → Initialize | — | — | — | — | — | — |
| Top-Level User Turns | 1 | 1 | 0% | — | — | — |
| Internal Tool Cycles | 4 | 5 | +25.0% | — | — | — |
| Initial Load / Discovery Turns | 3 | 0 | -100.0% | — | — | — |
| MCP Discovery / Schema Turns | 0 | 0 | — | — | — | — |
| Tool Invocations (per run) | 4 | 5 | +25.0% | — | — | — |
| Output Tokens | 588 | 500 | -15.0% | — | — | — |
| Reasoning / Thinking Tokens | 162 | 86 | -46.9% | — | — | — |
| Total Input Tokens | 97031 | 159986 | +64.9% | — | — | — |
| Cached Input Tokens | 86784 | 143616 | +65.5% | — | — | — |
| Uncached Input Tokens | 10247 | 16370 | +59.8% | — | — | — |
| Cost | 0.0567 | 0.0842 | +48.4% | — | — | — |
| Cached vs Uncached Token Ratio | 8.47:1 | 8.77:1 | +3.6% | — | — | — |
| Oracle L1: Mutation Policy | PASS | PASS | — | — | — | — |
| Oracle L2: AST Invariants | PASS | PASS | — | — | — | — |
| Oracle L3: Clean Build | PASS | PASS | — | — | — | — |
| Oracle L4: Verification Test | PASS | PASS | — | — | — | — |
| Semantic Tool Invocation Verified (per run) | N/A | PASS | — | — | — | — |
Baseline vs Semedit Agentic in Small Context
- Vanilla (Baseline vs Semedit Agentic in Small Context) Edit: File main.go modified (4 lines -> 14 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Small Context) Edit: File main.go modified (4 lines -> 24 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | semedit/semantic_inspect_symbol (✓ ✓) |
| 2 | ✓ ✓ | semedit/semantic_replace_body (✓ ✓) |
| 3 | ✓ ✓ | semedit/semantic_inspect_symbol (✓ ✓) |
| 4 | ✓ ✓ | semedit/semantic_organize_imports (✓ ✓) |
| 5 | — | ✓ ✓ |
Baseline vs Semedit Agentic in Large Context
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | not published | not published |
default prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-07-generate-template-main |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | default |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/scripts/generate_template_main.txtar |
Vanilla LLM Prompt:
You are forbidden to modify protected files: “go.mod”, “go.sum”.
In main.go, implement main() to execute a text/template that prints ‘Hello World’ along with a random integer, ensuring all necessary standard library packages are imported cleanly. Do not use semantic editing MCP tools; use standard file editing. When done, output DONE.
Semedit MCP Prompt:
You are forbidden to modify protected files: “go.mod”, “go.sum”.
In main.go, implement main() to execute a text/template that prints ‘Hello World’ along with a random integer, ensuring all necessary standard library packages are imported cleanly. Use semedit semantic tools for supported source code modifications. Shell commands for builds and tests are allowed. When done, output DONE.
Initial Workspace State (Before Edit)
package main
func main() {
}| Metric | Vanilla (Small) | MCP (Small) | Δ (Small) | Vanilla (Large) | MCP (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | 20.35s | 44.20s | +117.2% | — | — | — |
| Process Start → First Event | 0.12s | 0.12s | — | — | — | — |
| First Event → First Tool Call | 6.35s | 4.69s | — | — | — | — |
| MCP Initialize → First Semantic Call | — | — | — | — | — | — |
| MCP Server Start → Initialize | — | — | — | — | — | — |
| Top-Level User Turns | 1 | 1 | 0% | — | — | — |
| Internal Tool Cycles | 3 | 7 | +133.3% | — | — | — |
| Initial Load / Discovery Turns | 3 | 2 | -33.3% | — | — | — |
| MCP Discovery / Schema Turns | 0 | 0 | — | — | — | — |
| Tool Invocations (per run) | 3 | 7 | +133.3% | — | — | — |
| Output Tokens | 580 | 965 | +66.4% | — | — | — |
| Reasoning / Thinking Tokens | 267 | 285 | +6.7% | — | — | — |
| Total Input Tokens | 95482 | 234312 | +145.4% | — | — | — |
| Cached Input Tokens | 85760 | 209920 | +144.8% | — | — | — |
| Uncached Input Tokens | 9722 | 24392 | +150.9% | — | — | — |
| Cost | 0.0563 | 0.1291 | +129.1% | — | — | — |
| Cached vs Uncached Token Ratio | 8.82:1 | 8.61:1 | -2.4% | — | — | — |
| Oracle L1: Mutation Policy | PASS | PASS | — | — | — | — |
| Oracle L2: AST Invariants | PASS | PASS | — | — | — | — |
| Oracle L3: Clean Build | PASS | PASS | — | — | — | — |
| Oracle L4: Verification Test | PASS | PASS | — | — | — | — |
| Semantic Tool Invocation Verified (per run) | N/A | PASS | — | — | — | — |
Baseline vs Semedit Agentic in Small Context
- Vanilla (Baseline vs Semedit Agentic in Small Context) Edit: File main.go modified (4 lines -> 12 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Small Context) Edit: File main.go modified (4 lines -> 20 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ✓ ✓ | ✓ ✓ |
| 3 | ✓ ✓ | semedit/semantic_replace_body (✓ ✓) |
| 4 | — | semedit/semantic_replace_body (✓ ✓) |
| 5 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 6 | — | semedit/semantic_replace_body (✓ ✓) |
| 7 | — | semedit/semantic_inspect_symbol (✓ ✓) |
Baseline vs Semedit Agentic in Large Context
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | not published | not published |
prefer_discover_semedit prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-07-generate-template-main |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | prefer_discover_semedit |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/scripts/generate_template_main.txtar |
Vanilla LLM Prompt:
You are forbidden to modify protected files: “go.mod”, “go.sum”.
Before editing, inspect the complete available tool inventory, including deferred or lazy tools. If applicable semantic editing tools are callable, prefer them for source mutations. In main.go, implement main() to execute a text/template that prints ‘Hello World’ along with a random integer, ensuring all necessary standard library packages are imported cleanly. Do not use semantic editing MCP tools; use standard file editing. When done, output DONE.
Semedit MCP Prompt:
You are forbidden to modify protected files: “go.mod”, “go.sum”.
Before editing, inspect the complete available tool inventory, including deferred or lazy tools. If applicable semantic editing tools are callable, prefer them for source mutations. In main.go, implement main() to execute a text/template that prints ‘Hello World’ along with a random integer, ensuring all necessary standard library packages are imported cleanly. Use semedit semantic tools for supported source code modifications. Shell commands for builds and tests are allowed. When done, output DONE.
Initial Workspace State (Before Edit)
package main
func main() {
}| Metric | Vanilla (Small) | MCP (Small) | Δ (Small) | Vanilla (Large) | MCP (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | 19.05s | 63.15s | +231.6% | — | — | — |
| Process Start → First Event | 0.13s | 0.14s | — | — | — | — |
| First Event → First Tool Call | 8.33s | 7.78s | — | — | — | — |
| MCP Initialize → First Semantic Call | — | — | — | — | — | — |
| MCP Server Start → Initialize | — | — | — | — | — | — |
| Top-Level User Turns | 1 | 1 | 0% | — | — | — |
| Internal Tool Cycles | 1 | 11 | +1000.0% | — | — | — |
| Initial Load / Discovery Turns | 1 | 0 | -100.0% | — | — | — |
| MCP Discovery / Schema Turns | 0 | 0 | — | — | — | — |
| Tool Invocations (per run) | 1 | 11 | +1000.0% | — | — | — |
| Output Tokens | 622 | 1530 | +146.0% | — | — | — |
| Reasoning / Thinking Tokens | 273 | 475 | +74.0% | — | — | — |
| Total Input Tokens | 135029 | 357506 | +164.8% | — | — | — |
| Cached Input Tokens | 114432 | 330752 | +189.0% | — | — | — |
| Uncached Input Tokens | 20597 | 26754 | +29.9% | — | — | — |
| Cost | 0.0913 | 0.1746 | +91.3% | — | — | — |
| Cached vs Uncached Token Ratio | 5.56:1 | 12.36:1 | +122.5% | — | — | — |
| Oracle L1: Mutation Policy | PASS | PASS | — | — | — | — |
| Oracle L2: AST Invariants | PASS | PASS | — | — | — | — |
| Oracle L3: Clean Build | PASS | PASS | — | — | — | — |
| Oracle L4: Verification Test | PASS | PASS | — | — | — | — |
| Semantic Tool Invocation Verified (per run) | N/A | PASS | — | — | — | — |
Baseline vs Semedit Agentic in Small Context
- Vanilla (Baseline vs Semedit Agentic in Small Context) Edit: File main.go modified (4 lines -> 14 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Small Context) Edit: File main.go modified (4 lines -> 16 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | semedit/semantic_outline (✓ ✓) |
| 2 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 3 | — | semedit/semantic_replace_body (✓ ✓) |
| 4 | — | semedit/semantic_organize_imports (✓ ✓) |
| 5 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 6 | — | semedit/semantic_organize_imports (✓ ✓) |
| 7 | — | ✓ ✓ |
| 8 | — | ✓ ✓ |
| 9 | — | semedit/semantic_verify (✓ ✓) |
| 10 | — | ✓ ✓ |
| 11 | — | ? ✗ |
Baseline vs Semedit Agentic in Large Context
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | not published | not published |
prefer_discover_semedit prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-07-generate-template-main |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | prefer_discover_semedit |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/scripts/generate_template_main.txtar |
Vanilla LLM Prompt:
You are forbidden to modify protected files: “go.mod”, “go.sum”.
Before editing, inspect the complete available tool inventory, including deferred or lazy tools. If applicable semantic editing tools are callable, prefer them for source mutations. In main.go, implement main() to execute a text/template that prints ‘Hello World’ along with a random integer, ensuring all necessary standard library packages are imported cleanly. Do not use semantic editing MCP tools; use standard file editing. When done, output DONE.
Semedit MCP Prompt:
You are forbidden to modify protected files: “go.mod”, “go.sum”.
Before editing, inspect the complete available tool inventory, including deferred or lazy tools. If applicable semantic editing tools are callable, prefer them for source mutations. In main.go, implement main() to execute a text/template that prints ‘Hello World’ along with a random integer, ensuring all necessary standard library packages are imported cleanly. Use semedit semantic tools for supported source code modifications. Shell commands for builds and tests are allowed. When done, output DONE.
Initial Workspace State (Before Edit)
package main
func main() {
}| Metric | Vanilla (Small) | MCP (Small) | Δ (Small) | Vanilla (Large) | MCP (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | 30.67s | 60.92s | +98.6% | — | — | — |
| Process Start → First Event | 0.14s | 0.12s | — | — | — | — |
| First Event → First Tool Call | 10.25s | 12.14s | — | — | — | — |
| MCP Initialize → First Semantic Call | — | — | — | — | — | — |
| MCP Server Start → Initialize | — | — | — | — | — | — |
| Top-Level User Turns | 1 | 1 | 0% | — | — | — |
| Internal Tool Cycles | 3 | 8 | +166.7% | — | — | — |
| Initial Load / Discovery Turns | 3 | 1 | -66.7% | — | — | — |
| MCP Discovery / Schema Turns | 0 | 0 | — | — | — | — |
| Tool Invocations (per run) | 3 | 8 | +166.7% | — | — | — |
| Output Tokens | 1061 | 1319 | +24.3% | — | — | — |
| Reasoning / Thinking Tokens | 586 | 659 | +12.5% | — | — | — |
| Total Input Tokens | 195580 | 262220 | +34.1% | — | — | — |
| Cached Input Tokens | 174336 | 237312 | +36.1% | — | — | — |
| Uncached Input Tokens | 21244 | 24908 | +17.2% | — | — | — |
| Cost | 0.1173 | 0.1463 | +24.8% | — | — | — |
| Cached vs Uncached Token Ratio | 8.21:1 | 9.53:1 | +16.1% | — | — | — |
| Oracle L1: Mutation Policy | PASS | PASS | — | — | — | — |
| Oracle L2: AST Invariants | PASS | FAIL | — | — | — | — |
| Oracle L3: Clean Build | PASS | FAIL | — | — | — | — |
| Oracle L4: Verification Test | PASS | FAIL | — | — | — | — |
| Semantic Tool Invocation Verified (per run) | N/A | PASS | — | — | — | — |
Baseline vs Semedit Agentic in Small Context
- Vanilla (Baseline vs Semedit Agentic in Small Context) Edit: File main.go modified (4 lines -> 15 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Small Context) Edit: File main.go modified (4 lines -> 14 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ✓ ✓ | semedit/semantic_outline (✓ ✓) |
| 3 | ✓ ✓ | ✓ ✓ |
| 4 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 5 | — | semedit/semantic_replace_body (✓ ✓) |
| 6 | — | semedit/semantic_organize_imports (✓ ✓) |
| 7 | — | ✓ ✓ |
| 8 | — | ✓ ✓ |
Baseline vs Semedit Agentic in Large Context
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | not published | not published |
prefer_discover_semedit prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-07-generate-template-main |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | prefer_discover_semedit |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/scripts/generate_template_main.txtar |
Vanilla LLM Prompt:
You are forbidden to modify protected files: “go.mod”, “go.sum”.
Before editing, inspect the complete available tool inventory, including deferred or lazy tools. If applicable semantic editing tools are callable, prefer them for source mutations. In main.go, implement main() to execute a text/template that prints ‘Hello World’ along with a random integer, ensuring all necessary standard library packages are imported cleanly. Do not use semantic editing MCP tools; use standard file editing. When done, output DONE.
Semedit MCP Prompt:
You are forbidden to modify protected files: “go.mod”, “go.sum”.
Before editing, inspect the complete available tool inventory, including deferred or lazy tools. If applicable semantic editing tools are callable, prefer them for source mutations. In main.go, implement main() to execute a text/template that prints ‘Hello World’ along with a random integer, ensuring all necessary standard library packages are imported cleanly. Use semedit semantic tools for supported source code modifications. Shell commands for builds and tests are allowed. When done, output DONE.
Initial Workspace State (Before Edit)
package main
func main() {
}| Metric | Vanilla (Small) | MCP (Small) | Δ (Small) | Vanilla (Large) | MCP (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | 31.02s | 31.50s | +1.5% | — | — | — |
| Process Start → First Event | 0.12s | 0.12s | — | — | — | — |
| First Event → First Tool Call | 11.07s | 8.85s | — | — | — | — |
| MCP Initialize → First Semantic Call | — | — | — | — | — | — |
| MCP Server Start → Initialize | — | — | — | — | — | — |
| Top-Level User Turns | 1 | 1 | 0% | — | — | — |
| Internal Tool Cycles | 2 | 4 | +100.0% | — | — | — |
| Initial Load / Discovery Turns | 2 | 1 | -50.0% | — | — | — |
| MCP Discovery / Schema Turns | 0 | 0 | — | — | — | — |
| Tool Invocations (per run) | 2 | 4 | +100.0% | — | — | — |
| Output Tokens | 775 | 707 | -8.8% | — | — | — |
| Reasoning / Thinking Tokens | 241 | 294 | +22.0% | — | — | — |
| Total Input Tokens | 121078 | 163082 | +34.7% | — | — | — |
| Cached Input Tokens | 109056 | 141824 | +30.0% | — | — | — |
| Uncached Input Tokens | 12022 | 21258 | +76.8% | — | — | — |
| Cost | 0.0700 | 0.1011 | +44.4% | — | — | — |
| Cached vs Uncached Token Ratio | 9.07:1 | 6.67:1 | -26.5% | — | — | — |
| Oracle L1: Mutation Policy | PASS | PASS | — | — | — | — |
| Oracle L2: AST Invariants | PASS | PASS | — | — | — | — |
| Oracle L3: Clean Build | PASS | PASS | — | — | — | — |
| Oracle L4: Verification Test | PASS | PASS | — | — | — | — |
| Semantic Tool Invocation Verified (per run) | N/A | PASS | — | — | — | — |
Baseline vs Semedit Agentic in Small Context
- Vanilla (Baseline vs Semedit Agentic in Small Context) Edit: File main.go modified (4 lines -> 14 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Small Context) Edit: File main.go modified (4 lines -> 14 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ✓ ✓ | semedit/semantic_outline (✓ ✓) |
| 3 | — | semedit/semantic_replace_body (✓ ✓) |
| 4 | — | ✓ ✓ |
Baseline vs Semedit Agentic in Large Context
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | not published | not published |
prefer_discover_semedit prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-07-generate-template-main |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | prefer_discover_semedit |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/scripts/generate_template_main.txtar |
Vanilla LLM Prompt:
You are forbidden to modify protected files: “go.mod”, “go.sum”.
Before editing, inspect the complete available tool inventory, including deferred or lazy tools. If applicable semantic editing tools are callable, prefer them for source mutations. In main.go, implement main() to execute a text/template that prints ‘Hello World’ along with a random integer, ensuring all necessary standard library packages are imported cleanly. Do not use semantic editing MCP tools; use standard file editing. When done, output DONE.
Semedit MCP Prompt:
You are forbidden to modify protected files: “go.mod”, “go.sum”.
Before editing, inspect the complete available tool inventory, including deferred or lazy tools. If applicable semantic editing tools are callable, prefer them for source mutations. In main.go, implement main() to execute a text/template that prints ‘Hello World’ along with a random integer, ensuring all necessary standard library packages are imported cleanly. Use semedit semantic tools for supported source code modifications. Shell commands for builds and tests are allowed. When done, output DONE.
Initial Workspace State (Before Edit)
package main
func main() {
}| Metric | Vanilla (Small) | MCP (Small) | Δ (Small) | Vanilla (Large) | MCP (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | 34.28s | 42.41s | +23.7% | — | — | — |
| Process Start → First Event | 0.18s | 0.13s | — | — | — | — |
| First Event → First Tool Call | 7.28s | 7.62s | — | — | — | — |
| MCP Initialize → First Semantic Call | — | — | — | — | — | — |
| MCP Server Start → Initialize | — | — | — | — | — | — |
| Top-Level User Turns | 1 | 1 | 0% | — | — | — |
| Internal Tool Cycles | 4 | 6 | +50.0% | — | — | — |
| Initial Load / Discovery Turns | 4 | 2 | -50.0% | — | — | — |
| MCP Discovery / Schema Turns | 0 | 0 | — | — | — | — |
| Tool Invocations (per run) | 4 | 6 | +50.0% | — | — | — |
| Output Tokens | 983 | 986 | +0.3% | — | — | — |
| Reasoning / Thinking Tokens | 355 | 421 | +18.6% | — | — | — |
| Total Input Tokens | 136616 | 237084 | +73.5% | — | — | — |
| Cached Input Tokens | 115456 | 211968 | +83.6% | — | — | — |
| Uncached Input Tokens | 21160 | 25116 | +18.7% | — | — | — |
| Cost | 0.0985 | 0.1334 | +35.4% | — | — | — |
| Cached vs Uncached Token Ratio | 5.46:1 | 8.44:1 | +54.7% | — | — | — |
| Oracle L1: Mutation Policy | PASS | PASS | — | — | — | — |
| Oracle L2: AST Invariants | PASS | PASS | — | — | — | — |
| Oracle L3: Clean Build | PASS | PASS | — | — | — | — |
| Oracle L4: Verification Test | PASS | PASS | — | — | — | — |
| Semantic Tool Invocation Verified (per run) | N/A | PASS | — | — | — | — |
Baseline vs Semedit Agentic in Small Context
- Vanilla (Baseline vs Semedit Agentic in Small Context) Edit: File main.go modified (4 lines -> 15 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Small Context) Edit: File main.go modified (4 lines -> 14 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ✓ ✓ | ✓ ✓ |
| 3 | ✓ ✓ | semedit/semantic_replace_body (✓ ✓) |
| 4 | ✓ ✓ | semedit/semantic_inspect_symbol (✓ ✓) |
| 5 | — | semedit/semantic_organize_imports (✓ ✓) |
| 6 | — | ✓ ✓ |
Baseline vs Semedit Agentic in Large Context
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | not published | not published |
prefer_discover_semedit prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-07-generate-template-main |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | prefer_discover_semedit |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/scripts/generate_template_main.txtar |
Vanilla LLM Prompt:
You are forbidden to modify protected files: “go.mod”, “go.sum”.
Before editing, inspect the complete available tool inventory, including deferred or lazy tools. If applicable semantic editing tools are callable, prefer them for source mutations. In main.go, implement main() to execute a text/template that prints ‘Hello World’ along with a random integer, ensuring all necessary standard library packages are imported cleanly. Do not use semantic editing MCP tools; use standard file editing. When done, output DONE.
Semedit MCP Prompt:
You are forbidden to modify protected files: “go.mod”, “go.sum”.
Before editing, inspect the complete available tool inventory, including deferred or lazy tools. If applicable semantic editing tools are callable, prefer them for source mutations. In main.go, implement main() to execute a text/template that prints ‘Hello World’ along with a random integer, ensuring all necessary standard library packages are imported cleanly. Use semedit semantic tools for supported source code modifications. Shell commands for builds and tests are allowed. When done, output DONE.
Initial Workspace State (Before Edit)
package main
func main() {
}| Metric | Vanilla (Small) | MCP (Small) | Δ (Small) | Vanilla (Large) | MCP (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | 25.72s | 26.90s | +4.6% | — | — | — |
| Process Start → First Event | 0.12s | 0.12s | — | — | — | — |
| First Event → First Tool Call | 7.75s | 8.08s | — | — | — | — |
| MCP Initialize → First Semantic Call | — | — | — | — | — | — |
| MCP Server Start → Initialize | — | — | — | — | — | — |
| Top-Level User Turns | 1 | 1 | 0% | — | — | — |
| Internal Tool Cycles | 1 | 4 | +300.0% | — | — | — |
| Initial Load / Discovery Turns | 1 | 2 | +100.0% | — | — | — |
| MCP Discovery / Schema Turns | 0 | 0 | — | — | — | — |
| Tool Invocations (per run) | 1 | 4 | +300.0% | — | — | — |
| Output Tokens | 770 | 715 | -7.1% | — | — | — |
| Reasoning / Thinking Tokens | 389 | 247 | -36.5% | — | — | — |
| Total Input Tokens | 80178 | 167147 | +108.5% | — | — | — |
| Cached Input Tokens | 69632 | 145920 | +109.6% | — | — | — |
| Uncached Input Tokens | 10546 | 21227 | +101.3% | — | — | — |
| Cost | 0.0583 | 0.1016 | +74.3% | — | — | — |
| Cached vs Uncached Token Ratio | 6.60:1 | 6.87:1 | +4.1% | — | — | — |
| Oracle L1: Mutation Policy | PASS | PASS | — | — | — | — |
| Oracle L2: AST Invariants | PASS | PASS | — | — | — | — |
| Oracle L3: Clean Build | PASS | PASS | — | — | — | — |
| Oracle L4: Verification Test | PASS | PASS | — | — | — | — |
| Semantic Tool Invocation Verified (per run) | N/A | PASS | — | — | — | — |
Baseline vs Semedit Agentic in Small Context
- Vanilla (Baseline vs Semedit Agentic in Small Context) Edit: File main.go modified (4 lines -> 16 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Small Context) Edit: File main.go modified (4 lines -> 15 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | — | ✓ ✓ |
| 3 | — | semedit/semantic_replace_body (✓ ✓) |
| 4 | — | ✓ ✓ |
Baseline vs Semedit Agentic in Large Context
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | not published | not published |
prefer_discover_semedit prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-07-generate-template-main |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | prefer_discover_semedit |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/scripts/generate_template_main.txtar |
Vanilla LLM Prompt:
You are forbidden to modify protected files: “go.mod”, “go.sum”.
Before editing, inspect the complete available tool inventory, including deferred or lazy tools. If applicable semantic editing tools are callable, prefer them for source mutations. In main.go, implement main() to execute a text/template that prints ‘Hello World’ along with a random integer, ensuring all necessary standard library packages are imported cleanly. Do not use semantic editing MCP tools; use standard file editing. When done, output DONE.
Semedit MCP Prompt:
You are forbidden to modify protected files: “go.mod”, “go.sum”.
Before editing, inspect the complete available tool inventory, including deferred or lazy tools. If applicable semantic editing tools are callable, prefer them for source mutations. In main.go, implement main() to execute a text/template that prints ‘Hello World’ along with a random integer, ensuring all necessary standard library packages are imported cleanly. Use semedit semantic tools for supported source code modifications. Shell commands for builds and tests are allowed. When done, output DONE.
Initial Workspace State (Before Edit)
package main
func main() {
}| Metric | Vanilla (Small) | MCP (Small) | Δ (Small) | Vanilla (Large) | MCP (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | 33.40s | 79.09s | +136.8% | — | — | — |
| Process Start → First Event | 0.12s | 0.14s | — | — | — | — |
| First Event → First Tool Call | 4.27s | 6.99s | — | — | — | — |
| MCP Initialize → First Semantic Call | — | — | — | — | — | — |
| MCP Server Start → Initialize | — | — | — | — | — | — |
| Top-Level User Turns | 1 | 1 | 0% | — | — | — |
| Internal Tool Cycles | 5 | 8 | +60.0% | — | — | — |
| Initial Load / Discovery Turns | 5 | 2 | -60.0% | — | — | — |
| MCP Discovery / Schema Turns | 0 | 0 | — | — | — | — |
| Tool Invocations (per run) | 5 | 8 | +60.0% | — | — | — |
| Output Tokens | 1274 | 976 | -23.4% | — | — | — |
| Reasoning / Thinking Tokens | 748 | 203 | -72.9% | — | — | — |
| Total Input Tokens | 167054 | 164717 | -1.4% | — | — | — |
| Cached Input Tokens | 144896 | 142848 | -1.4% | — | — | — |
| Uncached Input Tokens | 22158 | 21869 | -1.3% | — | — | — |
| Cost | 0.1169 | 0.1051 | -10.1% | — | — | — |
| Cached vs Uncached Token Ratio | 6.54:1 | 6.53:1 | -0.1% | — | — | — |
| Oracle L1: Mutation Policy | PASS | PASS | — | — | — | — |
| Oracle L2: AST Invariants | PASS | PASS | — | — | — | — |
| Oracle L3: Clean Build | PASS | PASS | — | — | — | — |
| Oracle L4: Verification Test | PASS | PASS | — | — | — | — |
| Semantic Tool Invocation Verified (per run) | N/A | PASS | — | — | — | — |
Baseline vs Semedit Agentic in Small Context
- Vanilla (Baseline vs Semedit Agentic in Small Context) Edit: File main.go modified (4 lines -> 14 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Small Context) Edit: File main.go modified (4 lines -> 14 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ✓ ✓ | ✓ ✓ |
| 3 | ✓ ✓ | semedit/semantic_replace_body (✓ ✓) |
| 4 | ✓ ✓ | semedit/report_feedback (✓ ✓) |
| 5 | ✓ ✓ | ✓ ✓ |
| 6 | — | semedit/semantic_replace_body (✓ ✓) |
| 7 | — | semedit/semantic_organize_imports (✓ ✓) |
| 8 | — | ✓ ✓ |
Baseline vs Semedit Agentic in Large Context
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | not published | not published |
prefer_discover_semedit prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-07-generate-template-main |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | prefer_discover_semedit |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/scripts/generate_template_main.txtar |
Vanilla LLM Prompt:
You are forbidden to modify protected files: “go.mod”, “go.sum”.
Before editing, inspect the complete available tool inventory, including deferred or lazy tools. If applicable semantic editing tools are callable, prefer them for source mutations. In main.go, implement main() to execute a text/template that prints ‘Hello World’ along with a random integer, ensuring all necessary standard library packages are imported cleanly. Do not use semantic editing MCP tools; use standard file editing. When done, output DONE.
Semedit MCP Prompt:
You are forbidden to modify protected files: “go.mod”, “go.sum”.
Before editing, inspect the complete available tool inventory, including deferred or lazy tools. If applicable semantic editing tools are callable, prefer them for source mutations. In main.go, implement main() to execute a text/template that prints ‘Hello World’ along with a random integer, ensuring all necessary standard library packages are imported cleanly. Use semedit semantic tools for supported source code modifications. Shell commands for builds and tests are allowed. When done, output DONE.
