Restore DistANN follow-up campaign and Task 238 regression coverage - #83
Open
kreneskyp wants to merge 40 commits into
Open
Restore DistANN follow-up campaign and Task 238 regression coverage#83kreneskyp wants to merge 40 commits into
kreneskyp wants to merge 40 commits into
Conversation
Close Task 223 on the measured whole-bucket ceiling
…ket 003 Gate arithmetic, timer scope, feature gating, exact-generation provenance, and the reuse invariant all verify. MAT-26 GO / MAT-25 no-advance stands under three normalizations. Blocking: 7043a5c falsified six artifact hashes, and skip_routed_delete_vacuum_drill reports a skipped drill as pass=true in results.jsonl. Three packet-001 gate amendments remain unlanded. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Re-review of a96bfdc and e9a907e: all eight seq-01 items land. All 13 manifest SHA-256 values verified byte-exact, the skipped-drill fix verified in emitter and parser with the focused test rerun at HEAD, and the extension is unchanged since the measured SHA so the 100k evidence carries over. Verdict: DONE. MAT-26 only, MAT-25 retired, packet 003 authorized under instrumentation parity and the 18.258830% serial bound. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Four blockers. The candidate step cannot complete as preregistered: the fail-closed zero-column guard fires on the recall child and the materialization-correctness matrix, both of which run id-only projections under the same session GUC. The extension installed on this host is the debug pg_test build the packet's own pgrx log installed, and allow_debug_extension=true suppresses both preflight guards. The byte-identical claim is false for arrays carrying a null bitmap with no nulls (array_send emits AARR_HASNULL, the candidate scans for actual nulls). Instrumentation parity is asymmetric: only the candidate arm carries the per-value timing shim. Artifact hashes all verify byte-exact; encoder, feature isolation, and coordinator-owner propagation reviewed clean. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
…ized Three of the four seq01 blockers are fixed: bitmap parity now falls back on dataoffset with the reproducer pinned, the session GUC degrades to native sends with fast/fallback/ineligible telemetry and fail-closed CLI assertions, and allow_debug_extension is gone from every step. The candidate step still cannot run: --reuse-fixture and --materialization-correctness are mutually exclusive at distann_multicluster.rs:830, and the preregistered candidate step sets both. Confirmed by executing the step's own dry-run argument vector. Also flags the owner-restart ordering between control A and the candidate, the unexecutable attribution-gate arithmetic, and one malformed manifest SHA-256.
…rized B1-B4 verified fixed by execution: candidate step argv now reaches the reuse-dir check at :1049 instead of the drill conflict at :830, no timing step mutates the reused fixture, the attribution gate resolves to real emitted stage/latency fields, and all 22 manifest hashes verify byte-exact. New blocker B5: the preregistered semantic gate demands exactly seven physical_materialization_correctness rows, but the mandated attribution build emits nine (seven core scenarios plus mixed_local_remote and post_first_batch_remote_failure). Two real runs confirm nine per step. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
…ized Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Closeout review of the MAT-26 live screen at run checkpoint b834b7f. Evidence integrity is clean: all 38 live artifacts verify against live-artifact-sha256.txt, all 22 pre-run manifest hashes still verify, both suite manifests record the exact release SHA and config hashes, every cited latency/recall value and the prediction identity match the raw logs, and the semantic failure string is verbatim. Blocker B1: the preregistered activation gate was unsatisfiable in every runnable configuration. The CLI requires --owner-fast-real-array-send to be paired with --skip-owner-locality-profile, which pins owner_requested_tids to 0, which suppresses the entire coordinator block that records owner_projected_values, owner_binary_send_bytes and the owner_fast_real_array_* counters the gate asserts on. The candidate arm demonstrably drove the owner endpoint 6,328 times; the zeros are the closed telemetry channel, not the sender. The packet must record this. Rulings: STOP accepted, no packet 004 or replacement run, restated as MAT-26 unmeasured rather than screened-and-failed; raw eager 26.3 ms barred, with the same-configuration 44.6 -> 26.3 ms position/warmth swing shown to be ~8x the 5% usefulness bar; feature-only candidate retained default-off with mutual-exclusion comments required; the native 12/10 bounded-read over-read needs its own P1 follow-up (6+4=10 at Tasks 198 and 191 on the same corpus and query sha, now 8+4=12) and blocks the mandatory matrix for Tasks 229-233; proceed to Task 229, with Task 225 left conditional on its own measured premise rather than on the absent finalist. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
…lockers Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
… merge authorized Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Gate Task 224 owner payload locality
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.
Restores the canonical Tasks 223-235 and 237-238 campaign ledger on current main without overwriting accepted Tasks 222/226/227/236. Reconciles Task 234 as NOT DONE and Task 238 as an already-landed lifetime fix with missing regression/bookkeeping.
Task 238 integration adds deterministic forced-retry coverage and repairs two pg_test-only main integration gaps discovered during verification: missing Task 234 delay GUC definitions and non-explicit loopback plaintext after Task 236.
Validation:
Outside review requested in reviews/task-238/001-retry-snapshot-uaf/request.md.