feat(metrics): count a plain re-assignment in python/embedded_python/ruby state_mutation (#2817)

[squid-protocol]

Summary

state_mutation (feeds risk_state_flux) counted only self.x =/@ivar =, :=, global/nonlocal, container mutators and (ruby) bang methods for python/embedded_python/ruby — never a plain x = v. That is narrower than the #2765 contract's corollary 1 (a language with no declaration syntax counts the assignment statement as the write), so count = count + 1 scored 1 in lua/php/shell/go and 0 in python — the cross-language inequality #2812 removes. The contract doc recorded these three as deferred.

Fix: one house-style assignment arm (mirroring lua/javascript) added to each rule, plus re.M (inert for the existing arms). Anchored on the black/PEP8 convention that a statement assignment writes with a space before = (x = 1) while a keyword argument does not (x=1):

(?:^|;)[ \t]*<lvalue>(?:\.attr|[idx])*[ \t]+=(?![=])(?![^\n(]{0,300},[ \t]*$)

• Counts x = v, obj.attr = v, d[k] = v.
• Excludes foo(x=1), def f(x=1) (no space before =, not statement-start), x == 1, and annotated x: int = 1 (declaration-with-initializer).
• ruby uses a lowercase-initial lvalue (so CONST = 1 stays freeze_hits) and drops =~/=>.
• Out of scope (documented): compound x += 1 and unspaced x=1.

Crucible incidence table (language-crucible @ v1.2.0)

language	before	after	new-arm hits	multi-line-call FP candidates	true false positives
python	1490	11647	10988	6 (0.05%)	~1–3 (last spaced kwarg with no magic trailing comma in a non-black multi-line call; 2 were in a .rst doc file)
embedded_python	191	894	838	0	0
ruby	64	140	76	0	0

The space-before-= anchor + the trailing-comma guard reduce kwargs/multi-line-call false positives to ~0.02%. Sampled matches are otherwise all genuine writes.

Golden rebless (this PR)

Both masters regenerated per-leg in isolated venvs (full-precision / zero-dep). bless_scope.py on both legs: ~4133 diffs = 2656 corpus-wide topological (X/Y/Z) re-solve + ~1478 substantive, all attributable to the reprice — Variable Reassignments/State Flux Exposure/Amplified Cascading Flux directly, plus normalized exposures / Structural Magnitude / directory-group aggregates that are relative to a file's total magnitude. Every raw count change is on a .py/.rb file (incl. python files nested in mixed-language projects like cpp/NVDA); no unchanged-rule language moves its own raw count, and no files were newly parsed or excluded. The scoring layer absorbs the raw increase gracefully (e.g. braille.py reassignments 617→1801 but State Flux Exposure only 55.3→58.3).

Verification

• Rule/contract/strict suites: 372 passed.
• tests/extraction/ + tests/core_engine/: 8337 passed, 0 failed.
• tests/signal_contract_audit.py --ci: no new findings (contract stays stated; only counts move).
• tests/ruff_audit.py --ci (pinned ruff 0.16.0): no new lint findings; the 3 engine files are format-clean.

Docs

docs/state_mutation_rule_contract.md: python/ruby/embedded_python rows moved deferred → agrees with before → after counts; the "what the contract cannot do with a regex" note rewritten to "landed in #2817".

Label: rosetta:rebless-owed — the keyword-rosetta corpus re-bless PR opens against engine main after merge.

Closes #2817

🤖 Generated with Claude Code

[github-actions]

🐦‍⬛ Muninn Security Scan

✅ No security issues found.

🐦‍⬛ Powered by Muninn · Skald Lab