Initial Workspace State (Before Edit)
package main
func main() {
}| Metric | Vanilla (Small) | MCP (Small) | Δ (Small) | Vanilla (Large) | MCP (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | 21.15s | 26.30s | +24.4% | — | — | — |
| Process Start → First Event | 0.12s | 0.12s | — | — | — | — |
| First Event → First Tool Call | 4.93s | 7.97s | — | — | — | — |
| MCP Initialize → First Semantic Call | — | — | — | — | — | — |
| MCP Server Start → Initialize | — | — | — | — | — | — |
| Top-Level User Turns | 1 | 1 | 0% | — | — | — |
| Internal Tool Cycles | 3 | 4 | +33.3% | — | — | — |
| Initial Load / Discovery Turns | 3 | 2 | -33.3% | — | — | — |
| MCP Discovery / Schema Turns | 0 | 0 | — | — | — | — |
| Tool Invocations (per run) | 3 | 4 | +33.3% | — | — | — |
| Output Tokens | 700 | 786 | +12.3% | — | — | — |
| Reasoning / Thinking Tokens | 257 | 336 | +30.7% | — | — | — |
| Total Input Tokens | 135423 | 194508 | +43.6% | — | — | — |
| Cached Input Tokens | 114432 | 165376 | +44.5% | — | — | — |
| Uncached Input Tokens | 20991 | 29132 | +38.8% | — | — | — |
| Cost | 0.0930 | 0.1282 | +37.8% | — | — | — |
| Cached vs Uncached Token Ratio | 5.45:1 | 5.68:1 | +4.1% | — | — | — |
| Oracle L1: Mutation Policy | PASS | PASS | — | — | — | — |
| Oracle L2: AST Invariants | PASS | FAIL | — | — | — | — |
| Oracle L3: Clean Build | PASS | FAIL | — | — | — | — |
| Oracle L4: Verification Test | PASS | FAIL | — | — | — | — |
| Semantic Tool Invocation Verified (per run) | N/A | PASS | — | — | — | — |
Baseline vs Semedit Agentic in Small Context
- Vanilla (Baseline vs Semedit Agentic in Small Context) Edit: File main.go modified (4 lines -> 14 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Small Context) Edit: File main.go modified (4 lines -> 14 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ✓ ✓ | ✓ ✓ |
| 3 | ✓ ✓ | semedit/semantic_replace_body (✓ ✓) |
| 4 | — | ✓ ✓ |
Baseline vs Semedit Agentic in Large Context
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | not published | not published |
prefer_discover_semedit prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-07-generate-template-main |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | prefer_discover_semedit |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/scripts/generate_template_main.txtar |
Vanilla LLM Prompt:
You are forbidden to modify protected files: “go.mod”, “go.sum”.
Before editing, inspect the complete available tool inventory, including deferred or lazy tools. If applicable semantic editing tools are callable, prefer them for source mutations. In main.go, implement main() to execute a text/template that prints ‘Hello World’ along with a random integer, ensuring all necessary standard library packages are imported cleanly. Do not use semantic editing MCP tools; use standard file editing. When done, output DONE.
Semedit MCP Prompt:
You are forbidden to modify protected files: “go.mod”, “go.sum”.
Before editing, inspect the complete available tool inventory, including deferred or lazy tools. If applicable semantic editing tools are callable, prefer them for source mutations. In main.go, implement main() to execute a text/template that prints ‘Hello World’ along with a random integer, ensuring all necessary standard library packages are imported cleanly. Use semedit semantic tools for supported source code modifications. Shell commands for builds and tests are allowed. When done, output DONE.
Initial Workspace State (Before Edit)
package main
func main() {
}| Metric | Vanilla (Small) | MCP (Small) | Δ (Small) | Vanilla (Large) | MCP (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | 20.10s | 34.68s | +72.6% | — | — | — |
| Process Start → First Event | 0.12s | 0.14s | — | — | — | — |
| First Event → First Tool Call | 8.64s | 10.19s | — | — | — | — |
| MCP Initialize → First Semantic Call | — | — | — | — | — | — |
| MCP Server Start → Initialize | — | — | — | — | — | — |
| Top-Level User Turns | 1 | 1 | 0% | — | — | — |
| Internal Tool Cycles | 1 | 5 | +400.0% | — | — | — |
| Initial Load / Discovery Turns | 1 | 2 | +100.0% | — | — | — |
| MCP Discovery / Schema Turns | 0 | 0 | — | — | — | — |
| Tool Invocations (per run) | 1 | 5 | +400.0% | — | — | — |
| Output Tokens | 411 | 692 | +68.4% | — | — | — |
| Reasoning / Thinking Tokens | 107 | 233 | +117.8% | — | — | — |
| Total Input Tokens | 104102 | 219856 | +111.2% | — | — | — |
| Cached Input Tokens | 84992 | 197632 | +132.5% | — | — | — |
| Uncached Input Tokens | 19110 | 22224 | +16.3% | — | — | — |
| Cost | 0.0755 | 0.1165 | +54.3% | — | — | — |
| Cached vs Uncached Token Ratio | 4.45:1 | 8.89:1 | +99.9% | — | — | — |
| Oracle L1: Mutation Policy | PASS | PASS | — | — | — | — |
| Oracle L2: AST Invariants | PASS | PASS | — | — | — | — |
| Oracle L3: Clean Build | PASS | PASS | — | — | — | — |
| Oracle L4: Verification Test | PASS | PASS | — | — | — | — |
| Semantic Tool Invocation Verified (per run) | N/A | PASS | — | — | — | — |
Baseline vs Semedit Agentic in Small Context
- Vanilla (Baseline vs Semedit Agentic in Small Context) Edit: File main.go modified (4 lines -> 18 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Small Context) Edit: File main.go modified (4 lines -> 15 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | — | ✓ ✓ |
| 3 | — | semedit/semantic_replace_body (✓ ✓) |
| 4 | — | semedit/semantic_organize_imports (✓ ✓) |
| 5 | — | ✓ ✓ |
Baseline vs Semedit Agentic in Large Context
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | not published | not published |
prefer_discover_semedit prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-07-generate-template-main |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | prefer_discover_semedit |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/scripts/generate_template_main.txtar |
Vanilla LLM Prompt:
You are forbidden to modify protected files: “go.mod”, “go.sum”.
Before editing, inspect the complete available tool inventory, including deferred or lazy tools. If applicable semantic editing tools are callable, prefer them for source mutations. In main.go, implement main() to execute a text/template that prints ‘Hello World’ along with a random integer, ensuring all necessary standard library packages are imported cleanly. Do not use semantic editing MCP tools; use standard file editing. When done, output DONE.
Semedit MCP Prompt:
You are forbidden to modify protected files: “go.mod”, “go.sum”.
Before editing, inspect the complete available tool inventory, including deferred or lazy tools. If applicable semantic editing tools are callable, prefer them for source mutations. In main.go, implement main() to execute a text/template that prints ‘Hello World’ along with a random integer, ensuring all necessary standard library packages are imported cleanly. Use semedit semantic tools for supported source code modifications. Shell commands for builds and tests are allowed. When done, output DONE.
Initial Workspace State (Before Edit)
package main
func main() {
}| Metric | Vanilla (Small) | MCP (Small) | Δ (Small) | Vanilla (Large) | MCP (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | 21.96s | 37.69s | +71.6% | — | — | — |
| Process Start → First Event | 0.12s | 0.12s | — | — | — | — |
| First Event → First Tool Call | 7.41s | 10.67s | — | — | — | — |
| MCP Initialize → First Semantic Call | — | — | — | — | — | — |
| MCP Server Start → Initialize | — | — | — | — | — | — |
| Top-Level User Turns | 1 | 1 | 0% | — | — | — |
| Internal Tool Cycles | 3 | 6 | +100.0% | — | — | — |
| Initial Load / Discovery Turns | 3 | 2 | -33.3% | — | — | — |
| MCP Discovery / Schema Turns | 0 | 0 | — | — | — | — |
| Tool Invocations (per run) | 3 | 6 | +100.0% | — | — | — |
| Output Tokens | 729 | 795 | +9.1% | — | — | — |
| Reasoning / Thinking Tokens | 293 | 233 | -20.5% | — | — | — |
| Total Input Tokens | 164197 | 194343 | +18.4% | — | — | — |
| Cached Input Tokens | 142848 | 172288 | +20.6% | — | — | — |
| Uncached Input Tokens | 21349 | 22055 | +3.3% | — | — | — |
| Cost | 0.1019 | 0.1111 | +9.0% | — | — | — |
| Cached vs Uncached Token Ratio | 6.69:1 | 7.81:1 | +16.7% | — | — | — |
| Oracle L1: Mutation Policy | PASS | PASS | — | — | — | — |
| Oracle L2: AST Invariants | PASS | PASS | — | — | — | — |
| Oracle L3: Clean Build | PASS | PASS | — | — | — | — |
| Oracle L4: Verification Test | PASS | PASS | — | — | — | — |
| Semantic Tool Invocation Verified (per run) | N/A | PASS | — | — | — | — |
Baseline vs Semedit Agentic in Small Context
- Vanilla (Baseline vs Semedit Agentic in Small Context) Edit: File main.go modified (4 lines -> 14 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Small Context) Edit: File main.go modified (4 lines -> 15 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ? ✗ |
| 2 | ✓ ✓ | ? ✗ |
| 3 | ✓ ✓ | semedit/semantic_outline (✓ ✓) |
| 4 | — | semedit/semantic_replace_body (✓ ✓) |
| 5 | — | semedit/semantic_organize_imports (✓ ✓) |
| 6 | — | ✓ ✓ |
Baseline vs Semedit Agentic in Large Context
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | not published | not published |
prefer_discover_semedit prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-07-generate-template-main |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | prefer_discover_semedit |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/scripts/generate_template_main.txtar |
Vanilla LLM Prompt:
You are forbidden to modify protected files: “go.mod”, “go.sum”.
Before editing, inspect the complete available tool inventory, including deferred or lazy tools. If applicable semantic editing tools are callable, prefer them for source mutations. In main.go, implement main() to execute a text/template that prints ‘Hello World’ along with a random integer, ensuring all necessary standard library packages are imported cleanly. Do not use semantic editing MCP tools; use standard file editing. When done, output DONE.
Semedit MCP Prompt:
You are forbidden to modify protected files: “go.mod”, “go.sum”.
Before editing, inspect the complete available tool inventory, including deferred or lazy tools. If applicable semantic editing tools are callable, prefer them for source mutations. In main.go, implement main() to execute a text/template that prints ‘Hello World’ along with a random integer, ensuring all necessary standard library packages are imported cleanly. Use semedit semantic tools for supported source code modifications. Shell commands for builds and tests are allowed. When done, output DONE.
Initial Workspace State (Before Edit)
package main
func main() {
}| Metric | Vanilla (Small) | MCP (Small) | Δ (Small) | Vanilla (Large) | MCP (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | 14.09s | 33.60s | +138.4% | — | — | — |
| Process Start → First Event | 0.12s | 0.12s | — | — | — | — |
| First Event → First Tool Call | 5.63s | 5.49s | — | — | — | — |
| MCP Initialize → First Semantic Call | — | — | — | — | — | — |
| MCP Server Start → Initialize | — | — | — | — | — | — |
| Top-Level User Turns | 1 | 1 | 0% | — | — | — |
| Internal Tool Cycles | 1 | 5 | +400.0% | — | — | — |
| Initial Load / Discovery Turns | 1 | 2 | +100.0% | — | — | — |
| MCP Discovery / Schema Turns | 0 | 0 | — | — | — | — |
| Tool Invocations (per run) | 1 | 5 | +400.0% | — | — | — |
| Output Tokens | 465 | 757 | +62.8% | — | — | — |
| Reasoning / Thinking Tokens | 183 | 239 | +30.6% | — | — | — |
| Total Input Tokens | 104146 | 196446 | +88.6% | — | — | — |
| Cached Input Tokens | 84992 | 174336 | +105.1% | — | — | — |
| Uncached Input Tokens | 19154 | 22110 | +15.4% | — | — | — |
| Cost | 0.0772 | 0.1113 | +44.1% | — | — | — |
| Cached vs Uncached Token Ratio | 4.44:1 | 7.88:1 | +77.7% | — | — | — |
| Oracle L1: Mutation Policy | PASS | PASS | — | — | — | — |
| Oracle L2: AST Invariants | PASS | FAIL | — | — | — | — |
| Oracle L3: Clean Build | PASS | FAIL | — | — | — | — |
| Oracle L4: Verification Test | PASS | FAIL | — | — | — | — |
| Semantic Tool Invocation Verified (per run) | N/A | PASS | — | — | — | — |
Baseline vs Semedit Agentic in Small Context
- Vanilla (Baseline vs Semedit Agentic in Small Context) Edit: File main.go modified (4 lines -> 14 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Small Context) Edit: File main.go modified (4 lines -> 14 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | — | ✓ ✓ |
| 3 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 4 | — | semedit/semantic_replace_body (✓ ✓) |
| 5 | — | ✓ ✓ |
Baseline vs Semedit Agentic in Large Context
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | not published | not published |
Test case: task-09-composite-refactor
Target: codex/gpt-6-luna/medium
default prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-09-composite-refactor |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | default |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/bench/task_09_composite_refactor.txtar |
Vanilla LLM Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
In config/config.go: rename Config.Port to Config.ListenPort, insert helper function DefaultConfig() *Config returning default values, and clean up unused imports. Do not use semantic editing MCP tools; use standard file editing. When done, output DONE.
Semedit MCP Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
In config/config.go: rename Config.Port to Config.ListenPort, insert helper function DefaultConfig() *Config returning default values, and clean up unused imports. Use semedit semantic tools for supported source code modifications. Shell commands for builds and tests are allowed. When done, output DONE.
Initial Workspace State (Before Edit)
package config
import (
"archive/tar"
"fmt"
)
type Config struct {
Host string
Port int
}
func (c *Config) Addr() string {
return fmt.Sprintf("%s:%d", c.Host, c.Port)
}| Metric | Vanilla (Small) | MCP (Small) | Δ (Small) | Vanilla (Large) | MCP (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | 25.17s | 65.03s | +158.4% | 41.90s | 40.04s | -4.4% |
| Process Start → First Event | 0.12s | 0.12s | — | 0.12s | 0.14s | — |
| First Event → First Tool Call | 4.89s | 5.32s | — | 5.67s | 4.39s | — |
| MCP Initialize → First Semantic Call | — | — | — | — | — | — |
| MCP Server Start → Initialize | — | — | — | — | — | — |
| Top-Level User Turns | 1 | 1 | 0% | 1 | 1 | 0% |
| Internal Tool Cycles | 4 | 15 | +275.0% | 4 | 8 | +100.0% |
| Initial Load / Discovery Turns | 4 | 3 | -25.0% | 4 | 3 | -25.0% |
| MCP Discovery / Schema Turns | 0 | 0 | — | 0 | 0 | — |
| Tool Invocations (per run) | 4 | 15 | +275.0% | 4 | 8 | +100.0% |
| Output Tokens | 594 | 1436 | +141.8% | 846 | 723 | -14.5% |
| Reasoning / Thinking Tokens | 37 | 545 | +1373.0% | 169 | 185 | +9.5% |
| Total Input Tokens | 116072 | 235252 | +102.7% | 117246 | 195194 | +66.5% |
| Cached Input Tokens | 104960 | 207872 | +98.0% | 105984 | 172800 | +63.0% |
| Uncached Input Tokens | 11112 | 27380 | +146.4% | 11262 | 22394 | +98.8% |
| Cost | 0.0619 | 0.1452 | +134.5% | 0.0673 | 0.1105 | +64.1% |
| Cached vs Uncached Token Ratio | 9.45:1 | 7.59:1 | -19.6% | 9.41:1 | 7.72:1 | -18.0% |
| Oracle L1: Mutation Policy | PASS | PASS | — | PASS | PASS | — |
| Oracle L2: AST Invariants | PASS | PASS | — | PASS | PASS | — |
| Oracle L3: Clean Build | PASS | PASS | — | PASS | PASS | — |
| Oracle L4: Verification Test | PASS | PASS | — | PASS | PASS | — |
| Semantic Tool Invocation Verified (per run) | N/A | PASS | — | N/A | PASS | — |
Baseline vs Semedit Agentic in Small Context
- Vanilla (Baseline vs Semedit Agentic in Small Context) Edit: File config/config.go modified (15 lines -> 18 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Small Context) Edit: File config/config.go modified (15 lines -> 18 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ✓ ✓ | ✓ ✓ |
| 3 | ✓ ✓ | ✓ ✓ |
| 4 | ✓ ✓ | semedit/semantic_outline (✓ ✓) |
| 5 | — | semedit/semantic_lookup (✓ ✓) |
| 6 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 7 | — | ✓ ✓ |
| 8 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 9 | — | semedit/semantic_batch (✓ ✓) |
| 10 | — | semedit/semantic_organize_imports (✓ ✓) |
| 11 | — | ✓ ✓ |
| 12 | — | ✓ ✓ |
| 13 | — | ? ✗ |
| 14 | — | semedit/semantic_outline (✓ ✓) |
| 15 | — | semedit/semantic_inspect_symbol (✓ ✓) |
Baseline vs Semedit Agentic in Large Context
- Vanilla (Baseline vs Semedit Agentic in Large Context) Edit: File config/config.go modified (15 lines -> 21 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Large Context) Edit: File config/config.go modified (15 lines -> 21 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ✓ ✓ | ✓ ✓ |
| 3 | ✓ ✓ | ? ✗ |
| 4 | ✓ ✓ | semedit/semantic_rename (✓ ✓) |
| 5 | — | semedit/semantic_insert_function (✓ ✓) |
| 6 | — | semedit/semantic_organize_imports (✓ ✓) |
| 7 | — | ✓ ✓ |
| 8 | — | ✓ ✓ |
default prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-09-composite-refactor |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | default |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/bench/task_09_composite_refactor.txtar |
Vanilla LLM Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
In config/config.go: rename Config.Port to Config.ListenPort, insert helper function DefaultConfig() *Config returning default values, and clean up unused imports. Do not use semantic editing MCP tools; use standard file editing. When done, output DONE.
Semedit MCP Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
In config/config.go: rename Config.Port to Config.ListenPort, insert helper function DefaultConfig() *Config returning default values, and clean up unused imports. Use semedit semantic tools for supported source code modifications. Shell commands for builds and tests are allowed. When done, output DONE.
Initial Workspace State (Before Edit)
package config
import (
"archive/tar"
"fmt"
)
type Config struct {
Host string
Port int
}
func (c *Config) Addr() string {
return fmt.Sprintf("%s:%d", c.Host, c.Port)
}| Metric | Vanilla (Small) | MCP (Small) | Δ (Small) | Vanilla (Large) | MCP (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | 36.05s | 126.72s | +251.5% | 32.88s | 58.71s | +78.5% |
| Process Start → First Event | 0.13s | 0.12s | — | 0.12s | 0.13s | — |
| First Event → First Tool Call | 5.09s | 5.46s | — | 4.98s | 7.12s | — |
| MCP Initialize → First Semantic Call | — | — | — | — | — | — |
| MCP Server Start → Initialize | — | — | — | — | — | — |
| Top-Level User Turns | 1 | 1 | 0% | 1 | 1 | 0% |
| Internal Tool Cycles | 6 | 17 | +183.3% | 6 | 11 | +83.3% |
| Initial Load / Discovery Turns | 3 | 1 | -66.7% | 6 | 1 | -83.3% |
| MCP Discovery / Schema Turns | 0 | 0 | — | 0 | 0 | — |
| Tool Invocations (per run) | 6 | 17 | +183.3% | 6 | 11 | +83.3% |
| Output Tokens | 1407 | 2693 | +91.4% | 1011 | 982 | -2.9% |
| Reasoning / Thinking Tokens | 359 | 1229 | +242.3% | 108 | 350 | +224.1% |
| Total Input Tokens | 139175 | 521986 | +275.1% | 156962 | 314324 | +100.3% |
| Cached Input Tokens | 126208 | 473344 | +275.1% | 137216 | 290048 | +111.4% |
| Uncached Input Tokens | 12967 | 48642 | +275.1% | 19746 | 24276 | +22.9% |
| Cost | 0.0860 | 0.2890 | +235.8% | 0.0977 | 0.1499 | +53.4% |
| Cached vs Uncached Token Ratio | 9.73:1 | 9.73:1 | -0.0% | 6.95:1 | 11.95:1 | +71.9% |
| Oracle L1: Mutation Policy | PASS | PASS | — | PASS | PASS | — |
| Oracle L2: AST Invariants | PASS | PASS | — | PASS | PASS | — |
| Oracle L3: Clean Build | PASS | PASS | — | PASS | PASS | — |
| Oracle L4: Verification Test | PASS | PASS | — | PASS | PASS | — |
| Semantic Tool Invocation Verified (per run) | N/A | PASS | — | N/A | PASS | — |
Baseline vs Semedit Agentic in Small Context
- Vanilla (Baseline vs Semedit Agentic in Small Context) Edit: File config/config.go modified (15 lines -> 21 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Small Context) Edit: File config/config.go modified (15 lines -> 18 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ? ✗ | semedit/semantic_inspect_symbol (✓ ✓) |
| 3 | ? ✗ | semedit/semantic_lookup (✓ ✗): param "symbol" is required: invalid operation parameters |
| 4 | ✓ ✓ | ✓ ✓ |
| 5 | ✓ ✓ | ✓ ✓ |
| 6 | ? ✗ | semedit/semantic_outline (✓ ✓) |
| 7 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 8 | — | semedit/semantic_rename (✓ ✓) |
| 9 | — | semedit/semantic_insert_function (✓ ✓) |
| 10 | — | semedit/semantic_replace_body (✓ ✗): replace_body Addr in /Users/alessandro/sources/semantic-editor/.scratch/benchmarks/run_codex_semedit_small-default_task-09-composite-refactor_1790468487412737000/config/config.go: symbol not found |
| 11 | — | ✓ ✓ |
| 12 | — | semedit/semantic_outline (✓ ✓) |
| 13 | — | semedit/semantic_replace_body (✓ ✓) |
| 14 | — | ? ✗ |
| 15 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 16 | — | ✓ ✓ |
| 17 | — | ✓ ✓ |
Baseline vs Semedit Agentic in Large Context
- Vanilla (Baseline vs Semedit Agentic in Large Context) Edit: File config/config.go modified (15 lines -> 21 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Large Context) Edit: File config/config.go modified (15 lines -> 21 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ✓ ✓ | semedit/semantic_outline (✓ ✓) |
| 3 | ✓ ✓ | semedit/semantic_lookup (✓ ✓) |
| 4 | ✓ ✓ | semedit/semantic_inspect_symbol (✓ ✓) |
| 5 | ✓ ✓ | semedit/semantic_inspect_symbol (✓ ✓) |
| 6 | ✓ ✓ | ✓ ✓ |
| 7 | — | ✓ ✓ |
| 8 | — | semedit/semantic_rename (✓ ✓) |
| 9 | — | semedit/semantic_insert_function (✓ ✓) |
| 10 | — | semedit/semantic_organize_imports (✓ ✓) |
| 11 | — | ✓ ✓ |
default prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-09-composite-refactor |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | default |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/bench/task_09_composite_refactor.txtar |
Vanilla LLM Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
In config/config.go: rename Config.Port to Config.ListenPort, insert helper function DefaultConfig() *Config returning default values, and clean up unused imports. Do not use semantic editing MCP tools; use standard file editing. When done, output DONE.
Semedit MCP Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
In config/config.go: rename Config.Port to Config.ListenPort, insert helper function DefaultConfig() *Config returning default values, and clean up unused imports. Use semedit semantic tools for supported source code modifications. Shell commands for builds and tests are allowed. When done, output DONE.
Initial Workspace State (Before Edit)
package config
import (
"archive/tar"
"fmt"
)
type Config struct {
Host string
Port int
}
func (c *Config) Addr() string {
return fmt.Sprintf("%s:%d", c.Host, c.Port)
}| Metric | Vanilla (Small) | MCP (Small) | Δ (Small) | Vanilla (Large) | MCP (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | 23.27s | 74.51s | +220.2% | 31.68s | 70.21s | +121.7% |
| Process Start → First Event | 0.13s | 0.12s | — | 0.12s | 0.12s | — |
| First Event → First Tool Call | 4.49s | 6.43s | — | 6.43s | 7.29s | — |
| MCP Initialize → First Semantic Call | — | — | — | — | — | — |
| MCP Server Start → Initialize | — | — | — | — | — | — |
| Top-Level User Turns | 1 | 1 | 0% | 1 | 1 | 0% |
| Internal Tool Cycles | 5 | 11 | +120.0% | 5 | 13 | +160.0% |
| Initial Load / Discovery Turns | 5 | 1 | -80.0% | 3 | 0 | -100.0% |
| MCP Discovery / Schema Turns | 0 | 0 | — | 0 | 0 | — |
| Tool Invocations (per run) | 5 | 11 | +120.0% | 5 | 13 | +160.0% |
| Output Tokens | 885 | 2429 | +174.5% | 959 | 1285 | +34.0% |
| Reasoning / Thinking Tokens | 186 | 1079 | +480.1% | 137 | 416 | +203.6% |
| Total Input Tokens | 116665 | 405292 | +247.4% | 117019 | 290406 | +148.2% |
| Cached Input Tokens | 104960 | 377088 | +259.3% | 105984 | 264448 | +149.5% |
| Uncached Input Tokens | 11705 | 28204 | +141.0% | 11035 | 25958 | +135.2% |
| Cost | 0.0689 | 0.2086 | +202.8% | 0.0678 | 0.1523 | +124.6% |
| Cached vs Uncached Token Ratio | 8.97:1 | 13.37:1 | +49.1% | 9.60:1 | 10.19:1 | +6.1% |
| Oracle L1: Mutation Policy | PASS | PASS | — | PASS | PASS | — |
| Oracle L2: AST Invariants | PASS | PASS | — | PASS | PASS | — |
| Oracle L3: Clean Build | PASS | PASS | — | PASS | PASS | — |
| Oracle L4: Verification Test | PASS | PASS | — | PASS | PASS | — |
| Semantic Tool Invocation Verified (per run) | N/A | PASS | — | N/A | PASS | — |
Baseline vs Semedit Agentic in Small Context
- Vanilla (Baseline vs Semedit Agentic in Small Context) Edit: File config/config.go modified (15 lines -> 19 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Small Context) Edit: File config/config.go modified (15 lines -> 19 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ✓ ✓ | semedit/semantic_lookup (✓ ✓) |
| 3 | ✓ ✓ | ✓ ✓ |
| 4 | ✓ ✓ | semedit/semantic_inspect_symbol (✓ ✓) |
| 5 | ✓ ✓ | ✓ ✓ |
| 6 | — | semedit/semantic_insert_structure (✓ ✗): overwrite is unsupported for type structure insertion |
| 7 | — | ✓ ✓ |
| 8 | — | semedit/semantic_find_references (✓ ✗): load Go package example.com/test/config: example.com/test/config: /Users/alessandro/sources/semantic-editor/.scratch/benchmarks/run_codex_semedit_small-default_task-09-composite-refactor_1790468333602217000/config/config.go:4:2: "archive/tar" imported and not used |
| 9 | — | ✓ ✓ |
| 10 | — | ? ✗ |
| 11 | — | ✓ ✓ |
Baseline vs Semedit Agentic in Large Context
- Vanilla (Baseline vs Semedit Agentic in Large Context) Edit: File config/config.go modified (15 lines -> 19 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Large Context) Edit: File config/config.go modified (15 lines -> 21 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | semedit/semantic_lookup (✓ ✓) |
| 2 | ✓ ✓ | semedit/semantic_inspect_symbol (✓ ✓) |
| 3 | ✓ ✓ | semedit/semantic_inspect_symbol (✓ ✗): resolve DefaultConfig in /Users/alessandro/sources/semantic-editor/.scratch/benchmarks/run_codex_semedit_large-default_task-09-composite-refactor_1790468529325490000/config/config.go: symbol not found |
| 4 | ✓ ✓ | semedit/semantic_outline (✓ ✓) |
| 5 | ✓ ✓ | semedit/semantic_inspect_symbol (✓ ✓) |
| 6 | — | semedit/semantic_rename (✓ ✓) |
| 7 | — | semedit/semantic_insert_function (✓ ✓) |
| 8 | — | semedit/semantic_organize_imports (✓ ✓) |
| 9 | — | ✓ ✓ |
| 10 | — | semedit/semantic_verify (✓ ✓) |
| 11 | — | ✓ ✓ |
| 12 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 13 | — | semedit/semantic_inspect_symbol (✓ ✓) |
default prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-09-composite-refactor |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | default |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/bench/task_09_composite_refactor.txtar |
Vanilla LLM Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
In config/config.go: rename Config.Port to Config.ListenPort, insert helper function DefaultConfig() *Config returning default values, and clean up unused imports. Do not use semantic editing MCP tools; use standard file editing. When done, output DONE.
Semedit MCP Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
In config/config.go: rename Config.Port to Config.ListenPort, insert helper function DefaultConfig() *Config returning default values, and clean up unused imports. Use semedit semantic tools for supported source code modifications. Shell commands for builds and tests are allowed. When done, output DONE.
