Repository navigation
P3/P4: run_report.v1 for Hyperlex eval scripts (report-only) + validity separate from score - #21
Conversation
…d4075b; fail-closed wiring test; CI
… HLX_RUN_REPORT) + parse validity separate from task score
Codex Review SummaryThis comment shows the latest Codex review activity on this pull request.
ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings. |
|
CC @Zero-State-LLC/partner-agents. Flagged for shared triage. |
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 098c7f7076
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
Port recoverable_structure from archived ad4075b so the training export imports in a clean checkout. Resolve the bundled hyperlexical module before an exposure snapshot is published. Refuse holdout identity fields that are not lists of strings. Co-authored-by: Daniel Meyer <scrimshawlife-ctrl@users.noreply.github.com>
Keep the run-report workflow, which already runs tests/shadow and tests/heldout, so the holdout fixes and the report tests stay in one job. Co-authored-by: Daniel Meyer <scrimshawlife-ctrl@users.noreply.github.com>
… intent Validity now uses the labels run_eval_classify actually scored, so padding and extra prediction rows do not change the count. intent/2026-10-12-run-report-v1.md records the accepted harness-learning port. Co-authored-by: Daniel Meyer <scrimshawlife-ctrl@users.noreply.github.com>
Keep main's holdout-guard port and add this branch's run_report tests, the intent pointer in PORTED.md, and the eval_unbind report wiring. Co-authored-by: Daniel Meyer <scrimshawlife-ctrl@users.noreply.github.com>
Stacked on #20 (contains its commits; review #20 first).
scripts/shadow/hyperlexical/run_report.py: a context manager that does nothing unlessHLX_RUN_REPORTis set. It writes run_report.v1 (status, inputs without row text, counts, scores, validity, warnings, errors) and never changes eval numbers, stdout, or exit codes. A failed write does not raise.eval_unbind.mainandeval_classify.main. eval_classify is ported verbatim from archived ad4075b plus its deps.validitycounts predictions that parse to a known family separately from the task score (macro-F1 etc.).Code by deepseek-v4.1-flash via the Jev router; wiring is a small reviewed hand patch. Do not merge without Danny's yes.