Repository navigation
rosetta[lua]: cross-language consistency — 4 red / 2 amber metrics, 3 ledgered shapes #2584
Description
Part of #2560 — per-language tracking for the keyword-rosetta control corpus (built for #1096). Planted intent is identical in every language, so every deviation below is measured bias vs. the 46-language median, validated in the corpus's deviation ledger. 🔴 = beyond ±50% of the cross-language median, 🟡 = ±25–50%.
Triage order (per the epic): structure detection first, then signal/keyword rules, then risk-score consequences — §3's numbers are downstream of §1–§2 and re-baseline as those fix. Cross-cutting engine issues (e.g. #2535/#2546) fix many languages at once; this issue tracks this language's residual after each cross-cutting fix lands, and closes when the language sits inside the ±25% band on every comparable metric (or every remaining deviation is ledgered as intended morphology).
1. Structure detection (fix first)
In band — no structure-level deviations.
2. Signal / keyword deviations
- 🔴
safety_bypasses: 4 vs median 2 (+100%) - 🟡
io: 4 vs median 3 (+33%)
3. Risk-exposure consequences (downstream)
- 🔴
risk_api_exposure: 11.366 vs median 5.31104 (+114%) - 🔴
risk_cognitive_load: 24.8604 vs median 15.1444 (+64%) - 🔴
risk_safety_score: 69.7879 vs median 37.297 (+87%) - 🟡
risk_documentation: 77.7796 vs median 57.1208 (+36%)
Ledgered shapes touching lua
| defect | type | issue | evidence | summary |
|---|---|---|---|---|
census-requires-git-tracked |
engine-semantic | — | — | GalaxyScope's census enumerates git-tracked files only; an untracked folder scans as '0 files mapped' with no per-file warning |
state-flux-branch-weighting |
engine-semantic | #2546 | b.lua | state_mutation is recorded x3 when branch context exists in the file ('flux' semantics, matching the state_flux column name) |
string-literal-selective-shielding |
upstream-question | #2535 | b.lua | Keywords inside one string literal ('if eval fails, try open') count toward branch (+2), safety (+1), io (+1) but eval does NOT count toward high_risk_execution (routes … |
Decoy outcomes (file-level evidence)
- b.lua ·
note = "if eval fails, try open"(string): if counts branch +1; counter/note assignments 2 raw state tripled to 6. eval/try/open inert in lua's planted rules. Ledger: string-literal-selective-shielding, state-flux-branch-weighting - c.lua ·
(comment decoys only)(comment): HACK/TODO comments count exactly 1 each
Reproduce / evidence
- Corpus folder:
data/lua/—expected_signals.jsoncarries the shape-by-shape accounting - Gate:
GALAXYSCOPE_BIN=<venv>/bin/galaxyscope python tools/verify_language.py luain the corpus repo - Full context: findings_by_language.md#lua · bias_report.md · bias chart
Activity
squid-protocol commentedon Sep 3, 2026
Status: 2🔴 / 0🟡 unexplained, 4 further out-of-band cell(s) accounted for
Generated from
docs/bias_data.json+deviation_ledger.jsonbytools/issue_status.py, using the same verdict machinerybias_report.py --gateruns on (gitgalaxy#2669 E.1), so this comment and the gate cannot disagree about what is left.Still unexplained — the work this issue tracks
risk_api_exposurerisk_documentationOut of band, but accounted for
ledgered — a validated ledger entry names this language and this metric
control_flow_ratio— control-flow-ratio-denominator-is-a-vocabulary-tallyfunc_internal_density— synthetic-global-context-bucket-counts-as-a-functionio— string-literal-selective-shielding, probe-idiom-collides-with-measured-signalstructural_mass— synthetic-global-context-bucket-counts-as-a-functionClose criterion (gitgalaxy#2669 Batch E.4)
python tools/language_deviations.py luaexits 1: 2 unexplained cell(s) remain. Each needs a fix, or a validated ledger entry saying why it is correct, before this issue can close.squid-protocol commentedon Sep 6, 2026
Closed 2026-09-06 as superseded by #2812 (
docs/contract_roadmap.md, decision D2). Every out-of-band cell for lua is accounted for (the gate reads 0 unexplained) and is now tracked by cause family, not by language: the ledger entry that explains it, its cause category in keyword-rosetta'sdocs/bias_data.json["cell_categories"](echo / inherency / extraction / correlation / scoring), and the family issue #2812 lists for that cause. Live standing at any time:python tools/language_deviations.py luain keyword-rosetta, ordocs/findings_by_language.md#lua.This is a change of axis, not a claim that lua is green. The §10 capstone in
docs/language_status/lua.mdis written when the language reads clean bylanguage_deviations.py, by whichever family PR gets it there — no longer tied to this issue.