Initial Workspace State (Before Edit)
package config
import (
"archive/tar"
"fmt"
)
type Config struct {
Host string
Port int
}
func (c *Config) Addr() string {
return fmt.Sprintf("%s:%d", c.Host, c.Port)
}| Metric | Vanilla (Small) | MCP (Small) | Δ (Small) | Vanilla (Large) | MCP (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | 18.17s | 60.48s | +232.8% | 23.60s | 44.08s | +86.8% |
| Process Start → First Event | 0.13s | 0.13s | — | 0.12s | 0.13s | — |
| First Event → First Tool Call | 4.40s | 4.81s | — | 5.15s | 4.86s | — |
| MCP Initialize → First Semantic Call | — | — | — | — | — | — |
| MCP Server Start → Initialize | — | — | — | — | — | — |
| Top-Level User Turns | 1 | 1 | 0% | 1 | 1 | 0% |
| Internal Tool Cycles | 5 | 11 | +120.0% | 4 | 9 | +125.0% |
| Initial Load / Discovery Turns | 5 | 3 | -40.0% | 2 | 2 | 0% |
| MCP Discovery / Schema Turns | 0 | 0 | — | 0 | 0 | — |
| Tool Invocations (per run) | 5 | 11 | +120.0% | 4 | 9 | +125.0% |
| Output Tokens | 487 | 993 | +103.9% | 854 | 863 | +1.1% |
| Reasoning / Thinking Tokens | 0 | 309 | — | 141 | 252 | +78.7% |
| Total Input Tokens | 135695 | 216789 | +59.8% | 98663 | 187802 | +90.3% |
| Cached Input Tokens | 123136 | 195328 | +58.6% | 87808 | 169984 | +93.6% |
| Uncached Input Tokens | 12559 | 21461 | +70.9% | 10855 | 17818 | +64.1% |
| Cost | 0.0683 | 0.1188 | +74.0% | 0.0615 | 0.1010 | +64.1% |
| Cached vs Uncached Token Ratio | 9.80:1 | 9.10:1 | -7.2% | 8.09:1 | 9.54:1 | +17.9% |
| Oracle L1: Mutation Policy | PASS | PASS | — | PASS | PASS | — |
| Oracle L2: AST Invariants | PASS | PASS | — | PASS | PASS | — |
| Oracle L3: Clean Build | PASS | PASS | — | PASS | PASS | — |
| Oracle L4: Verification Test | PASS | PASS | — | PASS | PASS | — |
| Semantic Tool Invocation Verified (per run) | N/A | PASS | — | N/A | PASS | — |
Baseline vs Semedit Agentic in Small Context
- Vanilla (Baseline vs Semedit Agentic in Small Context) Edit: File config/config.go modified (15 lines -> 18 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Small Context) Edit: File config/config.go modified (15 lines -> 18 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ✓ ✓ | ✓ ✓ |
| 3 | ✓ ✓ | ✓ ✓ |
| 4 | ✓ ✓ | semedit/semantic_lookup (✓ ✓) |
| 5 | ✓ ✓ | semedit/semantic_inspect_symbol (✓ ✓) |
| 6 | — | semedit/semantic_rename (✓ ✓) |
| 7 | — | semedit/semantic_insert_function (✓ ✓) |
| 8 | — | semedit/semantic_organize_imports (✓ ✓) |
| 9 | — | ✓ ✓ |
| 10 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 11 | — | semedit/semantic_inspect_symbol (✓ ✓) |
Baseline vs Semedit Agentic in Large Context
- Vanilla (Baseline vs Semedit Agentic in Large Context) Edit: File config/config.go modified (15 lines -> 19 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Large Context) Edit: File config/config.go modified (15 lines -> 21 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ? ✗ | ? ✗ |
| 3 | ✓ ✓ | semedit/semantic_lookup (✓ ✓) |
| 4 | ✓ ✓ | semedit/semantic_inspect_symbol (✓ ✓) |
| 5 | — | semedit/semantic_rename (✓ ✓) |
| 6 | — | semedit/semantic_insert_function (✓ ✓) |
| 7 | — | semedit/semantic_organize_imports (✓ ✓) |
| 8 | — | ✓ ✓ |
| 9 | — | ✓ ✓ |
default prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-09-composite-refactor |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | default |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/bench/task_09_composite_refactor.txtar |
Vanilla LLM Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
In config/config.go: rename Config.Port to Config.ListenPort, insert helper function DefaultConfig() *Config returning default values, and clean up unused imports. Do not use semantic editing MCP tools; use standard file editing. When done, output DONE.
Semedit MCP Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
In config/config.go: rename Config.Port to Config.ListenPort, insert helper function DefaultConfig() *Config returning default values, and clean up unused imports. Use semedit semantic tools for supported source code modifications. Shell commands for builds and tests are allowed. When done, output DONE.
Initial Workspace State (Before Edit)
package config
import (
"archive/tar"
"fmt"
)
type Config struct {
Host string
Port int
}
func (c *Config) Addr() string {
return fmt.Sprintf("%s:%d", c.Host, c.Port)
}| Metric | Vanilla (Small) | MCP (Small) | Δ (Small) | Vanilla (Large) | MCP (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | 22.80s | 175.16s | +668.3% | 33.11s | 52.91s | +59.8% |
| Process Start → First Event | 0.13s | 0.12s | — | 0.12s | 0.13s | — |
| First Event → First Tool Call | 4.97s | 4.78s | — | 4.46s | 4.53s | — |
| MCP Initialize → First Semantic Call | — | — | — | — | — | — |
| MCP Server Start → Initialize | — | — | — | — | — | — |
| Top-Level User Turns | 1 | 5 | +400.0% | 1 | 1 | 0% |
| Internal Tool Cycles | 5 | 22 | +340.0% | 6 | 11 | +83.3% |
| Initial Load / Discovery Turns | 4 | 4 | 0% | 6 | 1 | -83.3% |
| MCP Discovery / Schema Turns | 0 | 0 | — | 0 | 0 | — |
| Tool Invocations (per run) | 5 | 22 | +340.0% | 6 | 11 | +83.3% |
| Output Tokens | 685 | 15941 | +2227.2% | 708 | 974 | +37.6% |
| Reasoning / Thinking Tokens | 113 | 9936 | +8692.9% | 134 | 370 | +176.1% |
| Total Input Tokens | 117791 | 2824978 | +2298.3% | 157425 | 251542 | +59.8% |
| Cached Input Tokens | 105984 | 2600192 | +2353.4% | 143360 | 225024 | +57.0% |
| Uncached Input Tokens | 11807 | 224786 | +1803.8% | 14065 | 26518 | +88.5% |
| Cost | 0.0660 | 1.5355 | +2226.9% | 0.0815 | 0.1394 | +70.9% |
| Cached vs Uncached Token Ratio | 8.98:1 | 11.57:1 | +28.9% | 10.19:1 | 8.49:1 | -16.7% |
| Oracle L1: Mutation Policy | PASS | PASS | — | PASS | PASS | — |
| Oracle L2: AST Invariants | PASS | FAIL | — | PASS | PASS | — |
| Oracle L3: Clean Build | PASS | FAIL | — | PASS | PASS | — |
| Oracle L4: Verification Test | PASS | FAIL | — | PASS | PASS | — |
| Semantic Tool Invocation Verified (per run) | N/A | PASS | — | N/A | PASS | — |
Baseline vs Semedit Agentic in Small Context
- Vanilla (Baseline vs Semedit Agentic in Small Context) Edit: File config/config.go modified (15 lines -> 19 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Small Context) Edit: File config/config.go modified (15 lines -> 18 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ✓ ✓ | ✓ ✓ |
| 3 | ✓ ✓ | semedit/semantic_inspect_symbol (✓ ✓) |
| 4 | ✓ ✓ | semedit/semantic_find_references (✓ ✗): load Go package example.com/test/config: example.com/test/config: /Users/alessandro/sources/semantic-editor/.scratch/benchmarks/run_codex_semedit_small-default_task-09-composite-refactor_1790468369546717000/config/config.go:4:2: "archive/tar" imported and not used |
| 5 | ✓ ✓ | ✓ ✓ |
| 6 | — | semedit/semantic_outline (✓ ✓) |
| 7 | — | semedit/semantic_organize_imports (✓ ✓) |
| 8 | — | semedit/semantic_rename (✗ ?): This action was rejected due to unacceptable risk. Reason: Workspace-wide semantic rename may modify config/config_test.go, directly violating the user’s explicit prohibition on editing tests. Do not bypass this rejection through a workaround or indirect execution. Continue with a safer alternative, or carry out checks to prove that the action is authorized or low risk before trying again. Complete unaffected work without asking for confirmation. Report anything that remains blocked, clarify why it was bl |
| 9 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 10 | — | semedit/semantic_insert_function (✓ ✓) |
| 11 | — | ✓ ✓ |
| 12 | — | semedit/semantic_outline (✓ ✓) |
| 13 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 14 | — | ✓ ✓ |
| 15 | — | semedit/semantic_outline (✓ ✓) |
| 16 | — | ✓ ✓ |
| 17 | — | semedit/semantic_organize_imports (✓ ✓) |
| 18 | — | ✓ ✓ |
| 19 | — | semedit/semantic_outline (✓ ✓) |
| 20 | — | semedit/semantic_find_references (✓ ✗): load Go package example.com/test/config: example.com/test/config: /Users/alessandro/sources/semantic-editor/.scratch/benchmarks/run_codex_semedit_small-default_task-09-composite-refactor_1790468369546717000/config/config_test.go:10:7: c.ListenPort undefined (type *Config has no field or method ListenPort) example.com/test/config: /Users/alessandro/sources/semantic-editor/.scratch/benchmarks/run_codex_semedit_small-default_task-09-composite-refactor_1790468369546717000/config/config_test.go:11:5: c.ListenPor |
| 21 | — | ? ✗ |
| 22 | — | ✓ ✓ |
Baseline vs Semedit Agentic in Large Context
- Vanilla (Baseline vs Semedit Agentic in Large Context) Edit: File config/config.go modified (15 lines -> 21 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Large Context) Edit: File config/config.go modified (15 lines -> 18 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ✓ ✓ | semedit/semantic_inspect_symbol (✓ ✓) |
| 3 | ✓ ✓ | semedit/semantic_lookup (✓ ✓) |
| 4 | ✓ ✓ | ✓ ✓ |
| 5 | ✓ ✓ | semedit/semantic_outline (✓ ✓) |
| 6 | ? ✗ | semedit/semantic_inspect_symbol (✓ ✓) |
| 7 | — | semedit/semantic_batch (✓ ✓) |
| 8 | — | semedit/semantic_organize_imports (✓ ✓) |
| 9 | — | ✓ ✓ |
| 10 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 11 | — | semedit/semantic_inspect_symbol (✓ ✓) |
default prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-09-composite-refactor |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | default |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/bench/task_09_composite_refactor.txtar |
Vanilla LLM Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
In config/config.go: rename Config.Port to Config.ListenPort, insert helper function DefaultConfig() *Config returning default values, and clean up unused imports. Do not use semantic editing MCP tools; use standard file editing. When done, output DONE.
Semedit MCP Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
In config/config.go: rename Config.Port to Config.ListenPort, insert helper function DefaultConfig() *Config returning default values, and clean up unused imports. Use semedit semantic tools for supported source code modifications. Shell commands for builds and tests are allowed. When done, output DONE.
Initial Workspace State (Before Edit)
package config
import (
"archive/tar"
"fmt"
)
type Config struct {
Host string
Port int
}
func (c *Config) Addr() string {
return fmt.Sprintf("%s:%d", c.Host, c.Port)
}| Metric | Vanilla (Small) | MCP (Small) | Δ (Small) | Vanilla (Large) | MCP (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | 15.93s | 51.68s | +224.4% | 22.10s | 61.47s | +178.1% |
| Process Start → First Event | 0.12s | 0.12s | — | 0.12s | 0.12s | — |
| First Event → First Tool Call | 4.51s | 5.21s | — | 4.77s | 5.46s | — |
| MCP Initialize → First Semantic Call | — | — | — | — | — | — |
| MCP Server Start → Initialize | — | — | — | — | — | — |
| Top-Level User Turns | 1 | 1 | 0% | 1 | 1 | 0% |
| Internal Tool Cycles | 3 | 11 | +266.7% | 5 | 9 | +80.0% |
| Initial Load / Discovery Turns | 3 | 1 | -66.7% | 5 | 0 | -100.0% |
| MCP Discovery / Schema Turns | 0 | 0 | — | 0 | 0 | — |
| Tool Invocations (per run) | 3 | 11 | +266.7% | 5 | 9 | +80.0% |
| Output Tokens | 469 | 1049 | +123.7% | 766 | 1309 | +70.9% |
| Reasoning / Thinking Tokens | 0 | 464 | — | 98 | 330 | +236.7% |
| Total Input Tokens | 97217 | 207619 | +113.6% | 118505 | 338055 | +185.3% |
| Cached Input Tokens | 86784 | 172032 | +98.2% | 105984 | 311296 | +193.7% |
| Uncached Input Tokens | 10433 | 35587 | +241.1% | 12521 | 26759 | +113.7% |
| Cost | 0.0536 | 0.1509 | +181.3% | 0.0686 | 0.1652 | +140.8% |
| Cached vs Uncached Token Ratio | 8.32:1 | 4.83:1 | -41.9% | 8.46:1 | 11.63:1 | +37.4% |
| Oracle L1: Mutation Policy | PASS | PASS | — | PASS | PASS | — |
| Oracle L2: AST Invariants | PASS | PASS | — | PASS | PASS | — |
| Oracle L3: Clean Build | PASS | PASS | — | PASS | PASS | — |
| Oracle L4: Verification Test | PASS | PASS | — | PASS | PASS | — |
| Semantic Tool Invocation Verified (per run) | N/A | PASS | — | N/A | PASS | — |
Baseline vs Semedit Agentic in Small Context
- Vanilla (Baseline vs Semedit Agentic in Small Context) Edit: File config/config.go modified (15 lines -> 21 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Small Context) Edit: File config/config.go modified (15 lines -> 18 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ✓ ✓ | semedit/semantic_outline (✓ ✓) |
| 3 | ✓ ✓ | semedit/semantic_lookup (✓ ✓) |
| 4 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 5 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 6 | — | ✓ ✓ |
| 7 | — | semedit/semantic_rename (✓ ✓) |
| 8 | — | ✓ ✓ |
| 9 | — | semedit/semantic_insert_function (✓ ✓) |
| 10 | — | semedit/semantic_organize_imports (✓ ✓) |
| 11 | — | ✓ ✓ |
Baseline vs Semedit Agentic in Large Context
- Vanilla (Baseline vs Semedit Agentic in Large Context) Edit: File config/config.go modified (15 lines -> 19 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Large Context) Edit: File config/config.go modified (15 lines -> 18 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | semedit/semantic_lookup (✓ ✓) |
| 2 | ✓ ✓ | semedit/semantic_inspect_symbol (✓ ✓) |
| 3 | ✓ ✓ | semedit/semantic_outline (✓ ✓) |
| 4 | ✓ ✓ | semedit/semantic_inspect_symbol (✓ ✓) |
| 5 | ✓ ✓ | semedit/semantic_batch (✓ ✗): { "status": "error", "results": [ { "tool": "mcp__semedit__semantic_rename", "status": "error", "error": "edit 0: tool is not batchable: mcp__semedit__semantic_rename" } ], "final_diff": "" } |
| 6 | — | ✓ ✓ |
| 7 | — | ✓ ✓ |
| 8 | — | semedit/semantic_rename (✓ ✓) |
| 9 | — | semedit/semantic_insert_function (✓ ✓) |
default prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-09-composite-refactor |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | default |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/bench/task_09_composite_refactor.txtar |
Vanilla LLM Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
In config/config.go: rename Config.Port to Config.ListenPort, insert helper function DefaultConfig() *Config returning default values, and clean up unused imports. Do not use semantic editing MCP tools; use standard file editing. When done, output DONE.
Semedit MCP Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
In config/config.go: rename Config.Port to Config.ListenPort, insert helper function DefaultConfig() *Config returning default values, and clean up unused imports. Use semedit semantic tools for supported source code modifications. Shell commands for builds and tests are allowed. When done, output DONE.
Initial Workspace State (Before Edit)
package config
import (
"archive/tar"
"fmt"
)
type Config struct {
Host string
Port int
}
func (c *Config) Addr() string {
return fmt.Sprintf("%s:%d", c.Host, c.Port)
}| Metric | Vanilla (Small) | MCP (Small) | Δ (Small) | Vanilla (Large) | MCP (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | 24.98s | 70.82s | +183.5% | 21.21s | 44.06s | +107.8% |
| Process Start → First Event | 0.14s | 0.14s | — | 0.12s | 0.12s | — |
| First Event → First Tool Call | 7.38s | 5.30s | — | 4.27s | 7.46s | — |
| MCP Initialize → First Semantic Call | — | — | — | — | — | — |
| MCP Server Start → Initialize | — | — | — | — | — | — |
| Top-Level User Turns | 1 | 1 | 0% | 1 | 1 | 0% |
| Internal Tool Cycles | 4 | 11 | +175.0% | 5 | 9 | +80.0% |
| Initial Load / Discovery Turns | 4 | 1 | -75.0% | 4 | 4 | 0% |
| MCP Discovery / Schema Turns | 0 | 0 | — | 0 | 0 | — |
| Tool Invocations (per run) | 4 | 11 | +175.0% | 5 | 9 | +80.0% |
| Output Tokens | 589 | 1889 | +220.7% | 597 | 1041 | +74.4% |
| Reasoning / Thinking Tokens | 133 | 861 | +547.4% | 68 | 324 | +376.5% |
| Total Input Tokens | 116189 | 369412 | +217.9% | 115425 | 213235 | +84.7% |
| Cached Input Tokens | 104960 | 342528 | +226.3% | 103936 | 173056 | +66.5% |
| Uncached Input Tokens | 11229 | 26884 | +139.4% | 11489 | 40179 | +249.7% |
| Cost | 0.0633 | 0.1872 | +195.6% | 0.0630 | 0.1608 | +155.1% |
| Cached vs Uncached Token Ratio | 9.35:1 | 12.74:1 | +36.3% | 9.05:1 | 4.31:1 | -52.4% |
| Oracle L1: Mutation Policy | PASS | PASS | — | PASS | PASS | — |
| Oracle L2: AST Invariants | PASS | PASS | — | PASS | PASS | — |
| Oracle L3: Clean Build | PASS | PASS | — | PASS | PASS | — |
| Oracle L4: Verification Test | PASS | PASS | — | PASS | PASS | — |
| Semantic Tool Invocation Verified (per run) | N/A | PASS | — | N/A | PASS | — |
Baseline vs Semedit Agentic in Small Context
- Vanilla (Baseline vs Semedit Agentic in Small Context) Edit: File config/config.go modified (15 lines -> 21 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Small Context) Edit: File config/config.go modified (15 lines -> 18 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ✓ ✓ | semedit/semantic_lookup (✓ ✓) |
| 3 | ✓ ✓ | semedit/semantic_inspect_symbol (✓ ✓) |
| 4 | ✓ ✓ | semedit/semantic_outline (✓ ✓) |
| 5 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 6 | — | ✓ ✓ |
| 7 | — | ✓ ✓ |
| 8 | — | semedit/semantic_batch (✓ ✓) |
| 9 | — | semedit/semantic_organize_imports (✓ ✓) |
| 10 | — | ✓ ✓ |
| 11 | — | ✓ ✓ |
Baseline vs Semedit Agentic in Large Context
- Vanilla (Baseline vs Semedit Agentic in Large Context) Edit: File config/config.go modified (15 lines -> 19 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Large Context) Edit: File config/config.go modified (15 lines -> 21 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ? ✗ |
| 2 | ✓ ✓ | ✓ ✓ |
| 3 | ✓ ✓ | ✓ ✓ |
| 4 | ✓ ✓ | ✓ ✓ |
| 5 | ✓ ✓ | semedit/semantic_rename (✓ ✓) |
| 6 | — | semedit/semantic_insert_function (✓ ✓) |
| 7 | — | semedit/semantic_organize_imports (✓ ✓) |
| 8 | — | ✓ ✓ |
| 9 | — | ✓ ✓ |
default prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-09-composite-refactor |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | default |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/bench/task_09_composite_refactor.txtar |
Vanilla LLM Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
In config/config.go: rename Config.Port to Config.ListenPort, insert helper function DefaultConfig() *Config returning default values, and clean up unused imports. Do not use semantic editing MCP tools; use standard file editing. When done, output DONE.
Semedit MCP Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
In config/config.go: rename Config.Port to Config.ListenPort, insert helper function DefaultConfig() *Config returning default values, and clean up unused imports. Use semedit semantic tools for supported source code modifications. Shell commands for builds and tests are allowed. When done, output DONE.
Initial Workspace State (Before Edit)
package config
import (
"archive/tar"
"fmt"
)
type Config struct {
Host string
Port int
}
func (c *Config) Addr() string {
return fmt.Sprintf("%s:%d", c.Host, c.Port)
}| Metric | Vanilla (Small) | MCP (Small) | Δ (Small) | Vanilla (Large) | MCP (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | 19.71s | 59.57s | +202.3% | 23.47s | 166.08s | +607.7% |
| Process Start → First Event | 0.12s | 0.13s | — | 0.13s | 0.14s | — |
| First Event → First Tool Call | 4.47s | 5.29s | — | 4.85s | 11.54s | — |
| MCP Initialize → First Semantic Call | — | — | — | — | — | — |
| MCP Server Start → Initialize | — | — | — | — | — | — |
| Top-Level User Turns | 1 | 1 | 0% | 1 | 1 | 0% |
| Internal Tool Cycles | 4 | 8 | +100.0% | 5 | 16 | +220.0% |
| Initial Load / Discovery Turns | 3 | 1 | -66.7% | 3 | 2 | -33.3% |
| MCP Discovery / Schema Turns | 0 | 0 | — | 0 | 0 | — |
| Tool Invocations (per run) | 4 | 8 | +100.0% | 5 | 16 | +220.0% |
| Output Tokens | 715 | 954 | +33.4% | 563 | 1940 | +244.6% |
| Reasoning / Thinking Tokens | 54 | 416 | +670.4% | 76 | 795 | +946.1% |
| Total Input Tokens | 96190 | 247823 | +157.6% | 116119 | 341233 | +193.9% |
| Cached Input Tokens | 85760 | 225792 | +163.3% | 104960 | 315392 | +200.5% |
| Uncached Input Tokens | 10430 | 22031 | +111.2% | 11159 | 25841 | +131.6% |
| Cost | 0.0571 | 0.1287 | +125.2% | 0.0621 | 0.1776 | +185.9% |
| Cached vs Uncached Token Ratio | 8.22:1 | 10.25:1 | +24.6% | 9.41:1 | 12.21:1 | +29.8% |
| Oracle L1: Mutation Policy | PASS | PASS | — | PASS | PASS | — |
| Oracle L2: AST Invariants | PASS | PASS | — | PASS | PASS | — |
| Oracle L3: Clean Build | PASS | PASS | — | PASS | PASS | — |
| Oracle L4: Verification Test | PASS | PASS | — | PASS | PASS | — |
| Semantic Tool Invocation Verified (per run) | N/A | PASS | — | N/A | PASS | — |
Baseline vs Semedit Agentic in Small Context
- Vanilla (Baseline vs Semedit Agentic in Small Context) Edit: File config/config.go modified (15 lines -> 19 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Small Context) Edit: File config/config.go modified (15 lines -> 18 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ✓ ✓ | semedit/semantic_inspect_symbol (✓ ✓) |
| 3 | ✓ ✓ | semedit/semantic_outline (✓ ✓) |
| 4 | ✓ ✓ | semedit/semantic_inspect_symbol (✓ ✓) |
| 5 | — | semedit/semantic_rename (✓ ✓) |
| 6 | — | semedit/semantic_insert_function (✓ ✓) |
| 7 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 8 | — | semedit/semantic_inspect_symbol (✓ ✓) |
Baseline vs Semedit Agentic in Large Context
- Vanilla (Baseline vs Semedit Agentic in Large Context) Edit: File config/config.go modified (15 lines -> 18 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Large Context) Edit: File config/config.go modified (15 lines -> 21 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ✓ ✓ | ✓ ✓ |
| 3 | ? ✗ | semedit/semantic_inspect_symbol (✓ ✓) |
| 4 | ✓ ✓ | semedit/semantic_outline (✓ ✓) |
| 5 | ✓ ✓ | semedit/semantic_lookup (✓ ✗): resolve Port in config/config.go: symbol not found |
| 6 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 7 | — | semedit/semantic_find_references (✓ ✗): load Go package example.com/test/config: example.com/test/config: /Users/alessandro/sources/semantic-editor/.scratch/benchmarks/run_codex_semedit_large-default_task-09-composite-refactor_1790468625696068000/config/config.go:4:2: "archive/tar" imported and not used |
| 8 | — | semedit/semantic_organize_imports (✓ ✓) |
| 9 | — | ? ✗ |
| 10 | — | ✓ ✓ |
| 11 | — | ✓ ✓ |
| 12 | — | semedit/semantic_rename (✓ ✓) |
| 13 | — | semedit/semantic_insert_function (✓ ✓) |
| 14 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 15 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 16 | — | ✓ ✓ |
default prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-09-composite-refactor |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | default |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/bench/task_09_composite_refactor.txtar |
Vanilla LLM Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
In config/config.go: rename Config.Port to Config.ListenPort, insert helper function DefaultConfig() *Config returning default values, and clean up unused imports. Do not use semantic editing MCP tools; use standard file editing. When done, output DONE.
Semedit MCP Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
In config/config.go: rename Config.Port to Config.ListenPort, insert helper function DefaultConfig() *Config returning default values, and clean up unused imports. Use semedit semantic tools for supported source code modifications. Shell commands for builds and tests are allowed. When done, output DONE.
Initial Workspace State (Before Edit)
package config
import (
"archive/tar"
"fmt"
)
type Config struct {
Host string
Port int
}
func (c *Config) Addr() string {
return fmt.Sprintf("%s:%d", c.Host, c.Port)
}| Metric | Vanilla (Small) | MCP (Small) | Δ (Small) | Vanilla (Large) | MCP (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | 22.21s | 39.77s | +79.0% | 23.77s | 81.85s | +244.4% |
| Process Start → First Event | 0.12s | 0.12s | — | 0.13s | 0.12s | — |
| First Event → First Tool Call | 4.72s | 5.19s | — | 5.61s | 5.55s | — |
| MCP Initialize → First Semantic Call | — | — | — | — | — | — |
| MCP Server Start → Initialize | — | — | — | — | — | — |
| Top-Level User Turns | 1 | 1 | 0% | 1 | 1 | 0% |
| Internal Tool Cycles | 4 | 8 | +100.0% | 5 | 15 | +200.0% |
| Initial Load / Discovery Turns | 3 | 2 | -33.3% | 2 | 2 | 0% |
| MCP Discovery / Schema Turns | 0 | 0 | — | 0 | 0 | — |
| Tool Invocations (per run) | 4 | 8 | +100.0% | 5 | 15 | +200.0% |
| Output Tokens | 757 | 847 | +11.9% | 825 | 1755 | +112.7% |
| Reasoning / Thinking Tokens | 101 | 0 | -100.0% | 180 | 496 | +175.6% |
| Total Input Tokens | 96876 | 202987 | +109.5% | 119238 | 399992 | +235.5% |
| Cached Input Tokens | 85760 | 185088 | +115.8% | 108032 | 365568 | +238.4% |
| Uncached Input Tokens | 11116 | 17899 | +61.0% | 11206 | 34424 | +207.2% |
| Cost | 0.0600 | 0.1016 | +69.5% | 0.0676 | 0.2056 | +204.2% |
| Cached vs Uncached Token Ratio | 7.72:1 | 10.34:1 | +34.0% | 9.64:1 | 10.62:1 | +10.2% |
| Oracle L1: Mutation Policy | PASS | PASS | — | PASS | PASS | — |
| Oracle L2: AST Invariants | PASS | PASS | — | PASS | PASS | — |
| Oracle L3: Clean Build | PASS | PASS | — | PASS | PASS | — |
| Oracle L4: Verification Test | PASS | PASS | — | PASS | PASS | — |
| Semantic Tool Invocation Verified (per run) | N/A | PASS | — | N/A | PASS | — |
Baseline vs Semedit Agentic in Small Context
- Vanilla (Baseline vs Semedit Agentic in Small Context) Edit: File config/config.go modified (15 lines -> 19 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Small Context) Edit: File config/config.go modified (15 lines -> 21 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ✓ ✓ | ✓ ✓ |
| 3 | ✓ ✓ | semedit/semantic_rename (✓ ✓) |
| 4 | ✓ ✓ | semedit/semantic_insert_function (✓ ✓) |
| 5 | — | ✓ ✓ |
| 6 | — | semedit/semantic_organize_imports (✓ ✓) |
| 7 | — | ✓ ✓ |
| 8 | — | ? ✗ |
Baseline vs Semedit Agentic in Large Context
- Vanilla (Baseline vs Semedit Agentic in Large Context) Edit: File config/config.go modified (15 lines -> 21 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Large Context) Edit: File config/config.go modified (15 lines -> 18 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ✓ ✓ | ✓ ✓ |
| 3 | ✓ ✓ | semedit/semantic_inspect_symbol (✓ ✓) |
| 4 | ✓ ✓ | semedit/semantic_lookup (✓ ✓) |
| 5 | ✓ ✓ | semedit/semantic_find_references (✓ ✗): load Go package example.com/test/config: example.com/test/config: /Users/alessandro/sources/semantic-editor/.scratch/benchmarks/run_codex_semedit_large-default_task-09-composite-refactor_1790468645814542000/config/config.go:4:2: "archive/tar" imported and not used |
| 6 | — | semedit/semantic_find_references (✓ ✗): load Go package example.com/test/config: example.com/test/config: /Users/alessandro/sources/semantic-editor/.scratch/benchmarks/run_codex_semedit_large-default_task-09-composite-refactor_1790468645814542000/config/config.go:4:2: "archive/tar" imported and not used |
| 7 | — | semedit/semantic_organize_imports (✓ ✓) |
| 8 | — | semedit/semantic_find_references (✓ ✗): load Go package example.com/test/config: example.com/test/config: /Users/alessandro/sources/semantic-editor/.scratch/benchmarks/run_codex_semedit_large-default_task-09-composite-refactor_1790468645814542000/config/config_test.go:6:7: undefined: DefaultConfig |
| 9 | — | ✓ ✓ |
| 10 | — | semedit/semantic_insert_function (✓ ✓) |
| 11 | — | semedit/semantic_rename (✓ ✗): resolve Port in config/config.go: symbol not found |
| 12 | — | ? ✗ |
| 13 | — | ✓ ✓ |
| 14 | — | ✓ ✓ |
| 15 | — | semedit/semantic_verify (✓ ✓) |
default prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-09-composite-refactor |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | default |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/bench/task_09_composite_refactor.txtar |
Vanilla LLM Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
In config/config.go: rename Config.Port to Config.ListenPort, insert helper function DefaultConfig() *Config returning default values, and clean up unused imports. Do not use semantic editing MCP tools; use standard file editing. When done, output DONE.
