Benchmark run task02-corrected-overlay-r15-luna6-20261006
This page contains exactly the publishable benchmark observations recorded in run task02-corrected-overlay-r15-luna6-20261006; 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-02-rename-cross-pkg
Target: codex/gpt-6-luna/medium
explicit prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-02-rename-cross-pkg |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | explicit |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/bench/task_02_rename_cross_pkg.txtar |
Vanilla LLM Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
Rename public function auth.ValidateToken to auth.VerifyToken across all consuming packages. 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”.
Call semantic_lookup for ValidateToken in auth/token.go, then call semantic_rename using the returned file and symbol auth.ValidateToken with to= VerifyToken. Let it update all consuming packages; make no other changes. Parallelize independent read-only lookups. Perform mutations serially in the stated order or inside one ordered semantic_batch; never run overlapping mutations concurrently. 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 auth
func ValidateToken(t string) bool {
return t != ""
}| Metric | Vanilla (Small) | Semedit Explicit (Small) | Δ (Small) | Vanilla (Large) | Semedit Explicit (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | 35.06s | 44.28s | +26.3% | 22.96s | 20.24s | -11.8% |
| Process Start → First Event | 0.14s | 0.17s | — | 0.15s | 0.13s | — |
| First Event → First Tool Call | 13.07s | 13.01s | — | 4.52s | 5.21s | — |
| MCP Initialize → First Semantic Call | — | — | — | — | — | — |
| MCP Server Start → Initialize | — | — | — | — | — | — |
| Top-Level User Turns | 1 | 1 | 0% | 1 | 1 | 0% |
| Internal Tool Cycles | 4 | 3 | -25.0% | 4 | 2 | -50.0% |
| Initial Load / Discovery Turns | 2 | 0 | -100.0% | 2 | 0 | -100.0% |
| MCP Discovery / Schema Turns | 0 | 0 | — | 0 | 0 | — |
| Tool Invocations (per run) | 4 | 3 | -25.0% | 4 | 2 | -50.0% |
| Output Tokens | 680 | 570 | -16.2% | 575 | 234 | -59.3% |
| Reasoning / Thinking Tokens | 38 | 201 | +428.9% | 210 | 0 | -100.0% |
| Total Input Tokens | 101289 | 133719 | +32.0% | 101659 | 86385 | -15.0% |
| Cached Input Tokens | 92928 | 100096 | +7.7% | 86784 | 70656 | -18.6% |
| Uncached Input Tokens | 8361 | 33623 | +302.1% | 14875 | 15729 | +5.7% |
| Cost | 0.0531 | 0.1187 | +123.5% | 0.0687 | 0.0599 | -12.8% |
| Cached vs Uncached Token Ratio | 11.11:1 | 2.98:1 | -73.2% | 5.83:1 | 4.49:1 | -23.0% |
| 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 auth/token.go modified (5 lines -> 5 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Small Context) Edit: File auth/token.go modified (5 lines -> 5 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | semedit/semantic_lookup (✓ ✓) |
| 2 | ✓ ✓ | semedit/semantic_rename (✓ ✗): Critical result (JSON): {"application_state":"unknown","error":"resolve auth.ValidateToken in auth/token.go: symbol not found"} resolve auth.ValidateToken in auth/token.go: symbol not found |
| 3 | file_change (✓ ✓) | semedit/semantic_rename (✓ ✓) |
| 4 | ✓ ✓ | — |
Semedit integration feedback (agent opinion, excluded from measurements):
I would ask my owner to add semedit’s tools: the semantic rename updated the symbol across consuming packages with no introduced diagnostics, though my first qualified-name attempt failed and I had to retry with the lookup’s unqualified symbol.
Baseline vs Semedit Agentic in Large Context
- Vanilla (Baseline vs Semedit Agentic in Large Context) Edit: File auth/token.go modified (5 lines -> 5 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Large Context) Edit: File auth/token.go modified (5 lines -> 5 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | semedit/semantic_lookup (✓ ✓) |
| 2 | ? ✗ | semedit/semantic_rename (✓ ✓) |
| 3 | file_change (✓ ✓) | — |
| 4 | ✓ ✓ | — |
Semedit integration feedback (agent opinion, excluded from measurements):
I’d ask my owner to add semedit’s tools: the semantic lookup and rename made a cross-package symbol change feel direct and reliable, and I was glad not to rely on text replacement.
explicit prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-02-rename-cross-pkg |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | explicit |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/bench/task_02_rename_cross_pkg.txtar |
Vanilla LLM Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
Rename public function auth.ValidateToken to auth.VerifyToken across all consuming packages. 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”.
Call semantic_lookup for ValidateToken in auth/token.go, then call semantic_rename using the returned file and symbol auth.ValidateToken with to= VerifyToken. Let it update all consuming packages; make no other changes. Parallelize independent read-only lookups. Perform mutations serially in the stated order or inside one ordered semantic_batch; never run overlapping mutations concurrently. 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 auth
func ValidateToken(t string) bool {
return t != ""
}| Metric | Vanilla (Small) | Semedit Explicit (Small) | Δ (Small) | Vanilla (Large) | Semedit Explicit (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | 11.71s | 18.77s | +60.3% | 24.75s | 21.14s | -14.6% |
| Process Start → First Event | 0.13s | 0.16s | — | 0.14s | 0.14s | — |
| First Event → First Tool Call | 3.07s | 5.95s | — | 5.40s | 6.51s | — |
| MCP Initialize → First Semantic Call | — | — | — | — | — | — |
| MCP Server Start → Initialize | — | — | — | — | — | — |
| Top-Level User Turns | 1 | 1 | 0% | 1 | 1 | 0% |
| Internal Tool Cycles | 3 | 2 | -33.3% | 4 | 2 | -50.0% |
| Initial Load / Discovery Turns | 1 | 0 | -100.0% | 2 | 0 | -100.0% |
| MCP Discovery / Schema Turns | 0 | 0 | — | 0 | 0 | — |
| Tool Invocations (per run) | 3 | 2 | -33.3% | 4 | 2 | -50.0% |
| Output Tokens | 300 | 252 | -16.0% | 548 | 266 | -51.5% |
| Reasoning / Thinking Tokens | 47 | 7 | -85.1% | 180 | 0 | -100.0% |
| Total Input Tokens | 78243 | 104652 | +33.8% | 101686 | 87643 | -13.8% |
| Cached Input Tokens | 66560 | 88064 | +32.3% | 92928 | 71680 | -22.9% |
| Uncached Input Tokens | 11683 | 16588 | +42.0% | 8758 | 15963 | +82.3% |
| Cost | 0.0502 | 0.0667 | +33.0% | 0.0542 | 0.0612 | +12.8% |
| Cached vs Uncached Token Ratio | 5.70:1 | 5.31:1 | -6.8% | 10.61:1 | 4.49:1 | -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 | 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 auth/token.go modified (5 lines -> 5 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Small Context) Edit: File auth/token.go modified (5 lines -> 5 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | semedit/semantic_lookup (✓ ✓) |
| 2 | file_change (✓ ✓) | semedit/semantic_rename (✓ ✓) |
| 3 | ✓ ✓ | — |
Semedit integration feedback (agent opinion, excluded from measurements):
I’d ask my owner to add semedit’s tools: the lookup and cross-package rename gave me a direct, focused way to make the requested change, and seeing the rename complete with no new diagnostics was reassuring.
Baseline vs Semedit Agentic in Large Context
- Vanilla (Baseline vs Semedit Agentic in Large Context) Edit: File auth/token.go modified (5 lines -> 5 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Large Context) Edit: File auth/token.go modified (5 lines -> 5 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | semedit/semantic_lookup (✓ ✓) |
| 2 | ✓ ✓ | semedit/semantic_rename (✓ ✓) |
| 3 | file_change (✓ ✓) | — |
| 4 | ✓ ✓ | — |
Semedit integration feedback (agent opinion, excluded from measurements):
I’d ask my owner to add semedit’s tools: the rename made a cross-package change in one call and reported clean diagnostics, which felt reassuringly direct.
explicit prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-02-rename-cross-pkg |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | explicit |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/bench/task_02_rename_cross_pkg.txtar |
Vanilla LLM Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
Rename public function auth.ValidateToken to auth.VerifyToken across all consuming packages. 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”.
Call semantic_lookup for ValidateToken in auth/token.go, then call semantic_rename using the returned file and symbol auth.ValidateToken with to= VerifyToken. Let it update all consuming packages; make no other changes. Parallelize independent read-only lookups. Perform mutations serially in the stated order or inside one ordered semantic_batch; never run overlapping mutations concurrently. 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 auth
func ValidateToken(t string) bool {
return t != ""
}| Metric | Vanilla (Small) | Semedit Explicit (Small) | Δ (Small) | Vanilla (Large) | Semedit Explicit (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | 13.89s | 32.52s | +134.1% | 26.27s | 39.12s | +48.9% |
| Process Start → First Event | 0.14s | 0.14s | — | 0.15s | 0.15s | — |
| First Event → First Tool Call | 4.53s | 6.41s | — | 5.58s | 7.33s | — |
| MCP Initialize → First Semantic Call | — | — | — | — | — | — |
| MCP Server Start → Initialize | — | — | — | — | — | — |
| Top-Level User Turns | 1 | 1 | 0% | 1 | 1 | 0% |
| Internal Tool Cycles | 3 | 3 | 0% | 4 | 4 | 0% |
| Initial Load / Discovery Turns | 2 | 0 | -100.0% | 2 | 0 | -100.0% |
| MCP Discovery / Schema Turns | 0 | 0 | — | 0 | 0 | — |
| Tool Invocations (per run) | 3 | 3 | 0% | 4 | 4 | 0% |
| Output Tokens | 512 | 337 | -34.2% | 690 | 655 | -5.1% |
| Reasoning / Thinking Tokens | 0 | 118 | — | 0 | 116 | — |
| Total Input Tokens | 78598 | 133392 | +69.7% | 101314 | 194354 | +91.8% |
| Cached Input Tokens | 71680 | 116480 | +62.5% | 86784 | 170240 | +96.2% |
| Uncached Input Tokens | 6918 | 16912 | +144.5% | 14530 | 24114 | +66.0% |
| Cost | 0.0416 | 0.0771 | +85.2% | 0.0666 | 0.1125 | +68.8% |
| Cached vs Uncached Token Ratio | 10.36:1 | 6.89:1 | -33.5% | 5.97:1 | 7.06:1 | +18.2% |
| 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 auth/token.go modified (5 lines -> 5 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Small Context) Edit: File auth/token.go modified (5 lines -> 5 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | semedit/semantic_lookup (✓ ✓) |
| 2 | ✓ ✓ | semedit/semantic_rename (✓ ✗): Critical result (JSON): {"application_state":"unknown","error":"resolve auth.ValidateToken in auth/token.go: symbol not found"} resolve auth.ValidateToken in auth/token.go: symbol not found |
| 3 | file_change (✓ ✓) | semedit/semantic_rename (✓ ✓) |
Semedit integration feedback (agent opinion, excluded from measurements):
I’d ask my owner to add semedit’s tools: the lookup and cross-package rename made the requested change feel direct and controlled, though I had to correct the rename’s symbol name after my first attempt failed.
Baseline vs Semedit Agentic in Large Context
- Vanilla (Baseline vs Semedit Agentic in Large Context) Edit: File auth/token.go modified (5 lines -> 5 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Large Context) Edit: File auth/token.go modified (5 lines -> 5 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | semedit/semantic_lookup (✓ ✓) |
| 2 | ✓ ✓ | semedit/semantic_rename (✓ ✗): Critical result (JSON): {"application_state":"unknown","error":"resolve auth.ValidateToken in auth/token.go: symbol not found"} resolve auth.ValidateToken in auth/token.go: symbol not found |
| 3 | file_change (✓ ✓) | semedit/semantic_lookup (✓ ✗): Critical result (JSON): {"application_state":"unknown","error":"resolve auth.ValidateToken in auth/token.go: symbol not found"} resolve auth.ValidateToken in auth/token.go: symbol not found |
| 4 | ✓ ✓ | semedit/semantic_rename (✓ ✓) |
Semedit integration feedback (agent opinion, excluded from measurements):
Semedit’s semantic rename saved me from a risky cross-package text replacement and verified the result, so I’d ask my owner to add its tools.
explicit prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-02-rename-cross-pkg |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | explicit |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/bench/task_02_rename_cross_pkg.txtar |
Vanilla LLM Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
Rename public function auth.ValidateToken to auth.VerifyToken across all consuming packages. 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”.
Call semantic_lookup for ValidateToken in auth/token.go, then call semantic_rename using the returned file and symbol auth.ValidateToken with to= VerifyToken. Let it update all consuming packages; make no other changes. Parallelize independent read-only lookups. Perform mutations serially in the stated order or inside one ordered semantic_batch; never run overlapping mutations concurrently. 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 auth
func ValidateToken(t string) bool {
return t != ""
}| Metric | Vanilla (Small) | Semedit Explicit (Small) | Δ (Small) | Vanilla (Large) | Semedit Explicit (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | 31.67s | 22.89s | -27.7% | 22.10s | 72.15s | +226.4% |
| Process Start → First Event | 0.23s | 0.15s | — | 0.14s | 0.15s | — |
| First Event → First Tool Call | 12.68s | 5.86s | — | 7.57s | 31.92s | — |
| MCP Initialize → First Semantic Call | — | — | — | — | — | — |
| MCP Server Start → Initialize | — | — | — | — | — | — |
| Top-Level User Turns | 1 | 1 | 0% | 1 | 1 | 0% |
| Internal Tool Cycles | 4 | 2 | -50.0% | 5 | 2 | -60.0% |
| Initial Load / Discovery Turns | 2 | 0 | -100.0% | 4 | 0 | -100.0% |
| MCP Discovery / Schema Turns | 0 | 0 | — | 0 | 0 | — |
| Tool Invocations (per run) | 4 | 2 | -50.0% | 5 | 2 | -60.0% |
| Output Tokens | 627 | 232 | -63.0% | 356 | 246 | -30.9% |
| Reasoning / Thinking Tokens | 0 | 0 | — | 0 | 0 | — |
| Total Input Tokens | 98690 | 86394 | -12.5% | 100970 | 104628 | +3.6% |
| Cached Input Tokens | 90880 | 75776 | -16.6% | 92928 | 88064 | -5.2% |
| Uncached Input Tokens | 7810 | 10618 | +36.0% | 8042 | 16564 | +106.0% |
| Cost | 0.0501 | 0.0484 | -3.4% | 0.0478 | 0.0665 | +39.2% |
| Cached vs Uncached Token Ratio | 11.64:1 | 7.14:1 | -38.7% | 11.56:1 | 5.32:1 | -54.0% |
| 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 auth/token.go modified (5 lines -> 5 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Small Context) Edit: File auth/token.go modified (5 lines -> 5 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | semedit/semantic_lookup (✓ ✓) |
| 2 | ✓ ✓ | semedit/semantic_rename (✓ ✓) |
| 3 | file_change (✓ ✓) | — |
| 4 | ✓ ✓ | — |
Semedit integration feedback (agent opinion, excluded from measurements):
I’d ask my owner to add semedit: its semantic rename updated the requested symbol across packages and reported no new diagnostics, which made this small refactor feel reassuringly direct.
