Add a blind, noise-calibrated retrieval eval harness#548
Closed
tobocop2 wants to merge 1 commit into
Closed
Conversation
Retrieval tuning has been graded by an ad-hoc kit living outside the repo; every future change needs a versioned gate. evals/retrieval generates a question battery from any indexed library (topical questions from sampled passages, known-item questions from sampled documents, count questions with exact streaming-scan oracles), answers it against two lilbee servers over the same index with per-row checkpointing, judges answers blind under opaque ids with one arm duplicated to measure the judge's noise floor, and scores mechanically: per-dimension means with the noise floor as the error bar plus exact pass/fail for counts and document lookups. Streaming scans and reservoir sampling keep memory flat on very large libraries. Lives outside src/ so it never ships in the package; lint and format gates cover it. bb-yjbr
tobocop2
marked this pull request as draft
July 18, 2026 04:55
Owner
Author
|
Superseded by the consolidated retrieval PR #557 — review and merge there. Converted to draft to keep it off the review queue; the branch and history stay intact for reference. |
Owner
Author
|
Superseded by the consolidated retrieval PR #557, which contains this change. Closing; the branch stays intact for reference. |
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Problem
Retrieval quality changes have been graded by an ad-hoc kit living outside the repo, so every tuning decision starts from scratch and nothing gates regressions.
Solution
evals/retrieval(repo-level, never shipped in the wheel) generates a question battery from any indexed library: topical questions from reservoir-sampled passages, known-item questions from sampled documents, and count questions with exact streaming-scan oracles. It answers the battery against two lilbee servers over the same index with per-row checkpointing so killed pod runs resume, judges answers blind under opaque ids with one arm duplicated to measure the judge's noise floor, and scores mechanically: per-dimension means with the noise floor as the error bar plus exact pass/fail for counts and document lookups. Streaming scans keep memory flat on very large libraries. Lint and format gates now cover the directory.