Repository navigation
feat(experiments): bring the recall experiment (#29) onto master - #30
Merged
Merged
Conversation
feat(experiments): union of embedding and BM25 candidates, a one-question probe and a hard set
This was referenced Oct 4, 2026
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
This carries the recall experiment from #29 to
master. #29 was merged while its base was stillexp/jev-reranker(the stacked base was not retargeted after #28 merged), so its content is on that branch but not onmaster. The diff here is the same 12 files, all underexperiments/recall/plus one.gitignoreline. Nothing inquarq/changes.What it adds, with the full results and caveats in #29 and in
experiments/recall/README.md:synthetic-draft(not human-reviewed).Headline, with the caveats from #29: the union puts the gold page in the first 10 chunks for 38 of 38 questions (embedding 36, RRF 35), and with Jev as the re-ranker scores 38/38 Hit@1 against a 36/38 baseline. That set has been iterated on, so treat it as "loses nothing and fixes the two known misses", not as a general accuracy figure.
After merging, delete
exp/jev-reranker.