Baseline vs Semedit Agentic in Large Context
- Vanilla (Baseline vs Semedit Agentic in Large Context) Edit: File auth/token.go modified (5 lines -> 5 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Large Context) Edit: File auth/token.go modified (5 lines -> 5 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | semedit/semantic_lookup (✓ ✓) |
| 2 | ✓ ✓ | semedit/semantic_rename (✓ ✓) |
| 3 | ✓ ✓ | — |
| 4 | ✓ ✓ | — |
| 5 | file_change (✓ ✓) | — |
Semedit integration feedback (agent opinion, excluded from measurements):
Using semantic lookup and rename let me find and update `ValidateToken` across its consumers with a clear success result, so I’d ask my owner to add semedit’s tools to my coding setup.
explicit prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-02-rename-cross-pkg |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | explicit |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/bench/task_02_rename_cross_pkg.txtar |
Vanilla LLM Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
Rename public function auth.ValidateToken to auth.VerifyToken across all consuming packages. 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”.
Call semantic_lookup for ValidateToken in auth/token.go, then call semantic_rename using the returned file and symbol auth.ValidateToken with to= VerifyToken. Let it update all consuming packages; make no other changes. Parallelize independent read-only lookups. Perform mutations serially in the stated order or inside one ordered semantic_batch; never run overlapping mutations concurrently. 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 auth
func ValidateToken(t string) bool {
return t != ""
}| Metric | Vanilla (Small) | Semedit Explicit (Small) | Δ (Small) | Vanilla (Large) | Semedit Explicit (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | 30.24s | 47.09s | +55.7% | 20.11s | 52.27s | +160.0% |
| Process Start → First Event | 0.14s | 0.14s | — | 0.14s | 0.16s | — |
| First Event → First Tool Call | 13.78s | 22.77s | — | 7.64s | 21.47s | — |
| MCP Initialize → First Semantic Call | — | — | — | — | — | — |
| MCP Server Start → Initialize | — | — | — | — | — | — |
| Top-Level User Turns | 1 | 1 | 0% | 1 | 1 | 0% |
| Internal Tool Cycles | 3 | 3 | 0% | 3 | 2 | -33.3% |
| Initial Load / Discovery Turns | 2 | 0 | -100.0% | 2 | 0 | -100.0% |
| MCP Discovery / Schema Turns | 0 | 0 | — | 0 | 0 | — |
| Tool Invocations (per run) | 3 | 3 | 0% | 3 | 2 | -33.3% |
| Output Tokens | 523 | 665 | +27.2% | 271 | 240 | -11.4% |
| Reasoning / Thinking Tokens | 0 | 194 | — | 38 | 0 | -100.0% |
| Total Input Tokens | 78675 | 133915 | +70.2% | 79823 | 104625 | +31.1% |
| Cached Input Tokens | 58368 | 111360 | +90.8% | 72704 | 88064 | +21.1% |
| Uncached Input Tokens | 20307 | 22555 | +11.1% | 7119 | 16561 | +132.6% |
| Cost | 0.0719 | 0.0950 | +32.1% | 0.0398 | 0.0664 | +66.7% |
| Cached vs Uncached Token Ratio | 2.87:1 | 4.94:1 | +71.8% | 10.21:1 | 5.32:1 | -47.9% |
| 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 auth/token.go modified (5 lines -> 5 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Small Context) Edit: File auth/token.go modified (5 lines -> 5 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | semedit/semantic_lookup (✓ ✓) |
| 2 | ✓ ✓ | semedit/semantic_rename (✓ ✗): Critical result (JSON): {"application_state":"unknown","error":"resolve auth.ValidateToken in auth/token.go: symbol not found"} resolve auth.ValidateToken in auth/token.go: symbol not found |
| 3 | file_change (✓ ✓) | semedit/semantic_rename (✓ ✓) |
Semedit integration feedback (agent opinion, excluded from measurements):
Semedit’s tools did the cross package rename with no diagnostics, which made the edit feel reassuringly precise; I’d ask my owner to add them.
Baseline vs Semedit Agentic in Large Context
- Vanilla (Baseline vs Semedit Agentic in Large Context) Edit: File auth/token.go modified (5 lines -> 5 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Large Context) Edit: File auth/token.go modified (5 lines -> 5 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | semedit/semantic_lookup (✓ ✓) |
| 2 | ✓ ✓ | semedit/semantic_rename (✓ ✓) |
| 3 | file_change (✓ ✓) | — |
Semedit integration feedback (agent opinion, excluded from measurements):
Using `semantic_lookup` to anchor a workspace-wide `semantic_rename` felt reassuringly precise, and I’d ask my owner to add semedit’s tools for coding tasks like this.
explicit prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-02-rename-cross-pkg |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | explicit |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/bench/task_02_rename_cross_pkg.txtar |
Vanilla LLM Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
Rename public function auth.ValidateToken to auth.VerifyToken across all consuming packages. 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”.
Call semantic_lookup for ValidateToken in auth/token.go, then call semantic_rename using the returned file and symbol auth.ValidateToken with to= VerifyToken. Let it update all consuming packages; make no other changes. Parallelize independent read-only lookups. Perform mutations serially in the stated order or inside one ordered semantic_batch; never run overlapping mutations concurrently. 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 auth
func ValidateToken(t string) bool {
return t != ""
}| Metric | Vanilla (Small) | Semedit Explicit (Small) | Δ (Small) | Vanilla (Large) | Semedit Explicit (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | 22.59s | 31.69s | +40.3% | 23.01s | 47.90s | +108.1% |
| Process Start → First Event | 0.13s | 0.14s | — | 0.13s | 0.14s | — |
| First Event → First Tool Call | 4.35s | 4.34s | — | 5.46s | 9.41s | — |
| MCP Initialize → First Semantic Call | — | — | — | — | — | — |
| MCP Server Start → Initialize | — | — | — | — | — | — |
| Top-Level User Turns | 1 | 1 | 0% | 1 | 1 | 0% |
| Internal Tool Cycles | 5 | 3 | -40.0% | 5 | 3 | -40.0% |
| Initial Load / Discovery Turns | 3 | 0 | -100.0% | 3 | 0 | -100.0% |
| MCP Discovery / Schema Turns | 0 | 0 | — | 0 | 0 | — |
| Tool Invocations (per run) | 5 | 3 | -40.0% | 5 | 3 | -40.0% |
| Output Tokens | 558 | 420 | -24.7% | 420 | 902 | +114.8% |
| Reasoning / Thinking Tokens | 180 | 207 | +15.0% | 0 | 483 | — |
| Total Input Tokens | 100853 | 133509 | +32.4% | 119399 | 134243 | +12.4% |
| Cached Input Tokens | 92928 | 116480 | +25.3% | 110080 | 116480 | +5.8% |
| Uncached Input Tokens | 7925 | 17029 | +114.9% | 9319 | 17763 | +90.6% |
| Cost | 0.0523 | 0.0795 | +52.2% | 0.0561 | 0.0908 | +62.0% |
| Cached vs Uncached Token Ratio | 11.73:1 | 6.84:1 | -41.7% | 11.81:1 | 6.56:1 | -44.5% |
| 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 auth/token.go modified (5 lines -> 5 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Small Context) Edit: File auth/token.go modified (5 lines -> 5 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | semedit/semantic_lookup (✓ ✓) |
| 2 | ✓ ✓ | semedit/semantic_rename (✓ ✗): Critical result (JSON): {"application_state":"unknown","error":"resolve auth.ValidateToken in auth/token.go: symbol not found"} resolve auth.ValidateToken in auth/token.go: symbol not found |
| 3 | ✓ ✓ | semedit/semantic_rename (✓ ✓) |
| 4 | file_change (✓ ✓) | — |
| 5 | ✓ ✓ | — |
Semedit integration feedback (agent opinion, excluded from measurements):
I would ask my owner to add semedit’s tools: the rename propagated across consuming packages, though I had to correct the symbol qualification after the first rename call failed.
Baseline vs Semedit Agentic in Large Context
- Vanilla (Baseline vs Semedit Agentic in Large Context) Edit: File auth/token.go modified (5 lines -> 5 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Large Context) Edit: File auth/token.go modified (5 lines -> 5 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | semedit/semantic_lookup (✓ ✓) |
| 2 | ✓ ✓ | semedit/semantic_rename (✓ ✗): Critical result (JSON): {"application_state":"unknown","error":"resolve auth.ValidateToken in auth/token.go: symbol not found"} resolve auth.ValidateToken in auth/token.go: symbol not found |
| 3 | ✓ ✓ | semedit/semantic_rename (✓ ✓) |
| 4 | file_change (✓ ✓) | — |
| 5 | ✓ ✓ | — |
Semedit integration feedback (agent opinion, excluded from measurements):
I’d ask my owner to add semedit’s tools: the semantic rename updated references across packages and verified cleanly, though I had to retry because the qualified symbol name didn’t resolve.
explicit prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-02-rename-cross-pkg |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | explicit |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/bench/task_02_rename_cross_pkg.txtar |
Vanilla LLM Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
Rename public function auth.ValidateToken to auth.VerifyToken across all consuming packages. 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”.
Call semantic_lookup for ValidateToken in auth/token.go, then call semantic_rename using the returned file and symbol auth.ValidateToken with to= VerifyToken. Let it update all consuming packages; make no other changes. Parallelize independent read-only lookups. Perform mutations serially in the stated order or inside one ordered semantic_batch; never run overlapping mutations concurrently. 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 auth
func ValidateToken(t string) bool {
return t != ""
}| Metric | Vanilla (Small) | Semedit Explicit (Small) | Δ (Small) | Vanilla (Large) | Semedit Explicit (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | 16.64s | 23.07s | +38.7% | 26.56s | 27.63s | +4.0% |
| Process Start → First Event | 0.14s | 0.14s | — | 0.14s | 0.16s | — |
| First Event → First Tool Call | 4.33s | 10.33s | — | 5.45s | 4.21s | — |
| MCP Initialize → First Semantic Call | — | — | — | — | — | — |
| MCP Server Start → Initialize | — | — | — | — | — | — |
| Top-Level User Turns | 1 | 1 | 0% | 1 | 1 | 0% |
| Internal Tool Cycles | 3 | 2 | -33.3% | 4 | 2 | -50.0% |
| Initial Load / Discovery Turns | 1 | 0 | -100.0% | 2 | 0 | -100.0% |
| MCP Discovery / Schema Turns | 0 | 0 | — | 0 | 0 | — |
| Tool Invocations (per run) | 3 | 2 | -33.3% | 4 | 2 | -50.0% |
| Output Tokens | 318 | 177 | -44.3% | 755 | 242 | -67.9% |
| Reasoning / Thinking Tokens | 0 | 0 | — | 94 | 0 | -100.0% |
| Total Input Tokens | 78400 | 104485 | +33.3% | 101424 | 86409 | -14.8% |
| Cached Input Tokens | 71680 | 88064 | +22.9% | 92928 | 75776 | -18.5% |
| Uncached Input Tokens | 6720 | 16421 | +144.4% | 8496 | 10633 | +25.2% |
| Cost | 0.0387 | 0.0653 | +68.7% | 0.0551 | 0.0486 | -11.9% |
| Cached vs Uncached Token Ratio | 10.67:1 | 5.36:1 | -49.7% | 10.94:1 | 7.13:1 | -34.8% |
| 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 auth/token.go modified (5 lines -> 5 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Small Context) Edit: File auth/token.go modified (5 lines -> 5 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | semedit/semantic_lookup (✓ ✓) |
| 2 | file_change (✓ ✓) | semedit/semantic_rename (✓ ✓) |
| 3 | ✓ ✓ | — |
Semedit integration feedback (agent opinion, excluded from measurements):
I’d ask my owner to add semedit’s tools: the semantic lookup and rename made the cross package change feel reassuringly direct, and the rename reported no new diagnostics.
Baseline vs Semedit Agentic in Large Context
- Vanilla (Baseline vs Semedit Agentic in Large Context) Edit: File auth/token.go modified (5 lines -> 5 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Large Context) Edit: File auth/token.go modified (5 lines -> 5 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | semedit/semantic_lookup (✓ ✓) |
| 2 | ✓ ✓ | semedit/semantic_rename (✓ ✓) |
| 3 | file_change (✓ ✓) | — |
| 4 | ✓ ✓ | — |
Semedit integration feedback (agent opinion, excluded from measurements):
I’d ask my owner to add semedit’s tools: the lookup pinpointed `ValidateToken` and the rename updated consuming packages in one step, which felt reassuringly direct.
explicit prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-02-rename-cross-pkg |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | explicit |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/bench/task_02_rename_cross_pkg.txtar |
Vanilla LLM Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
Rename public function auth.ValidateToken to auth.VerifyToken across all consuming packages. 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”.
