[tob-scott-a]

Python entrypoint discovery built lexical scopes for files without decorated functions and scanned every file for replacement writes even when no local identity decorator needed proving. Select decorated functions with a cached Tree-sitter query and defer the replacement scan until an identity proof needs it, preserving binding evidence and source order.

On a fixed 143-file Python corpus, seven alternating A/B runs reduced median full Trailmark analysis time from 2.268s to 1.978s (12.8%), with identical complete graph output: 4,308 nodes, 12,996 edges, and all serialized metadata. Small fixture and TypeScript controls showed no material timing change; the gain is specific to Python decorator/entrypoint analysis.

Add regression cases for decorator bindings, source ordering, file edits, argument boundaries, Python imports, and PHP callable-reference context and dispatch. Focused mutation checks now kill 21 unchanged mutants that survived the latest main CI campaign. Add a reusable analysis benchmark and correct component test ordering for private helpers while retaining every fallback test.

Validation:

• Full suite: 3,261 passed.
• Ruff lint and formatting, ty, source-distribution smoke check, and whitespace checks passed.
• Focused mutation campaign: clean and instrumented baselines passed; all 658 mutants evaluated, with 510 killed, 148 survivors, and no errors or timeouts. This uses two test files and is not a full-repository mutation score.
• The complete repository mutation score has not been recomputed; the baseline is main's completed CI campaign. No mutation exemptions were added.