Repository navigation
bench: reproducible 50-question recall comparison with Engram - #39
Merged
Merged
Conversation
7 tasks done
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Adds a frozen 50-question synthetic recall benchmark against Engram v2.2.1 using identical title/body memories, both Engram match modes, and the real ATS Core hybrid RRF path with pinned local MiniLM embeddings. No vector database or hosted model key is used. The runner isolates the Engram store, saves per-question results and provenance, fails on degraded retrieval, and trips if ATS hybrid recall falls below the stronger Engram mode.
Measured Recall@5: ATS keyword 12/50, ATS hybrid 47/50, Engram default
all20/50, Engramany32/50. README/report publish the keyword-only result, bucket breakdown, missed questions, raw output, reproduction steps, and limitations. Engram and ATS hybrid tie on exact-title and terse queries. This is ten distinct target decisions with five correlated questions each, not a production-scale product-superiority claim or an RRF ablation.Validation: full repository gate passed all 454 tests and required proofs;
check:publishpassed. Added scorer and dataset integrity checks. The frozen dataset was not tuned after the measured run. Runtime/model caches and ordinary run output are ignored; only reviewed synthetic output is committed.