Repository navigation
perf(context): filter capture noise and scope content retrieval - #502
Conversation
…en-reranker-ablation
…en-batch-memory-reads
…en-response-context-scope
…en-embedding-prefix-audit
…en-batch-embedding-paths
…en-storage-calibration
…en-indexed-recall-pools
…en-batch-memory-reads
…en-response-context-scope
# Conflicts: # benchmarks/reproduce.sh
cdeust
left a comment
cdeust
left a comment
There was a problem hiding this comment.
Completed source review of this implemented scope and its parent integration. No unresolved correctness, security or lifecycle finding. The description records limitations rather than claiming unmet optimization targets. Merge remains gated on current required CI and applicable baseline quality evidence. Review includes independent read-only review of the per-PR diff and combined locked local tests.
cdeust
left a comment
cdeust
left a comment
There was a problem hiding this comment.
Reviewed the CI-trigger commit: git diff from 5000f75 is empty and Git tree 2abc1aa73492b4cbf2b1be51705aa73c302ed21a is unchanged. Previous source review and test evidence apply to this identical tree. Required main-targeted checks are now running; no protection bypass is used.
cdeust
left a comment
cdeust
left a comment
There was a problem hiding this comment.
Reviewed current head e7b86b4 including inherited test fix75fb7740.
The only source delta since the prior review is two PostgreSQL semantic test helpers. Both execute real production SQL with transaction-local exact scans; the scoring helper pins the real executor to the same pooled connection. All assertions are unchanged. Production runtime, scripts and benchmark trees are identical to the reviewed benchmark candidate.
Validation:35 native PostgreSQL tests passed (scoring-debias, near-duplicate pairs, indexed pools), zero skips; planner settings restore on both helpers. A rollback-only mutation removing the two real SQL auto-capture exclusions reverses curated/auto ranking and is rejected by the unchanged assertion. Ruff and formatting pass.
Empirical control: with2000 varied deleted vectors, default near-duplicate search missed a pair which exact search recovered. Forced ANN also changed recall score contributions. The precise CI curated-order inversion was not reproduced; it remains consistent with incomplete ANN membership, rather than proven as that incident's cause. These fixtures assert exact scoring/pair semantics, while ANN plans are covered separately. Fresh CI and the full baseline/candidate quality comparison remain merge gates.
Review recorded as COMMENT because this account owns the PR; no independent approval is claimed.
Filter capture noise from methodology context and restrict fetch-by-ID access to the intended agent scope.
Includes current #501 slot/read fixes. Token ceilings remain unchanged; the separate 100-payload provider-token reduction claim is not established.
Validation and review (2026-09-08):
af717ab76d936065233d7d4ec3202b5105ff00aa.ca32ba50is running on separate disposable databases with identical public datasets. It uses the repository's existing 0.005 tolerance; no acceptance result is claimed before completion.Historical measurements apply to their recorded revisions and scopes. The current PR is authorized for merge after applicable checks; remaining optimization targets above are not presented as completed benefits.
The latest commit only triggers main-targeted security checks. Its tree
2abc1aa73492b4cbf2b1be51705aa73c302ed21ais identical to5000f75e5d1ee14c5dd62db709a99f86792dab2d. The composed benchmark uses frozen commit5000f75e5d1ee14c5dd62db709a99f86792dab2d, whose tree matches final PR #502.