test(inference): close nonfinite uncertainty review evidence gaps - #729
test(inference): close nonfinite uncertainty review evidence gaps#729seonghobae wants to merge 3 commits into
Conversation
|
Important Review skippedDraft detected. Please check the settings in the CodeRabbit UI or the ⚙️ Run configurationConfiguration used: Path: .coderabbit.yaml Review profile: CHILL Plan: Pro Plus Run ID: You can disable this status message by setting the Use the checkbox below for a quick retry:
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 |
|
Superseded by #730. Fresh comparison shows both branches target the same post-#725 non-finite uncertainty evidence gap. #730 now contains the signed -∞ Rust/Python regressions, current rendered CHANGELOG, and the non-duplicative AERA/NIST/Rust/source-specific doctoring rationale from this branch. No checks or reviews transfer across heads. Closing this Draft to preserve one canonical merge path. |
Post-merge review-completeness correction
Protected
mainat branch creation:3bc1222bf9552bb74f98cf34c89eba0b8d29c0dc.PR #725 correctly landed the production behavior: non-finite Hessians fail closed and non-finite covariance diagonals are preserved instead of becoming false zero uncertainty. A valid current-head CodeRabbit review identified two evidence gaps that remained after that protected merge:
NaNand+∞but not-∞, despite the public/Rust contract explicitly promising signed-infinity preservation; andBounded correction
-∞preservation assertions to public Python↔Rust parity and the Rust unit contract;f64NaN/±∞ representation documentation;No production numerical code, estimator semantics, dependencies, workflow, version, or release behavior changes in this slice. The finite-negative-diagonal clamp remains accepted compatibility behavior; changing it would require a separate scientific/API decision.
Require focused inference tests, Rust workspace/PyO3, Python 3.12/3.14 and all ordinary package/security gates on one unchanged head before merge. This closes the still-valid review findings from #725 without reopening the already-merged production correction.