Fixed-query caller prerequisites — run 01

Recorded 2026-10-12 for unchanged frozen artifact source a5dbda7a4fd6465c94f6505b3214cbcbc2130c82. These are harness controls and disposable smoke evidence; the four full successor query cells have not run. The six main cells on 8484493696a3e3e743889a83f14a60df54305aa9 retain their original measured results and clearing-v2 selection. Their older sampled-v1 edge diagnostic failed in five cells; no stationarity claim is made.

Candidate and source identity

• Debian SHA-256: 54cddf548c1a307aade8c3ac4332ede2007e2bc9603ab0cba72d8297287d6cee.
• RPM SHA-256: 982ad6923e17f64b35700dcfafb2994ed68e32b2c6b368589a8b6a0d550cac79.
• Minimal compiled source snapshot: f23830d5f70dd3029d540def41fdaf50fe0facf3.
• Source manifest SHA-256: b9944726e191e1bec0fe6abdcf3fbbbd905504c90a076167c09c33860e6f09e2.
• Source bundle SHA-256: abd1ea6458b16931af226f84b29ec226298a9084c4b5a4062e9166a386d1d9fc.
• Build provenance SHA-256: b02f2ebbf66e61b4b2b9133174fd787516af1a54701a6c02fd26041454b59f90.

No production or package byte changed. The prospective page schedule and corpus protocols precede their respective implementation commits. Four unpublished page commits had their message bodies corrected to the required trust-boundary prefix; identical trees and the old measured revisions remain retained on the private observations branch. This changes neither published history nor a receipt's source attribution.

Retained failures and causal checks

service-opening-r3-query-segment-s1-01 on 8484493 exited 1 with HTTP 429 inside the 43rd logical query, source-counter repetition 12 for query0036. Forty-two complete queries produced 66 HTTP pages; the failed history requires five pages. The old caller waits 0.1 seconds between logical queries but sends continuation pages without pacing. The error body was not captured, so no exact refusal message is claimed. Summary SHA-256: 116317cb2ab44c232c7136ad348433c7bc8c83fc4150760dedeace19f948403e; independent original source SHA-256: 6b5327176f9f7d16fc37f9a04242bf9b3196d25f072fb3c141ca8943ece942f5.

The new fixed-query reader dispatches page attempts at least 250 ms apart and inherits the complete-query timer. Every wait counts against the existing two-second latency gate. Credentials, the eight-per-second/eight-burst peer budget, query list, page/reply bounds, 0.1-second inter-query pause, UI and refusal behavior stay fixed; there is no retry. Fifty-three targeted tests on clean 2ba989216076b57e9987e21eb14144024b45ca62 exited 0. A separate unpaced module exited 1 with exactly the expected admission-budget and whole-query-timer failures. The finite offered-request model supplements the native counterexample; it does not reconstruct missing packet timestamps.

Paced smoke service-opening-r3-query-segment-s1-smoke-02 on 2ba9892 exited 1. Ten complete answers had no refusal, but four fleet/text verdicts failed retained-window and freshness. The obsolete adapter expected fixture-only grants. The accepted three-role contract makes readers installation-wide: returned metadata includes fabric-server-self. Its independently inspected Spool held two contiguous Batches without gaps. A contained diagnostic on successor 6bffe9a matched all ten retained answers against independent bounded companion prefixes (exit 0); the original native smoke stays failed.

The corpus repair keeps original Spool bytes and verifies recovered identities/bytes before using receive metadata. Native same-boot ACK/source events around the first page bound the possible companion snapshot prefix. Each admitted prefix is graded with the unchanged query oracle including freshness, retention, rows and continuation consistency. Missing, future, forged, unexplained and gap-bearing inputs fail closed. No expected telemetry comes from query answers. All 59 targeted tests on clean 6bffe9a7bb4ce921284d54c90a0fb8c84f45a198 exited 0. A separately injected companion omission exited 1 with the two expected failures. The other controls reject changed recovery bytes/labels, a missing prefix, forged metadata, future sources and incorrect node filtering.

Fresh native smoke

Command, run through the strictly observed resource launcher:

python3 -B tools/resource_group.py --delegate -- python3 -B tools/qualification/release_query.py \
  --deb-receipt /run/media/kmosoti/data/FabricO11y/results/installation-package-build/service-opening-deb-a-retry/receipt.json \
  --rpm-receipt /run/media/kmosoti/data/FabricO11y/results/packages/service-opening-rpm-a/receipt.json \
  --server-cpus 4-5 --worker-cpus 2-3 \
  --out /run/media/kmosoti/data/FabricO11y/results/service-opening-r3-query-segment-s1-smoke-03 \
  --seed 0xA11FA001 --mode segment --smoke

Clean 6bffe9a exited 0. All eight smoke gates were true: exact oracle answers, all fixture ACKs, delivery oracle, complete-query latency, server RSS, immutable source/harness and no OOM. Both phases retained actual UI readiness and completed native responses. The reduced fixture has two Batches, 100 logs, 100 metrics and two separately accounted three-span traces. All ten page attempts completed; maximum whole-query latency was 0.16222709801513702 seconds. This supplies no million-row or full repetition/Segment qualification.

Summary SHA-256: e84be3fb6ad0a24bd6a1e7d69c6f14c9f2b7a0e06b1ad39b23200132b52a3817; runtime SHA-256: 186a8151d4ee4ebb681b679bb506c6e53024d51a6f30322c944ac6281b5afe71; contained command receipt SHA-256: 8a34c7c191cdec64ca00e24033c685e5526e70daa24255e6e1f2ef576a88f053. Both new modules are in the archived immutable harness and explicit input hashes. query_reader.py SHA-256: ac66fddcea9a01ebce2444bb54fa2a3b2225201f038afe85975601ccb7f47ee1; query_corpus.py: f081afe884c3ef32a7e719ba582e8aad2a5ae4f44722bda820cbc7d2b92df217.

Containment, cleanup and limits

All checks used tools/resource_group.py, zero swap and a 1,800-second deadline; controls used default 16/20 GiB caps. Native fixtures tightened the whole group to 6 GiB, server/companion to 4,000,000,000 bytes (page-rounded enforcement 3,999,997,952), browser to 1 GiB and supervisor to 2 GiB. Server CPUs were 4–5; workers were 2–3. Each native admission reserved 5 GiB below the 95 GB stop and 100 GB aggregate ceiling. Storage was the mounted data drive. Every outer group stopped. Both failed native runs confirmed process cleanup but retained owned scratch; the fresh smoke confirmed scratch removal. Private originals, stopped-Spool dumps, runtime logs and raw answers remain under the data drive's results/evidence trees and are not uploaded here.

The original failures remain failed. No release, remote/outage, prolonged soak or browser qualification follows. Full query successors must run all seeds, full fixtures and both phases under the registered gates.