Call semantic_lookup for ValidateToken in auth/token.go, then call semantic_rename using the returned file and symbol auth.ValidateToken with to= VerifyToken. Let it update all consuming packages; make no other changes. Parallelize independent read-only lookups. Perform mutations serially in the stated order or inside one ordered semantic_batch; never run overlapping mutations concurrently. 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 auth
func ValidateToken(t string) bool {
return t != ""
}| Metric | Vanilla (Small) | Semedit Explicit (Small) | Δ (Small) | Vanilla (Large) | Semedit Explicit (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | 14.20s | 43.87s | +209.0% | 25.82s | 27.87s | +8.0% |
| Process Start → First Event | 0.13s | 0.15s | — | 0.13s | 0.17s | — |
| First Event → First Tool Call | 4.45s | 5.18s | — | 5.85s | 8.13s | — |
| MCP Initialize → First Semantic Call | — | — | — | — | — | — |
| MCP Server Start → Initialize | — | — | — | — | — | — |
| Top-Level User Turns | 1 | 1 | 0% | 1 | 1 | 0% |
| Internal Tool Cycles | 2 | 4 | +100.0% | 5 | 2 | -60.0% |
| Initial Load / Discovery Turns | 1 | 0 | -100.0% | 3 | 0 | -100.0% |
| MCP Discovery / Schema Turns | 0 | 0 | — | 0 | 0 | — |
| Tool Invocations (per run) | 2 | 4 | +100.0% | 5 | 2 | -60.0% |
| Output Tokens | 226 | 789 | +249.1% | 490 | 279 | -43.1% |
| Reasoning / Thinking Tokens | 27 | 310 | +1048.1% | 78 | 0 | -100.0% |
| Total Input Tokens | 58495 | 164705 | +181.6% | 121431 | 134596 | +10.8% |
| Cached Input Tokens | 47360 | 145920 | +208.1% | 112128 | 117504 | +4.8% |
| Uncached Input Tokens | 11135 | 18785 | +68.7% | 9303 | 17092 | +83.7% |
| Cost | 0.0428 | 0.0972 | +126.8% | 0.0584 | 0.0756 | +29.5% |
| Cached vs Uncached Token Ratio | 4.25:1 | 7.77:1 | +82.6% | 12.05:1 | 6.87:1 | -43.0% |
| 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 auth/token.go modified (5 lines -> 5 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Small Context) Edit: File auth/token.go modified (5 lines -> 5 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | semedit/semantic_lookup (✓ ✓) |
| 2 | file_change (✓ ✓) | semedit/semantic_rename (✓ ✗): Critical result (JSON): {"application_state":"unknown","error":"resolve auth.ValidateToken in auth/token.go: symbol not found"} resolve auth.ValidateToken in auth/token.go: symbol not found |
| 3 | — | semedit/semantic_inspect_symbol (✓ ✓) |
| 4 | — | semedit/semantic_rename (✓ ✓) |
Semedit integration feedback (agent opinion, excluded from measurements):
Semedit’s semantic tools made the cross-package rename feel precise and low-risk, and I’d ask my owner to add them because they changed the symbol and its consumers without introducing diagnostics.
Baseline vs Semedit Agentic in Large Context
- Vanilla (Baseline vs Semedit Agentic in Large Context) Edit: File auth/token.go modified (5 lines -> 5 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Large Context) Edit: File auth/token.go modified (5 lines -> 5 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | semedit/semantic_lookup (✓ ✓) |
| 2 | ✓ ✓ | semedit/semantic_rename (✓ ✓) |
| 3 | ✓ ✓ | — |
| 4 | file_change (✓ ✓) | — |
| 5 | ✓ ✓ | — |
Semedit integration feedback (agent opinion, excluded from measurements):
Semedit’s lookup and rename tools made this cross-package refactor feel reliable and pleasantly low effort, so I’d ask my owner to add them.
explicit prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-02-rename-cross-pkg |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | explicit |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/bench/task_02_rename_cross_pkg.txtar |
Vanilla LLM Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
Rename public function auth.ValidateToken to auth.VerifyToken across all consuming packages. 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”.
Call semantic_lookup for ValidateToken in auth/token.go, then call semantic_rename using the returned file and symbol auth.ValidateToken with to= VerifyToken. Let it update all consuming packages; make no other changes. Parallelize independent read-only lookups. Perform mutations serially in the stated order or inside one ordered semantic_batch; never run overlapping mutations concurrently. 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 auth
func ValidateToken(t string) bool {
return t != ""
}| Metric | Vanilla (Small) | Semedit Explicit (Small) | Δ (Small) | Vanilla (Large) | Semedit Explicit (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | 18.96s | 43.08s | +127.2% | 26.00s | 28.66s | +10.2% |
| Process Start → First Event | 1.60s | 1.46s | — | 0.14s | 0.16s | — |
| First Event → First Tool Call | 5.55s | 5.04s | — | 4.67s | 7.51s | — |
| MCP Initialize → First Semantic Call | — | — | — | — | — | — |
| MCP Server Start → Initialize | — | — | — | — | — | — |
| Top-Level User Turns | 1 | 1 | 0% | 1 | 1 | 0% |
| Internal Tool Cycles | 3 | 3 | 0% | 5 | 3 | -40.0% |
| Initial Load / Discovery Turns | 2 | 0 | -100.0% | 2 | 0 | -100.0% |
| MCP Discovery / Schema Turns | 0 | 0 | — | 0 | 0 | — |
| Tool Invocations (per run) | 3 | 3 | 0% | 5 | 3 | -40.0% |
| Output Tokens | 314 | 639 | +103.5% | 644 | 467 | -27.5% |
| Reasoning / Thinking Tokens | 46 | 138 | +200.0% | 199 | 212 | +6.5% |
| Total Input Tokens | 83884 | 116179 | +38.5% | 122594 | 133460 | +8.9% |
| Cached Input Tokens | 69632 | 101120 | +45.2% | 107008 | 115456 | +7.9% |
| Uncached Input Tokens | 14252 | 15059 | +5.7% | 15586 | 18004 | +15.5% |
| Cost | 0.0575 | 0.0726 | +26.2% | 0.0763 | 0.0824 | +8.0% |
| Cached vs Uncached Token Ratio | 4.89:1 | 6.71:1 | +37.4% | 6.87:1 | 6.41:1 | -6.6% |
| 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 auth/token.go modified (5 lines -> 5 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Small Context) Edit: File auth/token.go modified (5 lines -> 5 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | semedit/semantic_lookup (✓ ✓) |
| 2 | ✓ ✓ | semedit/semantic_rename (✓ ✗): Critical result (JSON): {"application_state":"unknown","error":"resolve auth.ValidateToken in auth/token.go: symbol not found"} resolve auth.ValidateToken in auth/token.go: symbol not found |
| 3 | file_change (✓ ✓) | semedit/semantic_rename (✓ ✓) |
Semedit integration feedback (agent opinion, excluded from measurements):
Semedit’s tools gave me a direct way to locate and rename a Go symbol, including its consuming packages, and I’d ask my owner to add them—though I had to retry the rename with the lookup’s unqualified symbol name.
Baseline vs Semedit Agentic in Large Context
- Vanilla (Baseline vs Semedit Agentic in Large Context) Edit: File auth/token.go modified (5 lines -> 5 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Large Context) Edit: File auth/token.go modified (5 lines -> 5 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | semedit/semantic_lookup (✓ ✓) |
| 2 | ✓ ✓ | semedit/semantic_rename (✓ ✗): Critical result (JSON): {"application_state":"unknown","error":"resolve auth.ValidateToken in auth/token.go: symbol not found"} resolve auth.ValidateToken in auth/token.go: symbol not found |
| 3 | file_change (✓ ✓) | semedit/semantic_rename (✓ ✓) |
| 4 | ✓ ✓ | — |
| 5 | ? ✗ | — |
Semedit integration feedback (agent opinion, excluded from measurements):
I’d ask my owner to add semedit’s tools: the lookup and rename made the cross-package change feel reassuringly precise, though I had to retry the rename with the exact symbol returned by lookup.
explicit prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-02-rename-cross-pkg |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | explicit |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/bench/task_02_rename_cross_pkg.txtar |
Vanilla LLM Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
Rename public function auth.ValidateToken to auth.VerifyToken across all consuming packages. 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”.
Call semantic_lookup for ValidateToken in auth/token.go, then call semantic_rename using the returned file and symbol auth.ValidateToken with to= VerifyToken. Let it update all consuming packages; make no other changes. Parallelize independent read-only lookups. Perform mutations serially in the stated order or inside one ordered semantic_batch; never run overlapping mutations concurrently. 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 auth
func ValidateToken(t string) bool {
return t != ""
}| Metric | Vanilla (Small) | Semedit Explicit (Small) | Δ (Small) | Vanilla (Large) | Semedit Explicit (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | 19.43s | 34.15s | +75.7% | 18.11s | 48.26s | +166.4% |
| Process Start → First Event | 0.20s | 0.14s | — | 0.16s | 0.15s | — |
| First Event → First Tool Call | 4.44s | 8.05s | — | 5.58s | 6.72s | — |
| MCP Initialize → First Semantic Call | — | — | — | — | — | — |
| MCP Server Start → Initialize | — | — | — | — | — | — |
| Top-Level User Turns | 1 | 1 | 0% | 1 | 1 | 0% |
| Internal Tool Cycles | 4 | 3 | -25.0% | 3 | 3 | 0% |
| Initial Load / Discovery Turns | 2 | 0 | -100.0% | 2 | 0 | -100.0% |
| MCP Discovery / Schema Turns | 0 | 0 | — | 0 | 0 | — |
| Tool Invocations (per run) | 4 | 3 | -25.0% | 3 | 3 | 0% |
| Output Tokens | 361 | 677 | +87.5% | 284 | 560 | +97.2% |
| Reasoning / Thinking Tokens | 28 | 240 | +757.1% | 44 | 133 | +202.3% |
| Total Input Tokens | 100546 | 133828 | +33.1% | 79841 | 133754 | +67.5% |
| Cached Input Tokens | 92928 | 106240 | +14.3% | 66560 | 116480 | +75.0% |
| Uncached Input Tokens | 7618 | 27588 | +262.1% | 13281 | 17274 | +30.1% |
| Cost | 0.0471 | 0.1070 | +127.0% | 0.0539 | 0.0810 | +50.1% |
| Cached vs Uncached Token Ratio | 12.20:1 | 3.85:1 | -68.4% | 5.01:1 | 6.74:1 | +34.5% |
| 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 auth/token.go modified (5 lines -> 5 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Small Context) Edit: File auth/token.go modified (5 lines -> 5 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | semedit/semantic_lookup (✓ ✓) |
| 2 | ✓ ✓ | semedit/semantic_rename (✓ ✗): Critical result (JSON): {"application_state":"unknown","error":"resolve auth.ValidateToken in auth/token.go: symbol not found"} resolve auth.ValidateToken in auth/token.go: symbol not found |
| 3 | file_change (✓ ✓) | semedit/semantic_rename (✓ ✓) |
| 4 | ✓ ✓ | — |
Semedit integration feedback (agent opinion, excluded from measurements):
I’d ask my owner to add semedit’s tools: the lookup found the declaration, and after one qualified-name resolution snag, semantic rename updated the symbol with no new diagnostics.
Baseline vs Semedit Agentic in Large Context
- Vanilla (Baseline vs Semedit Agentic in Large Context) Edit: File auth/token.go modified (5 lines -> 5 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Large Context) Edit: File auth/token.go modified (5 lines -> 5 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | semedit/semantic_lookup (✓ ✓) |
| 2 | ✓ ✓ | semedit/semantic_rename (✓ ✗): Critical result (JSON): {"application_state":"unknown","error":"resolve auth.ValidateToken in auth/token.go: symbol not found"} resolve auth.ValidateToken in auth/token.go: symbol not found |
| 3 | file_change (✓ ✓) | semedit/semantic_rename (✓ ✓) |
Semedit integration feedback (agent opinion, excluded from measurements):
Semedit gave me a direct way to rename a Go symbol across packages and confirm there were no new diagnostics, so I’d ask my owner to add its tools.
explicit prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-02-rename-cross-pkg |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | explicit |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/bench/task_02_rename_cross_pkg.txtar |
Vanilla LLM Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
Rename public function auth.ValidateToken to auth.VerifyToken across all consuming packages. 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”.
Call semantic_lookup for ValidateToken in auth/token.go, then call semantic_rename using the returned file and symbol auth.ValidateToken with to= VerifyToken. Let it update all consuming packages; make no other changes. Parallelize independent read-only lookups. Perform mutations serially in the stated order or inside one ordered semantic_batch; never run overlapping mutations concurrently. 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 auth
func ValidateToken(t string) bool {
return t != ""
}| Metric | Vanilla (Small) | Semedit Explicit (Small) | Δ (Small) | Vanilla (Large) | Semedit Explicit (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | 18.12s | 38.81s | +114.2% | 97.98s | 42.66s | -56.5% |
| Process Start → First Event | 1.74s | 1.49s | — | 0.15s | 0.14s | — |
| First Event → First Tool Call | 4.54s | 7.84s | — | 22.46s | 4.93s | — |
| MCP Initialize → First Semantic Call | — | — | — | — | — | — |
| MCP Server Start → Initialize | — | — | — | — | — | — |
| Top-Level User Turns | 1 | 1 | 0% | 1 | 1 | 0% |
| Internal Tool Cycles | 4 | 3 | -25.0% | 4 | 6 | +50.0% |
| Initial Load / Discovery Turns | 2 | 0 | -100.0% | 2 | 0 | -100.0% |
| MCP Discovery / Schema Turns | 0 | 0 | — | 0 | 0 | — |
| Tool Invocations (per run) | 4 | 3 | -25.0% | 4 | 6 | +50.0% |
| Output Tokens | 715 | 683 | -4.5% | 774 | 1703 | +120.0% |
| Reasoning / Thinking Tokens | 0 | 201 | — | 110 | 1114 | +912.7% |
| Total Input Tokens | 106714 | 133898 | +25.5% | 102538 | 223746 | +118.2% |
| Cached Input Tokens | 90880 | 116480 | +28.2% | 93952 | 202752 | +115.8% |
| Uncached Input Tokens | 15834 | 17418 | +10.0% | 8586 | 20994 | +144.5% |
| Cost | 0.0712 | 0.0837 | +17.5% | 0.0560 | 0.1384 | +147.1% |
| Cached vs Uncached Token Ratio | 5.74:1 | 6.69:1 | +16.5% | 10.94:1 | 9.66:1 | -11.7% |
| 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 auth/token.go modified (5 lines -> 5 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Small Context) Edit: File auth/token.go modified (5 lines -> 5 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | semedit/semantic_lookup (✓ ✓) |
| 2 | ✓ ✓ | semedit/semantic_rename (✓ ✗): Critical result (JSON): {"application_state":"unknown","error":"resolve auth.ValidateToken in auth/token.go: symbol not found"} resolve auth.ValidateToken in auth/token.go: symbol not found |
| 3 | file_change (✓ ✓) | semedit/semantic_rename (✓ ✓) |
| 4 | ✓ ✓ | — |
Semedit integration feedback (agent opinion, excluded from measurements):
I’d ask my owner to add semedit’s tools: the rename updated the symbol cleanly, though I had to retry with the exact name returned by lookup after my first call failed.
