fix(subscores): seal scientific evidence admission - #1363
Conversation
|
Warning Review limit reachedNext included review available in 1 second. View limit detailsLimit details: You’ve used the included review currently available. You've used all free OSS reviews for now. Wait for the free limit to reset to keep reviewing this public repository. Review configuration: ⚙️ Run configurationConfiguration used: Path: .coderabbit.yaml Review profile: CHILL Plan: Pro Plus Run ID: 📒 Files selected for processing (5)
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
|
@coderabbitai review |
|
Pin the baseline to protected main@9c12eab1 (v0.9.0). The open-PR queue collapsed from 74 to 6 since the previous snapshot; replace the stale PR table with the current six lanes (#1363, #1345, #1279, #1029, #1005, #998) including their merge-forward state. Update gap-matrix evidence columns for landed work: automatic Rust backend (#951/#1070), crossed multiple-membership (#1014), the Figma design-boundary ADR now at docs/adr/0016 via merged #1130, and cross-engine conformance (#1082) after #1077 closed as completed. No production code, formula, estimator, or release surface changed.
|
Merge-forward: replayed this draft onto |
![devin-ai-integration[bot]](https://avatars.githubusercontent.com/in/811515?s=60&v=4){kind=small}
2 participants
Closes #1362.
Closes #1364.
Defects
subscore_analysis()originally admittedresponsesandgroupsthrough unrestrictednp.asarray(...), so caller-defined__array__or nested numeric conversion providers could synthesize or replace the person×item score matrix or item→subscale partition while Python validation was deciding what the Rust-owned Haberman/Sinharay analysis would receive.After that callback boundary was sealed, the exact built-in list/tuple preflight still had no logical-cell, structural-work, or cycle bound. A self-referential container could therefore fail to terminate, malformed zero-cell fan-out could consume unbounded traversal work, and an exact NumPy broadcast leaf with tiny backing storage could expose an oversized logical array before parent-container materialization.
RED → GREEN
Callback-free scientific evidence — #1362
2e3462f3f42d3a977a8973468d13f9becfaa4288: hostile top-level response/group array providers execute zero callbacks and fail before compiled-core discovery.1c9797483478ef6ec65a0e7b21b1bd2e85a6f12d: admit only exact NumPy numeric arrays or exact built-in list/tuple trees of package-trusted concrete Python/NumPy numeric evidence before NumPy materialization.2550d3a17927efaeda6189141b4b254d8005058c: nested hostile numeric providers execute zero callbacks while ordinary built-in sequences containing concrete NumPy scalar values still reach Rust as canonical evidence.e06e0f7bfad8a8ddd5cbbb90f18bacf030c83aad→ GREENbfc8ad20eaaf39a42dcf66c3b16b66b4f8dea895: preserve exact inert NumPy numeric row leaves inside built-in response containers without admitting ndarray subclasses or generic array protocols.Bounded/cycle-safe traversal — #1364
ced9f5bf79d099a70e7d059d6c19e337156c4ae5: exact 20,000,001-cell response evidence and exact broadcast NumPy leaves nested in built-in containers must fail beforenp.asarray(...)or compiled-core discovery.c04778c58893354c25035daa66fc3701c236740a: enforce a 20,000,000 logical-cell envelope, charge exact NumPy leaves from inert.sizemetadata, traverse built-in containers with indexed frames, reject active-path cycles, and cap structural work independently before parent materialization.373f6af9a277bdc14ca8547ae87b1ed5698ce096: true self-cycles fail before NumPy, zero-cell fan-out obeys the structural budget, and repeated shared acyclic rows remain valid rather than being mistaken for cycles.d232423d64279697996005d229cbdca997fcea0erecords both trust and bounded-materialization contracts.Existing complex-specific diagnostics, response shape/completeness checks, integer/nonnegative group contracts, group-allocation bound, Rust dispatch, and result schema remain unchanged.
Scientific ownership
No Cronbach-alpha, observed/disattenuated-correlation, Haberman PRMSE, augmented-score weight, added-value decision, likelihood, uncertainty, or other production psychometric/statistical arithmetic changes. Those computations remain Rust-owned; this PR changes Python validation, bounded materialization, marshalling, and regression evidence only.
Fresh integration boundary
The branch moved compatibly from
d232423d64279697996005d229cbdca997fcea0eto exact current head9982b85abb3664763c59eb2e9dab72251254271cby absorbing unrelated protected-base multilevel/CAT/fitstats work; the effective diff against current protectedmain@d77505e9dc8eee5780bf354e13dd4f4bf83df453remains five subscore files. Current ancestry is 12 commits ahead / 2 behind with merge basef5d6189227545b42d3f384b4728c8d6f611063c7.Exact-current-head CI, Security Scan, CodeQL and SAST Semgrep are all terminal success, there are no formal reviews or unresolved review threads, and the PR is now Ready for independent review. Merge still requires fresh merge-result/base-sensitive evidence, every applicable protected-central coverage/review/package/fuzz/SBOM/provenance/dependency context, resolved conversations, and the independent approvals required by active rulesets. No gate weakening, self-approval, force update, destructive rebase, or Python numerical substitute is used.