Repository navigation
worker: claims can say how they are known; reads mark them apart from bare assertions - #86
Merged
Merged
Conversation
…ted claims before LIMIT
…-ghkdsb # Conflicts: # docs/features/hosted-worker.md # worker/src/mcp.ts # worker/src/memory.ts
vu1n
commented
Nov 7, 2026
vu1n
left a comment
Owner
Author
There was a problem hiding this comment.
VPS review of 313e89a: 0 finding(s), 0 blocking.
The change adds claim evidence to stored data, read responses, and immediate and delayed judging. The reported test coverage gap does not establish a reachable defect, so this finding does not block merging.
1 finding(s) dropped by the check
worker/test/evidence.test.ts:53Test evidence through the write gate and delayed judging: Although the test only checks gatedText directly, both repo and room claim callers pass gatedText(content, evidence) to judge, and judgeLate receives evidence hydrated by D1Store.unjudged, so no concrete failure is demonstrated.
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Requested by Vu · project thread
claimtakes an optionalevidence(≤500 chars): what was run or read and what it showed. It is stored inmetadata.evidence(no migration) and lifted out byclaimColumnsandunjudged. Recall and briefing lines with evidence end in[checked: …](clipped to 100 chars);expandreturns all of it. Roomclaimtakes it too.Agents read evidence, so the write gate and
judgeLatejudgegatedText(content, evidence). Note: when evidence is present the keep score now sees it as well as the content;DROP_BELOWwas tuned on content alone. Purge resetsmetadata, so it erases evidence.From the review of Anthropic's "components of agent memory" post: the models that gained most from memory wrote how they knew each rule.
Contracts touched:
claiminput schema (repo and room) gainsevidence;expandoutput gains optionalevidence; compact line gains a suffix.Verification:
npm run typecheck;npm test388/388; newtest/evidence.test.ts(4 cases: mark and expand, clip and limit, gate text, purge);featuremap.py checkclean.(The branch carries a merge of main because it was reused after #85 squash-merged; the diff against main is only this change.)
Generated by Claude Code