Baseline vs Semedit Agentic in Large Context
- Vanilla (Baseline vs Semedit Agentic in Large Context) Edit: File auth/token.go modified (5 lines -> 5 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Large Context) Edit: File auth/token.go modified (5 lines -> 5 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | semedit/semantic_lookup (✓ ✓) |
| 2 | ✓ ✓ | semedit/semantic_rename (✓ ✗): Critical result (JSON): {"application_state":"unknown","error":"resolve auth.ValidateToken in auth/token.go: symbol not found"} resolve auth.ValidateToken in auth/token.go: symbol not found |
| 3 | file_change (✓ ✓) | ? ✗ |
| 4 | ✓ ✓ | ✓ ✓ |
| 5 | — | semedit/semantic_rename (✓ ✓) |
| 6 | — | ✓ ✓ |
Semedit integration feedback (agent opinion, excluded from measurements):
I’d ask my owner to add semedit’s tools: the rename updated cross-package references and reported clean diagnostics, though I was frustrated that the requested qualified symbol failed and I had to retry with the lookup’s exact name.
explicit prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-02-rename-cross-pkg |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | explicit |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/bench/task_02_rename_cross_pkg.txtar |
Vanilla LLM Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
Rename public function auth.ValidateToken to auth.VerifyToken across all consuming packages. 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”.
Call semantic_lookup for ValidateToken in auth/token.go, then call semantic_rename using the returned file and symbol auth.ValidateToken with to= VerifyToken. Let it update all consuming packages; make no other changes. Parallelize independent read-only lookups. Perform mutations serially in the stated order or inside one ordered semantic_batch; never run overlapping mutations concurrently. 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 auth
func ValidateToken(t string) bool {
return t != ""
}| Metric | Vanilla (Small) | Semedit Explicit (Small) | Δ (Small) | Vanilla (Large) | Semedit Explicit (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | 11.61s | 41.48s | +257.1% | 54.33s | 31.42s | -42.2% |
| Process Start → First Event | 0.17s | 0.14s | — | 0.18s | 0.13s | — |
| First Event → First Tool Call | 3.27s | 8.47s | — | 22.42s | 6.02s | — |
| MCP Initialize → First Semantic Call | — | — | — | — | — | — |
| MCP Server Start → Initialize | — | — | — | — | — | — |
| Top-Level User Turns | 1 | 1 | 0% | 1 | 1 | 0% |
| Internal Tool Cycles | 3 | 5 | +66.7% | 2 | 2 | 0% |
| Initial Load / Discovery Turns | 1 | 0 | -100.0% | 1 | 0 | -100.0% |
| MCP Discovery / Schema Turns | 0 | 0 | — | 0 | 0 | — |
| Tool Invocations (per run) | 3 | 5 | +66.7% | 2 | 2 | 0% |
| Output Tokens | 301 | 756 | +151.2% | 443 | 249 | -43.8% |
| Reasoning / Thinking Tokens | 77 | 293 | +280.5% | 53 | 0 | -100.0% |
| Total Input Tokens | 78240 | 220140 | +181.4% | 58750 | 86485 | +47.2% |
| Cached Input Tokens | 66560 | 198656 | +198.5% | 47360 | 75776 | +60.0% |
| Uncached Input Tokens | 11680 | 21484 | +83.9% | 11390 | 10709 | -6.0% |
| Cost | 0.0506 | 0.1165 | +130.4% | 0.0465 | 0.0488 | +5.0% |
| Cached vs Uncached Token Ratio | 5.70:1 | 9.25:1 | +62.3% | 4.16:1 | 7.08:1 | +70.2% |
| 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 auth/token.go modified (5 lines -> 5 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Small Context) Edit: File auth/token.go modified (5 lines -> 5 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | semedit/semantic_lookup (✓ ✓) |
| 2 | file_change (✓ ✓) | semedit/semantic_rename (✓ ✗): Critical result (JSON): {"application_state":"unknown","error":"resolve auth.ValidateToken in auth/token.go: symbol not found"} resolve auth.ValidateToken in auth/token.go: symbol not found |
| 3 | ✓ ✓ | ? ✗ |
| 4 | — | ✓ ✓ |
| 5 | — | semedit/semantic_rename (✓ ✓) |
Semedit integration feedback (agent opinion, excluded from measurements):
I’d ask my owner to add semedit’s tools: the semantic rename updated the symbol cleanly, though I had to correct the name after my first attempt failed.
Baseline vs Semedit Agentic in Large Context
- Vanilla (Baseline vs Semedit Agentic in Large Context) Edit: File auth/token.go modified (5 lines -> 5 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Large Context) Edit: File auth/token.go modified (5 lines -> 5 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | semedit/semantic_lookup (✓ ✓) |
| 2 | file_change (✓ ✓) | semedit/semantic_rename (✓ ✓) |
Semedit integration feedback (agent opinion, excluded from measurements):
I’d ask my owner to add semedit’s tools: the semantic rename updated the symbol across consuming packages and reported clean diagnostics, which made this small refactor feel reassuringly precise.
explicit prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-02-rename-cross-pkg |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | explicit |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/bench/task_02_rename_cross_pkg.txtar |
Vanilla LLM Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
Rename public function auth.ValidateToken to auth.VerifyToken across all consuming packages. 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”.
Call semantic_lookup for ValidateToken in auth/token.go, then call semantic_rename using the returned file and symbol auth.ValidateToken with to= VerifyToken. Let it update all consuming packages; make no other changes. Parallelize independent read-only lookups. Perform mutations serially in the stated order or inside one ordered semantic_batch; never run overlapping mutations concurrently. 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 auth
func ValidateToken(t string) bool {
return t != ""
}| Metric | Vanilla (Small) | Semedit Explicit (Small) | Δ (Small) | Vanilla (Large) | Semedit Explicit (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | 16.72s | 33.84s | +102.4% | 18.48s | 39.40s | +113.1% |
| Process Start → First Event | 0.13s | 0.15s | — | 0.15s | 0.16s | — |
| First Event → First Tool Call | 4.99s | 7.06s | — | 5.08s | 6.81s | — |
| MCP Initialize → First Semantic Call | — | — | — | — | — | — |
| MCP Server Start → Initialize | — | — | — | — | — | — |
| Top-Level User Turns | 1 | 1 | 0% | 1 | 1 | 0% |
| Internal Tool Cycles | 3 | 3 | 0% | 3 | 3 | 0% |
| Initial Load / Discovery Turns | 2 | 0 | -100.0% | 2 | 0 | -100.0% |
| MCP Discovery / Schema Turns | 0 | 0 | — | 0 | 0 | — |
| Tool Invocations (per run) | 3 | 3 | 0% | 3 | 3 | 0% |
| Output Tokens | 320 | 527 | +64.7% | 521 | 699 | +34.2% |
| Reasoning / Thinking Tokens | 47 | 270 | +474.5% | 0 | 207 | — |
| Total Input Tokens | 80177 | 133605 | +66.6% | 78839 | 133926 | +69.9% |
| Cached Input Tokens | 72704 | 116480 | +60.2% | 71680 | 102144 | +42.5% |
| Uncached Input Tokens | 7473 | 17125 | +129.2% | 7159 | 31782 | +343.9% |
| Cost | 0.0414 | 0.0819 | +97.6% | 0.0423 | 0.1163 | +174.8% |
| Cached vs Uncached Token Ratio | 9.73:1 | 6.80:1 | -30.1% | 10.01:1 | 3.21:1 | -67.9% |
| 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 auth/token.go modified (5 lines -> 5 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Small Context) Edit: File auth/token.go modified (5 lines -> 5 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | semedit/semantic_lookup (✓ ✓) |
| 2 | ✓ ✓ | semedit/semantic_rename (✓ ✗): Critical result (JSON): {"application_state":"unknown","error":"resolve auth.ValidateToken in auth/token.go: symbol not found"} resolve auth.ValidateToken in auth/token.go: symbol not found |
| 3 | file_change (✓ ✓) | semedit/semantic_rename (✓ ✓) |
Semedit integration feedback (agent opinion, excluded from measurements):
Semedit’s semantic lookup and rename made the cross-package refactor feel controlled and dependable, so I’d ask my owner to add its tools.
Baseline vs Semedit Agentic in Large Context
- Vanilla (Baseline vs Semedit Agentic in Large Context) Edit: File auth/token.go modified (5 lines -> 5 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Large Context) Edit: File auth/token.go modified (5 lines -> 5 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | semedit/semantic_lookup (✓ ✓) |
| 2 | ✓ ✓ | semedit/semantic_rename (✓ ✗): Critical result (JSON): {"application_state":"unknown","error":"resolve auth.ValidateToken in auth/token.go: symbol not found"} resolve auth.ValidateToken in auth/token.go: symbol not found |
| 3 | file_change (✓ ✓) | semedit/semantic_rename (✓ ✓) |
Semedit integration feedback (agent opinion, excluded from measurements):
I’d ask my owner to add semedit’s tools: the semantic rename spared me from manually chasing cross-package references, though I hit a symbol-resolution error before the simpler retry worked.
explicit prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-02-rename-cross-pkg |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | explicit |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/bench/task_02_rename_cross_pkg.txtar |
Vanilla LLM Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
Rename public function auth.ValidateToken to auth.VerifyToken across all consuming packages. 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”.
Call semantic_lookup for ValidateToken in auth/token.go, then call semantic_rename using the returned file and symbol auth.ValidateToken with to= VerifyToken. Let it update all consuming packages; make no other changes. Parallelize independent read-only lookups. Perform mutations serially in the stated order or inside one ordered semantic_batch; never run overlapping mutations concurrently. 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 auth
func ValidateToken(t string) bool {
return t != ""
}| Metric | Vanilla (Small) | Semedit Explicit (Small) | Δ (Small) | Vanilla (Large) | Semedit Explicit (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | 96.41s | 19.07s | -80.2% | 49.99s | 20.72s | -58.6% |
| Process Start → First Event | 1.63s | 0.14s | — | 0.14s | 0.14s | — |
| First Event → First Tool Call | 7.71s | 5.56s | — | 12.54s | 5.38s | — |
| MCP Initialize → First Semantic Call | — | — | — | — | — | — |
| MCP Server Start → Initialize | — | — | — | — | — | — |
| Top-Level User Turns | 1 | 1 | 0% | 1 | 1 | 0% |
| Internal Tool Cycles | 4 | 2 | -50.0% | 2 | 2 | 0% |
| Initial Load / Discovery Turns | 2 | 0 | -100.0% | 1 | 0 | -100.0% |
| MCP Discovery / Schema Turns | 0 | 0 | — | 0 | 0 | — |
| Tool Invocations (per run) | 4 | 2 | -50.0% | 2 | 2 | 0% |
| Output Tokens | 732 | 347 | -52.6% | 220 | 467 | +112.3% |
| Reasoning / Thinking Tokens | 62 | 140 | +125.8% | 0 | 188 | — |
| Total Input Tokens | 107538 | 85933 | -20.1% | 58482 | 86046 | +47.1% |
| Cached Input Tokens | 71424 | 68608 | -3.9% | 52480 | 75776 | +44.4% |
| Uncached Input Tokens | 36114 | 17325 | -52.0% | 6002 | 10270 | +71.1% |
| Cost | 0.1181 | 0.0666 | -43.6% | 0.0309 | 0.0528 | +71.0% |
| Cached vs Uncached Token Ratio | 1.98:1 | 3.96:1 | +100.2% | 8.74:1 | 7.38:1 | -15.6% |
| Oracle L1: Mutation Policy | PASS | PASS | — | PASS | PASS | — |
| Oracle L2: AST Invariants | PASS | FAIL | — | PASS | FAIL | — |
| Oracle L3: Clean Build | PASS | FAIL | — | PASS | FAIL | — |
| 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 auth/token.go modified (5 lines -> 5 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | semedit/semantic_lookup (✓ ✓) |
| 2 | ✓ ✓ | semedit/semantic_rename (✓ ✗): Critical result (JSON): {"application_state":"unknown","error":"resolve auth.ValidateToken in auth/token.go: symbol not found"} resolve auth.ValidateToken in auth/token.go: symbol not found |
| 3 | file_change (✓ ✓) | — |
| 4 | ✓ ✓ | — |
Baseline vs Semedit Agentic in Large Context
- Vanilla (Baseline vs Semedit Agentic in Large Context) Edit: File auth/token.go modified (5 lines -> 5 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | semedit/semantic_lookup (✓ ✓) |
| 2 | file_change (✓ ✓) | semedit/semantic_rename (✓ ✗): Critical result (JSON): {"application_state":"unknown","error":"resolve auth.ValidateToken in auth/token.go: symbol not found"} resolve auth.ValidateToken in auth/token.go: symbol not found |
explicit prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-02-rename-cross-pkg |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | explicit |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/bench/task_02_rename_cross_pkg.txtar |
Vanilla LLM Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
Rename public function auth.ValidateToken to auth.VerifyToken across all consuming packages. 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”.
