Repository navigation
feat(review): investigate plausible findings before final KEEP - #14
Conversation
Classify specialist output as KEEP / PLAUSIBLE / REJECTED after normalize. Investigate KEEP and PLAUSIBLE (cap 3 hyps, 6 Graphify calls); re-validate so a symbol-in-diff hypothesis can gain a real path and promote to KEEP. REJECTED and trivia never hop. UNRESOLVED and PLAUSIBLE never post or go inline. Lockfile-only still skips investigate. Offline goldens qw-538 and qw-571 pass. No new LLM agent.
![greptile-apps[bot]](https://avatars.githubusercontent.com/in/867647?s=60&v=4){kind=small}
There was a problem hiding this comment.
Your trial has ended. Reactivate Greptile to resume code reviews.
![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: 41161296f2
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".
| blob = " ".join( | ||
| [(e.path or "") + " " + (e.text or "") for e in items] | ||
| ) | ||
| paths = _pr_paths_in_blob(blob, files_changed) | ||
| if paths: | ||
| f["file"] = paths[0] | ||
| ev = list(f.get("evidence") or []) | ||
| if paths[0] not in ev: | ||
| ev = [paths[0]] + ev | ||
| f["evidence"] = ev | ||
| f["hypothesis_status"] = KEEP |
There was a problem hiding this comment.
When the first hypothesis is PLAUSIBLE, this treats any investigation response that happens to mention a changed filename as confirmation, without verifying that the response supports that finding's symbol or claim. In particular, the unconditional get_pr_impact call is attached to the first hypothesis and can contain PR-wide paths; that makes the finding pass the subsequent path validator and be included in the review body even if the graph result is unrelated. Restrict promotion to evidence that establishes the hypothesis-to-path relationship (rather than a generic changed-path match).
Useful? React with 👍 / 👎.
1 participant
Classify specialist output as KEEP / PLAUSIBLE / REJECTED after normalize. Investigate KEEP and PLAUSIBLE (cap 3 hyps, 6 Graphify calls); re-validate so a symbol-in-diff hypothesis can gain a real path and promote to KEEP. REJECTED and trivia never hop. UNRESOLVED and PLAUSIBLE never post or go inline.
Lockfile-only still skips investigate. Offline goldens qw-538 and qw-571 pass. No new LLM agent.