Semedit MCP Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
In config/config.go: rename Config.Port to Config.ListenPort, insert helper function DefaultConfig() *Config returning default values, and clean up unused imports. Use semedit semantic tools for supported source code modifications. Shell commands for builds and tests are allowed. When done, output DONE.
Initial Workspace State (Before Edit)
package config
import (
"archive/tar"
"fmt"
)
type Config struct {
Host string
Port int
}
func (c *Config) Addr() string {
return fmt.Sprintf("%s:%d", c.Host, c.Port)
}| Metric | Vanilla (Small) | MCP (Small) | Δ (Small) | Vanilla (Large) | MCP (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | 19.73s | 50.13s | +154.0% | 29.20s | 66.95s | +129.2% |
| Process Start → First Event | 0.13s | 0.15s | — | 0.13s | 0.12s | — |
| First Event → First Tool Call | 4.72s | 9.70s | — | 4.50s | 6.10s | — |
| MCP Initialize → First Semantic Call | — | — | — | — | — | — |
| MCP Server Start → Initialize | — | — | — | — | — | — |
| Top-Level User Turns | 1 | 1 | 0% | 1 | 1 | 0% |
| Internal Tool Cycles | 4 | 11 | +175.0% | 6 | 13 | +116.7% |
| Initial Load / Discovery Turns | 2 | 0 | -100.0% | 6 | 0 | -100.0% |
| MCP Discovery / Schema Turns | 0 | 0 | — | 0 | 0 | — |
| Tool Invocations (per run) | 4 | 11 | +175.0% | 6 | 13 | +116.7% |
| Output Tokens | 591 | 828 | +40.1% | 1020 | 1209 | +18.5% |
| Reasoning / Thinking Tokens | 106 | 151 | +42.5% | 188 | 427 | +127.1% |
| Total Input Tokens | 96837 | 206181 | +112.9% | 158487 | 331443 | +109.1% |
| Cached Input Tokens | 85760 | 173056 | +101.8% | 145408 | 310784 | +113.7% |
| Uncached Input Tokens | 11077 | 33125 | +199.0% | 13079 | 20659 | +58.0% |
| Cost | 0.0578 | 0.1383 | +139.1% | 0.0841 | 0.1498 | +78.0% |
| Cached vs Uncached Token Ratio | 7.74:1 | 5.22:1 | -32.5% | 11.12:1 | 15.04:1 | +35.3% |
| Oracle L1: Mutation Policy | PASS | PASS | — | PASS | PASS | — |
| Oracle L2: AST Invariants | PASS | PASS | — | PASS | PASS | — |
| Oracle L3: Clean Build | PASS | PASS | — | PASS | PASS | — |
| Oracle L4: Verification Test | PASS | PASS | — | PASS | PASS | — |
| Semantic Tool Invocation Verified (per run) | N/A | PASS | — | N/A | PASS | — |
Baseline vs Semedit Agentic in Small Context
- Vanilla (Baseline vs Semedit Agentic in Small Context) Edit: File config/config.go modified (15 lines -> 19 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Small Context) Edit: File config/config.go modified (15 lines -> 21 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | semedit/semantic_inspect_symbol (✓ ✓) |
| 2 | ✓ ✓ | ✓ ✓ |
| 3 | ✓ ✓ | semedit/semantic_lookup (✓ ✗): resolve Port in config/config.go: symbol not found |
| 4 | ✓ ✓ | ✓ ✓ |
| 5 | — | semedit/semantic_inspect_symbol (✓ ✗): resolve Validate in /Users/alessandro/sources/semantic-editor/.scratch/benchmarks/run_codex_semedit_small-default_task-09-composite-refactor_1790468479102313000/config/config.go: symbol not found |
| 6 | — | semedit/semantic_inspect_symbol (✓ ✗): resolve LoadConfig in /Users/alessandro/sources/semantic-editor/.scratch/benchmarks/run_codex_semedit_small-default_task-09-composite-refactor_1790468479102313000/config/config.go: symbol not found |
| 7 | — | semedit/semantic_batch (✓ ✓) |
| 8 | — | semedit/semantic_organize_imports (✓ ✓) |
| 9 | — | ✓ ✓ |
| 10 | — | ✓ ✓ |
| 11 | — | ✓ ✓ |
Baseline vs Semedit Agentic in Large Context
- Vanilla (Baseline vs Semedit Agentic in Large Context) Edit: File config/config.go modified (15 lines -> 21 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Large Context) Edit: File config/config.go modified (15 lines -> 21 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | semedit/semantic_lookup (✓ ✓) |
| 2 | ✓ ✓ | semedit/semantic_inspect_symbol (✓ ✓) |
| 3 | ✓ ✓ | ✓ ✓ |
| 4 | ✓ ✓ | ✓ ✓ |
| 5 | ✓ ✓ | semedit/semantic_inspect_symbol (✓ ✗): resolve Address in /Users/alessandro/sources/semantic-editor/.scratch/benchmarks/run_codex_semedit_large-default_task-09-composite-refactor_1790468663117392000/config/config.go: symbol not found |
| 6 | ✓ ✓ | semedit/semantic_lookup (✓ ✗): resolve Config.Address in config/config.go: symbol not found |
| 7 | — | ? ✗ |
| 8 | — | ✓ ✓ |
| 9 | — | ✓ ✓ |
| 10 | — | semedit/semantic_rename (✓ ✓) |
| 11 | — | semedit/semantic_insert_function (✓ ✓) |
| 12 | — | ✓ ✓ |
| 13 | — | ✓ ✓ |
Test case: task-11-mixed-sink-api-migration
Target: codex/gpt-6-luna/medium
default prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-11-mixed-sink-api-migration |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | default |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/bench/task_11_mixed_sink_api_migration.txtar |
Vanilla LLM Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
Audit event kinds may contain inconsistent whitespace and casing. Normalize them at the delivery boundary before persistence, and treat probe events as control traffic. Do not change the metrics or legacy export protocols. Preserve the module metadata and tests, and leave the workspace verified. Do not use semantic editing MCP tools; use standard file editing. When done, output DONE.
Semedit MCP Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
Audit event kinds may contain inconsistent whitespace and casing. Normalize them at the delivery boundary before persistence, and treat probe events as control traffic. Do not change the metrics or legacy export protocols. Preserve the module metadata and tests, and leave the workspace verified. Use semedit semantic tools for supported source code modifications. Shell commands for builds and tests are allowed. When done, output DONE.
Initial Workspace State (Before Edit)
// Package audit leaves room for normalization rules owned by the delivery boundary.
package audit| Metric | Vanilla (Small) | MCP (Small) | Δ (Small) | Vanilla (Large) | MCP (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | 86.85s | 212.85s | +145.1% | 131.06s | 127.17s | -3.0% |
| Process Start → First Event | 0.12s | 0.12s | — | 0.13s | 0.13s | — |
| First Event → First Tool Call | 4.74s | 4.80s | — | 4.24s | 4.30s | — |
| MCP Initialize → First Semantic Call | — | — | — | — | — | — |
| MCP Server Start → Initialize | — | — | — | — | — | — |
| Top-Level User Turns | 2 | 5 | +150.0% | 5 | 4 | -20.0% |
| Internal Tool Cycles | 11 | 41 | +272.7% | 14 | 23 | +64.3% |
| Initial Load / Discovery Turns | 1 | 6 | +500.0% | 2 | 3 | +50.0% |
| MCP Discovery / Schema Turns | 0 | 0 | — | 0 | 0 | — |
| Tool Invocations (per run) | 11 | 41 | +272.7% | 14 | 23 | +64.3% |
| Output Tokens | 4230 | 16354 | +286.6% | 17728 | 10379 | -41.5% |
| Reasoning / Thinking Tokens | 1307 | 7451 | +470.1% | 6355 | 4310 | -32.2% |
| Total Input Tokens | 429052 | 3859040 | +799.4% | 1568928 | 2228118 | +42.0% |
| Cached Input Tokens | 380928 | 3658496 | +860.4% | 1448704 | 2072320 | +43.0% |
| Uncached Input Tokens | 48124 | 200544 | +316.7% | 120224 | 155798 | +29.6% |
| Cost | 0.2848 | 1.7135 | +501.8% | 0.9638 | 1.0912 | +13.2% |
| Cached vs Uncached Token Ratio | 7.92:1 | 18.24:1 | +130.5% | 12.05:1 | 13.30:1 | +10.4% |
| Oracle L1: Mutation Policy | PASS | PASS | — | PASS | PASS | — |
| Oracle L2: AST Invariants | PASS | PASS | — | PASS | PASS | — |
| Oracle L3: Clean Build | PASS | PASS | — | PASS | PASS | — |
| Oracle L4: Verification Test | PASS | FAIL | — | PASS | PASS | — |
| Semantic Tool Invocation Verified (per run) | N/A | PASS | — | N/A | PASS | — |
Baseline vs Semedit Agentic in Small Context
- Vanilla (Baseline vs Semedit Agentic in Small Context) Edit: File audit/normalize.go modified (2 lines -> 8 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Small Context) Edit: File audit/normalize.go modified (2 lines -> 8 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ✓ ✓ | ? ✗ |
| 3 | ✓ ✓ | ✓ ✓ |
| 4 | ✓ ✓ | semedit/semantic_inspect_symbol (✓ ✓) |
| 5 | ✓ ✓ | semedit/semantic_inspect_symbol (✓ ✓) |
| 6 | ✓ ✓ | semedit/semantic_inspect_symbol (✓ ✓) |
| 7 | ✓ ✓ | semedit/semantic_inspect_symbol (✓ ✓) |
| 8 | ✓ ✓ | ✓ ✓ |
| 9 | ✓ ✓ | semedit/semantic_inspect_symbol (✓ ✗): resolve NormalizeKind: symbol not found |
| 10 | ✓ ✓ | semedit/semantic_insert_function (✓ ✓) |
| 11 | ✓ ✓ | semedit/semantic_replace_body (✓ ✓) |
| 12 | — | semedit/semantic_replace_body (✓ ✓) |
| 13 | — | ? ✗ |
| 14 | — | semedit/semantic_verify (✓ ✓) |
| 15 | — | ✓ ✓ |
| 16 | — | ✓ ✓ |
| 17 | — | ✓ ✓ |
| 18 | — | ✓ ✓ |
| 19 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 20 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 21 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 22 | — | ✓ ✓ |
| 23 | — | semedit/semantic_verify (✓ ✓) |
| 24 | — | ✓ ✓ |
| 25 | — | ✓ ✓ |
| 26 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 27 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 28 | — | semedit/semantic_replace_body (✓ ✓) |
| 29 | — | ✓ ✓ |
| 30 | — | semedit/semantic_verify (✓ ✓) |
| 31 | — | ✓ ✓ |
| 32 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 33 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 34 | — | ✓ ✓ |
| 35 | — | semedit/semantic_verify (✗ ?): This action was rejected due to unacceptable risk. Reason: semantic_verify on the entire Go workspace may format and write files, potentially modifying tests despite the user's explicit prohibition. Do not bypass this rejection through a workaround or indirect execution. Continue with a safer alternative, or carry out checks to prove that the action is authorized or low risk before trying again. Complete unaffected work without asking for confirmation. Report anything that remains blocked, clarify why it wa |
| 36 | — | semedit/semantic_verify (✓ ✓) |
| 37 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 38 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 39 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 40 | — | ✓ ✓ |
| 41 | — | ✓ ✓ |
Baseline vs Semedit Agentic in Large Context
- Vanilla (Baseline vs Semedit Agentic in Large Context) Edit: File audit/normalize.go modified (2 lines -> 10 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Large Context) Edit: File audit/normalize.go modified (2 lines -> 8 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ✓ ✓ | ? ✗ |
| 3 | ✓ ✓ | ✓ ✓ |
| 4 | ✓ ✓ | ✓ ✓ |
| 5 | ✓ ✓ | ✓ ✓ |
| 6 | ✓ ✓ | semedit/semantic_outline (✓ ✓) |
| 7 | ✓ ✓ | semedit/semantic_insert_function (✓ ✓) |
| 8 | ✓ ✓ | semedit/semantic_replace_body (✓ ✓) |
| 9 | ✓ ✓ | semedit/semantic_replace_body (✓ ✓) |
| 10 | ✓ ✓ | ✓ ✓ |
| 11 | ✓ ✓ | ✓ ✓ |
| 12 | ✓ ✓ | ✓ ✓ |
| 13 | ✓ ✓ | ✓ ✓ |
| 14 | ✓ ✓ | semedit/semantic_verify (✓ ✓) |
| 15 | — | ✓ ✓ |
| 16 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 17 | — | ✓ ✓ |
| 18 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 19 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 20 | — | ✓ ✓ |
| 21 | — | semedit/semantic_replace_body (✓ ✓) |
| 22 | — | semedit/semantic_replace_body (✓ ✓) |
| 23 | — | ✓ ✓ |
default prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-11-mixed-sink-api-migration |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | default |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/bench/task_11_mixed_sink_api_migration.txtar |
Vanilla LLM Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
Audit event kinds may contain inconsistent whitespace and casing. Normalize them at the delivery boundary before persistence, and treat probe events as control traffic. Do not change the metrics or legacy export protocols. Preserve the module metadata and tests, and leave the workspace verified. Do not use semantic editing MCP tools; use standard file editing. When done, output DONE.
Semedit MCP Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
Audit event kinds may contain inconsistent whitespace and casing. Normalize them at the delivery boundary before persistence, and treat probe events as control traffic. Do not change the metrics or legacy export protocols. Preserve the module metadata and tests, and leave the workspace verified. Use semedit semantic tools for supported source code modifications. Shell commands for builds and tests are allowed. When done, output DONE.
Initial Workspace State (Before Edit)
// Package audit leaves room for normalization rules owned by the delivery boundary.
package audit| Metric | Vanilla (Small) | MCP (Small) | Δ (Small) | Vanilla (Large) | MCP (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | 65.35s | 96.06s | +47.0% | 94.30s | 218.80s | +132.0% |
| Process Start → First Event | 0.12s | 0.12s | — | 0.13s | 0.15s | — |
| First Event → First Tool Call | 3.97s | 4.54s | — | 5.94s | 4.77s | — |
| MCP Initialize → First Semantic Call | — | — | — | — | — | — |
| MCP Server Start → Initialize | — | — | — | — | — | — |
| Top-Level User Turns | 2 | 1 | -50.0% | 5 | 5 | 0% |
| Internal Tool Cycles | 10 | 19 | +90.0% | 11 | 34 | +209.1% |
| Initial Load / Discovery Turns | 1 | 3 | +200.0% | 3 | 3 | 0% |
| MCP Discovery / Schema Turns | 0 | 0 | — | 0 | 0 | — |
| Tool Invocations (per run) | 10 | 19 | +90.0% | 11 | 34 | +209.1% |
| Output Tokens | 3751 | 2482 | -33.8% | 12307 | 18790 | +52.7% |
| Reasoning / Thinking Tokens | 1600 | 1263 | -21.1% | 4854 | 8658 | +78.4% |
| Total Input Tokens | 425204 | 403593 | -5.1% | 1540327 | 3033229 | +96.9% |
| Cached Input Tokens | 392192 | 367616 | -6.3% | 1395712 | 2842368 | +103.7% |
| Uncached Input Tokens | 33012 | 35977 | +9.0% | 144615 | 190861 | +32.0% |
| Cost | 0.2475 | 0.2287 | -7.6% | 0.9250 | 1.5308 | +65.5% |
| Cached vs Uncached Token Ratio | 11.88:1 | 10.22:1 | -14.0% | 9.65:1 | 14.89:1 | +54.3% |
| Oracle L1: Mutation Policy | PASS | PASS | — | PASS | PASS | — |
| Oracle L2: AST Invariants | PASS | PASS | — | PASS | PASS | — |
| Oracle L3: Clean Build | PASS | PASS | — | PASS | PASS | — |
| Oracle L4: Verification Test | PASS | PASS | — | PASS | FAIL | — |
| Semantic Tool Invocation Verified (per run) | N/A | PASS | — | N/A | PASS | — |
Baseline vs Semedit Agentic in Small Context
- Vanilla (Baseline vs Semedit Agentic in Small Context) Edit: File audit/normalize.go modified (2 lines -> 8 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Small Context) Edit: File audit/normalize.go modified (2 lines -> 8 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ✓ ✓ | ✓ ✓ |
| 3 | ✓ ✓ | ✓ ✓ |
| 4 | ✓ ✓ | semedit/semantic_outline (✓ ✓) |
| 5 | ✓ ✓ | semedit/semantic_outline (✓ ✓) |
| 6 | ✓ ✓ | semedit/semantic_inspect_symbol (✓ ✓) |
| 7 | ✓ ✓ | semedit/semantic_inspect_symbol (✓ ✓) |
| 8 | ✓ ✓ | semedit/semantic_inspect_symbol (✓ ✓) |
| 9 | ✓ ✓ | ✓ ✓ |
| 10 | ✓ ✓ | semedit/semantic_inspect_symbol (✓ ✓) |
| 11 | — | ✓ ✓ |
| 12 | — | semedit/semantic_batch (✓ ✓) |
| 13 | — | semedit/semantic_organize_imports (✓ ✓) |
| 14 | — | semedit/semantic_verify (✓ ✓) |
| 15 | — | semedit/semantic_verify (✓ ✓) |
| 16 | — | semedit/semantic_verify (✓ ✓) |
| 17 | — | ✓ ✓ |
| 18 | — | ✓ ✓ |
| 19 | — | ✓ ✓ |
Baseline vs Semedit Agentic in Large Context
- Vanilla (Baseline vs Semedit Agentic in Large Context) Edit: File audit/normalize.go modified (2 lines -> 17 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Large Context) Edit: File audit/normalize.go modified (2 lines -> 9 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ✓ ✓ | ✓ ✓ |
| 3 | ✓ ✓ | ✓ ✓ |
| 4 | ✓ ✓ | semedit/semantic_insert_function (✓ ✓) |
| 5 | ✓ ✓ | ✓ ✓ |
| 6 | ✓ ✓ | semedit/semantic_replace_body (✓ ✓) |
| 7 | ✓ ✓ | semedit/semantic_replace_body (✓ ✓) |
| 8 | ✓ ✓ | semedit/semantic_replace_body (✓ ✓) |
| 9 | ✓ ✓ | semedit/semantic_replace_body (✓ ✓) |
| 10 | ✓ ✓ | semedit/semantic_insert_function (✓ ✓) |
| 11 | ✓ ✓ | semedit/semantic_replace_body (✓ ✓) |
| 12 | — | semedit/semantic_replace_body (✓ ✓) |
| 13 | — | semedit/semantic_replace_body (✓ ✓) |
| 14 | — | ✓ ✓ |
| 15 | — | ✓ ✓ |
| 16 | — | semedit/semantic_verify (✓ ✓) |
| 17 | — | ✓ ✓ |
| 18 | — | ✓ ✓ |
| 19 | — | ✓ ✓ |
| 20 | — | semedit/semantic_verify (✓ ✓) |
| 21 | — | ✓ ✓ |
| 22 | — | semedit/semantic_replace_body (✓ ✓) |
| 23 | — | ? ✗ |
| 24 | — | ✓ ✓ |
| 25 | — | semedit/semantic_verify (✓ ✓) |
| 26 | — | ✓ ✓ |
| 27 | — | ✓ ✓ |
| 28 | — | semedit/semantic_verify (✓ ✓) |
| 29 | — | semedit/semantic_insert_function (✓ ✓) |
| 30 | — | semedit/semantic_replace_body (✓ ✓) |
| 31 | — | semedit/semantic_replace_body (✓ ✓) |
| 32 | — | semedit/semantic_organize_imports (✓ ✓) |
| 33 | — | semedit/semantic_replace_body (✓ ✓) |
| 34 | — | ✓ ✓ |
default prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-11-mixed-sink-api-migration |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | default |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/bench/task_11_mixed_sink_api_migration.txtar |
Vanilla LLM Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
Audit event kinds may contain inconsistent whitespace and casing. Normalize them at the delivery boundary before persistence, and treat probe events as control traffic. Do not change the metrics or legacy export protocols. Preserve the module metadata and tests, and leave the workspace verified. Do not use semantic editing MCP tools; use standard file editing. When done, output DONE.
Semedit MCP Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
Audit event kinds may contain inconsistent whitespace and casing. Normalize them at the delivery boundary before persistence, and treat probe events as control traffic. Do not change the metrics or legacy export protocols. Preserve the module metadata and tests, and leave the workspace verified. Use semedit semantic tools for supported source code modifications. Shell commands for builds and tests are allowed. When done, output DONE.
Initial Workspace State (Before Edit)
// Package audit leaves room for normalization rules owned by the delivery boundary.
package audit| Metric | Vanilla (Small) | MCP (Small) | Δ (Small) | Vanilla (Large) | MCP (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | 109.72s | 223.14s | +103.4% | 162.89s | — | — |
| Process Start → First Event | 0.12s | 0.13s | — | 0.13s | — | — |
| First Event → First Tool Call | 5.32s | 4.67s | — | 4.39s | — | — |
| MCP Initialize → First Semantic Call | — | — | — | — | — | — |
| MCP Server Start → Initialize | — | — | — | — | — | — |
| Top-Level User Turns | 5 | 5 | 0% | 5 | — | — |
| Internal Tool Cycles | 12 | 33 | +175.0% | 14 | — | — |
| Initial Load / Discovery Turns | 2 | 3 | +50.0% | 3 | — | — |
| MCP Discovery / Schema Turns | 0 | 0 | — | 0 | — | — |
| Tool Invocations (per run) | 12 | 33 | +175.0% | 14 | — | — |
| Output Tokens | 13066 | 17274 | +32.2% | 23044 | — | — |
| Reasoning / Thinking Tokens | 4300 | 9107 | +111.8% | 11491 | — | — |
| Total Input Tokens | 1538763 | 4077498 | +165.0% | 2488465 | — | — |
| Cached Input Tokens | 1439232 | 3755008 | +160.9% | 2309888 | — | — |
| Uncached Input Tokens | 99531 | 322490 | +224.0% | 178577 | — | — |
| Cost | 0.8257 | 2.0747 | +151.3% | 1.4556 | — | — |
| Cached vs Uncached Token Ratio | 14.46:1 | 11.64:1 | -19.5% | 12.93:1 | — | — |
| Oracle L1: Mutation Policy | PASS | PASS | — | PASS | — | — |
| Oracle L2: AST Invariants | PASS | PASS | — | PASS | — | — |
| Oracle L3: Clean Build | PASS | PASS | — | PASS | — | — |
| Oracle L4: Verification Test | FAIL | FAIL | — | FAIL | — | — |
| Semantic Tool Invocation Verified (per run) | N/A | PASS | — | N/A | — | — |
Baseline vs Semedit Agentic in Small Context
- Vanilla (Baseline vs Semedit Agentic in Small Context) Edit: File audit/normalize.go modified (2 lines -> 8 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Small Context) Edit: File audit/normalize.go modified (2 lines -> 8 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ✓ ✓ | ✓ ✓ |
| 3 | ✓ ✓ | ✓ ✓ |
| 4 | ✓ ✓ | semedit/semantic_outline (✓ ✓) |
| 5 | ✓ ✓ | semedit/semantic_outline (✓ ✓) |
| 6 | ✓ ✓ | semedit/semantic_inspect_symbol (✓ ✓) |
| 7 | ✓ ✓ | semedit/semantic_inspect_symbol (✓ ✓) |
| 8 | ✓ ✓ | semedit/semantic_inspect_symbol (✓ ✓) |
| 9 | ✓ ✓ | semedit/semantic_inspect_symbol (✓ ✗): resolve Normalize in /Users/alessandro/sources/semantic-editor/.scratch/benchmarks/run_codex_semedit_small-default_task-11-mixed-sink-api-migration_1790469189984016000/audit/normalize.go: symbol not found |
| 10 | ✓ ✓ | ✓ ✓ |
| 11 | ✓ ✓ | semedit/semantic_insert_function (✓ ✓) |
| 12 | ✓ ✓ | semedit/semantic_replace_body (✓ ✓) |
| 13 | — | semedit/semantic_verify (✓ ✓) |
| 14 | — | ✓ ✓ |
| 15 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 16 | — | ✓ ✓ |
| 17 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 18 | — | ✓ ✓ |
| 19 | — | semedit/semantic_replace_body (✓ ✓) |
| 20 | — | semedit/semantic_replace_body (✓ ✓) |
| 21 | — | ✓ ✓ |
| 22 | — | semedit/semantic_verify (✓ ✓) |
| 23 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 24 | — | semedit/semantic_replace_body (✓ ✓) |
| 25 | — | ✓ ✓ |
| 26 | — | semedit/semantic_verify (✓ ✓) |
| 27 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 28 | — | ✓ ✓ |
| 29 | — | semedit/semantic_verify (✓ ✓) |
| 30 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 31 | — | semedit/semantic_replace_body (✓ ✓) |
| 32 | — | ✓ ✓ |
| 33 | — | semedit/semantic_verify (✓ ✓) |
Baseline vs Semedit Agentic in Large Context
- Vanilla (Baseline vs Semedit Agentic in Large Context) Edit: File audit/normalize.go modified (2 lines -> 27 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | not published |
| 2 | ✓ ✓ | not published |
| 3 | ✓ ✓ | not published |
| 4 | ✓ ✓ | not published |
| 5 | ✓ ✓ | not published |
| 6 | ✓ ✓ | not published |
| 7 | ✓ ✓ | not published |
| 8 | ✓ ✓ | not published |
| 9 | ✓ ✓ | not published |
| 10 | ✓ ✓ | not published |
| 11 | ✓ ✓ | not published |
| 12 | ✓ ✓ | not published |
| 13 | ✓ ✓ | not published |
| 14 | ✓ ✓ | not published |
default prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-11-mixed-sink-api-migration |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | default |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/bench/task_11_mixed_sink_api_migration.txtar |
Vanilla LLM Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
Audit event kinds may contain inconsistent whitespace and casing. Normalize them at the delivery boundary before persistence, and treat probe events as control traffic. Do not change the metrics or legacy export protocols. Preserve the module metadata and tests, and leave the workspace verified. Do not use semantic editing MCP tools; use standard file editing. When done, output DONE.
Semedit MCP Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
Audit event kinds may contain inconsistent whitespace and casing. Normalize them at the delivery boundary before persistence, and treat probe events as control traffic. Do not change the metrics or legacy export protocols. Preserve the module metadata and tests, and leave the workspace verified. Use semedit semantic tools for supported source code modifications. Shell commands for builds and tests are allowed. When done, output DONE.
