Rust-hybrid pre-release: TS/JS and Go Agent Sufficiency refresh

[jununfly]

This was generated by AI during triage.

Parent

• Rust-hybrid pre-release polish tracker #275

What to build

Refresh README-backed Agent Sufficiency claims with current rust-hybrid evidence for one TS/JS representative and one Go representative. This is release-claim hygiene, not #165 performance optimization. Do not use deterministic smoke as a substitute for agent A/B metrics.

Default representatives are Excalidraw for TS/JS and Gin for Go unless current corpus availability requires an explicit documented substitution.

Acceptance criteria

• TS/JS representative repo is selected and recorded, defaulting to Excalidraw.
• Go representative repo is selected and recorded, defaulting to Gin.
• Each repo runs 2 flow prompts.
• Each prompt runs 2 runs per arm, WITH ZCodeGraph vs WITHOUT ZCodeGraph.
• Read/Grep fallback, tool calls, wall time, and cost/tokens when available are recorded.
• Corpus commits, corpus paths, ZCodeGraph commit, runner version/context, and unavailable reasons are recorded.
• README metrics are replaced only for claims covered by this evidence.
• If agent A/B cannot run, evidence records the unavailable reason and README metric replacement is blocked rather than invented.
• Full 7-repo scoreboard is not run.
• Evidence is written under docs/benchmarks/.

Blocked by

• Rust-hybrid pre-release: API surface cleanup #276
• Rust-hybrid pre-release: docs, PRD, and changelog alignment #277

[jununfly]

Completed in 08a2434 with an explicit caveat. Ran real Claude Code headless A/B spot-checks after rust-hybrid indexing: Excalidraw TS/React showed strong exploration reduction (61s/7 tools/2 fallbacks with ZCodeGraph vs 222s/54 tools/53 fallbacks without); Gin route overview was clean (24s/1 tool/0 fallbacks vs 78s/28 tools/27 fallbacks); Gin POST /upload did not improve (29s/5 tools/4 fallbacks vs 26s/4 tools/4 fallbacks). README now reflects this as targeted pre-release evidence, not a full benchmark replacement. Evidence: docs/benchmarks/2026-06-19-rust-hybrid-pre-release-agent-sufficiency.md. Validation: full npm test passed.