bench: directional date bounds for deductive scope - #163
Merged
Merged
Conversation
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
Questions with a directional cue + date ("the week before August 3, 2023",
"as of November 2023", "after his trip in August 2023") previously bound
sessions by substring equality — scoping to the bound period itself while
demoting the earlier/later sessions where the evidence actually lives.
parseQuestion now extracts a one-sided bound (closest cue-date pair wins,
nonexistent dates rejected), and in-scope binds sessions on the correct
side of the bound, keeping the bound period itself in scope. No contract
change: in-scope semantics are computed harness-side and the question
digest already binds the resulting session set.
Validated on conv-50 (3 affected questions): deductive 0.834 vs 0.8275
pre-fix; cat2 temporal 0.818. Artifacts regenerate under the fix.
Generated with [Devin](https://devin.ai)
Co-Authored-By: Devin <158243242+devin-ai-integration[bot]@users.noreply.github.com>
Dev: deductive 0.8343 (+2.05 vs bm25-block, CI [0.011, 0.030]). Test: deductive 0.8341 (+4.67 vs bm25-window, CI [0.036, 0.056]; +0.55 vs bm25-block, CI [-0.001, 0.011]). All ten: deductive 0.8341 (+0.86 vs bm25-block, CI [0.0015, 0.0155] — excludes zero; union arm also excludes zero). Test artifact now carries the per-question rows audit surface like dev/all. Generated with [Devin](https://devin.ai) Co-Authored-By: Devin <158243242+devin-ai-integration[bot]@users.noreply.github.com>
0thernet
force-pushed
the
bench/directional-scope
branch
from
September 22, 2026 03:49
e5f2cbb to
67b77e6
Compare
This branch was successfully deployed
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
parseQuestionextracts a one-sided bound — the closest cue–date pair wins (as of/until/before→ before-side,after/since/later than/following→ after-side); nonexistent dates fail closed to equality scope.in-scopebinds sessions on the evidence side, keeping the bound period itself in scope. No contract change:in-scopesemantics are computed harness-side and the pinned question digest already binds the resulting session set.Test plan
tests/memory-benchmark-deductive-recall.test.ts— 9/9 pass, new parse + directional-scope tests (before/after/as-of bounds, month granularity, fail-closed on "February 30")bunx tsc --noEmitcleanGenerated with Devin