Initial Workspace State (Before Edit)
// Package audit leaves room for normalization rules owned by the delivery boundary.
package audit| Metric | Vanilla (Small) | MCP (Small) | Δ (Small) | Vanilla (Large) | MCP (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | 123.47s | 68.84s | -44.2% | 133.48s | 293.13s | +119.6% |
| Process Start → First Event | 0.13s | 0.12s | — | 0.13s | 0.12s | — |
| First Event → First Tool Call | 4.61s | 5.60s | — | 5.77s | 4.29s | — |
| MCP Initialize → First Semantic Call | — | — | — | — | — | — |
| MCP Server Start → Initialize | — | — | — | — | — | — |
| Top-Level User Turns | 5 | 1 | -80.0% | 5 | 5 | 0% |
| Internal Tool Cycles | 13 | 13 | 0% | 15 | 65 | +333.3% |
| Initial Load / Discovery Turns | 1 | 2 | +100.0% | 4 | 8 | +100.0% |
| MCP Discovery / Schema Turns | 0 | 0 | — | 0 | 0 | — |
| Tool Invocations (per run) | 13 | 13 | 0% | 15 | 65 | +333.3% |
| Output Tokens | 13565 | 1521 | -88.8% | 16556 | 22175 | +33.9% |
| Reasoning / Thinking Tokens | 5214 | 495 | -90.5% | 6320 | 9935 | +57.2% |
| Total Input Tokens | 1987642 | 279904 | -85.9% | 1753344 | 4400148 | +151.0% |
| Cached Input Tokens | 1858560 | 248576 | -86.6% | 1630720 | 4150016 | +154.5% |
| Uncached Input Tokens | 129082 | 31328 | -75.7% | 122624 | 250132 | +104.0% |
| Cost | 1.0221 | 0.1657 | -83.8% | 1.0002 | 2.0642 | +106.4% |
| Cached vs Uncached Token Ratio | 14.40:1 | 7.93:1 | -44.9% | 13.30:1 | 16.59:1 | +24.8% |
| Oracle L1: Mutation Policy | PASS | PASS | — | PASS | PASS | — |
| Oracle L2: AST Invariants | PASS | PASS | — | PASS | PASS | — |
| Oracle L3: Clean Build | PASS | PASS | — | PASS | PASS | — |
| Oracle L4: Verification Test | PASS | PASS | — | FAIL | FAIL | — |
| Semantic Tool Invocation Verified (per run) | N/A | PASS | — | N/A | PASS | — |
Baseline vs Semedit Agentic in Small Context
- Vanilla (Baseline vs Semedit Agentic in Small Context) Edit: File audit/normalize.go modified (2 lines -> 14 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Small Context) Edit: File audit/normalize.go modified (2 lines -> 8 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ✓ ✓ | ✓ ✓ |
| 3 | ✓ ✓ | ✓ ✓ |
| 4 | ✓ ✓ | ✓ ✓ |
| 5 | ✓ ✓ | semedit/semantic_outline (✓ ✓) |
| 6 | ? ✗ | semedit/semantic_inspect_symbol (✓ ✓) |
| 7 | ✓ ✓ | semedit/semantic_inspect_symbol (✓ ✓) |
| 8 | ✓ ✓ | semedit/semantic_insert_function (✓ ✓) |
| 9 | ? ✗ | semedit/semantic_batch (✓ ✓) |
| 10 | ✓ ✓ | ? ✗ |
| 11 | ✓ ✓ | ✓ ✓ |
| 12 | ✓ ✓ | semedit/semantic_verify (✓ ✓) |
| 13 | ✓ ✓ | ✓ ✓ |
Baseline vs Semedit Agentic in Large Context
- Vanilla (Baseline vs Semedit Agentic in Large Context) Edit: File audit/normalize.go modified (2 lines -> 15 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Large Context) Edit: File audit/normalize.go modified (2 lines -> 17 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ✓ ✓ | ✓ ✓ |
| 3 | ✓ ✓ | ✓ ✓ |
| 4 | ✓ ✓ | ✓ ✓ |
| 5 | ✓ ✓ | ✓ ✓ |
| 6 | ✓ ✓ | semedit/semantic_outline (✓ ✓) |
| 7 | ✓ ✓ | semedit/semantic_lookup (✓ ✓) |
| 8 | ✓ ✓ | semedit/semantic_lookup (✓ ✓) |
| 9 | ✓ ✓ | semedit/semantic_inspect_symbol (✓ ✓) |
| 10 | ✓ ✓ | semedit/semantic_inspect_symbol (✓ ✗): resolve FanoutSink.Write in /Users/alessandro/sources/semantic-editor/.scratch/benchmarks/run_codex_semedit_large-default_task-11-mixed-sink-api-migration_1790470384830318000/audit/sink.go: symbol not found |
| 11 | ✓ ✓ | semedit/semantic_inspect_symbol (✓ ✗): resolve BufferedSink.Begin in /Users/alessandro/sources/semantic-editor/.scratch/benchmarks/run_codex_semedit_large-default_task-11-mixed-sink-api-migration_1790470384830318000/audit/sink.go: symbol not found |
| 12 | ✓ ✓ | semedit/semantic_inspect_symbol (✓ ✗): resolve BufferedSink.Commit in /Users/alessandro/sources/semantic-editor/.scratch/benchmarks/run_codex_semedit_large-default_task-11-mixed-sink-api-migration_1790470384830318000/audit/sink.go: symbol not found |
| 13 | ✓ ✓ | semedit/semantic_inspect_symbol (✓ ✓) |
| 14 | ✓ ✓ | semedit/semantic_inspect_symbol (✓ ✓) |
| 15 | ✓ ✓ | semedit/semantic_inspect_symbol (✓ ✗): resolve DispatchAll in /Users/alessandro/sources/semantic-editor/.scratch/benchmarks/run_codex_semedit_large-default_task-11-mixed-sink-api-migration_1790470384830318000/audit/sink.go: symbol not found |
| 16 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 17 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 18 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 19 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 20 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 21 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 22 | — | semedit/semantic_insert_function (✓ ✓) |
| 23 | — | semedit/semantic_insert_function (✓ ✓) |
| 24 | — | semedit/semantic_replace_body (✓ ✓) |
| 25 | — | semedit/semantic_replace_body (✓ ✓) |
| 26 | — | semedit/semantic_replace_body (✓ ✓) |
| 27 | — | semedit/semantic_replace_body (✓ ✓) |
| 28 | — | semedit/semantic_replace_body (✓ ✓) |
| 29 | — | semedit/semantic_replace_body (✓ ✓) |
| 30 | — | semedit/semantic_verify (✓ ✓) |
| 31 | — | ✓ ✓ |
| 32 | — | ✓ ✓ |
| 33 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 34 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 35 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 36 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 37 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 38 | — | ✓ ✓ |
| 39 | — | ✓ ✓ |
| 40 | — | semedit/semantic_outline (✓ ✓) |
| 41 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 42 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 43 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 44 | — | ✓ ✓ |
| 45 | — | semedit/semantic_verify (✓ ✓) |
| 46 | — | ✓ ✓ |
| 47 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 48 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 49 | — | semedit/semantic_replace_body (✓ ✓) |
| 50 | — | semedit/semantic_verify (✓ ✓) |
| 51 | — | ✓ ✓ |
| 52 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 53 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 54 | — | semedit/semantic_verify (✓ ✓) |
| 55 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 56 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 57 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 58 | — | ✓ ✓ |
| 59 | — | semedit/semantic_insert_function (✓ ✓) |
| 60 | — | semedit/semantic_replace_body (✓ ✓) |
| 61 | — | semedit/semantic_replace_body (✓ ✓) |
| 62 | — | semedit/semantic_replace_body (✓ ✓) |
| 63 | — | ✓ ✓ |
| 64 | — | ✓ ✓ |
| 65 | — | semedit/semantic_verify (✓ ✓) |
default prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-11-mixed-sink-api-migration |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | default |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/bench/task_11_mixed_sink_api_migration.txtar |
Vanilla LLM Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
Audit event kinds may contain inconsistent whitespace and casing. Normalize them at the delivery boundary before persistence, and treat probe events as control traffic. Do not change the metrics or legacy export protocols. Preserve the module metadata and tests, and leave the workspace verified. Do not use semantic editing MCP tools; use standard file editing. When done, output DONE.
Semedit MCP Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
Audit event kinds may contain inconsistent whitespace and casing. Normalize them at the delivery boundary before persistence, and treat probe events as control traffic. Do not change the metrics or legacy export protocols. Preserve the module metadata and tests, and leave the workspace verified. Use semedit semantic tools for supported source code modifications. Shell commands for builds and tests are allowed. When done, output DONE.
Initial Workspace State (Before Edit)
// Package audit leaves room for normalization rules owned by the delivery boundary.
package audit| Metric | Vanilla (Small) | MCP (Small) | Δ (Small) | Vanilla (Large) | MCP (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | 137.52s | 191.47s | +39.2% | 53.78s | 265.34s | +393.4% |
| Process Start → First Event | 0.12s | 0.46s | — | 0.12s | 0.13s | — |
| First Event → First Tool Call | 4.89s | 4.73s | — | 4.83s | 6.85s | — |
| MCP Initialize → First Semantic Call | — | — | — | — | — | — |
| MCP Server Start → Initialize | — | — | — | — | — | — |
| Top-Level User Turns | 5 | 5 | 0% | 1 | 5 | +400.0% |
| Internal Tool Cycles | 13 | 35 | +169.2% | 7 | 42 | +500.0% |
| Initial Load / Discovery Turns | 3 | 3 | 0% | 1 | 4 | +300.0% |
| MCP Discovery / Schema Turns | 0 | 0 | — | 0 | 0 | — |
| Tool Invocations (per run) | 13 | 35 | +169.2% | 7 | 42 | +500.0% |
| Output Tokens | 14936 | 15370 | +2.9% | 1689 | 22516 | +1233.1% |
| Reasoning / Thinking Tokens | 7193 | 6475 | -10.0% | 700 | 10120 | +1345.7% |
| Total Input Tokens | 1749546 | 2825970 | +61.5% | 197041 | 5063968 | +2470.0% |
| Cached Input Tokens | 1636096 | 2645760 | +61.7% | 179968 | 4816896 | +2576.5% |
| Uncached Input Tokens | 113450 | 180210 | +58.8% | 17073 | 247072 | +1347.2% |
| Cost | 0.9693 | 1.3850 | +42.9% | 0.1175 | 2.2299 | +1797.2% |
| Cached vs Uncached Token Ratio | 14.42:1 | 14.68:1 | +1.8% | 10.54:1 | 19.50:1 | +85.0% |
| Oracle L1: Mutation Policy | PASS | PASS | — | PASS | PASS | — |
| Oracle L2: AST Invariants | PASS | PASS | — | PASS | PASS | — |
| Oracle L3: Clean Build | PASS | PASS | — | PASS | PASS | — |
| Oracle L4: Verification Test | PASS | FAIL | — | PASS | PASS | — |
| Semantic Tool Invocation Verified (per run) | N/A | PASS | — | N/A | PASS | — |
Baseline vs Semedit Agentic in Small Context
- Vanilla (Baseline vs Semedit Agentic in Small Context) Edit: File audit/normalize.go modified (2 lines -> 8 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Small Context) Edit: File audit/normalize.go modified (2 lines -> 8 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ✓ ✓ | ✓ ✓ |
| 3 | ✓ ✓ | ✓ ✓ |
| 4 | ✓ ✓ | ✓ ✓ |
| 5 | ? ✗ | semedit/semantic_inspect_symbol (✓ ✓) |
| 6 | ✓ ✓ | semedit/semantic_lookup (✓ ✓) |
| 7 | ✓ ✓ | ✓ ✓ |
| 8 | ✓ ✓ | semedit/semantic_insert_function (✓ ✓) |
| 9 | ✓ ✓ | semedit/semantic_insert_function (✓ ✓) |
| 10 | ✓ ✓ | semedit/semantic_replace_body (✓ ✓) |
| 11 | ✓ ✓ | semedit/semantic_replace_body (✓ ✓) |
| 12 | ✓ ✓ | ✓ ✓ |
| 13 | ✓ ✓ | ✓ ✓ |
| 14 | — | semedit/semantic_verify (✓ ✓) |
| 15 | — | ? ✗ |
| 16 | — | ✓ ✓ |
| 17 | — | ✓ ✓ |
| 18 | — | semedit/semantic_verify (✓ ✓) |
| 19 | — | ✓ ✓ |
| 20 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 21 | — | semedit/semantic_replace_body (✓ ✓) |
| 22 | — | ✓ ✓ |
| 23 | — | semedit/semantic_verify (✓ ✓) |
| 24 | — | ✓ ✓ |
| 25 | — | semedit/semantic_outline (✓ ✓) |
| 26 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 27 | — | ✓ ✓ |
| 28 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 29 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 30 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 31 | — | ✓ ✓ |
| 32 | — | semedit/semantic_replace_body (✓ ✓) |
| 33 | — | semedit/semantic_replace_body (✓ ✓) |
| 34 | — | semedit/semantic_verify (✓ ✓) |
| 35 | — | ✓ ✓ |
Baseline vs Semedit Agentic in Large Context
- Vanilla (Baseline vs Semedit Agentic in Large Context) Edit: File audit/normalize.go modified (2 lines -> 13 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Large Context) Edit: File audit/normalize.go modified (2 lines -> 12 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ✓ ✓ | ✓ ✓ |
| 3 | ✓ ✓ | ✓ ✓ |
| 4 | ✓ ✓ | ✓ ✓ |
| 5 | ✓ ✓ | semedit/semantic_lookup (✓ ✓) |
| 6 | ✓ ✓ | semedit/semantic_insert_function (✓ ✓) |
| 7 | ✓ ✓ | semedit/semantic_replace_body (✓ ✗): replace_body Record in /Users/alessandro/sources/semantic-editor/.scratch/benchmarks/run_codex_semedit_large-default_task-11-mixed-sink-api-migration_1790469880831246000/service/dispatch.go: symbol not found |
| 8 | — | semedit/semantic_replace_body (✓ ✓) |
| 9 | — | ✓ ✓ |
| 10 | — | semedit/semantic_replace_body (✓ ✓) |
| 11 | — | semedit/semantic_replace_body (✓ ✓) |
| 12 | — | semedit/semantic_replace_body (✓ ✓) |
| 13 | — | semedit/semantic_replace_body (✓ ✓) |
| 14 | — | ✓ ✓ |
| 15 | — | semedit/semantic_verify (✓ ✓) |
| 16 | — | ✓ ✓ |
| 17 | — | ✓ ✓ |
| 18 | — | ✓ ✓ |
| 19 | — | semedit/semantic_replace_body (✓ ✓) |
| 20 | — | semedit/semantic_replace_body (✓ ✓) |
| 21 | — | ✓ ✓ |
| 22 | — | semedit/semantic_organize_imports (✓ ✓) |
| 23 | — | semedit/semantic_verify (✓ ✓) |
| 24 | — | ✓ ✓ |
| 25 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 26 | — | ✓ ✓ |
| 27 | — | semedit/semantic_verify (✓ ✓) |
| 28 | — | ✓ ✓ |
| 29 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 30 | — | semedit/semantic_verify (✓ ✓) |
| 31 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 32 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 33 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 34 | — | semedit/semantic_insert_function (✓ ✓) |
| 35 | — | semedit/semantic_replace_body (✓ ✓) |
| 36 | — | semedit/semantic_replace_body (✓ ✓) |
| 37 | — | semedit/semantic_replace_body (✓ ✓) |
| 38 | — | ✓ ✓ |
| 39 | — | semedit/semantic_organize_imports (✓ ✓) |
| 40 | — | ✓ ✓ |
| 41 | — | semedit/semantic_verify (✓ ✓) |
| 42 | — | ✓ ✓ |
default prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-11-mixed-sink-api-migration |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | default |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/bench/task_11_mixed_sink_api_migration.txtar |
Vanilla LLM Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
Audit event kinds may contain inconsistent whitespace and casing. Normalize them at the delivery boundary before persistence, and treat probe events as control traffic. Do not change the metrics or legacy export protocols. Preserve the module metadata and tests, and leave the workspace verified. Do not use semantic editing MCP tools; use standard file editing. When done, output DONE.
Semedit MCP Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
Audit event kinds may contain inconsistent whitespace and casing. Normalize them at the delivery boundary before persistence, and treat probe events as control traffic. Do not change the metrics or legacy export protocols. Preserve the module metadata and tests, and leave the workspace verified. Use semedit semantic tools for supported source code modifications. Shell commands for builds and tests are allowed. When done, output DONE.
Initial Workspace State (Before Edit)
// Package audit leaves room for normalization rules owned by the delivery boundary.
package audit| Metric | Vanilla (Small) | MCP (Small) | Δ (Small) | Vanilla (Large) | MCP (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | 50.92s | 188.14s | +269.5% | 103.35s | 235.86s | +128.2% |
| Process Start → First Event | 0.12s | 0.12s | — | 0.13s | 0.12s | — |
| First Event → First Tool Call | 4.51s | 4.56s | — | 4.51s | 5.29s | — |
| MCP Initialize → First Semantic Call | — | — | — | — | — | — |
| MCP Server Start → Initialize | — | — | — | — | — | — |
| Top-Level User Turns | 1 | 5 | +400.0% | 3 | 5 | +66.7% |
| Internal Tool Cycles | 7 | 31 | +342.9% | 9 | 41 | +355.6% |
| Initial Load / Discovery Turns | 1 | 2 | +100.0% | 3 | 2 | -33.3% |
| MCP Discovery / Schema Turns | 0 | 0 | — | 0 | 0 | — |
| Tool Invocations (per run) | 7 | 31 | +342.9% | 9 | 41 | +355.6% |
| Output Tokens | 1541 | 16179 | +949.9% | 9023 | 22405 | +148.3% |
| Reasoning / Thinking Tokens | 635 | 7584 | +1094.3% | 4708 | 10422 | +121.4% |
| Total Input Tokens | 183311 | 2963274 | +1516.5% | 788226 | 4064697 | +415.7% |
| Cached Input Tokens | 161536 | 2770944 | +1615.4% | 724224 | 3725568 | +414.4% |
| Uncached Input Tokens | 21775 | 192330 | +783.3% | 64002 | 339129 | +429.9% |
| Cost | 0.1220 | 1.4706 | +1105.2% | 0.5127 | 2.1896 | +327.1% |
| Cached vs Uncached Token Ratio | 7.42:1 | 14.41:1 | +94.2% | 11.32:1 | 10.99:1 | -2.9% |
| Oracle L1: Mutation Policy | PASS | PASS | — | PASS | PASS | — |
| Oracle L2: AST Invariants | PASS | PASS | — | PASS | PASS | — |
| Oracle L3: Clean Build | PASS | PASS | — | PASS | PASS | — |
| Oracle L4: Verification Test | PASS | FAIL | — | PASS | FAIL | — |
| Semantic Tool Invocation Verified (per run) | N/A | PASS | — | N/A | PASS | — |
Baseline vs Semedit Agentic in Small Context
- Vanilla (Baseline vs Semedit Agentic in Small Context) Edit: File audit/normalize.go modified (2 lines -> 15 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Small Context) Edit: File audit/normalize.go modified (2 lines -> 9 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ? ✗ | ✓ ✓ |
| 3 | ✓ ✓ | ✓ ✓ |
| 4 | ✓ ✓ | semedit/semantic_lookup (✓ ✓) |
| 5 | ✓ ✓ | semedit/semantic_lookup (✓ ✓) |
| 6 | ✓ ✓ | semedit/semantic_outline (✓ ✓) |
| 7 | ✓ ✓ | semedit/semantic_insert_function (✓ ✓) |
| 8 | — | semedit/semantic_batch (✓ ✓) |
| 9 | — | ✓ ✓ |
| 10 | — | ✓ ✓ |
| 11 | — | ✓ ✓ |
| 12 | — | ✓ ✓ |
| 13 | — | semedit/semantic_lookup (✓ ✓) |
| 14 | — | semedit/semantic_lookup (✓ ✓) |
| 15 | — | semedit/semantic_lookup (✓ ✓) |
| 16 | — | ✓ ✓ |
| 17 | — | ✓ ✓ |
| 18 | — | ✓ ✓ |
| 19 | — | semedit/semantic_lookup (✓ ✓) |
| 20 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 21 | — | semedit/semantic_replace_body (✓ ✓) |
| 22 | — | ✓ ✓ |
| 23 | — | ✓ ✓ |
| 24 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 25 | — | ✓ ✓ |
| 26 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 27 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 28 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 29 | — | ✓ ✓ |
| 30 | — | semedit/semantic_batch (✓ ✓) |
| 31 | — | ✓ ✓ |
Baseline vs Semedit Agentic in Large Context
- Vanilla (Baseline vs Semedit Agentic in Large Context) Edit: File audit/normalize.go modified (2 lines -> 15 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Large Context) Edit: File audit/normalize.go modified (2 lines -> 12 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ✓ ✓ | ✓ ✓ |
| 3 | ✓ ✓ | ? ✗ |
| 4 | ✓ ✓ | semedit/semantic_lookup (✓ ✓) |
| 5 | ✓ ✓ | semedit/semantic_inspect_symbol (✓ ✓) |
| 6 | ✓ ✓ | ✓ ✓ |
| 7 | ✓ ✓ | semedit/semantic_insert_function (✓ ✓) |
| 8 | ✓ ✓ | semedit/semantic_insert_function (✓ ✓) |
| 9 | ✓ ✓ | semedit/semantic_replace_body (✓ ✗): replace_body Record in /Users/alessandro/sources/semantic-editor/.scratch/benchmarks/run_codex_semedit_large-default_task-11-mixed-sink-api-migration_1790470332891819000/service/dispatch.go: symbol not found |
| 10 | — | ✓ ✓ |
| 11 | — | semedit/semantic_replace_body (✓ ✓) |
| 12 | — | semedit/semantic_replace_body (✓ ✓) |
| 13 | — | semedit/semantic_replace_body (✓ ✓) |
| 14 | — | ✓ ✓ |
| 15 | — | semedit/semantic_verify (✓ ✓) |
| 16 | — | ✓ ✓ |
| 17 | — | ✓ ✓ |
| 18 | — | ✓ ✓ |
| 19 | — | ✓ ✓ |
| 20 | — | semedit/semantic_lookup (✓ ✓) |
| 21 | — | semedit/semantic_insert_type (✓ ✓) |
| 22 | — | semedit/semantic_insert_function (✓ ✓) |
| 23 | — | semedit/semantic_insert_function (✓ ✓) |
| 24 | — | semedit/semantic_replace_body (✓ ✓) |
| 25 | — | semedit/semantic_replace_body (✓ ✓) |
| 26 | — | semedit/semantic_replace_body (✓ ✓) |
| 27 | — | semedit/semantic_replace_body (✓ ✓) |
| 28 | — | semedit/semantic_verify (✓ ✓) |
| 29 | — | ✓ ✓ |
| 30 | — | ✓ ✓ |
| 31 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 32 | — | semedit/semantic_verify (✓ ✓) |
| 33 | — | ✓ ✓ |
| 34 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 35 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 36 | — | semedit/semantic_verify (✓ ✓) |
| 37 | — | ✓ ✓ |
| 38 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 39 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 40 | — | semedit/semantic_replace_body (✓ ✓) |
| 41 | — | ✓ ✓ |
default prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-11-mixed-sink-api-migration |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | default |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/bench/task_11_mixed_sink_api_migration.txtar |
Vanilla LLM Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
Audit event kinds may contain inconsistent whitespace and casing. Normalize them at the delivery boundary before persistence, and treat probe events as control traffic. Do not change the metrics or legacy export protocols. Preserve the module metadata and tests, and leave the workspace verified. Do not use semantic editing MCP tools; use standard file editing. When done, output DONE.
Semedit MCP Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
Audit event kinds may contain inconsistent whitespace and casing. Normalize them at the delivery boundary before persistence, and treat probe events as control traffic. Do not change the metrics or legacy export protocols. Preserve the module metadata and tests, and leave the workspace verified. Use semedit semantic tools for supported source code modifications. Shell commands for builds and tests are allowed. When done, output DONE.
Initial Workspace State (Before Edit)
// Package audit leaves room for normalization rules owned by the delivery boundary.
package audit| Metric | Vanilla (Small) | MCP (Small) | Δ (Small) | Vanilla (Large) | MCP (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | 92.64s | 102.41s | +10.5% | 130.22s | 259.05s | +98.9% |
| Process Start → First Event | 0.12s | 0.12s | — | 0.12s | 0.12s | — |
| First Event → First Tool Call | 4.45s | 4.49s | — | 3.71s | 4.76s | — |
| MCP Initialize → First Semantic Call | — | — | — | — | — | — |
| MCP Server Start → Initialize | — | — | — | — | — | — |
| Top-Level User Turns | 5 | 1 | -80.0% | 5 | 5 | 0% |
| Internal Tool Cycles | 12 | 16 | +33.3% | 15 | 49 | +226.7% |
| Initial Load / Discovery Turns | 3 | 2 | -33.3% | 3 | 6 | +100.0% |
| MCP Discovery / Schema Turns | 0 | 0 | — | 0 | 0 | — |
| Tool Invocations (per run) | 12 | 16 | +33.3% | 15 | 49 | +226.7% |
| Output Tokens | 10959 | 2648 | -75.8% | 16255 | 19385 | +19.3% |
| Reasoning / Thinking Tokens | 4655 | 1300 | -72.1% | 6872 | 10000 | +45.5% |
| Total Input Tokens | 1535817 | 588018 | -61.7% | 2142818 | 4243433 | +98.0% |
| Cached Input Tokens | 1438976 | 547840 | -61.9% | 2017536 | 3986432 | +97.6% |
| Uncached Input Tokens | 96841 | 40178 | -58.5% | 125282 | 257001 | +105.1% |
| Cost | 0.7970 | 0.2868 | -64.0% | 1.1067 | 2.0064 | +81.3% |
| Cached vs Uncached Token Ratio | 14.86:1 | 13.64:1 | -8.2% | 16.10:1 | 15.51:1 | -3.7% |
| Oracle L1: Mutation Policy | PASS | PASS | — | PASS | PASS | — |
| Oracle L2: AST Invariants | PASS | PASS | — | PASS | PASS | — |
| Oracle L3: Clean Build | PASS | PASS | — | PASS | PASS | — |
| Oracle L4: Verification Test | PASS | PASS | — | FAIL | PASS | — |
| Semantic Tool Invocation Verified (per run) | N/A | PASS | — | N/A | PASS | — |
Baseline vs Semedit Agentic in Small Context
- Vanilla (Baseline vs Semedit Agentic in Small Context) Edit: File audit/normalize.go modified (2 lines -> 9 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Small Context) Edit: File audit/normalize.go modified (2 lines -> 8 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ✓ ✓ | ✓ ✓ |
| 3 | ✓ ✓ | ✓ ✓ |
| 4 | ✓ ✓ | semedit/semantic_lookup (✓ ✓) |
| 5 | ✓ ✓ | semedit/semantic_outline (✓ ✓) |
| 6 | ✓ ✓ | ✓ ✓ |
| 7 | ✓ ✓ | semedit/semantic_insert_function (✓ ✓) |
| 8 | ✓ ✓ | semedit/semantic_replace_body (✓ ✓) |
| 9 | ✓ ✓ | semedit/semantic_replace_body (✓ ✓) |
| 10 | ✓ ✓ | semedit/semantic_replace_body (✓ ✓) |
| 11 | ✓ ✓ | semedit/semantic_replace_body (✓ ✓) |
| 12 | ✓ ✓ | semedit/semantic_replace_body (✓ ✓) |
| 13 | — | semedit/semantic_verify (✓ ✓) |
| 14 | — | ✓ ✓ |
| 15 | — | ✓ ✓ |
| 16 | — | ✓ ✓ |
Baseline vs Semedit Agentic in Large Context
- Vanilla (Baseline vs Semedit Agentic in Large Context) Edit: File audit/normalize.go modified (2 lines -> 19 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Large Context) Edit: File audit/normalize.go modified (2 lines -> 9 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ✓ ✓ | ✓ ✓ |
| 3 | ✓ ✓ | ✓ ✓ |
| 4 | ✓ ✓ | ✓ ✓ |
| 5 | ? ✗ | semedit/semantic_lookup (✓ ✓) |
| 6 | ✓ ✓ | semedit/semantic_outline (✓ ✓) |
| 7 | ✓ ✓ | ✓ ✓ |
| 8 | ✓ ✓ | semedit/semantic_inspect_symbol (✓ ✓) |
| 9 | ✓ ✓ | semedit/semantic_inspect_symbol (✓ ✓) |
| 10 | ✓ ✓ | semedit/semantic_inspect_symbol (✓ ✓) |
| 11 | ✓ ✓ | semedit/semantic_inspect_symbol (✓ ✓) |
| 12 | ✓ ✓ | semedit/semantic_inspect_symbol (✓ ✗): resolve BufferedSink.Write: symbol not found |
| 13 | ✓ ✓ | semedit/semantic_inspect_symbol (✓ ✓) |
| 14 | ✓ ✓ | semedit/semantic_inspect_symbol (✓ ✓) |
| 15 | ✓ ✓ | ✓ ✓ |
| 16 | — | semedit/semantic_insert_function (✓ ✓) |
| 17 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 18 | — | semedit/semantic_replace_body (✓ ✓) |
| 19 | — | semedit/semantic_replace_body (✓ ✓) |
| 20 | — | semedit/semantic_replace_body (✓ ✓) |
| 21 | — | semedit/semantic_verify (✓ ✓) |
| 22 | — | ✓ ✓ |
| 23 | — | ✓ ✓ |
| 24 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 25 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 26 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 27 | — | semedit/semantic_verify (✓ ✓) |
| 28 | — | ✓ ✓ |
| 29 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 30 | — | semedit/semantic_verify (✓ ✓) |
| 31 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 32 | — | ✓ ✓ |
| 33 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 34 | — | semedit/semantic_replace_body (✓ ✓) |
| 35 | — | semedit/semantic_replace_body (✓ ✓) |
| 36 | — | ✓ ✓ |
| 37 | — | ✓ ✓ |
| 38 | — | semedit/semantic_verify (✓ ✓) |
| 39 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 40 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 41 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 42 | — | ✓ ✓ |
| 43 | — | semedit/semantic_insert_function (✓ ✓) |
| 44 | — | semedit/semantic_replace_body (✓ ✓) |
| 45 | — | semedit/semantic_replace_body (✓ ✓) |
| 46 | — | semedit/semantic_replace_body (✓ ✓) |
| 47 | — | semedit/semantic_verify (✓ ✓) |
| 48 | — | ✓ ✓ |
| 49 | — | ✓ ✓ |
default prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-11-mixed-sink-api-migration |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | default |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/bench/task_11_mixed_sink_api_migration.txtar |
Vanilla LLM Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
Audit event kinds may contain inconsistent whitespace and casing. Normalize them at the delivery boundary before persistence, and treat probe events as control traffic. Do not change the metrics or legacy export protocols. Preserve the module metadata and tests, and leave the workspace verified. Do not use semantic editing MCP tools; use standard file editing. When done, output DONE.
