feat: filesystem-grep and full-context baseline providers#18
Merged
Conversation
Published memory-system comparisons are only credible read against two cheap floors: - baseline-grep: deterministic TF-with-log-damping term matching over the raw corpus markdown — no index, no embeddings, no LLM. Squared coverage factor so a doc matching every distinct query term outranks a doc spamming one term. A memory system that can't beat this isn't adding retrieval value. - baseline-fullcontext: no retrieval at all; the whole corpus is returned as a single hit (capped at 600K chars ≈ 150K tokens) for the QA stage to answer over. Published results repeatedly show full-context beating dedicated memory systems on corpora that fit a context window; QA accuracy and token cost should be read against this provider's. Makes no doc-id claims, so retrieval metrics are meaningless for it by design. Both registered in the provider factory; 9 new tests; smoke-verified on the synthetic corpus (grep recall@5 0.75; full-context recall 0 / content-hit 0.75 as expected). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Signed-off-by: Drew Cain <groksrc@gmail.com>
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Summary
Adds the two honesty-floor providers every credible comparison needs:
baseline-grep— deterministic TF (log-damped) term matching over raw corpus markdown. No index, no embeddings, no LLM. Squared coverage factor so full-coverage docs outrank single-term spam. The Letta-style argument: a memory system that can't beat grep isn't adding retrieval value.baseline-fullcontext— no retrieval; the whole corpus is one hit (capped 600K chars) for the QA stage to answer over. Full-context repeatedly beats dedicated memory systems on window-sized corpora, so QA accuracy and token cost should be read against it. Returns no doc ids — retrieval metrics are meaningless for it by design, and it says so.Both work in grouped (LongMemEval) and flat (LoCoMo) modes since they implement the standard provider interface.
Verification
🤖 Generated with Claude Code