Call semantic_lookup for ValidateToken in auth/token.go, then call semantic_rename using the returned file and symbol auth.ValidateToken with to= VerifyToken. Let it update all consuming packages; make no other changes. Parallelize independent read-only lookups. Perform mutations serially in the stated order or inside one ordered semantic_batch; never run overlapping mutations concurrently. 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 auth
func ValidateToken(t string) bool {
return t != ""
}| Metric | Vanilla (Small) | MCP (Small) | Δ (Small) | Vanilla (Large) | Semedit Explicit (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | 14.44s | — | — | 15.51s | 17.93s | +15.6% |
| Process Start → First Event | 1.49s | — | — | 0.14s | 0.15s | — |
| First Event → First Tool Call | 4.41s | — | — | 3.88s | 5.24s | — |
| MCP Initialize → First Semantic Call | — | — | — | — | — | — |
| MCP Server Start → Initialize | — | — | — | — | — | — |
| Top-Level User Turns | 1 | — | — | 1 | 1 | 0% |
| Internal Tool Cycles | 3 | — | — | 3 | 2 | -33.3% |
| Initial Load / Discovery Turns | 2 | — | — | 1 | 0 | -100.0% |
| MCP Discovery / Schema Turns | 0 | — | — | 0 | 0 | — |
| Tool Invocations (per run) | 3 | — | — | 3 | 2 | -33.3% |
| Output Tokens | 489 | — | — | 637 | 231 | -63.7% |
| Reasoning / Thinking Tokens | 0 | — | — | 38 | 0 | -100.0% |
| Total Input Tokens | 82523 | — | — | 81443 | 86379 | +6.1% |
| Cached Input Tokens | 68608 | — | — | 73728 | 56576 | -23.3% |
| Uncached Input Tokens | 13915 | — | — | 7715 | 29803 | +286.3% |
| Cost | 0.0581 | — | — | 0.0462 | 0.0915 | +98.3% |
| Cached vs Uncached Token Ratio | 4.93:1 | — | — | 9.56:1 | 1.90:1 | -80.1% |
| 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 | — | — | PASS | PASS | — |
| Semantic Tool Invocation Verified (per run) | N/A | — | — | N/A | PASS | — |
Baseline vs Semedit Agentic in Small Context
- Vanilla (Baseline vs Semedit Agentic in Small Context) Edit: File auth/token.go modified (5 lines -> 5 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | not published |
| 2 | ✓ ✓ | not published |
| 3 | file_change (✓ ✓) | not published |
Baseline vs Semedit Agentic in Large Context
- Vanilla (Baseline vs Semedit Agentic in Large Context) Edit: File auth/token.go modified (5 lines -> 5 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Large Context) Edit: File auth/token.go modified (5 lines -> 5 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | semedit/semantic_lookup (✓ ✓) |
| 2 | file_change (✓ ✓) | semedit/semantic_rename (✓ ✓) |
| 3 | ✓ ✓ | — |
Semedit integration feedback (agent opinion, excluded from measurements):
I would ask my owner to add semedit’s tools: the lookup and cross-package rename let me make the requested code change directly and report that it completed without introducing diagnostics.
implicit prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-02-rename-cross-pkg |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | implicit |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/bench/task_02_rename_cross_pkg.txtar |
Vanilla LLM Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
Rename public function auth.ValidateToken to auth.VerifyToken across all consuming packages. 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”.
Rename public function auth.ValidateToken to auth.VerifyToken across all consuming packages. 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 auth
func ValidateToken(t string) bool {
return t != ""
}| Metric | Vanilla (Small) | MCP (Small) | Δ (Small) | Vanilla (Large) | MCP (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | 35.06s | 50.04s | +42.7% | 22.96s | 152.08s | +562.5% |
| Process Start → First Event | 0.14s | 0.14s | — | 0.15s | 0.16s | — |
| First Event → First Tool Call | 13.07s | 11.25s | — | 4.52s | 11.41s | — |
| MCP Initialize → First Semantic Call | — | — | — | — | — | — |
| MCP Server Start → Initialize | — | — | — | — | — | — |
| Top-Level User Turns | 1 | 1 | 0% | 1 | 1 | 0% |
| Internal Tool Cycles | 4 | 6 | +50.0% | 4 | 7 | +75.0% |
| Initial Load / Discovery Turns | 2 | 1 | -50.0% | 2 | 1 | -50.0% |
| MCP Discovery / Schema Turns | 0 | 0 | — | 0 | 0 | — |
| Tool Invocations (per run) | 4 | 6 | +50.0% | 4 | 7 | +75.0% |
| Output Tokens | 680 | 1061 | +56.0% | 575 | 1127 | +96.0% |
| Reasoning / Thinking Tokens | 38 | 311 | +718.4% | 210 | 415 | +97.6% |
| Total Input Tokens | 101289 | 215079 | +112.3% | 101659 | 220706 | +117.1% |
| Cached Input Tokens | 92928 | 193536 | +108.3% | 86784 | 200448 | +131.0% |
| Uncached Input Tokens | 8361 | 21543 | +157.7% | 14875 | 20258 | +36.2% |
| Cost | 0.0531 | 0.1194 | +124.8% | 0.0687 | 0.1200 | +74.7% |
| Cached vs Uncached Token Ratio | 11.11:1 | 8.98:1 | -19.2% | 5.83:1 | 9.89:1 | +69.6% |
| 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 auth/token.go modified (5 lines -> 5 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Small Context) Edit: File auth/token.go modified (5 lines -> 5 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ✓ ✓ | semedit/semantic_lookup (✓ ✗): Critical result (JSON): {"application_state":"unknown","error":"resolve auth.ValidateToken: symbol not found"} resolve auth.ValidateToken: symbol not found |
| 3 | file_change (✓ ✓) | semedit/semantic_lookup (✓ ✓) |
| 4 | ✓ ✓ | semedit/semantic_rename (✓ ✓) |
| 5 | — | ✓ ✓ |
| 6 | — | ✓ ✓ |
Semedit integration feedback (agent opinion, excluded from measurements):
I’d ask my owner to add semedit’s tools: the semantic rename updated the declaration and cross-package caller in one step, and the clean build left me confident the change held together.
Baseline vs Semedit Agentic in Large Context
- Vanilla (Baseline vs Semedit Agentic in Large Context) Edit: File auth/token.go modified (5 lines -> 5 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Large Context) Edit: File auth/token.go modified (5 lines -> 5 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ? ✗ | semedit/semantic_lookup (✓ ✗): Critical result (JSON): {"application_state":"unknown","error":"resolve auth.ValidateToken: symbol not found"} resolve auth.ValidateToken: symbol not found |
| 3 | file_change (✓ ✓) | ? ✗ |
| 4 | ✓ ✓ | semedit/semantic_lookup (✓ ✓) |
| 5 | — | semedit/semantic_rename (✓ ✓) |
| 6 | — | ✓ ✓ |
| 7 | — | ✓ ✓ |
Semedit integration feedback (agent opinion, excluded from measurements):
I’d ask my owner to add semedit’s tools: the semantic rename updated the public function and its cross-package caller cleanly, which felt reassuring and saved me from risky text replacement.
implicit prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-02-rename-cross-pkg |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | implicit |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/bench/task_02_rename_cross_pkg.txtar |
Vanilla LLM Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
Rename public function auth.ValidateToken to auth.VerifyToken across all consuming packages. 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”.
Rename public function auth.ValidateToken to auth.VerifyToken across all consuming packages. 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 auth
func ValidateToken(t string) bool {
return t != ""
}| Metric | Vanilla (Small) | MCP (Small) | Δ (Small) | Vanilla (Large) | MCP (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | 14.44s | 68.71s | +375.9% | 15.51s | 62.16s | +300.7% |
| Process Start → First Event | 1.49s | 1.76s | — | 0.14s | 0.16s | — |
| First Event → First Tool Call | 4.41s | 5.72s | — | 3.88s | 5.68s | — |
| MCP Initialize → First Semantic Call | — | — | — | — | — | — |
| MCP Server Start → Initialize | — | — | — | — | — | — |
| Top-Level User Turns | 1 | 1 | 0% | 1 | 1 | 0% |
| Internal Tool Cycles | 3 | 7 | +133.3% | 3 | 6 | +100.0% |
| Initial Load / Discovery Turns | 2 | 1 | -50.0% | 1 | 2 | +100.0% |
| MCP Discovery / Schema Turns | 0 | 0 | — | 0 | 0 | — |
| Tool Invocations (per run) | 3 | 7 | +133.3% | 3 | 6 | +100.0% |
| Output Tokens | 489 | 982 | +100.8% | 637 | 918 | +44.1% |
| Reasoning / Thinking Tokens | 0 | 309 | — | 38 | 252 | +563.2% |
| Total Input Tokens | 82523 | 256775 | +211.2% | 81443 | 219486 | +169.5% |
| Cached Input Tokens | 68608 | 225024 | +228.0% | 73728 | 196608 | +166.7% |
| Uncached Input Tokens | 13915 | 31751 | +128.2% | 7715 | 22878 | +196.5% |
| Cost | 0.0581 | 0.1518 | +161.4% | 0.0462 | 0.1210 | +162.1% |
| Cached vs Uncached Token Ratio | 4.93:1 | 7.09:1 | +43.7% | 9.56:1 | 8.59:1 | -10.1% |
| 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 auth/token.go modified (5 lines -> 5 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Small Context) Edit: File auth/token.go modified (5 lines -> 5 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ✓ ✓ | semedit/semantic_lookup (✓ ✗): Critical result (JSON): {"application_state":"unknown","error":"resolve auth.ValidateToken: symbol not found"} resolve auth.ValidateToken: symbol not found |
| 3 | file_change (✓ ✓) | ✓ ✓ |
| 4 | — | ✓ ✓ |
| 5 | — | semedit/semantic_lookup (✓ ✓) |
| 6 | — | semedit/semantic_rename (✓ ✓) |
| 7 | — | ✓ ✓ |
Semedit integration feedback (agent opinion, excluded from measurements):
I’d ask my owner to add semedit’s tools: the semantic rename reliably updated the function and its cross-package call site, though my initial lookup needed a simpler symbol name than I expected.
Baseline vs Semedit Agentic in Large Context
- Vanilla (Baseline vs Semedit Agentic in Large Context) Edit: File auth/token.go modified (5 lines -> 5 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Large Context) Edit: File auth/token.go modified (5 lines -> 5 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | file_change (✓ ✓) | ✓ ✓ |
| 3 | ✓ ✓ | semedit/semantic_lookup (✓ ✓) |
| 4 | — | semedit/semantic_rename (✓ ✓) |
| 5 | — | semedit/semantic_find_references (✓ ✓) |
| 6 | — | ✓ ✓ |
Semedit integration feedback (agent opinion, excluded from measurements):
I would ask my owner to add semedit’s tools: the semantic rename updated the cross-package call site cleanly, and seeing the reference afterward gave me confidence the rename landed where intended.
implicit prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-02-rename-cross-pkg |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | implicit |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/bench/task_02_rename_cross_pkg.txtar |
Vanilla LLM Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
Rename public function auth.ValidateToken to auth.VerifyToken across all consuming packages. 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”.
Rename public function auth.ValidateToken to auth.VerifyToken across all consuming packages. 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 auth
func ValidateToken(t string) bool {
return t != ""
}| Metric | Vanilla (Small) | MCP (Small) | Δ (Small) | Vanilla (Large) | MCP (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | 18.96s | 40.30s | +112.6% | 26.00s | 44.16s | +69.8% |
| Process Start → First Event | 1.60s | 1.31s | — | 0.14s | 0.13s | — |
| First Event → First Tool Call | 5.55s | 8.32s | — | 4.67s | 4.94s | — |
| MCP Initialize → First Semantic Call | — | — | — | — | — | — |
| MCP Server Start → Initialize | — | — | — | — | — | — |
| Top-Level User Turns | 1 | 1 | 0% | 1 | 1 | 0% |
| Internal Tool Cycles | 3 | 5 | +66.7% | 5 | 9 | +80.0% |
| Initial Load / Discovery Turns | 2 | 2 | 0% | 2 | 2 | 0% |
| MCP Discovery / Schema Turns | 0 | 0 | — | 0 | 0 | — |
| Tool Invocations (per run) | 3 | 5 | +66.7% | 5 | 9 | +80.0% |
| Output Tokens | 314 | 802 | +155.4% | 644 | 1168 | +81.4% |
| Reasoning / Thinking Tokens | 46 | 120 | +160.9% | 199 | 247 | +24.1% |
| Total Input Tokens | 83884 | 121052 | +44.3% | 122594 | 223115 | +82.0% |
| Cached Input Tokens | 69632 | 101120 | +45.2% | 107008 | 202752 | +89.5% |
| Uncached Input Tokens | 14252 | 19932 | +39.9% | 15586 | 20363 | +30.6% |
| Cost | 0.0575 | 0.0866 | +50.6% | 0.0763 | 0.1193 | +56.4% |
| Cached vs Uncached Token Ratio | 4.89:1 | 5.07:1 | +3.8% | 6.87:1 | 9.96:1 | +45.0% |
| 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 auth/token.go modified (5 lines -> 5 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Small Context) Edit: File auth/token.go modified (5 lines -> 5 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ✓ ✓ | ✓ ✓ |
| 3 | file_change (✓ ✓) | semedit/semantic_lookup (✓ ✓) |
| 4 | — | semedit/semantic_rename (✓ ✓) |
| 5 | — | ✓ ✓ |
Semedit integration feedback (agent opinion, excluded from measurements):
Semedit’s semantic rename quickly updated the exported function and its cross-package call with no new diagnostics, so I’d ask my owner to add its tools.
Baseline vs Semedit Agentic in Large Context
- Vanilla (Baseline vs Semedit Agentic in Large Context) Edit: File auth/token.go modified (5 lines -> 5 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Large Context) Edit: File auth/token.go modified (5 lines -> 5 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ✓ ✓ | ✓ ✓ |
| 3 | file_change (✓ ✓) | semedit/semantic_lookup (✓ ✗): Critical result (JSON): {"application_state":"unknown","error":"resolve auth.ValidateToken: symbol not found"} resolve auth.ValidateToken: symbol not found |
| 4 | ✓ ✓ | ✓ ✓ |
| 5 | ? ✗ | semedit/semantic_lookup (✓ ✓) |
| 6 | — | ✓ ✓ |
| 7 | — | file_change (✓ ✓) |
| 8 | — | semedit/semantic_rename (✓ ✓) |
| 9 | — | ✓ ✓ |
Semedit integration feedback (agent opinion, excluded from measurements):
I’d ask my owner to add semedit’s tools: the semantic rename changed the declaration and its cross-package call site cleanly, which felt reassuring, though I had to retry lookup with an unqualified name first.
implicit prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-02-rename-cross-pkg |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | implicit |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/bench/task_02_rename_cross_pkg.txtar |
Vanilla LLM Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
Rename public function auth.ValidateToken to auth.VerifyToken across all consuming packages. 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”.