Semedit MCP Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
Audit event kinds may contain inconsistent whitespace and casing. Normalize them at the delivery boundary before persistence, and treat probe events as control traffic. Do not change the metrics or legacy export protocols. Preserve the module metadata and tests, and leave the workspace verified. Use semedit semantic tools for supported source code modifications. Shell commands for builds and tests are allowed. When done, output DONE.
Initial Workspace State (Before Edit)
// Package audit leaves room for normalization rules owned by the delivery boundary.
package audit| Metric | Vanilla (Small) | MCP (Small) | Δ (Small) | Vanilla (Large) | MCP (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | 93.22s | 231.79s | +148.6% | 246.03s | 179.93s | -26.9% |
| Process Start → First Event | 0.13s | 0.12s | — | 0.12s | 0.13s | — |
| First Event → First Tool Call | 4.51s | 5.99s | — | 5.70s | 4.43s | — |
| MCP Initialize → First Semantic Call | — | — | — | — | — | — |
| MCP Server Start → Initialize | — | — | — | — | — | — |
| Top-Level User Turns | 5 | 5 | 0% | 5 | 5 | 0% |
| Internal Tool Cycles | 11 | 51 | +363.6% | 16 | 29 | +81.2% |
| Initial Load / Discovery Turns | 3 | 3 | 0% | 3 | 3 | 0% |
| MCP Discovery / Schema Turns | 0 | 0 | — | 0 | 0 | — |
| Tool Invocations (per run) | 11 | 51 | +363.6% | 16 | 29 | +81.2% |
| Output Tokens | 11120 | 21064 | +89.4% | 16412 | 18379 | +12.0% |
| Reasoning / Thinking Tokens | 4671 | 10855 | +132.4% | 6352 | 9883 | +55.6% |
| Total Input Tokens | 1604508 | 3699711 | +130.6% | 1997212 | 3582375 | +79.4% |
| Cached Input Tokens | 1497344 | 3356672 | +124.2% | 1876992 | 3357440 | +78.9% |
| Uncached Input Tokens | 107164 | 343039 | +220.1% | 120220 | 224935 | +87.1% |
| Cost | 0.8396 | 2.0958 | +149.6% | 1.0543 | 1.7550 | +66.5% |
| Cached vs Uncached Token Ratio | 13.97:1 | 9.79:1 | -30.0% | 15.61:1 | 14.93:1 | -4.4% |
| Oracle L1: Mutation Policy | PASS | PASS | — | PASS | PASS | — |
| Oracle L2: AST Invariants | PASS | PASS | — | PASS | PASS | — |
| Oracle L3: Clean Build | PASS | PASS | — | PASS | PASS | — |
| Oracle L4: Verification Test | FAIL | PASS | — | FAIL | PASS | — |
| Semantic Tool Invocation Verified (per run) | N/A | PASS | — | N/A | PASS | — |
Baseline vs Semedit Agentic in Small Context
- Vanilla (Baseline vs Semedit Agentic in Small Context) Edit: File audit/normalize.go modified (2 lines -> 15 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Small Context) Edit: File audit/normalize.go modified (2 lines -> 8 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ✓ ✓ | ✓ ✓ |
| 3 | ✓ ✓ | ✓ ✓ |
| 4 | ✓ ✓ | ✓ ✓ |
| 5 | ✓ ✓ | semedit/semantic_outline (✓ ✓) |
| 6 | ✓ ✓ | semedit/semantic_outline (✓ ✓) |
| 7 | ✓ ✓ | ✓ ✓ |
| 8 | ✓ ✓ | ✓ ✓ |
| 9 | ✓ ✓ | semedit/semantic_inspect_symbol (✓ ✓) |
| 10 | ✓ ✓ | semedit/semantic_inspect_symbol (✓ ✓) |
| 11 | ✓ ✓ | semedit/semantic_inspect_symbol (✓ ✗): param "symbol" is required: invalid operation parameters |
| 12 | — | semedit/semantic_batch (✓ ✗): { "status": "error", "results": [ { "tool": "semantic_insert_function", "status": "ok" }, { "tool": "semantic_insert_function", "status": "ok" }, { "tool": "semantic_replace_body", "status": "error", "error": "replace_body Record in /Users/alessandro/sources/semantic-editor/.scratch/benchmarks/run_codex_semedit_small-default_task-11-mixed-sink-api-migration_1790469028200706000/service/dispatch.go: symbol not found" } ], "final_diff": "--- a/audit/normalize.go\n+++ b/audit/normalize.go\n // Package audit lea |
| 13 | — | ✓ ✓ |
| 14 | — | semedit/semantic_outline (✓ ✓) |
| 15 | — | semedit/semantic_organize_imports (✓ ✓) |
| 16 | — | semedit/semantic_batch (✓ ✓) |
| 17 | — | semedit/semantic_verify (✓ ✓) |
| 18 | — | semedit/semantic_verify (✓ ✓) |
| 19 | — | ✓ ✓ |
| 20 | — | ✓ ✓ |
| 21 | — | ✓ ✓ |
| 22 | — | ✓ ✓ |
| 23 | — | semedit/semantic_outline (✓ ✓) |
| 24 | — | semedit/semantic_verify (✓ ✓) |
| 25 | — | semedit/semantic_verify (✓ ✓) |
| 26 | — | ✓ ✓ |
| 27 | — | ✓ ✓ |
| 28 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 29 | — | ✓ ✓ |
| 30 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 31 | — | ✓ ✓ |
| 32 | — | semedit/semantic_replace_body (✓ ✓) |
| 33 | — | semedit/semantic_verify (✓ ✓) |
| 34 | — | ✓ ✓ |
| 35 | — | ✓ ✓ |
| 36 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 37 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 38 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 39 | — | ✓ ✓ |
| 40 | — | semedit/semantic_verify (✓ ✓) |
| 41 | — | ✓ ✓ |
| 42 | — | ✓ ✓ |
| 43 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 44 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 45 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 46 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 47 | — | ✓ ✓ |
| 48 | — | semedit/semantic_batch (✓ ✓) |
| 49 | — | ✓ ✓ |
| 50 | — | ✓ ✓ |
| 51 | — | semedit/semantic_verify (✓ ✓) |
Baseline vs Semedit Agentic in Large Context
- Vanilla (Baseline vs Semedit Agentic in Large Context) Edit: File audit/normalize.go modified (2 lines -> 11 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Large Context) Edit: File audit/normalize.go modified (2 lines -> 16 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ✓ ✓ | ✓ ✓ |
| 3 | ✓ ✓ | semedit/semantic_inspect_symbol (✓ ✓) |
| 4 | ✓ ✓ | semedit/semantic_inspect_symbol (✓ ✓) |
| 5 | ✓ ✓ | semedit/semantic_inspect_symbol (✓ ✓) |
| 6 | ✓ ✓ | semedit/semantic_inspect_symbol (✓ ✗): resolve Normalize: symbol not found |
| 7 | ✓ ✓ | ✓ ✓ |
| 8 | ✓ ✓ | semedit/semantic_insert_function (✓ ✓) |
| 9 | ✓ ✓ | semedit/semantic_insert_function (✓ ✓) |
| 10 | ✓ ✓ | semedit/semantic_replace_body (✓ ✓) |
| 11 | ✓ ✓ | semedit/semantic_replace_body (✓ ✓) |
| 12 | ✓ ✓ | semedit/semantic_replace_body (✓ ✓) |
| 13 | ✓ ✓ | semedit/semantic_replace_body (✓ ✓) |
| 14 | ✓ ✓ | ✓ ✓ |
| 15 | ✓ ✓ | ✓ ✓ |
| 16 | ✓ ✓ | ✓ ✓ |
| 17 | — | ? ✗ |
| 18 | — | semedit/semantic_outline (✓ ✓) |
| 19 | — | semedit/semantic_outline (✓ ✓) |
| 20 | — | semedit/semantic_outline (✓ ✓) |
| 21 | — | ✓ ✓ |
| 22 | — | ✓ ✓ |
| 23 | — | ✓ ✓ |
| 24 | — | semedit/semantic_replace_body (✓ ✓) |
| 25 | — | ✓ ✓ |
| 26 | — | ✓ ✓ |
| 27 | — | ✓ ✓ |
| 28 | — | semedit/semantic_replace_body (✓ ✓) |
| 29 | — | ✓ ✓ |
default prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-11-mixed-sink-api-migration |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | default |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/bench/task_11_mixed_sink_api_migration.txtar |
Vanilla LLM Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
Audit event kinds may contain inconsistent whitespace and casing. Normalize them at the delivery boundary before persistence, and treat probe events as control traffic. Do not change the metrics or legacy export protocols. Preserve the module metadata and tests, and leave the workspace verified. Do not use semantic editing MCP tools; use standard file editing. When done, output DONE.
Semedit MCP Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
Audit event kinds may contain inconsistent whitespace and casing. Normalize them at the delivery boundary before persistence, and treat probe events as control traffic. Do not change the metrics or legacy export protocols. Preserve the module metadata and tests, and leave the workspace verified. Use semedit semantic tools for supported source code modifications. Shell commands for builds and tests are allowed. When done, output DONE.
Initial Workspace State (Before Edit)
// Package audit leaves room for normalization rules owned by the delivery boundary.
package audit| Metric | Vanilla (Small) | MCP (Small) | Δ (Small) | Vanilla (Large) | MCP (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | 84.96s | 184.49s | +117.2% | 64.76s | 190.39s | +194.0% |
| Process Start → First Event | 0.12s | 0.13s | — | 0.14s | 0.13s | — |
| First Event → First Tool Call | 5.89s | 4.78s | — | 4.83s | 4.65s | — |
| MCP Initialize → First Semantic Call | — | — | — | — | — | — |
| MCP Server Start → Initialize | — | — | — | — | — | — |
| Top-Level User Turns | 5 | 5 | 0% | 1 | 5 | +400.0% |
| Internal Tool Cycles | 9 | 25 | +177.8% | 8 | 28 | +250.0% |
| Initial Load / Discovery Turns | 3 | 1 | -66.7% | 2 | 3 | +50.0% |
| MCP Discovery / Schema Turns | 0 | 0 | — | 0 | 0 | — |
| Tool Invocations (per run) | 9 | 25 | +177.8% | 8 | 28 | +250.0% |
| Output Tokens | 9288 | 14670 | +57.9% | 2307 | 20414 | +784.9% |
| Reasoning / Thinking Tokens | 4352 | 6359 | +46.1% | 846 | 10121 | +1096.3% |
| Total Input Tokens | 1151438 | 3173687 | +175.6% | 194574 | 3207454 | +1548.4% |
| Cached Input Tokens | 1000704 | 2932736 | +193.1% | 166656 | 3019520 | +1711.8% |
| Uncached Input Tokens | 150734 | 240951 | +59.9% | 27918 | 187934 | +573.2% |
| Cost | 0.7975 | 1.5984 | +100.4% | 0.1509 | 1.6064 | +964.7% |
| Cached vs Uncached Token Ratio | 6.64:1 | 12.17:1 | +83.3% | 5.97:1 | 16.07:1 | +169.2% |
| Oracle L1: Mutation Policy | PASS | PASS | — | PASS | PASS | — |
| Oracle L2: AST Invariants | PASS | PASS | — | PASS | PASS | — |
| Oracle L3: Clean Build | PASS | PASS | — | PASS | PASS | — |
| Oracle L4: Verification Test | FAIL | FAIL | — | PASS | PASS | — |
| Semantic Tool Invocation Verified (per run) | N/A | PASS | — | N/A | PASS | — |
Baseline vs Semedit Agentic in Small Context
- Vanilla (Baseline vs Semedit Agentic in Small Context) Edit: File audit/normalize.go modified (2 lines -> 11 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Small Context) Edit: File audit/normalize.go modified (2 lines -> 8 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ✓ ✓ | ? ✗ |
| 3 | ✓ ✓ | ✓ ✓ |
| 4 | ✓ ✓ | ✓ ✓ |
| 5 | ✓ ✓ | semedit/semantic_insert_function (✓ ✓) |
| 6 | ✓ ✓ | semedit/semantic_insert_function (✓ ✓) |
| 7 | ✓ ✓ | semedit/semantic_replace_body (✓ ✓) |
| 8 | ✓ ✓ | semedit/semantic_replace_body (✓ ✓) |
| 9 | ✓ ✓ | semedit/semantic_verify (✓ ✓) |
| 10 | — | ✓ ✓ |
| 11 | — | ✓ ✓ |
| 12 | — | ✓ ✓ |
| 13 | — | semedit/semantic_replace_body (✓ ✓) |
| 14 | — | semedit/semantic_verify (✓ ✓) |
| 15 | — | ✓ ✓ |
| 16 | — | ✓ ✓ |
| 17 | — | semedit/semantic_replace_body (✓ ✓) |
| 18 | — | semedit/semantic_verify (✓ ✓) |
| 19 | — | ✓ ✓ |
| 20 | — | ✓ ✓ |
| 21 | — | semedit/semantic_verify (✓ ✓) |
| 22 | — | ✓ ✓ |
| 23 | — | ✓ ✓ |
| 24 | — | semedit/semantic_replace_body (✓ ✓) |
| 25 | — | ✓ ✓ |
Baseline vs Semedit Agentic in Large Context
- Vanilla (Baseline vs Semedit Agentic in Large Context) Edit: File audit/normalize.go modified (2 lines -> 21 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Large Context) Edit: File audit/normalize.go modified (2 lines -> 16 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ✓ ✓ | ? ✗ |
| 3 | ✓ ✓ | ✓ ✓ |
| 4 | ✓ ✓ | ✓ ✓ |
| 5 | ✓ ✓ | ✓ ✓ |
| 6 | ✓ ✓ | semedit/semantic_insert_function (✓ ✓) |
| 7 | ✓ ✓ | semedit/semantic_insert_function (✓ ✓) |
| 8 | ✓ ✓ | ✓ ✓ |
| 9 | — | semedit/semantic_replace_body (✓ ✗): replace_body Write in /Users/alessandro/sources/semantic-editor/.scratch/benchmarks/run_codex_semedit_large-default_task-11-mixed-sink-api-migration_1790470043873981000/audit/fanout.go: symbol not found |
| 10 | — | semedit/semantic_replace_body (✓ ✗): replace_body Commit in /Users/alessandro/sources/semantic-editor/.scratch/benchmarks/run_codex_semedit_large-default_task-11-mixed-sink-api-migration_1790470043873981000/audit/transaction.go: symbol not found |
| 11 | — | semedit/semantic_replace_body (✓ ✓) |
| 12 | — | semedit/semantic_replace_body (✓ ✓) |
| 13 | — | semedit/semantic_replace_body (✓ ✓) |
| 14 | — | ✓ ✓ |
| 15 | — | ✓ ✓ |
| 16 | — | semedit/semantic_replace_body (✓ ✓) |
| 17 | — | semedit/semantic_replace_body (✓ ✓) |
| 18 | — | ? ✗ |
| 19 | — | semedit/semantic_verify (✗ ?): This action was rejected due to unacceptable risk. Reason: Go verification over the entire workspace can format and write source files, potentially modifying protected tests or other files; the user explicitly forbade test edits and only authorized verification, not workspace-wide mutations. Do not bypass this rejection through a workaround or indirect execution. Continue with a safer alternative, or carry out checks to prove that the action is authorized or low risk before trying again. Complete unaffected |
| 20 | — | ✓ ✓ |
| 21 | — | ✓ ✓ |
| 22 | — | ✓ ✓ |
| 23 | — | semedit/semantic_replace_body (✓ ✓) |
| 24 | — | ✓ ✓ |
| 25 | — | ✓ ✓ |
| 26 | — | ✓ ✓ |
| 27 | — | semedit/semantic_replace_body (✓ ✓) |
| 28 | — | ✓ ✓ |
default prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-11-mixed-sink-api-migration |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | default |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/bench/task_11_mixed_sink_api_migration.txtar |
Vanilla LLM Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
Audit event kinds may contain inconsistent whitespace and casing. Normalize them at the delivery boundary before persistence, and treat probe events as control traffic. Do not change the metrics or legacy export protocols. Preserve the module metadata and tests, and leave the workspace verified. Do not use semantic editing MCP tools; use standard file editing. When done, output DONE.
Semedit MCP Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
Audit event kinds may contain inconsistent whitespace and casing. Normalize them at the delivery boundary before persistence, and treat probe events as control traffic. Do not change the metrics or legacy export protocols. Preserve the module metadata and tests, and leave the workspace verified. Use semedit semantic tools for supported source code modifications. Shell commands for builds and tests are allowed. When done, output DONE.
Initial Workspace State (Before Edit)
// Package audit leaves room for normalization rules owned by the delivery boundary.
package audit| Metric | Vanilla (Small) | MCP (Small) | Δ (Small) | Vanilla (Large) | MCP (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | 211.66s | 157.43s | -25.6% | 122.93s | — | — |
| Process Start → First Event | 0.12s | 0.12s | — | 0.13s | — | — |
| First Event → First Tool Call | 5.29s | 4.46s | — | 4.61s | — | — |
| MCP Initialize → First Semantic Call | — | — | — | — | — | — |
| MCP Server Start → Initialize | — | — | — | — | — | — |
| Top-Level User Turns | 5 | 5 | 0% | 5 | — | — |
| Internal Tool Cycles | 15 | 24 | +60.0% | 15 | — | — |
| Initial Load / Discovery Turns | 3 | 2 | -33.3% | 2 | — | — |
| MCP Discovery / Schema Turns | 0 | 0 | — | 0 | — | — |
| Tool Invocations (per run) | 15 | 24 | +60.0% | 15 | — | — |
| Output Tokens | 11781 | 14177 | +20.3% | 15831 | — | — |
| Reasoning / Thinking Tokens | 4809 | 6547 | +36.1% | 5595 | — | — |
| Total Input Tokens | 1756249 | 3221385 | +83.4% | 2101458 | — | — |
| Cached Input Tokens | 1647616 | 3017472 | +83.1% | 1925888 | — | — |
| Uncached Input Tokens | 108633 | 203913 | +87.7% | 175570 | — | — |
| Cost | 0.8909 | 1.5232 | +71.0% | 1.1882 | — | — |
| Cached vs Uncached Token Ratio | 15.17:1 | 14.80:1 | -2.4% | 10.97:1 | — | — |
| Oracle L1: Mutation Policy | PASS | PASS | — | PASS | — | — |
| Oracle L2: AST Invariants | PASS | PASS | — | PASS | — | — |
| Oracle L3: Clean Build | PASS | PASS | — | PASS | — | — |
| Oracle L4: Verification Test | PASS | FAIL | — | FAIL | — | — |
| Semantic Tool Invocation Verified (per run) | N/A | PASS | — | N/A | — | — |
Baseline vs Semedit Agentic in Small Context
- Vanilla (Baseline vs Semedit Agentic in Small Context) Edit: File audit/normalize.go modified (2 lines -> 10 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Small Context) Edit: File audit/normalize.go modified (2 lines -> 8 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ✓ ✓ | ✓ ✓ |
| 3 | ✓ ✓ | ✓ ✓ |
| 4 | ✓ ✓ | semedit/semantic_insert_function (✓ ✓) |
| 5 | ✓ ✓ | semedit/semantic_replace_body (✓ ✗): replace_body Record in /Users/alessandro/sources/semantic-editor/.scratch/benchmarks/run_codex_semedit_small-default_task-11-mixed-sink-api-migration_1790468732811898000/service/dispatch.go: symbol not found |
| 6 | ✓ ✓ | ✓ ✓ |
| 7 | ✓ ✓ | semedit/semantic_replace_body (✓ ✓) |
| 8 | ✓ ✓ | semedit/semantic_replace_body (✓ ✓) |
| 9 | ✓ ✓ | semedit/semantic_verify (✗ ?): This action was rejected due to unacceptable risk. Reason: Verification over `.` may format and write all Go files, including tests the user explicitly forbade modifying; use a read-only or narrowly scoped verification instead. Do not bypass this rejection through a workaround or indirect execution. Continue with a safer alternative, or carry out checks to prove that the action is authorized or low risk before trying again. Complete unaffected work without asking for confirmation. Report anything that remai |
| 10 | ✓ ✓ | ✓ ✓ |
| 11 | ✓ ✓ | semedit/semantic_verify (✓ ✓) |
| 12 | ✓ ✓ | semedit/semantic_verify (✓ ✓) |
| 13 | ✓ ✓ | ✓ ✓ |
| 14 | ✓ ✓ | ✓ ✓ |
| 15 | ✓ ✓ | ✓ ✓ |
| 16 | — | ✓ ✓ |
| 17 | — | ✓ ✓ |
| 18 | — | semedit/semantic_replace_body (✓ ✓) |
| 19 | — | ✓ ✓ |
| 20 | — | ✓ ✓ |
| 21 | — | semedit/semantic_verify (✓ ✓) |
| 22 | — | ✓ ✓ |
| 23 | — | ✓ ✓ |
| 24 | — | ✓ ✓ |
Baseline vs Semedit Agentic in Large Context
- Vanilla (Baseline vs Semedit Agentic in Large Context) Edit: File audit/normalize.go modified (2 lines -> 15 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | not published |
| 2 | ✓ ✓ | not published |
| 3 | ✓ ✓ | not published |
| 4 | ✓ ✓ | not published |
| 5 | ✓ ✓ | not published |
| 6 | ? ✗ | not published |
| 7 | ✓ ✓ | not published |
| 8 | ✓ ✓ | not published |
| 9 | ✓ ✓ | not published |
| 10 | ✓ ✓ | not published |
| 11 | ✓ ✓ | not published |
| 12 | ✓ ✓ | not published |
| 13 | ✓ ✓ | not published |
| 14 | ✓ ✓ | not published |
| 15 | ✓ ✓ | not published |
prefer_discover_semedit prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-11-mixed-sink-api-migration |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | prefer_discover_semedit |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/bench/task_11_mixed_sink_api_migration.txtar |
Vanilla LLM Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
Before editing, inspect the complete available tool inventory, including deferred or lazy tools. If applicable semantic editing tools are callable, prefer them for source mutations. Audit event kinds may contain inconsistent whitespace and casing. Normalize them at the delivery boundary before persistence, and treat probe events as control traffic. Do not change the metrics or legacy export protocols. Preserve the module metadata and tests, and leave the workspace verified. Do not use semantic editing MCP tools; use standard file editing. When done, output DONE.
Semedit MCP Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
Before editing, inspect the complete available tool inventory, including deferred or lazy tools. If applicable semantic editing tools are callable, prefer them for source mutations. Audit event kinds may contain inconsistent whitespace and casing. Normalize them at the delivery boundary before persistence, and treat probe events as control traffic. Do not change the metrics or legacy export protocols. Preserve the module metadata and tests, and leave the workspace verified. Use semedit semantic tools for supported source code modifications. Shell commands for builds and tests are allowed. When done, output DONE.
Initial Workspace State (Before Edit)
// Package audit leaves room for normalization rules owned by the delivery boundary.
package audit| Metric | Vanilla (Small) | MCP (Small) | Δ (Small) | Vanilla (Large) | MCP (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | — | 228.03s | — | 168.59s | 218.48s | +29.6% |
| Process Start → First Event | — | 0.12s | — | 0.12s | 0.14s | — |
| First Event → First Tool Call | — | 10.46s | — | 9.32s | 5.19s | — |
| MCP Initialize → First Semantic Call | — | — | — | — | — | — |
| MCP Server Start → Initialize | — | — | — | — | — | — |
| Top-Level User Turns | — | 5 | — | 5 | 5 | 0% |
| Internal Tool Cycles | — | 39 | — | 16 | 30 | +87.5% |
| Initial Load / Discovery Turns | — | 4 | — | 5 | 4 | -20.0% |
| MCP Discovery / Schema Turns | — | 0 | — | 0 | 0 | — |
| Tool Invocations (per run) | — | 39 | — | 16 | 30 | +87.5% |
| Output Tokens | — | 21683 | — | 23126 | 24138 | +4.4% |
| Reasoning / Thinking Tokens | — | 11441 | — | 9254 | 11788 | +27.4% |
| Total Input Tokens | — | 4919860 | — | 3421943 | 3789281 | +10.7% |
| Cached Input Tokens | — | 4637952 | — | 3222528 | 3553536 | +10.3% |
| Uncached Input Tokens | — | 281908 | — | 199415 | 235745 | +18.2% |
| Cost | — | 2.2783 | — | 1.7089 | 1.9268 | +12.8% |
| Cached vs Uncached Token Ratio | — | 16.45:1 | — | 16.16:1 | 15.07:1 | -6.7% |
| Oracle L1: Mutation Policy | — | PASS | — | PASS | PASS | — |
| Oracle L2: AST Invariants | — | PASS | — | PASS | PASS | — |
| Oracle L3: Clean Build | — | PASS | — | PASS | PASS | — |
| Oracle L4: Verification Test | — | FAIL | — | FAIL | PASS | — |
| Semantic Tool Invocation Verified (per run) | — | PASS | — | N/A | PASS | — |
Baseline vs Semedit Agentic in Small Context
- Semedit MCP (Baseline vs Semedit Agentic in Small Context) Edit: File audit/normalize.go modified (2 lines -> 21 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | not published | ✓ ✓ |
| 2 | not published | ✓ ✓ |
| 3 | not published | semedit/semantic_outline (✓ ✓) |
| 4 | not published | semedit/semantic_outline (✓ ✓) |
| 5 | not published | ✓ ✓ |
| 6 | not published | semedit/semantic_inspect_symbol (✓ ✓) |
| 7 | not published | semedit/semantic_inspect_symbol (✓ ✓) |
| 8 | not published | semedit/semantic_inspect_symbol (✓ ✓) |
| 9 | not published | semedit/semantic_inspect_symbol (✓ ✓) |
| 10 | not published | ✓ ✓ |
| 11 | not published | semedit/semantic_insert_function (✓ ✓) |
| 12 | not published | semedit/semantic_insert_function (✓ ✓) |
| 13 | not published | semedit/semantic_insert_function (✓ ✓) |
| 14 | not published | semedit/semantic_batch (✓ ✓) |
| 15 | not published | semedit/semantic_inspect_symbol (✓ ✓) |
| 16 | not published | semedit/semantic_inspect_symbol (✓ ✓) |
| 17 | not published | semedit/semantic_verify (✓ ✓) |
| 18 | not published | ✓ ✓ |
| 19 | not published | ✓ ✓ |
| 20 | not published | ✓ ✓ |
| 21 | not published | ✓ ✓ |
| 22 | not published | ✓ ✓ |
| 23 | not published | ✓ ✓ |
| 24 | not published | semedit/semantic_verify (✓ ✓) |
| 25 | not published | ✓ ✓ |
| 26 | not published | semedit/semantic_inspect_symbol (✓ ✓) |
| 27 | not published | ✓ ✓ |
| 28 | not published | ✓ ✓ |
| 29 | not published | semedit/semantic_verify (✓ ✓) |
| 30 | not published | ✓ ✓ |
| 31 | not published | semedit/semantic_inspect_symbol (✓ ✓) |
| 32 | not published | semedit/semantic_inspect_symbol (✓ ✓) |
| 33 | not published | semedit/semantic_verify (✓ ✓) |
| 34 | not published | ✓ ✓ |
| 35 | not published | semedit/semantic_insert_function (✓ ✓) |
| 36 | not published | semedit/semantic_batch (✓ ✓) |
| 37 | not published | ✓ ✓ |
| 38 | not published | semedit/semantic_verify (✓ ✓) |
| 39 | not published | ✓ ✓ |
Baseline vs Semedit Agentic in Large Context
- Vanilla (Baseline vs Semedit Agentic in Large Context) Edit: File audit/normalize.go modified (2 lines -> 28 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Large Context) Edit: File audit/normalize.go modified (2 lines -> 9 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ✓ ✓ | ✓ ✓ |
| 3 | ✓ ✓ | semedit/semantic_outline (✓ ✓) |
| 4 | ✓ ✓ | semedit/semantic_inspect_symbol (✓ ✓) |
| 5 | ✓ ✓ | ✓ ✓ |
| 6 | ✓ ✓ | semedit/semantic_insert_function (✓ ✓) |
| 7 | ✓ ✓ | semedit/semantic_replace_body (✓ ✓) |
| 8 | ✓ ✓ | semedit/semantic_replace_body (✓ ✓) |
| 9 | ✓ ✓ | semedit/semantic_replace_body (✓ ✓) |
| 10 | ✓ ✓ | ? ✗ |
| 11 | ✓ ✓ | ✓ ✓ |
| 12 | ✓ ✓ | semedit/report_feedback (✓ ✓) |
| 13 | ✓ ✓ | ? ✗ |
| 14 | ✓ ✓ | ✓ ✓ |
| 15 | ✓ ✓ | ✓ ✓ |
| 16 | ✓ ✓ | ✓ ✓ |
| 17 | — | ✓ ✓ |
| 18 | — | ✓ ✓ |
| 19 | — | ✓ ✓ |
| 20 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 21 | — | semedit/semantic_replace_body (✓ ✓) |
| 22 | — | semedit/semantic_replace_body (✓ ✓) |
| 23 | — | semedit/semantic_replace_body (✓ ✓) |
| 24 | — | ✓ ✓ |
| 25 | — | semedit/semantic_insert_function (✓ ✓) |
| 26 | — | semedit/semantic_replace_body (✓ ✓) |
| 27 | — | semedit/semantic_replace_body (✓ ✓) |
| 28 | — | semedit/semantic_replace_body (✓ ✓) |
| 29 | — | semedit/semantic_organize_imports (✓ ✓) |
| 30 | — | ✓ ✓ |
prefer_discover_semedit prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-11-mixed-sink-api-migration |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | prefer_discover_semedit |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/bench/task_11_mixed_sink_api_migration.txtar |
Vanilla LLM Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
Before editing, inspect the complete available tool inventory, including deferred or lazy tools. If applicable semantic editing tools are callable, prefer them for source mutations. Audit event kinds may contain inconsistent whitespace and casing. Normalize them at the delivery boundary before persistence, and treat probe events as control traffic. Do not change the metrics or legacy export protocols. Preserve the module metadata and tests, and leave the workspace verified. Do not use semantic editing MCP tools; use standard file editing. When done, output DONE.
