Repository navigation
Support full init reruns with frozen discovery settings - #300
Conversation
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. |
![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: 11b378866f
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".
| raise ValueError | ||
| except (ValueError, TypeError): | ||
| raise InitError("KB_DISCOVERY_START_INTERVAL_S must be finite and between 1 and 60 seconds") from None | ||
| level = getattr(rt.gateway, "zcode_reasoning_level", "max") if rt.generator.provider == "zcode" else None |
There was a problem hiding this comment.
When KB_DISCOVERY_GENERATOR selects a non-Zcode provider but KB_DISCOVERY_JUDGE selects Zcode, this records zcode_reasoning_level as null even though the judge transport still uses that setting. Changing ZCODE_REASONING_LEVEL therefore leaves the checkpoint digest unchanged: a prepared publication can resume under a different declared configuration, while an unfinished batch can restore progress and then invalidate prior judge receipts as archive mismatches instead of rejecting the changed input up front. Include the effective level whenever either configured discovery role uses Zcode.
Useful? React with 👍 / 👎.
Existing knowledge bases can rerun modules from a merged owner map without manufacturing skeleton records. Actual stage records and discovered-catalog merge gates remain mandatory; an optional frozen branch suffix lets another batch publish without replacing old branches.
Discovery now accepts bounded packet size and start-spacing settings, preserving every indexed shard and adaptive rate-limit cooldown. Its effective Zcode reasoning configuration is archived and bound to resume identity, including prepared publication retries; candidate-limit saturation remains an explicit omission lead. Defaults are unchanged.
Validation: combined focused regressions, independent gate/resume review, documentation checks, and a native GLM-5.3 low-reasoning preflight with a 165,752-byte attachment. The preflight checked markers and paths; it does not measure feature-discovery quality or PR-review accuracy. Existing knowledge pages and approval records are untouched.