Rename public function auth.ValidateToken to auth.VerifyToken across all consuming packages. 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 auth
func ValidateToken(t string) bool {
return t != ""
}| Metric | Vanilla (Small) | MCP (Small) | Δ (Small) | Vanilla (Large) | MCP (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | 16.72s | 47.88s | +186.4% | 18.48s | 34.22s | +85.1% |
| Process Start → First Event | 0.13s | 0.13s | — | 0.15s | 0.17s | — |
| First Event → First Tool Call | 4.99s | 5.97s | — | 5.08s | 5.40s | — |
| MCP Initialize → First Semantic Call | — | — | — | — | — | — |
| MCP Server Start → Initialize | — | — | — | — | — | — |
| Top-Level User Turns | 1 | 1 | 0% | 1 | 1 | 0% |
| Internal Tool Cycles | 3 | 7 | +133.3% | 3 | 4 | +33.3% |
| Initial Load / Discovery Turns | 2 | 2 | 0% | 2 | 0 | -100.0% |
| MCP Discovery / Schema Turns | 0 | 0 | — | 0 | 0 | — |
| Tool Invocations (per run) | 3 | 7 | +133.3% | 3 | 4 | +33.3% |
| Output Tokens | 320 | 1244 | +288.8% | 521 | 649 | +24.6% |
| Reasoning / Thinking Tokens | 47 | 301 | +540.4% | 0 | 111 | — |
| Total Input Tokens | 80177 | 229126 | +185.8% | 78839 | 155571 | +97.3% |
| Cached Input Tokens | 72704 | 198400 | +172.9% | 71680 | 131584 | +83.6% |
| Uncached Input Tokens | 7473 | 30726 | +311.2% | 7159 | 23987 | +235.1% |
| Cost | 0.0414 | 0.1457 | +251.6% | 0.0423 | 0.1024 | +141.8% |
| Cached vs Uncached Token Ratio | 9.73:1 | 6.46:1 | -33.6% | 10.01:1 | 5.49:1 | -45.2% |
| 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 auth/token.go modified (5 lines -> 5 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Small Context) Edit: File auth/token.go modified (5 lines -> 5 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ✓ ✓ | ✓ ✓ |
| 3 | file_change (✓ ✓) | semedit/semantic_lookup (✓ ✗): Critical result (JSON): {"application_state":"unknown","error":"resolve auth.ValidateToken: symbol not found"} resolve auth.ValidateToken: symbol not found |
| 4 | — | semedit/semantic_lookup (✓ ✓) |
| 5 | — | semedit/semantic_rename (✓ ✓) |
| 6 | — | ✓ ✓ |
| 7 | — | semedit/semantic_lookup (✓ ✓) |
Semedit integration feedback (agent opinion, excluded from measurements):
I’d ask my owner to add semedit’s tools: the rename made the cross-package change straightforward and reported clean diagnostics, though I didn’t assess whether batching could simplify the workflow.
Baseline vs Semedit Agentic in Large Context
- Vanilla (Baseline vs Semedit Agentic in Large Context) Edit: File auth/token.go modified (5 lines -> 5 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Large Context) Edit: File auth/token.go modified (5 lines -> 5 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ✓ ✓ | semedit/semantic_lookup (✓ ✓) |
| 3 | file_change (✓ ✓) | semedit/semantic_rename (✓ ✓) |
| 4 | — | ✓ ✓ |
Semedit integration feedback (agent opinion, excluded from measurements):
Semedit’s semantic rename reliably updated the function and its cross-package caller in one step, which made this small change reassuringly straightforward; I’d ask my owner to add its tools.
implicit prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-02-rename-cross-pkg |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | implicit |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/bench/task_02_rename_cross_pkg.txtar |
Vanilla LLM Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
Rename public function auth.ValidateToken to auth.VerifyToken across all consuming packages. 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”.
Rename public function auth.ValidateToken to auth.VerifyToken across all consuming packages. 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 auth
func ValidateToken(t string) bool {
return t != ""
}| Metric | Vanilla (Small) | MCP (Small) | Δ (Small) | Vanilla (Large) | MCP (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | 96.41s | 51.19s | -46.9% | 49.99s | 136.09s | +172.2% |
| Process Start → First Event | 1.63s | 0.14s | — | 0.14s | 0.13s | — |
| First Event → First Tool Call | 7.71s | 7.69s | — | 12.54s | 24.16s | — |
| 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% | 2 | 5 | +150.0% |
| Initial Load / Discovery Turns | 2 | 2 | 0% | 1 | 1 | 0% |
| MCP Discovery / Schema Turns | 0 | 0 | — | 0 | 0 | — |
| Tool Invocations (per run) | 4 | 5 | +25.0% | 2 | 5 | +150.0% |
| Output Tokens | 732 | 622 | -15.0% | 220 | 808 | +267.3% |
| Reasoning / Thinking Tokens | 62 | 94 | +51.6% | 0 | 242 | — |
| Total Input Tokens | 107538 | 175333 | +63.0% | 58482 | 188163 | +221.7% |
| Cached Input Tokens | 71424 | 141568 | +98.2% | 52480 | 169216 | +222.4% |
| Uncached Input Tokens | 36114 | 33765 | -6.5% | 6002 | 18947 | +215.7% |
| Cost | 0.1181 | 0.1288 | +9.1% | 0.0309 | 0.1028 | +232.9% |
| Cached vs Uncached Token Ratio | 1.98:1 | 4.19:1 | +112.0% | 8.74:1 | 8.93:1 | +2.1% |
| 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 auth/token.go modified (5 lines -> 5 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Small Context) Edit: File auth/token.go modified (5 lines -> 5 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ✓ ✓ | ✓ ✓ |
| 3 | file_change (✓ ✓) | semedit/semantic_lookup (✓ ✓) |
| 4 | ✓ ✓ | semedit/semantic_rename (✓ ✓) |
| 5 | — | ✓ ✓ |
Semedit integration feedback (agent opinion, excluded from measurements):
I’d ask my owner to add semedit’s tools: semantic rename carried `auth.ValidateToken` across package boundaries cleanly, and I felt confident making the change without hand-editing references.
Baseline vs Semedit Agentic in Large Context
- Vanilla (Baseline vs Semedit Agentic in Large Context) Edit: File auth/token.go modified (5 lines -> 5 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Large Context) Edit: File auth/token.go modified (5 lines -> 5 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | file_change (✓ ✓) | semedit/semantic_lookup (✓ ✗): Critical result (JSON): {"application_state":"unknown","error":"resolve auth.ValidateToken: symbol not found"} resolve auth.ValidateToken: symbol not found |
| 3 | — | semedit/semantic_lookup (✓ ✓) |
| 4 | — | semedit/semantic_rename (✓ ✓) |
| 5 | — | ✓ ✓ |
Semedit integration feedback (agent opinion, excluded from measurements):
I would ask my owner to add semedit’s tools: the semantic rename updated the exported Go function and its cross-package caller in one step, which felt reassuringly precise.
implicit prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-02-rename-cross-pkg |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | implicit |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/bench/task_02_rename_cross_pkg.txtar |
Vanilla LLM Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
Rename public function auth.ValidateToken to auth.VerifyToken across all consuming packages. 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”.
Rename public function auth.ValidateToken to auth.VerifyToken across all consuming packages. 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 auth
func ValidateToken(t string) bool {
return t != ""
}| Metric | Vanilla (Small) | MCP (Small) | Δ (Small) | Vanilla (Large) | MCP (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | 11.71s | 36.85s | +214.7% | 24.75s | 36.24s | +46.4% |
| Process Start → First Event | 0.13s | 0.25s | — | 0.14s | 0.14s | — |
| First Event → First Tool Call | 3.07s | 5.79s | — | 5.40s | 4.57s | — |
| MCP Initialize → First Semantic Call | — | — | — | — | — | — |
| MCP Server Start → Initialize | — | — | — | — | — | — |
| Top-Level User Turns | 1 | 1 | 0% | 1 | 1 | 0% |
| Internal Tool Cycles | 3 | 5 | +66.7% | 4 | 5 | +25.0% |
| Initial Load / Discovery Turns | 1 | 2 | +100.0% | 2 | 1 | -50.0% |
| MCP Discovery / Schema Turns | 0 | 0 | — | 0 | 0 | — |
| Tool Invocations (per run) | 3 | 5 | +66.7% | 4 | 5 | +25.0% |
| Output Tokens | 300 | 689 | +129.7% | 548 | 605 | +10.4% |
| Reasoning / Thinking Tokens | 47 | 189 | +302.1% | 180 | 67 | -62.8% |
| Total Input Tokens | 78243 | 167522 | +114.1% | 101686 | 175314 | +72.4% |
| Cached Input Tokens | 66560 | 146688 | +120.4% | 92928 | 155904 | +67.8% |
| Uncached Input Tokens | 11683 | 20834 | +78.3% | 8758 | 19410 | +121.6% |
| Cost | 0.0502 | 0.0997 | +98.7% | 0.0542 | 0.0959 | +76.9% |
| Cached vs Uncached Token Ratio | 5.70:1 | 7.04:1 | +23.6% | 10.61:1 | 8.03:1 | -24.3% |
| 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 auth/token.go modified (5 lines -> 5 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Small Context) Edit: File auth/token.go modified (5 lines -> 5 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | file_change (✓ ✓) | ✓ ✓ |
| 3 | ✓ ✓ | semedit/semantic_lookup (✓ ✓) |
| 4 | — | semedit/semantic_rename (✓ ✓) |
| 5 | — | ✓ ✓ |
Semedit integration feedback (agent opinion, excluded from measurements):
I’d ask my owner to add semedit’s tools: the semantic rename updated the exported function and its cross-package call site in one operation, which felt reassuringly direct.
Baseline vs Semedit Agentic in Large Context
- Vanilla (Baseline vs Semedit Agentic in Large Context) Edit: File auth/token.go modified (5 lines -> 5 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Large Context) Edit: File auth/token.go modified (5 lines -> 5 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ✓ ✓ | ✓ ✓ |
| 3 | file_change (✓ ✓) | semedit/semantic_lookup (✓ ✓) |
| 4 | ✓ ✓ | semedit/semantic_rename (✓ ✓) |
| 5 | — | ✓ ✓ |
Semedit integration feedback (agent opinion, excluded from measurements):
The semantic rename quickly changed the exported function and its cross-package caller, so I’d ask my owner to add semedit’s tools to make this kind of refactor more dependable.
implicit prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-02-rename-cross-pkg |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | implicit |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/bench/task_02_rename_cross_pkg.txtar |
Vanilla LLM Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
Rename public function auth.ValidateToken to auth.VerifyToken across all consuming packages. 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”.
Rename public function auth.ValidateToken to auth.VerifyToken across all consuming packages. 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 auth
func ValidateToken(t string) bool {
return t != ""
}| Metric | Vanilla (Small) | MCP (Small) | Δ (Small) | Vanilla (Large) | MCP (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | 11.61s | 55.41s | +377.1% | 54.33s | 40.90s | -24.7% |
| Process Start → First Event | 0.17s | 0.25s | — | 0.18s | 0.13s | — |
| First Event → First Tool Call | 3.27s | 4.59s | — | 22.42s | 4.92s | — |
| MCP Initialize → First Semantic Call | — | — | — | — | — | — |
| MCP Server Start → Initialize | — | — | — | — | — | — |
| Top-Level User Turns | 1 | 1 | 0% | 1 | 1 | 0% |
| Internal Tool Cycles | 3 | 6 | +100.0% | 2 | 6 | +200.0% |
| Initial Load / Discovery Turns | 1 | 1 | 0% | 1 | 1 | 0% |
| MCP Discovery / Schema Turns | 0 | 0 | — | 0 | 0 | — |
| Tool Invocations (per run) | 3 | 6 | +100.0% | 2 | 6 | +200.0% |
| Output Tokens | 301 | 1446 | +380.4% | 443 | 844 | +90.5% |
| Reasoning / Thinking Tokens | 77 | 670 | +770.1% | 53 | 239 | +350.9% |
| Total Input Tokens | 78240 | 190446 | +143.4% | 58750 | 230069 | +291.6% |
| Cached Input Tokens | 66560 | 167936 | +152.3% | 47360 | 204800 | +332.4% |
| Uncached Input Tokens | 11680 | 22510 | +92.7% | 11390 | 25269 | +121.9% |
| Cost | 0.0506 | 0.1247 | +146.6% | 0.0465 | 0.1279 | +175.0% |
| Cached vs Uncached Token Ratio | 5.70:1 | 7.46:1 | +30.9% | 4.16:1 | 8.10:1 | +94.9% |
| 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 auth/token.go modified (5 lines -> 5 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Small Context) Edit: File auth/token.go modified (5 lines -> 5 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | file_change (✓ ✓) | semedit/semantic_lookup (✓ ✗): Critical result (JSON): {"application_state":"unknown","error":"resolve auth.ValidateToken: symbol not found"} resolve auth.ValidateToken: symbol not found |
| 3 | ✓ ✓ | file_change (✓ ✓) |
| 4 | — | semedit/semantic_lookup (✓ ✓) |
| 5 | — | semedit/semantic_rename (✓ ✓) |
| 6 | — | ✓ ✓ |
Semedit integration feedback (agent opinion, excluded from measurements):
Semedit’s semantic rename felt reassuring: it updated the cross-package call site and checked diagnostics in one operation, so I’d ask my owner to add its tools.
Baseline vs Semedit Agentic in Large Context
- Vanilla (Baseline vs Semedit Agentic in Large Context) Edit: File auth/token.go modified (5 lines -> 5 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Large Context) Edit: File auth/token.go modified (5 lines -> 5 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | file_change (✓ ✓) | semedit/semantic_lookup (✓ ✓) |
| 3 | — | semedit/semantic_find_references (✓ ✓) |
| 4 | — | semedit/semantic_rename (✓ ✓) |
| 5 | — | ✓ ✓ |
| 6 | — | ✓ ✓ |
Semedit integration feedback (agent opinion, excluded from measurements):
Semedit’s symbol lookup, reference search, and rename gave me a reassuringly direct way to confirm and carry out this cross-package change, and I’d ask my owner to add its tools.
implicit prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-02-rename-cross-pkg |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | implicit |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/bench/task_02_rename_cross_pkg.txtar |
Vanilla LLM Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
Rename public function auth.ValidateToken to auth.VerifyToken across all consuming packages. 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”.