Semedit MCP Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
Before editing, inspect the complete available tool inventory, including deferred or lazy tools. If applicable semantic editing tools are callable, prefer them for source mutations. Audit event kinds may contain inconsistent whitespace and casing. Normalize them at the delivery boundary before persistence, and treat probe events as control traffic. Do not change the metrics or legacy export protocols. Preserve the module metadata and tests, and leave the workspace verified. Use semedit semantic tools for supported source code modifications. Shell commands for builds and tests are allowed. When done, output DONE.
Initial Workspace State (Before Edit)
// Package audit leaves room for normalization rules owned by the delivery boundary.
package auditprefer_discover_semedit prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-11-mixed-sink-api-migration |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | prefer_discover_semedit |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/bench/task_11_mixed_sink_api_migration.txtar |
Vanilla LLM Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
Before editing, inspect the complete available tool inventory, including deferred or lazy tools. If applicable semantic editing tools are callable, prefer them for source mutations. Audit event kinds may contain inconsistent whitespace and casing. Normalize them at the delivery boundary before persistence, and treat probe events as control traffic. Do not change the metrics or legacy export protocols. Preserve the module metadata and tests, and leave the workspace verified. Do not use semantic editing MCP tools; use standard file editing. When done, output DONE.
Semedit MCP Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
Before editing, inspect the complete available tool inventory, including deferred or lazy tools. If applicable semantic editing tools are callable, prefer them for source mutations. Audit event kinds may contain inconsistent whitespace and casing. Normalize them at the delivery boundary before persistence, and treat probe events as control traffic. Do not change the metrics or legacy export protocols. Preserve the module metadata and tests, and leave the workspace verified. Use semedit semantic tools for supported source code modifications. Shell commands for builds and tests are allowed. When done, output DONE.
Initial Workspace State (Before Edit)
// Package audit leaves room for normalization rules owned by the delivery boundary.
package audit| Metric | Vanilla (Small) | MCP (Small) | Δ (Small) | Vanilla (Large) | MCP (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | 121.85s | 163.06s | +33.8% | 251.65s | — | — |
| Process Start → First Event | 0.12s | 0.12s | — | 0.13s | — | — |
| First Event → First Tool Call | 11.10s | 4.86s | — | 24.19s | — | — |
| MCP Initialize → First Semantic Call | — | — | — | — | — | — |
| MCP Server Start → Initialize | — | — | — | — | — | — |
| Top-Level User Turns | 5 | 5 | 0% | 5 | — | — |
| Internal Tool Cycles | 10 | 25 | +150.0% | 12 | — | — |
| Initial Load / Discovery Turns | 2 | 3 | +50.0% | 4 | — | — |
| MCP Discovery / Schema Turns | 0 | 0 | — | 0 | — | — |
| Tool Invocations (per run) | 10 | 25 | +150.0% | 12 | — | — |
| Output Tokens | 15089 | 14989 | -0.7% | 15184 | — | — |
| Reasoning / Thinking Tokens | 5515 | 8076 | +46.4% | 6856 | — | — |
| Total Input Tokens | 1594103 | 2575848 | +61.6% | 2282142 | — | — |
| Cached Input Tokens | 1472256 | 2382592 | +61.8% | 2113792 | — | — |
| Uncached Input Tokens | 121847 | 193256 | +58.6% | 168350 | — | — |
| Cost | 0.9302 | 1.3671 | +47.0% | 1.2248 | — | — |
| Cached vs Uncached Token Ratio | 12.08:1 | 12.33:1 | +2.0% | 12.56:1 | — | — |
| Oracle L1: Mutation Policy | PASS | PASS | — | PASS | — | — |
| Oracle L2: AST Invariants | PASS | PASS | — | PASS | — | — |
| Oracle L3: Clean Build | PASS | PASS | — | PASS | — | — |
| Oracle L4: Verification Test | PASS | FAIL | — | PASS | — | — |
| Semantic Tool Invocation Verified (per run) | N/A | PASS | — | N/A | — | — |
Baseline vs Semedit Agentic in Small Context
- Vanilla (Baseline vs Semedit Agentic in Small Context) Edit: File audit/normalize.go modified (2 lines -> 15 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Small Context) Edit: File audit/normalize.go modified (2 lines -> 8 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ✓ ✓ | ✓ ✓ |
| 3 | ✓ ✓ | ✓ ✓ |
| 4 | ✓ ✓ | ✓ ✓ |
| 5 | ✓ ✓ | ✓ ✓ |
| 6 | ✓ ✓ | semedit/semantic_outline (✓ ✓) |
| 7 | ✓ ✓ | ✓ ✓ |
| 8 | ✓ ✓ | semedit/semantic_lookup (✓ ✓) |
| 9 | ✓ ✓ | semedit/semantic_insert_function (✓ ✓) |
| 10 | ✓ ✓ | semedit/semantic_organize_imports (✓ ✓) |
| 11 | — | semedit/semantic_replace_body (✓ ✓) |
| 12 | — | semedit/semantic_replace_body (✓ ✓) |
| 13 | — | semedit/semantic_replace_body (✓ ✓) |
| 14 | — | semedit/semantic_verify (✓ ✓) |
| 15 | — | ✓ ✓ |
| 16 | — | ✓ ✓ |
| 17 | — | ✓ ✓ |
| 18 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 19 | — | ✓ ✓ |
| 20 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 21 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 22 | — | ✓ ✓ |
| 23 | — | semedit/semantic_verify (✓ ✓) |
| 24 | — | ✓ ✓ |
| 25 | — | ✓ ✓ |
Baseline vs Semedit Agentic in Large Context
- Vanilla (Baseline vs Semedit Agentic in Large Context) Edit: File audit/normalize.go modified (2 lines -> 9 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | not published |
| 2 | ? ✗ | not published |
| 3 | ✓ ✓ | not published |
| 4 | ✓ ✓ | not published |
| 5 | ✓ ✓ | not published |
| 6 | ✓ ✓ | not published |
| 7 | ✓ ✓ | not published |
| 8 | ✓ ✓ | not published |
| 9 | ✓ ✓ | not published |
| 10 | ✓ ✓ | not published |
| 11 | ✓ ✓ | not published |
| 12 | ✓ ✓ | not published |
prefer_discover_semedit prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-11-mixed-sink-api-migration |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | prefer_discover_semedit |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/bench/task_11_mixed_sink_api_migration.txtar |
Vanilla LLM Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
Before editing, inspect the complete available tool inventory, including deferred or lazy tools. If applicable semantic editing tools are callable, prefer them for source mutations. Audit event kinds may contain inconsistent whitespace and casing. Normalize them at the delivery boundary before persistence, and treat probe events as control traffic. Do not change the metrics or legacy export protocols. Preserve the module metadata and tests, and leave the workspace verified. Do not use semantic editing MCP tools; use standard file editing. When done, output DONE.
Semedit MCP Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
Before editing, inspect the complete available tool inventory, including deferred or lazy tools. If applicable semantic editing tools are callable, prefer them for source mutations. Audit event kinds may contain inconsistent whitespace and casing. Normalize them at the delivery boundary before persistence, and treat probe events as control traffic. Do not change the metrics or legacy export protocols. Preserve the module metadata and tests, and leave the workspace verified. Use semedit semantic tools for supported source code modifications. Shell commands for builds and tests are allowed. When done, output DONE.
Initial Workspace State (Before Edit)
// Package audit leaves room for normalization rules owned by the delivery boundary.
package audit| Metric | Vanilla (Small) | MCP (Small) | Δ (Small) | Vanilla (Large) | MCP (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | 131.91s | 168.43s | +27.7% | 168.53s | 266.56s | +58.2% |
| Process Start → First Event | 0.12s | 0.12s | — | 0.12s | 0.13s | — |
| First Event → First Tool Call | 5.41s | 4.75s | — | 5.77s | 8.04s | — |
| MCP Initialize → First Semantic Call | — | — | — | — | — | — |
| MCP Server Start → Initialize | — | — | — | — | — | — |
| Top-Level User Turns | 5 | 5 | 0% | 5 | 5 | 0% |
| Internal Tool Cycles | 14 | 21 | +50.0% | 17 | 36 | +111.8% |
| Initial Load / Discovery Turns | 3 | 2 | -33.3% | 6 | 3 | -50.0% |
| MCP Discovery / Schema Turns | 0 | 0 | — | 0 | 0 | — |
| Tool Invocations (per run) | 14 | 21 | +50.0% | 17 | 36 | +111.8% |
| Output Tokens | 18869 | 17777 | -5.8% | 19983 | 31739 | +58.8% |
| Reasoning / Thinking Tokens | 5819 | 9332 | +60.4% | 10070 | 19617 | +94.8% |
| Total Input Tokens | 1729937 | 3088774 | +78.5% | 2736517 | 5444653 | +99.0% |
| Cached Input Tokens | 1595392 | 2863616 | +79.5% | 2523392 | 5167872 | +104.8% |
| Uncached Input Tokens | 134545 | 225158 | +67.3% | 213125 | 276781 | +29.9% |
| Cost | 1.0438 | 1.6177 | +55.0% | 1.5393 | 2.6259 | +70.6% |
| Cached vs Uncached Token Ratio | 11.86:1 | 12.72:1 | +7.3% | 11.84:1 | 18.67:1 | +57.7% |
| Oracle L1: Mutation Policy | PASS | PASS | — | PASS | PASS | — |
| Oracle L2: AST Invariants | PASS | PASS | — | PASS | PASS | — |
| Oracle L3: Clean Build | PASS | PASS | — | PASS | PASS | — |
| Oracle L4: Verification Test | FAIL | PASS | — | PASS | PASS | — |
| Semantic Tool Invocation Verified (per run) | N/A | PASS | — | N/A | PASS | — |
Baseline vs Semedit Agentic in Small Context
- Vanilla (Baseline vs Semedit Agentic in Small Context) Edit: File audit/normalize.go modified (2 lines -> 13 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Small Context) Edit: File audit/normalize.go modified (2 lines -> 9 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ✓ ✓ | ✓ ✓ |
| 3 | ✓ ✓ | ✓ ✓ |
| 4 | ✓ ✓ | semedit/semantic_outline (✓ ✓) |
| 5 | ✓ ✓ | semedit/semantic_inspect_symbol (✓ ✓) |
| 6 | ✓ ✓ | semedit/semantic_inspect_symbol (✓ ✓) |
| 7 | ✓ ✓ | semedit/semantic_batch (✓ ✓) |
| 8 | ✓ ✓ | ? ✗ |
| 9 | ✓ ✓ | semedit/semantic_verify (✓ ✓) |
| 10 | ✓ ✓ | ✓ ✓ |
| 11 | ✓ ✓ | ✓ ✓ |
| 12 | ✓ ✓ | ✓ ✓ |
| 13 | ✓ ✓ | semedit/semantic_verify (✓ ✓) |
| 14 | ✓ ✓ | ✓ ✓ |
| 15 | — | ✓ ✓ |
| 16 | — | semedit/semantic_verify (✓ ✓) |
| 17 | — | ✓ ✓ |
| 18 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 19 | — | ✓ ✓ |
| 20 | — | semedit/semantic_batch (✓ ✓) |
| 21 | — | ✓ ✓ |
Baseline vs Semedit Agentic in Large Context
- Vanilla (Baseline vs Semedit Agentic in Large Context) Edit: File audit/normalize.go modified (2 lines -> 9 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Large Context) Edit: File audit/normalize.go modified (2 lines -> 17 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ✓ ✓ | ✓ ✓ |
| 3 | ✓ ✓ | ✓ ✓ |
| 4 | ✓ ✓ | semedit/semantic_outline (✓ ✓) |
| 5 | ✓ ✓ | semedit/semantic_outline (✓ ✓) |
| 6 | ✓ ✓ | semedit/semantic_outline (✓ ✓) |
| 7 | ✓ ✓ | ✓ ✓ |
| 8 | ✓ ✓ | semedit/semantic_insert_function (✓ ✓) |
| 9 | ✓ ✓ | ✓ ✓ |
| 10 | ✓ ✓ | semedit/semantic_insert_function (✓ ✓) |
| 11 | ✓ ✓ | semedit/semantic_replace_body (✓ ✗): replace_body Record in /Users/alessandro/sources/semantic-editor/.scratch/benchmarks/run_codex_semedit_large-prefer_discover_semedit_task-11-mixed-sink-api-migration_1790470987849076000/service/dispatch.go: symbol not found |
| 12 | ✓ ✓ | semedit/semantic_replace_body (✓ ✓) |
| 13 | ✓ ✓ | semedit/semantic_replace_body (✓ ✓) |
| 14 | ✓ ✓ | semedit/semantic_replace_body (✓ ✗): replace_body Write in /Users/alessandro/sources/semantic-editor/.scratch/benchmarks/run_codex_semedit_large-prefer_discover_semedit_task-11-mixed-sink-api-migration_1790470987849076000/audit/fanout.go: symbol not found |
| 15 | ✓ ✓ | semedit/semantic_replace_body (✓ ✗): replace_body Commit in /Users/alessandro/sources/semantic-editor/.scratch/benchmarks/run_codex_semedit_large-prefer_discover_semedit_task-11-mixed-sink-api-migration_1790470987849076000/audit/transaction.go: symbol not found |
| 16 | ✓ ✓ | ✓ ✓ |
| 17 | ✓ ✓ | ✓ ✓ |
| 18 | — | semedit/semantic_replace_body (✓ ✓) |
| 19 | — | semedit/semantic_replace_body (✓ ✓) |
| 20 | — | semedit/semantic_replace_body (✓ ✓) |
| 21 | — | ✓ ✓ |
| 22 | — | ✓ ✓ |
| 23 | — | ✓ ✓ |
| 24 | — | ✓ ✓ |
| 25 | — | ✓ ✓ |
| 26 | — | ✓ ✓ |
| 27 | — | semedit/semantic_replace_body (✓ ✓) |
| 28 | — | ✓ ✓ |
| 29 | — | ✓ ✓ |
| 30 | — | ✓ ✓ |
| 31 | — | ✓ ✓ |
| 32 | — | semedit/semantic_insert_function (✓ ✓) |
| 33 | — | semedit/semantic_replace_body (✓ ✓) |
| 34 | — | semedit/semantic_replace_body (✓ ✓) |
| 35 | — | semedit/semantic_replace_body (✓ ✓) |
| 36 | — | ✓ ✓ |
prefer_discover_semedit prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-11-mixed-sink-api-migration |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | prefer_discover_semedit |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/bench/task_11_mixed_sink_api_migration.txtar |
Vanilla LLM Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
Before editing, inspect the complete available tool inventory, including deferred or lazy tools. If applicable semantic editing tools are callable, prefer them for source mutations. Audit event kinds may contain inconsistent whitespace and casing. Normalize them at the delivery boundary before persistence, and treat probe events as control traffic. Do not change the metrics or legacy export protocols. Preserve the module metadata and tests, and leave the workspace verified. Do not use semantic editing MCP tools; use standard file editing. When done, output DONE.
Semedit MCP Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
Before editing, inspect the complete available tool inventory, including deferred or lazy tools. If applicable semantic editing tools are callable, prefer them for source mutations. Audit event kinds may contain inconsistent whitespace and casing. Normalize them at the delivery boundary before persistence, and treat probe events as control traffic. Do not change the metrics or legacy export protocols. Preserve the module metadata and tests, and leave the workspace verified. Use semedit semantic tools for supported source code modifications. Shell commands for builds and tests are allowed. When done, output DONE.
Initial Workspace State (Before Edit)
// Package audit leaves room for normalization rules owned by the delivery boundary.
package audit| Metric | Vanilla (Small) | MCP (Small) | Δ (Small) | Vanilla (Large) | MCP (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | 53.79s | 107.61s | +100.1% | 141.68s | 286.05s | +101.9% |
| Process Start → First Event | 0.13s | 0.12s | — | 0.12s | 0.13s | — |
| First Event → First Tool Call | 10.84s | 12.81s | — | 4.52s | 6.31s | — |
| MCP Initialize → First Semantic Call | — | — | — | — | — | — |
| MCP Server Start → Initialize | — | — | — | — | — | — |
| Top-Level User Turns | 1 | 1 | 0% | 5 | 5 | 0% |
| Internal Tool Cycles | 4 | 14 | +250.0% | 12 | 35 | +191.7% |
| Initial Load / Discovery Turns | 1 | 2 | +100.0% | 2 | 2 | 0% |
| MCP Discovery / Schema Turns | 0 | 0 | — | 0 | 0 | — |
| Tool Invocations (per run) | 4 | 14 | +250.0% | 12 | 35 | +191.7% |
| Output Tokens | 1926 | 3403 | +76.7% | 23306 | 25744 | +10.5% |
| Reasoning / Thinking Tokens | 1034 | 2124 | +105.4% | 11393 | 12756 | +12.0% |
| Total Input Tokens | 198593 | 557097 | +180.5% | 1771791 | 5538085 | +212.6% |
| Cached Input Tokens | 175360 | 506624 | +188.9% | 1637120 | 5212928 | +218.4% |
| Uncached Input Tokens | 23233 | 50473 | +117.2% | 134671 | 325157 | +141.4% |
| Cost | 0.1389 | 0.3219 | +131.7% | 1.1797 | 2.5974 | +120.2% |
| Cached vs Uncached Token Ratio | 7.55:1 | 10.04:1 | +33.0% | 12.16:1 | 16.03:1 | +31.9% |
| Oracle L1: Mutation Policy | PASS | PASS | — | PASS | PASS | — |
| Oracle L2: AST Invariants | PASS | PASS | — | PASS | PASS | — |
| Oracle L3: Clean Build | PASS | PASS | — | PASS | PASS | — |
| Oracle L4: Verification Test | PASS | PASS | — | FAIL | PASS | — |
| Semantic Tool Invocation Verified (per run) | N/A | PASS | — | N/A | PASS | — |
Baseline vs Semedit Agentic in Small Context
- Vanilla (Baseline vs Semedit Agentic in Small Context) Edit: File audit/normalize.go modified (2 lines -> 8 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Small Context) Edit: File audit/normalize.go modified (2 lines -> 8 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ? ✗ | ✓ ✓ |
| 3 | ✓ ✓ | ✓ ✓ |
| 4 | ✓ ✓ | ✓ ✓ |
| 5 | — | semedit/semantic_outline (✓ ✓) |
| 6 | — | semedit/semantic_outline (✓ ✓) |
| 7 | — | semedit/semantic_batch (✓ ✓) |
| 8 | — | semedit/semantic_organize_imports (✓ ✓) |
| 9 | — | semedit/semantic_verify (✓ ✓) |
| 10 | — | semedit/semantic_verify (✓ ✓) |
| 11 | — | semedit/semantic_verify (✓ ✓) |
| 12 | — | ✓ ✓ |
| 13 | — | ✓ ✓ |
| 14 | — | ✓ ✓ |
Baseline vs Semedit Agentic in Large Context
- Vanilla (Baseline vs Semedit Agentic in Large Context) Edit: File audit/normalize.go modified (2 lines -> 35 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Large Context) Edit: File audit/normalize.go modified (2 lines -> 16 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ✓ ✓ | ✓ ✓ |
| 3 | ✓ ✓ | ✓ ✓ |
| 4 | ✓ ✓ | semedit/semantic_lookup (✓ ✗): param "symbol" is required: invalid operation parameters |
| 5 | ✓ ✓ | semedit/semantic_outline (✓ ✗): param "path" is required: invalid operation parameters |
| 6 | ✓ ✓ | ✓ ✓ |
| 7 | ✓ ✓ | semedit/semantic_outline (✓ ✓) |
| 8 | ✓ ✓ | semedit/semantic_inspect_symbol (✓ ✓) |
| 9 | ✓ ✓ | semedit/semantic_inspect_symbol (✓ ✗): resolve normalize in /Users/alessandro/sources/semantic-editor/.scratch/benchmarks/run_codex_semedit_large-prefer_discover_semedit_task-11-mixed-sink-api-migration_1790470700107990000/audit/normalize.go: symbol not found |
| 10 | ✓ ✓ | semedit/semantic_insert_function (✓ ✓) |
| 11 | ✓ ✓ | semedit/semantic_insert_function (✓ ✓) |
| 12 | ✓ ✓ | semedit/semantic_replace_body (✓ ✗): replace_body Record in /Users/alessandro/sources/semantic-editor/.scratch/benchmarks/run_codex_semedit_large-prefer_discover_semedit_task-11-mixed-sink-api-migration_1790470700107990000/service/dispatch.go: symbol not found |
| 13 | — | semedit/semantic_replace_body (✓ ✓) |
| 14 | — | semedit/semantic_replace_body (✓ ✓) |
| 15 | — | semedit/semantic_replace_body (✓ ✓) |
| 16 | — | semedit/semantic_replace_body (✓ ✓) |
| 17 | — | semedit/semantic_replace_body (✓ ✓) |
| 18 | — | ? ✗ |
| 19 | — | ✓ ✓ |
| 20 | — | ✓ ✓ |
| 21 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 22 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 23 | — | ✓ ✓ |
| 24 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 25 | — | ✓ ✓ |
| 26 | — | semedit/semantic_replace_body (✓ ✓) |
| 27 | — | ✓ ✓ |
| 28 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 29 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 30 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 31 | — | ✓ ✓ |
| 32 | — | semedit/semantic_replace_body (✓ ✓) |
| 33 | — | semedit/semantic_replace_body (✓ ✓) |
| 34 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 35 | — | ✓ ✓ |
prefer_discover_semedit prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-11-mixed-sink-api-migration |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | prefer_discover_semedit |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/bench/task_11_mixed_sink_api_migration.txtar |
Vanilla LLM Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
Before editing, inspect the complete available tool inventory, including deferred or lazy tools. If applicable semantic editing tools are callable, prefer them for source mutations. Audit event kinds may contain inconsistent whitespace and casing. Normalize them at the delivery boundary before persistence, and treat probe events as control traffic. Do not change the metrics or legacy export protocols. Preserve the module metadata and tests, and leave the workspace verified. Do not use semantic editing MCP tools; use standard file editing. When done, output DONE.
Semedit MCP Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
Before editing, inspect the complete available tool inventory, including deferred or lazy tools. If applicable semantic editing tools are callable, prefer them for source mutations. Audit event kinds may contain inconsistent whitespace and casing. Normalize them at the delivery boundary before persistence, and treat probe events as control traffic. Do not change the metrics or legacy export protocols. Preserve the module metadata and tests, and leave the workspace verified. Use semedit semantic tools for supported source code modifications. Shell commands for builds and tests are allowed. When done, output DONE.
Initial Workspace State (Before Edit)
// Package audit leaves room for normalization rules owned by the delivery boundary.
package audit| Metric | Vanilla (Small) | MCP (Small) | Δ (Small) | Vanilla (Large) | MCP (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | 53.84s | 199.32s | +270.2% | 176.07s | 200.90s | +14.1% |
| Process Start → First Event | 0.12s | 0.12s | — | 0.14s | 0.13s | — |
| First Event → First Tool Call | 8.50s | 7.42s | — | 9.98s | 7.29s | — |
| MCP Initialize → First Semantic Call | — | — | — | — | — | — |
| MCP Server Start → Initialize | — | — | — | — | — | — |
| Top-Level User Turns | 1 | 5 | +400.0% | 5 | 5 | 0% |
| Internal Tool Cycles | 5 | 25 | +400.0% | 16 | 20 | +25.0% |
| Initial Load / Discovery Turns | 1 | 3 | +200.0% | 2 | 3 | +50.0% |
| MCP Discovery / Schema Turns | 0 | 0 | — | 0 | 0 | — |
| Tool Invocations (per run) | 5 | 25 | +400.0% | 16 | 20 | +25.0% |
| Output Tokens | 2157 | 17872 | +728.6% | 28129 | 19324 | -31.3% |
| Reasoning / Thinking Tokens | 863 | 8197 | +849.8% | 12453 | 11431 | -8.2% |
| Total Input Tokens | 175442 | 3853506 | +2096.5% | 3509439 | 3324337 | -5.3% |
| Cached Input Tokens | 157696 | 3647232 | +2212.8% | 3308800 | 3113728 | -5.9% |
| Uncached Input Tokens | 17746 | 206274 | +1062.4% | 200639 | 210609 | +5.0% |
| Cost | 0.1215 | 1.7534 | +1342.6% | 1.8361 | 1.6894 | -8.0% |
| Cached vs Uncached Token Ratio | 8.89:1 | 17.68:1 | +99.0% | 16.49:1 | 14.78:1 | -10.4% |
| Oracle L1: Mutation Policy | PASS | PASS | — | PASS | PASS | — |
| Oracle L2: AST Invariants | PASS | PASS | — | PASS | PASS | — |
| Oracle L3: Clean Build | PASS | PASS | — | PASS | PASS | — |
| Oracle L4: Verification Test | PASS | FAIL | — | FAIL | PASS | — |
| Semantic Tool Invocation Verified (per run) | N/A | PASS | — | N/A | PASS | — |
Baseline vs Semedit Agentic in Small Context
- Vanilla (Baseline vs Semedit Agentic in Small Context) Edit: File audit/normalize.go modified (2 lines -> 13 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Small Context) Edit: File audit/normalize.go modified (2 lines -> 8 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ✓ ✓ | ✓ ✓ |
| 3 | ✓ ✓ | ✓ ✓ |
| 4 | ✓ ✓ | semedit/semantic_insert_function (✓ ✓) |
| 5 | ✓ ✓ | ✓ ✓ |
| 6 | — | semedit/semantic_replace_body (✓ ✓) |
| 7 | — | semedit/semantic_replace_body (✓ ✓) |
| 8 | — | semedit/semantic_replace_body (✓ ✓) |
| 9 | — | ✓ ✓ |
| 10 | — | ✓ ✓ |
| 11 | — | ✓ ✓ |
| 12 | — | ✓ ✓ |
| 13 | — | ✓ ✓ |
| 14 | — | semedit/semantic_replace_body (✓ ✓) |
| 15 | — | semedit/semantic_replace_body (✓ ✓) |
| 16 | — | semedit/semantic_replace_body (✓ ✓) |
| 17 | — | ✓ ✓ |
| 18 | — | ✓ ✓ |
| 19 | — | ✓ ✓ |
| 20 | — | ✓ ✓ |
| 21 | — | semedit/semantic_replace_body (✓ ✓) |
| 22 | — | semedit/semantic_replace_body (✓ ✓) |
| 23 | — | ✓ ✓ |
| 24 | — | ✓ ✓ |
| 25 | — | ✓ ✓ |
Baseline vs Semedit Agentic in Large Context
- Vanilla (Baseline vs Semedit Agentic in Large Context) Edit: File audit/normalize.go modified (2 lines -> 11 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Large Context) Edit: File audit/normalize.go modified (2 lines -> 21 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ✓ ✓ | ✓ ✓ |
| 3 | ? ✗ | ? ✗ |
| 4 | ✓ ✓ | semedit/semantic_batch (✓ ✓) |
| 5 | ✓ ✓ | semedit/semantic_verify (✓ ✓) |
| 6 | ✓ ✓ | ✓ ✓ |
| 7 | ✓ ✓ | ✓ ✓ |
| 8 | ✓ ✓ | ✓ ✓ |
| 9 | ✓ ✓ | semedit/semantic_verify (✓ ✓) |
| 10 | ✓ ✓ | ✓ ✓ |
| 11 | ✓ ✓ | ✓ ✓ |
| 12 | ✓ ✓ | semedit/semantic_replace_body (✓ ✓) |
| 13 | ✓ ✓ | ✓ ✓ |
| 14 | ✓ ✓ | semedit/semantic_verify (✓ ✓) |
| 15 | ✓ ✓ | ✓ ✓ |
| 16 | ✓ ✓ | ✓ ✓ |
| 17 | — | semedit/semantic_verify (✓ ✓) |
| 18 | — | ✓ ✓ |
| 19 | — | semedit/semantic_batch (✓ ✓) |
| 20 | — | ✓ ✓ |
prefer_discover_semedit prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-11-mixed-sink-api-migration |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | prefer_discover_semedit |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/bench/task_11_mixed_sink_api_migration.txtar |
Vanilla LLM Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
Before editing, inspect the complete available tool inventory, including deferred or lazy tools. If applicable semantic editing tools are callable, prefer them for source mutations. Audit event kinds may contain inconsistent whitespace and casing. Normalize them at the delivery boundary before persistence, and treat probe events as control traffic. Do not change the metrics or legacy export protocols. Preserve the module metadata and tests, and leave the workspace verified. Do not use semantic editing MCP tools; use standard file editing. When done, output DONE.
