Skip to content

fix: harden intraday evidence provenance - #485

Merged
KCNyu merged 1 commit into
masterfrom
codex/intraday-evidence-hardening
Aug 11, 2026
Merged

fix: harden intraday evidence provenance#485
KCNyu merged 1 commit into
masterfrom
codex/intraday-evidence-hardening

Conversation

@KCNyu

@KCNyu KCNyu commented Aug 11, 2026

Copy link
Copy Markdown
Owner

What changed

  • normalize the intraday insights sidecar before dashboard publication, retaining only model-owned narrative and stamping a harness-owned UTC generated_at
  • keep missing or malformed sidecars warn-only so deterministic report delivery continues
  • extend the shared numeric provenance advisory to %, pp, x/×, and σ, with unit-specific context matching
  • update the shared skill and canonical cron payload so the model is no longer instructed to author infrastructure time
  • repair the numeric contract tests' stale import path so both real postflight integrations execute instead of skipping

Evidence

  • replaying trajectory 993d3509-08fb-43be-834a-91f13087f5b7 now produces exactly one advisory: 2.3x; all context-backed percentage and pp claims pass
  • the +8h model timestamp is overwritten with UTC before publish_data_plane() runs
  • malformed sidecar JSON still delivers the report and records insights_sidecar=false

Verification

  • pytest -q tests/test_numeric_claims.py tests/test_intraday_prose_assembly.py tests/test_sync_cron_payloads.py tests/test_cron_contracts.py — 73 passed
  • mutation checks: replacing unit-specific matching with the flat number set fails the 2.3x regression; trusting the model timestamp fails the future-time regression
  • required remote checks must pass before merge

Closes #482
Closes #483

@KCNyu
KCNyu merged commit b36109d into master Aug 11, 2026
8 checks passed
@KCNyu
KCNyu deleted the codex/intraday-evidence-hardening branch August 11, 2026 05:42
KCNyu added a commit that referenced this pull request Aug 11, 2026
## Outcome

`#485` merged minutes before the release PR, so the commit that will be
tagged `v0.1.2` contains it — while the entry claimed the release
changed no package behavior. It changes
`src/clawock/harness/validation.py`, which is observable to a package
consumer, so it belongs in the entry rather than in the next one.

## Risk

Changelog text only. Blocks the tag until merged: tagging first would
publish an artifact describing itself incompletely, which cannot be
corrected in place.

## Validation

- [x] `tests/test_versions_agree.py` — 6 passed, 1 environment-only
failure (this host's editable `clawock-kcnyu` metadata is still 0.1.1)
- [ ] Required GitHub Actions checks pass
- [x] No secrets, scratch files, or unrelated generated data are
included
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

1 participant