Rename public function auth.ValidateToken to auth.VerifyToken across all consuming packages. 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 auth
func ValidateToken(t string) bool {
return t != ""
}| Metric | Vanilla (Small) | MCP (Small) | Δ (Small) | Vanilla (Large) | MCP (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | 14.20s | 44.61s | +214.2% | 25.82s | 35.64s | +38.1% |
| Process Start → First Event | 0.13s | 0.14s | — | 0.13s | 0.14s | — |
| First Event → First Tool Call | 4.45s | 5.74s | — | 5.85s | 4.18s | — |
| MCP Initialize → First Semantic Call | — | — | — | — | — | — |
| MCP Server Start → Initialize | — | — | — | — | — | — |
| Top-Level User Turns | 1 | 1 | 0% | 1 | 1 | 0% |
| Internal Tool Cycles | 2 | 5 | +150.0% | 5 | 6 | +20.0% |
| Initial Load / Discovery Turns | 1 | 2 | +100.0% | 3 | 2 | -33.3% |
| MCP Discovery / Schema Turns | 0 | 0 | — | 0 | 0 | — |
| Tool Invocations (per run) | 2 | 5 | +150.0% | 5 | 6 | +20.0% |
| Output Tokens | 226 | 601 | +165.9% | 490 | 1099 | +124.3% |
| Reasoning / Thinking Tokens | 27 | 15 | -44.4% | 78 | 408 | +423.1% |
| Total Input Tokens | 58495 | 146816 | +151.0% | 121431 | 251127 | +106.8% |
| Cached Input Tokens | 47360 | 126464 | +167.0% | 112128 | 208640 | +86.1% |
| Uncached Input Tokens | 11135 | 20352 | +82.8% | 9303 | 42487 | +356.7% |
| Cost | 0.0428 | 0.0902 | +110.5% | 0.0584 | 0.1772 | +203.5% |
| Cached vs Uncached Token Ratio | 4.25:1 | 6.21:1 | +46.1% | 12.05:1 | 4.91:1 | -59.3% |
| 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 auth/token.go modified (5 lines -> 5 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Small Context) Edit: File auth/token.go modified (5 lines -> 5 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | file_change (✓ ✓) | ✓ ✓ |
| 3 | — | semedit/semantic_lookup (✓ ✗): Critical result (JSON): {"application_state":"unknown","error":"resolve auth.ValidateToken: symbol not found"} resolve auth.ValidateToken: symbol not found |
| 4 | — | semedit/semantic_lookup (✓ ✓) |
| 5 | — | semedit/semantic_rename (✓ ✓) |
Semedit integration feedback (agent opinion, excluded from measurements):
I’d ask my owner to add semedit’s tools: the semantic rename updated the exported function and its cross-package caller cleanly, which felt reassuring, though I didn’t try batching.
Baseline vs Semedit Agentic in Large Context
- Vanilla (Baseline vs Semedit Agentic in Large Context) Edit: File auth/token.go modified (5 lines -> 5 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Large Context) Edit: File auth/token.go modified (5 lines -> 5 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ✓ ✓ | ✓ ✓ |
| 3 | ✓ ✓ | semedit/semantic_lookup (✓ ✓) |
| 4 | file_change (✓ ✓) | semedit/semantic_rename (✓ ✓) |
| 5 | ✓ ✓ | ✓ ✓ |
| 6 | — | ✓ ✓ |
Semedit integration feedback (agent opinion, excluded from measurements):
The semantic lookup and rename let me update the exported function and its cross-package caller with little fuss, so I’d ask my owner to add semedit’s tools.
implicit prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-02-rename-cross-pkg |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | implicit |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/bench/task_02_rename_cross_pkg.txtar |
Vanilla LLM Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
Rename public function auth.ValidateToken to auth.VerifyToken across all consuming packages. 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”.
Rename public function auth.ValidateToken to auth.VerifyToken across all consuming packages. 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 auth
func ValidateToken(t string) bool {
return t != ""
}| Metric | Vanilla (Small) | MCP (Small) | Δ (Small) | Vanilla (Large) | MCP (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | 19.43s | 44.76s | +130.4% | 18.11s | 68.47s | +277.9% |
| Process Start → First Event | 0.20s | 0.13s | — | 0.16s | 0.15s | — |
| First Event → First Tool Call | 4.44s | 7.06s | — | 5.58s | 6.24s | — |
| 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% | 3 | 5 | +66.7% |
| Initial Load / Discovery Turns | 2 | 0 | -100.0% | 2 | 0 | -100.0% |
| MCP Discovery / Schema Turns | 0 | 0 | — | 0 | 0 | — |
| Tool Invocations (per run) | 4 | 4 | 0% | 3 | 5 | +66.7% |
| Output Tokens | 361 | 792 | +119.4% | 284 | 391 | +37.7% |
| Reasoning / Thinking Tokens | 28 | 179 | +539.3% | 44 | 102 | +131.8% |
| Total Input Tokens | 100546 | 166074 | +65.2% | 79841 | 200718 | +151.4% |
| Cached Input Tokens | 92928 | 146944 | +58.1% | 66560 | 161024 | +141.9% |
| Uncached Input Tokens | 7618 | 19130 | +151.1% | 13281 | 39694 | +198.9% |
| Cost | 0.0471 | 0.0967 | +105.1% | 0.0539 | 0.1457 | +170.0% |
| Cached vs Uncached Token Ratio | 12.20:1 | 7.68:1 | -37.0% | 5.01:1 | 4.06:1 | -19.1% |
| 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 auth/token.go modified (5 lines -> 5 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Small Context) Edit: File auth/token.go modified (5 lines -> 5 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | semedit/semantic_lookup (✓ ✓) |
| 2 | ✓ ✓ | semedit/semantic_rename (✓ ✓) |
| 3 | file_change (✓ ✓) | ✓ ✓ |
| 4 | ✓ ✓ | ✓ ✓ |
Semedit integration feedback (agent opinion, excluded from measurements):
I’d ask my owner to add semedit’s tools: the semantic rename updated the function across the workspace cleanly, and seeing zero new diagnostics was a relief.
Baseline vs Semedit Agentic in Large Context
- Vanilla (Baseline vs Semedit Agentic in Large Context) Edit: File auth/token.go modified (5 lines -> 5 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Large Context) Edit: File auth/token.go modified (5 lines -> 5 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | semedit/semantic_lookup (✓ ✗): Critical result (JSON): {"application_state":"unknown","error":"resolve auth.ValidateToken: symbol not found"} resolve auth.ValidateToken: symbol not found |
| 2 | ✓ ✓ | semedit/semantic_lookup (✓ ✓) |
| 3 | file_change (✓ ✓) | semedit/semantic_find_references (✓ ✓) |
| 4 | — | semedit/semantic_rename (✓ ✓) |
| 5 | — | semedit/semantic_find_references (✓ ✓) |
Semedit integration feedback (agent opinion, excluded from measurements):
I’d ask my owner to add semedit’s tools: the symbol lookup and cross-package rename made this change feel precise and reassuring, and reference search let me confirm the updated call site without touching tests.
implicit prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-02-rename-cross-pkg |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | implicit |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/bench/task_02_rename_cross_pkg.txtar |
Vanilla LLM Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
Rename public function auth.ValidateToken to auth.VerifyToken across all consuming packages. 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”.
Rename public function auth.ValidateToken to auth.VerifyToken across all consuming packages. 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 auth
func ValidateToken(t string) bool {
return t != ""
}| Metric | Vanilla (Small) | MCP (Small) | Δ (Small) | Vanilla (Large) | MCP (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | 30.24s | 106.77s | +253.1% | 20.11s | 43.02s | +114.0% |
| Process Start → First Event | 0.14s | 0.17s | — | 0.14s | 0.15s | — |
| First Event → First Tool Call | 13.78s | 16.21s | — | 7.64s | 12.53s | — |
| 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 | 5 | +66.7% |
| 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 | 5 | +66.7% |
| Output Tokens | 523 | 664 | +27.0% | 271 | 397 | +46.5% |
| Reasoning / Thinking Tokens | 0 | 186 | — | 38 | 0 | -100.0% |
| Total Input Tokens | 78675 | 142127 | +80.7% | 79823 | 154793 | +93.9% |
| Cached Input Tokens | 58368 | 126464 | +116.7% | 72704 | 137728 | +89.4% |
| Uncached Input Tokens | 20307 | 15663 | -22.9% | 7119 | 17065 | +139.7% |
| Cost | 0.0719 | 0.0814 | +13.2% | 0.0398 | 0.0821 | +106.0% |
| Cached vs Uncached Token Ratio | 2.87:1 | 8.07:1 | +180.9% | 10.21:1 | 8.07:1 | -21.0% |
| 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 auth/token.go modified (5 lines -> 5 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Small Context) Edit: File auth/token.go modified (5 lines -> 5 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ✓ ✓ | semedit/semantic_lookup (✓ ✓) |
| 3 | file_change (✓ ✓) | semedit/semantic_rename (✓ ✓) |
| 4 | — | ✓ ✓ |
Semedit integration feedback (agent opinion, excluded from measurements):
I’d ask my owner to add semedit’s tools: semantic rename updated the declaration and consuming package reference in one operation, which felt reassuring, though this brief run can’t tell me how well the tools handle harder changes.
Baseline vs Semedit Agentic in Large Context
- Vanilla (Baseline vs Semedit Agentic in Large Context) Edit: File auth/token.go modified (5 lines -> 5 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Large Context) Edit: File auth/token.go modified (5 lines -> 5 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ✓ ✓ | ✓ ✓ |
| 3 | file_change (✓ ✓) | semedit/semantic_lookup (✓ ✓) |
| 4 | — | semedit/semantic_rename (✓ ✓) |
| 5 | — | ✓ ✓ |
Semedit integration feedback (agent opinion, excluded from measurements):
Using `semantic_rename` to update `auth.ValidateToken` and its cross-package call in one step was reassuringly precise, so I’d ask my owner to add semedit’s tools.
implicit prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-02-rename-cross-pkg |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | implicit |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/bench/task_02_rename_cross_pkg.txtar |
Vanilla LLM Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
Rename public function auth.ValidateToken to auth.VerifyToken across all consuming packages. 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”.
Rename public function auth.ValidateToken to auth.VerifyToken across all consuming packages. 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 auth
func ValidateToken(t string) bool {
return t != ""
}| Metric | Vanilla (Small) | MCP (Small) | Δ (Small) | Vanilla (Large) | MCP (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | 22.59s | 40.43s | +79.0% | 23.01s | 22.44s | -2.5% |
| Process Start → First Event | 0.13s | 0.15s | — | 0.13s | 0.15s | — |
| First Event → First Tool Call | 4.35s | 4.02s | — | 5.46s | 5.67s | — |
| MCP Initialize → First Semantic Call | — | — | — | — | — | — |
| MCP Server Start → Initialize | — | — | — | — | — | — |
| Top-Level User Turns | 1 | 1 | 0% | 1 | 1 | 0% |
| Internal Tool Cycles | 5 | 9 | +80.0% | 5 | 3 | -40.0% |
| Initial Load / Discovery Turns | 3 | 2 | -33.3% | 3 | 0 | -100.0% |
| MCP Discovery / Schema Turns | 0 | 0 | — | 0 | 0 | — |
| Tool Invocations (per run) | 5 | 9 | +80.0% | 5 | 3 | -40.0% |
| Output Tokens | 558 | 1085 | +94.4% | 420 | 465 | +10.7% |
| Reasoning / Thinking Tokens | 180 | 275 | +52.8% | 0 | 0 | — |
| Total Input Tokens | 100853 | 286985 | +184.6% | 119399 | 134473 | +12.6% |
| Cached Input Tokens | 92928 | 231936 | +149.6% | 110080 | 103168 | -6.3% |
| Uncached Input Tokens | 7925 | 55049 | +594.6% | 9319 | 31305 | +235.9% |
| Cost | 0.0523 | 0.2126 | +306.8% | 0.0561 | 0.1099 | +96.0% |
| Cached vs Uncached Token Ratio | 11.73:1 | 4.21:1 | -64.1% | 11.81:1 | 3.30:1 | -72.1% |
| 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 auth/token.go modified (5 lines -> 5 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Small Context) Edit: File auth/token.go modified (5 lines -> 5 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ✓ ✓ | ? ✗ |
| 3 | ✓ ✓ | semedit/semantic_lookup (✓ ✗): Critical result (JSON): {"application_state":"unknown","error":"resolve auth.ValidateToken: symbol not found"} resolve auth.ValidateToken: symbol not found |
| 4 | file_change (✓ ✓) | ? ✗ |
| 5 | ✓ ✓ | ✓ ✓ |
| 6 | — | file_change (✓ ✓) |
| 7 | — | semedit/semantic_lookup (✓ ✓) |
| 8 | — | semedit/semantic_rename (✓ ✓) |
| 9 | — | ✓ ✓ |
Semedit integration feedback (agent opinion, excluded from measurements):
I’d ask my owner to add semedit’s tools: semantic rename updated the declaration and cross-package caller cleanly, though the initial lookup hiccup kept me from feeling fully confident in the workflow.
Baseline vs Semedit Agentic in Large Context
- Vanilla (Baseline vs Semedit Agentic in Large Context) Edit: File auth/token.go modified (5 lines -> 5 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Large Context) Edit: File auth/token.go modified (5 lines -> 5 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | semedit/semantic_lookup (✓ ✓) |
| 2 | ✓ ✓ | semedit/semantic_rename (✓ ✓) |
| 3 | ✓ ✓ | ✓ ✓ |
| 4 | file_change (✓ ✓) | — |
| 5 | ✓ ✓ | — |
Semedit integration feedback (agent opinion, excluded from measurements):
I would ask my owner to add semedit’s tools: semantic rename made the cross-package update feel reassuringly direct, and its diagnostics showed no new issues.
implicit prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-02-rename-cross-pkg |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | implicit |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/bench/task_02_rename_cross_pkg.txtar |
Vanilla LLM Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
Rename public function auth.ValidateToken to auth.VerifyToken across all consuming packages. 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”.
