Conversation
Refs #109 Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Moved the full empirical justification from retrieval-policy.ts (lines 16-19, orphaned after the constant extraction) to effort-profiles.ts as a doc comment above DEFAULT_EXPANSION_FLOOR, and deleted the stray remnant. The rationale explains the empirical basis (0.89–0.985 scale) and behavior-preserving value. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Refs #109 Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…eval pipeline Refs #109 Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
… constants Refs #109 Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Pin the advertisement-derivation test to registerTool(...).spec.description (the surface an MCP client actually sees) instead of the builder's raw .description, and extend the source-token assertions to cover every AXES interpolation so a revert to hardcoded literals fails the test. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…schema Fixes final-review findings on PR 1 of extract-retrieval-pipeline: the threshold/expansion_floor description numbers were pinned only against search-notes.ts's source text, so a swapped interpolation slot stayed green. Assert against the tool's JSON-Schema-converted inputSchema instead, and drop the vacuous toBeTruthy()/duplicate-constant assertions it was riding alongside. Also restores the "limit is not forwarded into executeRetrieval" invariant as a comment (lost when the old MERGED_CAP block was deleted) and corrects effort-profiles.ts's header, which claimed eval/ already reads these constants when it still restates them as literals. Refs #109 Co-Authored-By: Claude Fable 5 <noreply@anthropic.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.
Extracts the leg-assembly orchestration out of
search_notesintosrc/modules/semantic/retrieval-pipeline.ts, with the quick/deep effort profile single-sourced ineffort-profiles.tsand the advertised description numbers derived from it (byte-identical text).Behavior-preserving. No retrieval number changes value, no
search_notescontract change, and zero edits to any pre-existing test file — that last one is the regression proof: the existing gate-routedsearch_notessuites pass unchanged against the rewired handler.What moved:
effort-profiles.ts— the quick/deep profile (semantic pool + threshold, lexical note cap, expansion switch/limit, merged cap) plusLEXICAL_PER_NOTE_CAP,FALLBACK_THRESHOLD,DEFAULT_EXPANSION_FLOOR.retrieval-policy.ts'sMODE_DEFAULTSnow derives from it.retrieval-pipeline.ts— source narrowing, both legs, the semantic degrade, existence filtering (seeds and their expansion targets), fusion and the cap slice.executeRetrieval's four caller obligations are now module-internal, and the pipeline is its onlysrc/caller.search_noteskeeps only MCP shaping: it decides whether the semantic leg runs and how to word the outcome (semantic_status,DEGRADED_NOTES,query_stats,found_in).runSearchForEntryloses 149 lines.threshold/expansion_floorprose interpolate the profile constants, pinned by a new test against both the source (interpolated, not re-literalized) and the advertised schema text (the right numbers in the right slots).New direct tests cover fusion ordering, cap-slice truncation, leg-pool overflow, existence filtering, semantic degrade and leg skipping — with no temp vault, registry stub, or embedder mock.
Part 1 of 3 for the extract-retrieval-pipeline change (next: eval rewired onto the module + drift-pin removal; then the docs sweep). The intermediate state is safe:
eval/pipelines.tsandtest/eval/production-fusion-pin.test.tsstill hold their duplicate orchestration, and since production behavior is unchanged the pin keeps passing — acting as one more independent check on this refactor until PR 2 retires it.Gates:
npm test(1366 passing) ·npm run lint·npm run typecheck·npm run build— all green.Refs #109
🤖 Generated with Claude Code