Repository navigation
Rebuild research evidence stack (assurance, R10-R15, sidecar v2) - #13
Merged
Merged
Conversation
Rebuild soak collection so host/role matrix gaps and stale evidence fail closed instead of silently aggregating incomplete runs.
Digest-bound evidence and pass/fail/unknown verification keep replay and bundle rebuild honest when checks cannot decide.
Schema migration normalizes legacy bundles onto the current id so level inference stays consistent across replay and promotion gates.
Fail closed on tampered digests and incomplete evidence so promotion cannot ride on corrupted or gap-filled bundles.
Machine-readable status aggregation keeps research progress claims tied to current assurance evidence rather than stale narrative.
Ranked sampling and disagreement features focus shadow evaluation on consequential near-boundary regimes instead of uniform draws.
New regimes and sampling-study schema updates make CI fixtures exercise boundary disagreement coverage with reproducible corpus versioning.
Verified-forward gating and adapter catalog keep observatory differentiation claims scoped to implemented, checkable targets.
Structured transition probes and study runners replace ad-hoc sweeps so gradient-mode comparisons stay comparable across regimes.
CI and runbook require complete host/role matrices before soak aggregates can count as promotion evidence.
Artifact hash checks bind promotion evidence to governed manifests so swapped files cannot pass multi-host attestation.
Held-out and batch checks harden domain stage gates so CBF claims rest on reproducible feasibility and RH behavior.
Assembled claims, limitations, and replay evidence make flagship promotion decisions explicit and fail-closed when gates are unmet.
Versioned handshake and mode docs keep Moreau sidecar exchanges aligned so interrupted or mismatched sessions fail predictably.
Fail-closed comparison arms and scaffold reports prevent training execution until critical promotion gates clear.
Schema and conftest updates keep CI validation in lockstep with digest fields and three-valued verification status.
Docs and polish tests capture L0-L4 verification language so deliverable claims match the rebuilt evidence model.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Test plan
pytest tests/research tests/platform/sidecar -q