fix: improve search evidence and context selection (1.17.3) - #44
Merged
Merged
Conversation
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
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Improve rule and helper retrieval when constants or file diversity hide implementations, and select complementary implementation evidence within explicit context budgets. Add bounded local helper expansion and advisory weak-evidence diagnostics. Preserve the 12,000-token default and explicit limits.
Release 1.17.3 for core, CLI and MCP, with synchronized plugin/runtime versions and reproducible quality fixtures.
Validation: build and workspace typechecks; 266 tests; dependency audit (zero vulnerabilities); version consistency and packed release artifact verification. Frozen CuraAI evaluation: symbol top-5 8/11 -> 10/11, top-10 8/11 -> 11/11; content-bearing contexts 6/11 -> 11/11. Search median increased from 2.54s to 2.83s. Weak-evidence heuristics remain advisory and uncalibrated (3/4 negative controls flagged).