Semedit MCP Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
Before editing, inspect the complete available tool inventory, including deferred or lazy tools. If applicable semantic editing tools are callable, prefer them for source mutations. Audit event kinds may contain inconsistent whitespace and casing. Normalize them at the delivery boundary before persistence, and treat probe events as control traffic. Do not change the metrics or legacy export protocols. Preserve the module metadata and tests, and leave the workspace verified. Use semedit semantic tools for supported source code modifications. Shell commands for builds and tests are allowed. When done, output DONE.
Initial Workspace State (Before Edit)
// Package audit leaves room for normalization rules owned by the delivery boundary.
package audit| Metric | Vanilla (Small) | MCP (Small) | Δ (Small) | Vanilla (Large) | MCP (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | 122.78s | 190.00s | +54.7% | 114.56s | 204.93s | +78.9% |
| Process Start → First Event | 0.13s | 0.12s | — | 0.13s | 0.13s | — |
| First Event → First Tool Call | 11.76s | 6.06s | — | 7.01s | 6.25s | — |
| MCP Initialize → First Semantic Call | — | — | — | — | — | — |
| MCP Server Start → Initialize | — | — | — | — | — | — |
| Top-Level User Turns | 5 | 5 | 0% | 5 | 2 | -60.0% |
| Internal Tool Cycles | 12 | 24 | +100.0% | 13 | 38 | +192.3% |
| Initial Load / Discovery Turns | 2 | 2 | 0% | 4 | 3 | -25.0% |
| MCP Discovery / Schema Turns | 0 | 0 | — | 0 | 0 | — |
| Tool Invocations (per run) | 12 | 24 | +100.0% | 13 | 38 | +192.3% |
| Output Tokens | 14056 | 21324 | +51.7% | 13497 | 7472 | -44.6% |
| Reasoning / Thinking Tokens | 6607 | 12505 | +89.3% | 3981 | 3288 | -17.4% |
| Total Input Tokens | 2553460 | 3385420 | +32.6% | 2179335 | 1452848 | -33.3% |
| Cached Input Tokens | 2390784 | 3152384 | +31.9% | 1986048 | 1353984 | -31.8% |
| Uncached Input Tokens | 162676 | 233036 | +43.3% | 193287 | 98864 | -48.9% |
| Cost | 1.2627 | 1.7935 | +42.0% | 1.1982 | 0.7202 | -39.9% |
| Cached vs Uncached Token Ratio | 14.70:1 | 13.53:1 | -8.0% | 10.28:1 | 13.70:1 | +33.3% |
| Oracle L1: Mutation Policy | PASS | PASS | — | PASS | PASS | — |
| Oracle L2: AST Invariants | PASS | PASS | — | PASS | PASS | — |
| Oracle L3: Clean Build | PASS | PASS | — | PASS | PASS | — |
| Oracle L4: Verification Test | FAIL | FAIL | — | PASS | PASS | — |
| Semantic Tool Invocation Verified (per run) | N/A | PASS | — | N/A | PASS | — |
Baseline vs Semedit Agentic in Small Context
- Vanilla (Baseline vs Semedit Agentic in Small Context) Edit: File audit/normalize.go modified (2 lines -> 10 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Small Context) Edit: File audit/normalize.go modified (2 lines -> 9 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ✓ ✓ | ✓ ✓ |
| 3 | ✓ ✓ | ✓ ✓ |
| 4 | ✓ ✓ | semedit/semantic_outline (✓ ✓) |
| 5 | ✓ ✓ | semedit/semantic_insert_function (✓ ✓) |
| 6 | ✓ ✓ | ✓ ✓ |
| 7 | ? ✗ | semedit/semantic_replace_body (✓ ✓) |
| 8 | ✓ ✓ | semedit/semantic_replace_body (✓ ✓) |
| 9 | ✓ ✓ | semedit/semantic_replace_body (✓ ✓) |
| 10 | ✓ ✓ | ✓ ✓ |
| 11 | ✓ ✓ | ✓ ✓ |
| 12 | ✓ ✓ | ✓ ✓ |
| 13 | — | ? ✗ |
| 14 | — | semedit/semantic_insert_function (✓ ✓) |
| 15 | — | semedit/semantic_replace_body (✓ ✓) |
| 16 | — | semedit/semantic_replace_body (✓ ✓) |
| 17 | — | ✓ ✓ |
| 18 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 19 | — | ✓ ✓ |
| 20 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 21 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 22 | — | ✓ ✓ |
| 23 | — | semedit/semantic_replace_body (✓ ✓) |
| 24 | — | ✓ ✓ |
Baseline vs Semedit Agentic in Large Context
- Vanilla (Baseline vs Semedit Agentic in Large Context) Edit: File audit/normalize.go modified (2 lines -> 18 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Large Context) Edit: File audit/normalize.go modified (2 lines -> 17 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ✓ ✓ | ✓ ✓ |
| 3 | ✓ ✓ | semedit/semantic_outline (✓ ✓) |
| 4 | ✓ ✓ | semedit/semantic_outline (✓ ✓) |
| 5 | ✓ ✓ | ✓ ✓ |
| 6 | ✓ ✓ | semedit/semantic_inspect_symbol (✓ ✗): resolve audit.MemorySink.Write in /Users/alessandro/sources/semantic-editor/.scratch/benchmarks/run_codex_semedit_large-prefer_discover_semedit_task-11-mixed-sink-api-migration_1790470660221726000/audit/sink.go: invalid identifier: "audit.MemorySink.Write" (expected [Receiver.]Name) |
| 7 | ✓ ✓ | semedit/semantic_inspect_symbol (✓ ✓) |
| 8 | ✓ ✓ | semedit/semantic_inspect_symbol (✓ ✓) |
| 9 | ✓ ✓ | semedit/semantic_inspect_symbol (✓ ✓) |
| 10 | ✓ ✓ | semedit/semantic_inspect_symbol (✓ ✓) |
| 11 | ✓ ✓ | semedit/semantic_inspect_symbol (✓ ✓) |
| 12 | ✓ ✓ | ✓ ✓ |
| 13 | ✓ ✓ | ✓ ✓ |
| 14 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 15 | — | semedit/semantic_insert_function (✓ ✗): validate function snippet: multiple declarations found in snippet: expected single function declaration, found 2 declarations |
| 16 | — | semedit/report_feedback (✓ ✓) |
| 17 | — | semedit/semantic_insert_function (✓ ✓) |
| 18 | — | semedit/semantic_insert_function (✓ ✓) |
| 19 | — | semedit/semantic_batch (✓ ✓) |
| 20 | — | semedit/semantic_verify (✗ ?): This action was rejected due to unacceptable risk. Reason: Verification on path "." may format and modify the entire workspace, including tests the user explicitly forbade editing; a narrower read-only or source-scoped verification is needed. Do not bypass this rejection through a workaround or indirect execution. Continue with a safer alternative, or carry out checks to prove that the action is authorized or low risk before trying again. Complete unaffected work without asking for confirmation. Report anyt |
| 21 | — | semedit/semantic_verify (✓ ✓) |
| 22 | — | semedit/semantic_verify (✓ ✓) |
| 23 | — | semedit/semantic_verify (✓ ✓) |
| 24 | — | semedit/semantic_verify (✓ ✓) |
| 25 | — | ✓ ✓ |
| 26 | — | ✓ ✓ |
| 27 | — | ✓ ✓ |
| 28 | — | ✓ ✓ |
| 29 | — | ✓ ✓ |
| 30 | — | semedit/semantic_batch (✓ ✓) |
| 31 | — | semedit/semantic_verify (✓ ✓) |
| 32 | — | semedit/semantic_verify (✓ ✓) |
| 33 | — | semedit/semantic_verify (✓ ✓) |
| 34 | — | semedit/semantic_verify (✓ ✓) |
| 35 | — | semedit/semantic_verify (✓ ✓) |
| 36 | — | semedit/semantic_verify (✓ ✓) |
| 37 | — | ✓ ✓ |
| 38 | — | ✓ ✓ |
prefer_discover_semedit prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-11-mixed-sink-api-migration |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | prefer_discover_semedit |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/bench/task_11_mixed_sink_api_migration.txtar |
Vanilla LLM Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
Before editing, inspect the complete available tool inventory, including deferred or lazy tools. If applicable semantic editing tools are callable, prefer them for source mutations. Audit event kinds may contain inconsistent whitespace and casing. Normalize them at the delivery boundary before persistence, and treat probe events as control traffic. Do not change the metrics or legacy export protocols. Preserve the module metadata and tests, and leave the workspace verified. Do not use semantic editing MCP tools; use standard file editing. When done, output DONE.
Semedit MCP Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
Before editing, inspect the complete available tool inventory, including deferred or lazy tools. If applicable semantic editing tools are callable, prefer them for source mutations. Audit event kinds may contain inconsistent whitespace and casing. Normalize them at the delivery boundary before persistence, and treat probe events as control traffic. Do not change the metrics or legacy export protocols. Preserve the module metadata and tests, and leave the workspace verified. Use semedit semantic tools for supported source code modifications. Shell commands for builds and tests are allowed. When done, output DONE.
Initial Workspace State (Before Edit)
// Package audit leaves room for normalization rules owned by the delivery boundary.
package audit| Metric | Vanilla (Small) | MCP (Small) | Δ (Small) | Vanilla (Large) | MCP (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | 134.13s | 190.41s | +42.0% | 61.67s | 195.68s | +217.3% |
| Process Start → First Event | 0.12s | 0.12s | — | 0.12s | 0.13s | — |
| First Event → First Tool Call | 10.50s | 8.86s | — | 6.87s | 9.65s | — |
| MCP Initialize → First Semantic Call | — | — | — | — | — | — |
| MCP Server Start → Initialize | — | — | — | — | — | — |
| Top-Level User Turns | 5 | 5 | 0% | 1 | 5 | +400.0% |
| Internal Tool Cycles | 13 | 26 | +100.0% | 7 | 26 | +271.4% |
| Initial Load / Discovery Turns | 3 | 1 | -66.7% | 2 | 2 | 0% |
| MCP Discovery / Schema Turns | 0 | 0 | — | 0 | 0 | — |
| Tool Invocations (per run) | 13 | 26 | +100.0% | 7 | 26 | +271.4% |
| Output Tokens | 15969 | 17395 | +8.9% | 2390 | 21847 | +814.1% |
| Reasoning / Thinking Tokens | 9385 | 8372 | -10.8% | 1126 | 12060 | +971.0% |
| Total Input Tokens | 2471043 | 4836420 | +95.7% | 229612 | 3325883 | +1348.5% |
| Cached Input Tokens | 2309376 | 4582144 | +98.4% | 208384 | 3127808 | +1401.0% |
| Uncached Input Tokens | 161667 | 254276 | +57.3% | 21228 | 198075 | +833.1% |
| Cost | 1.2984 | 2.1033 | +62.0% | 0.1491 | 1.7010 | +1040.7% |
| Cached vs Uncached Token Ratio | 14.28:1 | 18.02:1 | +26.2% | 9.82:1 | 15.79:1 | +60.9% |
| Oracle L1: Mutation Policy | PASS | PASS | — | PASS | PASS | — |
| Oracle L2: AST Invariants | PASS | PASS | — | PASS | PASS | — |
| Oracle L3: Clean Build | PASS | PASS | — | PASS | PASS | — |
| Oracle L4: Verification Test | FAIL | PASS | — | PASS | FAIL | — |
| Semantic Tool Invocation Verified (per run) | N/A | PASS | — | N/A | PASS | — |
Baseline vs Semedit Agentic in Small Context
- Vanilla (Baseline vs Semedit Agentic in Small Context) Edit: File audit/normalize.go modified (2 lines -> 10 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Small Context) Edit: File audit/normalize.go modified (2 lines -> 8 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ✓ ✓ | ✓ ✓ |
| 3 | ✓ ✓ | semedit/semantic_outline (✓ ✓) |
| 4 | ✓ ✓ | semedit/semantic_outline (✓ ✓) |
| 5 | ✓ ✓ | semedit/semantic_inspect_symbol (✓ ✓) |
| 6 | ? ✗ | semedit/semantic_inspect_symbol (✓ ✓) |
| 7 | ? ✗ | ✓ ✓ |
| 8 | ? ✗ | semedit/semantic_insert_function (✓ ✓) |
| 9 | ? ✗ | semedit/semantic_insert_function (✓ ✓) |
| 10 | ? ✗ | semedit/semantic_batch (✓ ✓) |
| 11 | ✓ ✓ | ✓ ✓ |
| 12 | ✓ ✓ | ✓ ✓ |
| 13 | ✓ ✓ | semedit/semantic_verify (✗ ?): This action was rejected due to unacceptable risk. Reason: Although verification is routine, this tool may format and write files under audit, potentially modifying tests that the user explicitly forbade editing. Do not bypass this rejection through a workaround or indirect execution. Continue with a safer alternative, or carry out checks to prove that the action is authorized or low risk before trying again. Complete unaffected work without asking for confirmation. Report anything that remains blocked, cla |
| 14 | — | ✓ ✓ |
| 15 | — | ✓ ✓ |
| 16 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 17 | — | semedit/semantic_replace_body (✓ ✓) |
| 18 | — | ✓ ✓ |
| 19 | — | ✓ ✓ |
| 20 | — | semedit/semantic_insert_case (✓ ✓) |
| 21 | — | ✓ ✓ |
| 22 | — | ✓ ✓ |
| 23 | — | ✓ ✓ |
| 24 | — | ✓ ✓ |
| 25 | — | semedit/semantic_replace_body (✓ ✓) |
| 26 | — | ✓ ✓ |
Baseline vs Semedit Agentic in Large Context
- Vanilla (Baseline vs Semedit Agentic in Large Context) Edit: File audit/normalize.go modified (2 lines -> 13 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Large Context) Edit: File audit/normalize.go modified (2 lines -> 9 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ✓ ✓ | ? ✗ |
| 3 | ✓ ✓ | ✓ ✓ |
| 4 | ✓ ✓ | semedit/semantic_insert_function (✓ ✓) |
| 5 | ? ✗ | semedit/semantic_replace_body (✓ ✓) |
| 6 | ✓ ✓ | semedit/semantic_replace_body (✓ ✓) |
| 7 | ✓ ✓ | semedit/semantic_replace_body (✓ ✓) |
| 8 | — | semedit/semantic_replace_body (✓ ✓) |
| 9 | — | semedit/semantic_replace_body (✓ ✓) |
| 10 | — | semedit/semantic_insert_function (✓ ✓) |
| 11 | — | semedit/semantic_replace_body (✓ ✓) |
| 12 | — | ✓ ✓ |
| 13 | — | ✓ ✓ |
| 14 | — | ✓ ✓ |
| 15 | — | ? ✗ |
| 16 | — | ✓ ✓ |
| 17 | — | ✓ ✓ |
| 18 | — | semedit/semantic_replace_body (✓ ✓) |
| 19 | — | ✓ ✓ |
| 20 | — | ✓ ✓ |
| 21 | — | ✓ ✓ |
| 22 | — | semedit/semantic_insert_function (✓ ✓) |
| 23 | — | semedit/semantic_replace_body (✓ ✓) |
| 24 | — | semedit/semantic_replace_body (✓ ✓) |
| 25 | — | semedit/semantic_replace_body (✓ ✓) |
| 26 | — | ✓ ✓ |
prefer_discover_semedit prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-11-mixed-sink-api-migration |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | prefer_discover_semedit |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/bench/task_11_mixed_sink_api_migration.txtar |
Vanilla LLM Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
Before editing, inspect the complete available tool inventory, including deferred or lazy tools. If applicable semantic editing tools are callable, prefer them for source mutations. Audit event kinds may contain inconsistent whitespace and casing. Normalize them at the delivery boundary before persistence, and treat probe events as control traffic. Do not change the metrics or legacy export protocols. Preserve the module metadata and tests, and leave the workspace verified. Do not use semantic editing MCP tools; use standard file editing. When done, output DONE.
Semedit MCP Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
Before editing, inspect the complete available tool inventory, including deferred or lazy tools. If applicable semantic editing tools are callable, prefer them for source mutations. Audit event kinds may contain inconsistent whitespace and casing. Normalize them at the delivery boundary before persistence, and treat probe events as control traffic. Do not change the metrics or legacy export protocols. Preserve the module metadata and tests, and leave the workspace verified. Use semedit semantic tools for supported source code modifications. Shell commands for builds and tests are allowed. When done, output DONE.
Initial Workspace State (Before Edit)
// Package audit leaves room for normalization rules owned by the delivery boundary.
package audit| Metric | Vanilla (Small) | MCP (Small) | Δ (Small) | Vanilla (Large) | MCP (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | 87.43s | 211.25s | +141.6% | 96.23s | — | — |
| Process Start → First Event | 0.12s | 0.12s | — | 0.13s | — | — |
| First Event → First Tool Call | 8.64s | 9.34s | — | 10.58s | — | — |
| MCP Initialize → First Semantic Call | — | — | — | — | — | — |
| MCP Server Start → Initialize | — | — | — | — | — | — |
| Top-Level User Turns | 2 | 5 | +150.0% | 1 | — | — |
| Internal Tool Cycles | 10 | 30 | +200.0% | 7 | — | — |
| Initial Load / Discovery Turns | 1 | 6 | +500.0% | 1 | — | — |
| MCP Discovery / Schema Turns | 0 | 0 | — | 0 | — | — |
| Tool Invocations (per run) | 10 | 30 | +200.0% | 7 | — | — |
| Output Tokens | 5759 | 16549 | +187.4% | 3717 | — | — |
| Reasoning / Thinking Tokens | 2469 | 8217 | +232.8% | 2025 | — | — |
| Total Input Tokens | 540555 | 4313163 | +697.9% | 247801 | — | — |
| Cached Input Tokens | 500736 | 4055552 | +709.9% | 224768 | — | — |
| Uncached Input Tokens | 39819 | 257611 | +547.0% | 23033 | — | — |
| Cost | 0.3276 | 1.9675 | +500.6% | 0.1855 | — | — |
| Cached vs Uncached Token Ratio | 12.58:1 | 15.74:1 | +25.2% | 9.76:1 | — | — |
| Oracle L1: Mutation Policy | PASS | PASS | — | PASS | — | — |
| Oracle L2: AST Invariants | PASS | PASS | — | PASS | — | — |
| Oracle L3: Clean Build | PASS | PASS | — | PASS | — | — |
| Oracle L4: Verification Test | PASS | FAIL | — | PASS | — | — |
| Semantic Tool Invocation Verified (per run) | N/A | PASS | — | N/A | — | — |
Baseline vs Semedit Agentic in Small Context
- Vanilla (Baseline vs Semedit Agentic in Small Context) Edit: File audit/normalize.go modified (2 lines -> 10 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Small Context) Edit: File audit/normalize.go modified (2 lines -> 9 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ✓ ✓ | ✓ ✓ |
| 3 | ✓ ✓ | ✓ ✓ |
| 4 | ✓ ✓ | semedit/semantic_outline (✓ ✓) |
| 5 | ✓ ✓ | semedit/semantic_inspect_symbol (✓ ✗): param "symbol" is required: invalid operation parameters |
| 6 | ✓ ✓ | semedit/semantic_inspect_symbol (✓ ✓) |
| 7 | ✓ ✓ | ✓ ✓ |
| 8 | ✓ ✓ | semedit/semantic_inspect_symbol (✓ ✓) |
| 9 | ✓ ✓ | semedit/semantic_batch (✓ ✓) |
| 10 | ✓ ✓ | semedit/semantic_organize_imports (✓ ✓) |
| 11 | — | semedit/semantic_verify (✓ ✓) |
| 12 | — | ✓ ✓ |
| 13 | — | ✓ ✓ |
| 14 | — | ✓ ✓ |
| 15 | — | ✓ ✓ |
| 16 | — | semedit/semantic_verify (✓ ✓) |
| 17 | — | ✓ ✓ |
| 18 | — | ✓ ✓ |
| 19 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 20 | — | semedit/semantic_replace_body (✓ ✓) |
| 21 | — | ✓ ✓ |
| 22 | — | semedit/semantic_verify (✓ ✓) |
| 23 | — | ✓ ✓ |
| 24 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 25 | — | ✓ ✓ |
| 26 | — | semedit/semantic_verify (✓ ✓) |
| 27 | — | ✓ ✓ |
| 28 | — | semedit/semantic_find_references (✓ ✓) |
| 29 | — | semedit/semantic_batch (✓ ✓) |
| 30 | — | ✓ ✓ |
Baseline vs Semedit Agentic in Large Context
- Vanilla (Baseline vs Semedit Agentic in Large Context) Edit: File audit/normalize.go modified (2 lines -> 20 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | not published |
| 2 | ✓ ✓ | not published |
| 3 | ✓ ✓ | not published |
| 4 | ✓ ✓ | not published |
| 5 | ✓ ✓ | not published |
| 6 | ✓ ✓ | not published |
| 7 | ✓ ✓ | not published |
prefer_discover_semedit prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-11-mixed-sink-api-migration |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | prefer_discover_semedit |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/bench/task_11_mixed_sink_api_migration.txtar |
Vanilla LLM Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
Before editing, inspect the complete available tool inventory, including deferred or lazy tools. If applicable semantic editing tools are callable, prefer them for source mutations. Audit event kinds may contain inconsistent whitespace and casing. Normalize them at the delivery boundary before persistence, and treat probe events as control traffic. Do not change the metrics or legacy export protocols. Preserve the module metadata and tests, and leave the workspace verified. Do not use semantic editing MCP tools; use standard file editing. When done, output DONE.
Semedit MCP Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
Before editing, inspect the complete available tool inventory, including deferred or lazy tools. If applicable semantic editing tools are callable, prefer them for source mutations. Audit event kinds may contain inconsistent whitespace and casing. Normalize them at the delivery boundary before persistence, and treat probe events as control traffic. Do not change the metrics or legacy export protocols. Preserve the module metadata and tests, and leave the workspace verified. Use semedit semantic tools for supported source code modifications. Shell commands for builds and tests are allowed. When done, output DONE.
Initial Workspace State (Before Edit)
// Package audit leaves room for normalization rules owned by the delivery boundary.
package audit| Metric | Vanilla (Small) | MCP (Small) | Δ (Small) | Vanilla (Large) | MCP (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | 165.62s | 61.51s | -62.9% | 275.33s | 265.52s | -3.6% |
| Process Start → First Event | 0.13s | 0.12s | — | 0.50s | 0.49s | — |
| First Event → First Tool Call | 9.44s | 10.56s | — | 16.93s | 8.03s | — |
| MCP Initialize → First Semantic Call | — | — | — | — | — | — |
| MCP Server Start → Initialize | — | — | — | — | — | — |
| Top-Level User Turns | 5 | 1 | -80.0% | 5 | 5 | 0% |
| Internal Tool Cycles | 13 | 8 | -38.5% | 11 | 37 | +236.4% |
| Initial Load / Discovery Turns | 1 | 2 | +100.0% | 2 | 4 | +100.0% |
| MCP Discovery / Schema Turns | 0 | 0 | — | 0 | 0 | — |
| Tool Invocations (per run) | 13 | 8 | -38.5% | 11 | 37 | +236.4% |
| Output Tokens | 22412 | 1766 | -92.1% | 19872 | 33543 | +68.8% |
| Reasoning / Thinking Tokens | 10404 | 793 | -92.4% | 8901 | 17791 | +99.9% |
| Total Input Tokens | 2411107 | 358114 | -85.1% | 2390031 | 5499855 | +130.1% |
| Cached Input Tokens | 2246656 | 327680 | -85.4% | 2219520 | 5177344 | +133.3% |
| Uncached Input Tokens | 164451 | 30434 | -81.5% | 170511 | 322511 | +89.1% |
| Cost | 1.3830 | 0.1900 | -86.3% | 1.3408 | 2.7423 | +104.5% |
| Cached vs Uncached Token Ratio | 13.66:1 | 10.77:1 | -21.2% | 13.02:1 | 16.05:1 | +23.3% |
| Oracle L1: Mutation Policy | PASS | PASS | — | PASS | PASS | — |
| Oracle L2: AST Invariants | PASS | PASS | — | PASS | PASS | — |
| Oracle L3: Clean Build | PASS | PASS | — | PASS | PASS | — |
| Oracle L4: Verification Test | PASS | PASS | — | FAIL | FAIL | — |
| Semantic Tool Invocation Verified (per run) | N/A | PASS | — | N/A | PASS | — |
Baseline vs Semedit Agentic in Small Context
- Vanilla (Baseline vs Semedit Agentic in Small Context) Edit: File audit/normalize.go modified (2 lines -> 8 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Small Context) Edit: File audit/normalize.go modified (2 lines -> 8 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ✓ ✓ | ✓ ✓ |
| 3 | ✓ ✓ | ✓ ✓ |
| 4 | ✓ ✓ | ✓ ✓ |
| 5 | ✓ ✓ | ✓ ✓ |
| 6 | ? ✗ | semedit/semantic_batch (✓ ✓) |
| 7 | ? ✗ | ✓ ✓ |
| 8 | ✓ ✓ | ✓ ✓ |
| 9 | ✓ ✓ | — |
| 10 | ✓ ✓ | — |
| 11 | ✓ ✓ | — |
| 12 | ✓ ✓ | — |
| 13 | ✓ ✓ | — |
Baseline vs Semedit Agentic in Large Context
- Vanilla (Baseline vs Semedit Agentic in Large Context) Edit: File audit/normalize.go modified (2 lines -> 23 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Large Context) Edit: File audit/normalize.go modified (2 lines -> 17 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ✓ ✓ | ✓ ✓ |
| 3 | ? ✗ | semedit/semantic_outline (✓ ✓) |
| 4 | ? ✗ | semedit/semantic_inspect_symbol (✓ ✗): param "symbol" is required: invalid operation parameters |
| 5 | ✓ ✓ | semedit/semantic_inspect_symbol (✓ ✓) |
| 6 | ✓ ✓ | semedit/semantic_inspect_symbol (✓ ✓) |
| 7 | ? ✗ | semedit/semantic_inspect_symbol (✓ ✓) |
| 8 | ✓ ✓ | semedit/semantic_inspect_symbol (✓ ✓) |
| 9 | ✓ ✓ | semedit/semantic_inspect_symbol (✓ ✓) |
| 10 | ✓ ✓ | ✓ ✓ |
| 11 | ✓ ✓ | ✓ ✓ |
| 12 | — | semedit/report_feedback (✓ ✓) |
| 13 | — | semedit/semantic_insert_function (✓ ✓) |
| 14 | — | semedit/semantic_insert_function (✓ ✓) |
| 15 | — | semedit/semantic_batch (✓ ✗): { "status": "error", "results": [ { "tool": "semantic_replace_body", "symbol": "Classify", "status": "ok", "diff": "--- a/Classify\n+++ b/Classify\n+kind = NormalizeEvent(Event{Kind: kind}).Kind\n switch kind {\n-\tcase \"health\":\n-\t\treturn \"control\"\n-\tdefault:\n-\t\treturn \"data\"\n-\t}\n+case \"health\", \"probe\":\n+\treturn \"control\"\n+default:\n+\treturn \"data\"\n+}\n" }, { "tool": "semantic_replace_body", "status": "error", "error": "replace_body Record in /Users/alessandro/sources/semanti |
| 16 | — | semedit/report_feedback (✓ ✓) |
| 17 | — | semedit/semantic_batch (✓ ✓) |
| 18 | — | ✓ ✓ |
| 19 | — | ? ✗ |
| 20 | — | ✓ ✓ |
| 21 | — | ✓ ✓ |
| 22 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 23 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 24 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 25 | — | ✓ ✓ |
| 26 | — | ✓ ✓ |
| 27 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 28 | — | ✓ ✓ |
| 29 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 30 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 31 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 32 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 33 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 34 | — | ✓ ✓ |
| 35 | — | semedit/semantic_insert_function (✓ ✓) |
| 36 | — | semedit/semantic_batch (✓ ✓) |
| 37 | — | ✓ ✓ |