docs(contracts): contract roadmap, signal contract sheet + baseline-gated audit, rule-contract-audit skill

[squid-protocol]

Summary

Phase 1 of docs/contract_roadmap.md (the strategy approved 2026-09-06, companion artifact linked from the doc). No engine behaviour change — nothing at scan time imports the new module, so no golden-master delta.

What it adds:

• docs/contract_roadmap.md — the assessment: of the 267 gated red cells in the keyword-rosetta report, 102 are echoes, ~45 language inherency, ~50 extraction defects, 38 proximity-correlation effects, ~32 scoring choices; the consistency badge could not tell these apart. The three-layer diagnosis (stream / count / score), decisions D1–D4, phases 0–5, issue disposition. Corrects two earlier claims (prism/detector: string-literal shielding is selective by signal family (if/try/open count inside a literal, eval does not) #2535 closed as not-a-bug: there is no string shielding to make uniform; the ×3 flux lives in spatial_correlation.py, pinned by docs(core): document + pin the ×3 cascading-flux weighting (closes #2546) #2631).
• gitgalaxy/standards/signal_contracts.py — the contract sheet (D3): one SignalContract per registry rule key — 68 entries over the 80 keys the 59 language definitions declare, the 12 _-prefixed helpers described separately — each with kind (declaration / site / annotation / tally), unit, the sentence, status, doc, issue, planted. The stream contract and the count contract as module constants. api and args are stated; the other 66 are draft (the schema comment, verbatim).
• tests/signal_contract_audit.py + baseline + .github/workflows/signal-contract-audit.yml — baseline-gated like dead_key_audit.py: missing-contract, orphan-contract, schema-drift (the how_to_add_a_language.md comment must contain the sentence), missing-schema, draft. --render writes docs/signal_contracts.md.
• how_to_add_a_language.md — CRITICAL ENGINE RULE 18 (what text a rule sees) and a pointer above the OUTPUT SCHEMA.
• .claude/skills/rule-contract-audit/SKILL.md — the The api rule has no stated contract: 7 languages' own visibility idiom is invisible to it, and 5 others count any public token #2730/fix(core-engine): give the api rule a stated contract and audit all 46 languages against it (#2730) #2743 method as a repeatable workflow for taking one signal from draft to stated.
• docs/ecosystem.md — skills table; the rosetta workflow row is now a cause-family workflow.

Proof

$ python tests/signal_contract_audit.py --ci
signal contracts: 68 entries, 2 stated, 66 draft, 80 registry keys
signal_contract_audit: no new findings

ruff format clean; ruff_audit.py --ci / mypy_audit.py --ci: the only new hits are in untracked scratch files already sitting in the working tree (gitgalaxy/tests.py, gen_tests.py, parse_regex.py — not in this PR); dead_key_audit.py --ci clean; pytest tests/core_engine: 372 passed, 1 pre-existing failure unrelated to this diff (test_committed_fidelity_table_is_fresh_against_the_corpus — corpus-state dependent, see comment below).

Cross-repo

• Companion: keyword-rosetta branch tools/cell-categories (Phase 0: the cause roll-up in bias_report.py, GATING.md category semantics, rosetta-language-sweep re-scoped). Independent of this PR — neither side's CI reads the other's change; merge order does not matter. Corpus PR: tools(bias_report): say what each red cell is — cause categories, open-defect share, decayed-entry check keyword-rosetta#77.
• Nothing re-runs after merge. The sheet becomes importable by the corpus's _registry.py for Phase 4.
• Epic (handoff surface for phases 2–5): Epic: contracts at every layer — stream, count, score (the rosetta roadmap; child of #2560) #2812. Phase 2 engine issue (D1): spatial_correlation.py edits recorded signal counts in place — move the proximity weights to the formulas so a count is a count (roadmap Phase 2, D1) #2813.

🤖 Generated with Claude Code

[squid-protocol]

Pre-existing local failure, not this PR's: tests/core_engine/test_language_strictness.py::test_committed_fidelity_table_is_fresh_against_the_corpus fails on this branch and on origin/main alike — tests/tools/fidelity_table.py --check --corpus ../keyword-rosetta reports the committed gitgalaxy/standards/fidelity_table.py stale against corpus main (555604e, keyword-rosetta #74 merged today). This diff touches neither the table nor its inputs; the regen is owed by whichever PR next touches the fidelity layer (one layer per PR). Logged in the contract-roadmap epic's housekeeping list.

[github-actions]

🐦‍⬛ Muninn Security Scan

✅ No security issues found.

🐦‍⬛ Powered by Muninn · Skald Lab