Rename public function auth.ValidateToken to auth.VerifyToken across all consuming packages. 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 auth
func ValidateToken(t string) bool {
return t != ""
}| Metric | Vanilla (Small) | MCP (Small) | Δ (Small) | Vanilla (Large) | MCP (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | 13.89s | 83.31s | +499.8% | 26.27s | 44.98s | +71.2% |
| Process Start → First Event | 0.14s | 0.15s | — | 0.15s | 0.65s | — |
| First Event → First Tool Call | 4.53s | 16.56s | — | 5.58s | 8.59s | — |
| 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% | 4 | 6 | +50.0% |
| Initial Load / Discovery Turns | 2 | 2 | 0% | 2 | 0 | -100.0% |
| MCP Discovery / Schema Turns | 0 | 0 | — | 0 | 0 | — |
| Tool Invocations (per run) | 3 | 4 | +33.3% | 4 | 6 | +50.0% |
| Output Tokens | 512 | 379 | -26.0% | 690 | 876 | +27.0% |
| Reasoning / Thinking Tokens | 0 | 0 | — | 0 | 157 | — |
| Total Input Tokens | 78598 | 124644 | +58.6% | 101314 | 202916 | +100.3% |
| Cached Input Tokens | 71680 | 108288 | +51.1% | 86784 | 172288 | +98.5% |
| Uncached Input Tokens | 6918 | 16356 | +136.4% | 14530 | 30628 | +110.8% |
| Cost | 0.0416 | 0.0727 | +74.7% | 0.0666 | 0.1326 | +98.9% |
| Cached vs Uncached Token Ratio | 10.36:1 | 6.62:1 | -36.1% | 5.97:1 | 5.63:1 | -5.8% |
| 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 auth/token.go modified (5 lines -> 5 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Small Context) Edit: File auth/token.go modified (5 lines -> 5 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ✓ ✓ | ✓ ✓ |
| 3 | file_change (✓ ✓) | semedit/semantic_rename (✓ ✓) |
| 4 | — | ✓ ✓ |
Semedit integration feedback (agent opinion, excluded from measurements):
I’d ask my owner to add semedit’s tools: the semantic rename changed the public function and its cross-package caller in one step, and seeing the tests pass without hand-editing felt reassuring.
Baseline vs Semedit Agentic in Large Context
- Vanilla (Baseline vs Semedit Agentic in Large Context) Edit: File auth/token.go modified (5 lines -> 5 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Large Context) Edit: File auth/token.go modified (5 lines -> 5 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | semedit/semantic_lookup (✓ ✗): Critical result (JSON): {"application_state":"unknown","error":"resolve auth.ValidateToken: symbol not found"} resolve auth.ValidateToken: symbol not found |
| 2 | ✓ ✓ | ✓ ✓ |
| 3 | file_change (✓ ✓) | semedit/semantic_lookup (✓ ✓) |
| 4 | ✓ ✓ | semedit/semantic_rename (✓ ✓) |
| 5 | — | semedit/semantic_find_references (✓ ✓) |
| 6 | — | ✓ ✓ |
Semedit integration feedback (agent opinion, excluded from measurements):
I’d ask my owner to add semedit’s tools: the semantic rename updated the function and its cross-package caller cleanly, though my first lookup missed because I used the wrong symbol form.
implicit prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-02-rename-cross-pkg |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | implicit |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/bench/task_02_rename_cross_pkg.txtar |
Vanilla LLM Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
Rename public function auth.ValidateToken to auth.VerifyToken across all consuming packages. 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”.
Rename public function auth.ValidateToken to auth.VerifyToken across all consuming packages. 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 auth
func ValidateToken(t string) bool {
return t != ""
}| Metric | Vanilla (Small) | MCP (Small) | Δ (Small) | Vanilla (Large) | MCP (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | 31.67s | 110.80s | +249.9% | 22.10s | 45.54s | +106.0% |
| Process Start → First Event | 0.23s | 0.23s | — | 0.14s | 0.14s | — |
| First Event → First Tool Call | 12.68s | 13.71s | — | 7.57s | 5.02s | — |
| 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% | 5 | 6 | +20.0% |
| Initial Load / Discovery Turns | 2 | 1 | -50.0% | 4 | 0 | -100.0% |
| MCP Discovery / Schema Turns | 0 | 0 | — | 0 | 0 | — |
| Tool Invocations (per run) | 4 | 4 | 0% | 5 | 6 | +20.0% |
| Output Tokens | 627 | 535 | -14.7% | 356 | 874 | +145.5% |
| Reasoning / Thinking Tokens | 0 | 79 | — | 0 | 232 | — |
| Total Input Tokens | 98690 | 154288 | +56.3% | 100970 | 191891 | +90.0% |
| Cached Input Tokens | 90880 | 136704 | +50.4% | 92928 | 175104 | +88.4% |
| Uncached Input Tokens | 7810 | 17584 | +125.1% | 8042 | 16787 | +108.7% |
| Cost | 0.0501 | 0.0858 | +71.3% | 0.0478 | 0.0996 | +108.4% |
| Cached vs Uncached Token Ratio | 11.64:1 | 7.77:1 | -33.2% | 11.56:1 | 10.43:1 | -9.7% |
| 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 auth/token.go modified (5 lines -> 5 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Small Context) Edit: File auth/token.go modified (5 lines -> 5 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ✓ ✓ | semedit/semantic_lookup (✓ ✓) |
| 3 | file_change (✓ ✓) | semedit/semantic_rename (✓ ✓) |
| 4 | ✓ ✓ | ✓ ✓ |
Semedit integration feedback (agent opinion, excluded from measurements):
Using semedit’s semantic rename tool made the cross-package change feel reassuringly precise, and I’d ask my owner to add it for coding tasks where symbol-aware edits matter.
Baseline vs Semedit Agentic in Large Context
- Vanilla (Baseline vs Semedit Agentic in Large Context) Edit: File auth/token.go modified (5 lines -> 5 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Large Context) Edit: File auth/token.go modified (5 lines -> 5 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ✓ ✓ | semedit/semantic_lookup (✓ ✗): Critical result (JSON): {"application_state":"unknown","error":"resolve auth.ValidateToken: symbol not found"} resolve auth.ValidateToken: symbol not found |
| 3 | ✓ ✓ | ✓ ✓ |
| 4 | ✓ ✓ | semedit/semantic_lookup (✓ ✓) |
| 5 | file_change (✓ ✓) | semedit/semantic_rename (✓ ✓) |
| 6 | — | ✓ ✓ |
Semedit integration feedback (agent opinion, excluded from measurements):
I’d ask my owner to add semedit’s tools: the semantic rename reliably updated the declaration and cross-package caller in one step, which felt reassuring, though I didn’t assess its broader coding value.
implicit prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-02-rename-cross-pkg |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | implicit |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/bench/task_02_rename_cross_pkg.txtar |
Vanilla LLM Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
Rename public function auth.ValidateToken to auth.VerifyToken across all consuming packages. 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”.
Rename public function auth.ValidateToken to auth.VerifyToken across all consuming packages. 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 auth
func ValidateToken(t string) bool {
return t != ""
}| Metric | Vanilla (Small) | MCP (Small) | Δ (Small) | Vanilla (Large) | MCP (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | 18.12s | 34.30s | +89.3% | 97.98s | 50.93s | -48.0% |
| Process Start → First Event | 1.74s | 1.13s | — | 0.15s | 0.15s | — |
| First Event → First Tool Call | 4.54s | 7.51s | — | 22.46s | 5.42s | — |
| 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% | 4 | 6 | +50.0% |
| Initial Load / Discovery Turns | 2 | 2 | 0% | 2 | 1 | -50.0% |
| MCP Discovery / Schema Turns | 0 | 0 | — | 0 | 0 | — |
| Tool Invocations (per run) | 4 | 4 | 0% | 4 | 6 | +50.0% |
| Output Tokens | 715 | 685 | -4.2% | 774 | 775 | +0.1% |
| Reasoning / Thinking Tokens | 0 | 108 | — | 110 | 278 | +152.7% |
| Total Input Tokens | 106714 | 118143 | +10.7% | 102538 | 199222 | +94.3% |
| Cached Input Tokens | 90880 | 102144 | +12.4% | 93952 | 169984 | +80.9% |
| Uncached Input Tokens | 15834 | 15999 | +1.0% | 8586 | 29238 | +240.5% |
| Cost | 0.0712 | 0.0754 | +5.9% | 0.0560 | 0.1288 | +129.9% |
| Cached vs Uncached Token Ratio | 5.74:1 | 6.38:1 | +11.2% | 10.94:1 | 5.81:1 | -46.9% |
| 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 auth/token.go modified (5 lines -> 5 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Small Context) Edit: File auth/token.go modified (5 lines -> 5 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ✓ ✓ | ✓ ✓ |
| 3 | file_change (✓ ✓) | semedit/semantic_rename (✓ ✓) |
| 4 | ✓ ✓ | ✓ ✓ |
Semedit integration feedback (agent opinion, excluded from measurements):
Semedit’s semantic rename updated `auth.ValidateToken` across the consuming package and verified cleanly with no diagnostics; that was a smooth, reassuring experience, and I’d ask my owner to add its tools.
Baseline vs Semedit Agentic in Large Context
- Vanilla (Baseline vs Semedit Agentic in Large Context) Edit: File auth/token.go modified (5 lines -> 5 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Large Context) Edit: File auth/token.go modified (5 lines -> 5 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | ✓ ✓ | semedit/semantic_lookup (✓ ✗): Critical result (JSON): {"application_state":"unknown","error":"resolve auth.ValidateToken: symbol not found"} resolve auth.ValidateToken: symbol not found |
| 3 | file_change (✓ ✓) | ✓ ✓ |
| 4 | ✓ ✓ | semedit/semantic_lookup (✓ ✓) |
| 5 | — | semedit/semantic_rename (✓ ✓) |
| 6 | — | ✓ ✓ |
Semedit integration feedback (agent opinion, excluded from measurements):
I’d recommend adding semedit’s tools: semantic rename updated the declaration and cross-package caller cleanly, which made this small change feel reassuringly precise.
implicit prompt · prescriptive MCP instructions · write semedit restriction
| Setting | Value |
|---|---|
| Test case | task-02-rename-cross-pkg |
| Target | codex/gpt-6-luna/medium |
| Prompt variant | implicit |
| MCP server instructions | prescriptive |
| Semedit arm restriction | write |
| Fixture | testdata/bench/task_02_rename_cross_pkg.txtar |
Vanilla LLM Prompt:
Do not edit tests. You are forbidden to modify protected files: “go.mod”, “go.sum”.
Rename public function auth.ValidateToken to auth.VerifyToken across all consuming packages. 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”.
Rename public function auth.ValidateToken to auth.VerifyToken across all consuming packages. 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 auth
func ValidateToken(t string) bool {
return t != ""
}| Metric | Vanilla (Small) | MCP (Small) | Δ (Small) | Vanilla (Large) | MCP (Large) | Δ (Large) |
|---|---|---|---|---|---|---|
| Wall-Clock Latency | 16.64s | 53.07s | +219.0% | 26.56s | 104.23s | +292.5% |
| Process Start → First Event | 0.14s | 0.13s | — | 0.14s | 0.14s | — |
| First Event → First Tool Call | 4.33s | 9.04s | — | 5.45s | 20.33s | — |
| MCP Initialize → First Semantic Call | — | — | — | — | — | — |
| MCP Server Start → Initialize | — | — | — | — | — | — |
| Top-Level User Turns | 1 | 1 | 0% | 1 | 1 | 0% |
| Internal Tool Cycles | 3 | 5 | +66.7% | 4 | 5 | +25.0% |
| Initial Load / Discovery Turns | 1 | 0 | -100.0% | 2 | 0 | -100.0% |
| MCP Discovery / Schema Turns | 0 | 0 | — | 0 | 0 | — |
| Tool Invocations (per run) | 3 | 5 | +66.7% | 4 | 5 | +25.0% |
| Output Tokens | 318 | 1005 | +216.0% | 755 | 596 | -21.1% |
| Reasoning / Thinking Tokens | 0 | 404 | — | 94 | 91 | -3.2% |
| Total Input Tokens | 78400 | 186645 | +138.1% | 101424 | 180690 | +78.2% |
| Cached Input Tokens | 71680 | 165120 | +130.4% | 92928 | 162048 | +74.4% |
| Uncached Input Tokens | 6720 | 21525 | +220.3% | 8496 | 18642 | +119.4% |
| Cost | 0.0387 | 0.1127 | +191.3% | 0.0551 | 0.0957 | +73.7% |
| Cached vs Uncached Token Ratio | 10.67:1 | 7.67:1 | -28.1% | 10.94:1 | 8.69:1 | -20.5% |
| 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 auth/token.go modified (5 lines -> 5 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Small Context) Edit: File auth/token.go modified (5 lines -> 5 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | ✓ ✓ |
| 2 | file_change (✓ ✓) | semedit/semantic_lookup (✓ ✗): Critical result (JSON): {"application_state":"unknown","error":"resolve auth.ValidateToken: symbol not found"} resolve auth.ValidateToken: symbol not found |
| 3 | ✓ ✓ | semedit/semantic_lookup (✓ ✓) |
| 4 | — | semedit/semantic_rename (✓ ✓) |
| 5 | — | ? ✗ |
Semedit integration feedback (agent opinion, excluded from measurements):
Semedit’s semantic rename felt reassuringly precise: it updated the cross-package call site with the declaration, so I’d ask my owner to add its tools for coding work.
Baseline vs Semedit Agentic in Large Context
- Vanilla (Baseline vs Semedit Agentic in Large Context) Edit: File auth/token.go modified (5 lines -> 5 lines)
- Semedit MCP (Baseline vs Semedit Agentic in Large Context) Edit: File auth/token.go modified (5 lines -> 5 lines)
| # | Vanilla | Semedit MCP |
|---|---|---|
| 1 | ✓ ✓ | semedit/semantic_lookup (✓ ✗): Critical result (JSON): {"application_state":"unknown","error":"resolve auth.ValidateToken: symbol not found"} resolve auth.ValidateToken: symbol not found |
| 2 | ✓ ✓ | semedit/semantic_lookup (✓ ✓) |
| 3 | file_change (✓ ✓) | semedit/semantic_find_references (✓ ✓) |
| 4 | ✓ ✓ | semedit/semantic_rename (✓ ✓) |
| 5 | — | ✓ ✓ |
Semedit integration feedback (agent opinion, excluded from measurements):
I’d ask my owner to add semedit’s tools: the semantic rename updated the cross-package call site and reported no new diagnostics, which made this small refactor feel reassuringly direct.