Historical report imported 2026-10-10 from docs/experiments/formal/bounded-writer-default-run-01.md. Original SHA-256: d6437b7c3319ebc57992caf4e6e89d97693b6ded20923986baa92cedd9b0abfe. Measurements and outcomes retain their stated revisions. Links and local environment labels are normalized; $FABRIC_DATA is the mounted experiment drive. See evidence policy.

Bounded writer on default settings, run 01

This record captures finite local evidence for the normal, default bounded-writer change. It does not introduce or change a protocol, acceptance gate, or product guarantee. The run is not deployment qualification.

Evidence

The unflagged release build completed in launcher unit fabric-work-abcffb68bfb241b9af20cff00c36b1ca (exit 0; build.json). Evidence is under $FABRIC_DATA/results/readiness-continuation-default-01. The build took 11.223 seconds and reached 662.8 MiB cgroup memory. Four of five binaries have identical full-file hashes to the preceding trial. fabric-server differs in full-file hash and size; the loaded-ELF comparison found 30 allocated sections, 12 program headers and 11 program segments equivalent after validating and normalizing only the ELF section-table offset and GNU build-id descriptor. A .text byte-flip negative control was rejected. This establishes loaded-content equivalence, not whole-file identity. See loaded-elf-equivalence.json and checker unit fabric-work-3e1ff3b5898a47419968a2548cf4731f (exit 0).

The unflagged suite recorded 27 passing tests in launcher unit fabric-work-7d4f4c00fadc4075b4657ea8680dfa9d (exit 0). The named 13 kill-cut cases and the no-hit control completed in launcher unit fabric-work-9fe4e4065c8149fdaec846229aaca429 (exit 0). These results belong to this finite local run and do not replace other registered gates.

An earlier smoke was interrupted because its runner output was placed under repository target/ on the system disk rather than the data drive. Launcher unit fabric-work-68bfcf134fe243c6a2e02fcc810558b9 exited 2; the wrapper exited 0 and recorded the child state as started, with no child exit result. Its 2,825,697-byte owned tree was archived and hash-verified before exact-child cleanup. Archive SHA-256 is 74aceb744489e30d244090367fafda91cdab2a7e542a8008d74cd1882d3ab58a; cleanup launcher unit fabric-work-e06c0d0dc0c340dd8f428c938da8b9bf exited 0 and confirmed the child absent. This interrupted attempt is not an acceptance result.

A replacement freeze was prepared on the data drive at $FABRIC_DATA/scratch/soak-r2-default-01 in unit fabric-work-cd9a01b81e684a0ea2e7d4a5f830c073 (exit 0). Precheck unit fabric-work-a67a11a1bb53469abd1e075c8f09b16d and postcheck unit fabric-work-9f8c106179a64e438a365689ac83646d exited 0; both receipts record manifest SHA-256 aaed20affff02c6ed2d9e9e2be34433740704b769e6243c358ba798c93e5a7d7.

The replacement native smoke ran in launcher unit fabric-work-2f6e81a00f3f40efac2c39182a2a5bdb and exited 1 because no Segment was produced during its 70-second workload. The simulator delivered 7,000 exact batches; the companion delivered 13 exact batches. All nine other gates were true, backlog was zero, and both production processes exited 0. Seven queries completed, with p99 32.27 ms; combined server/companion cgroup peak was 38,924,288 bytes with zero memory events and zero swap. The workload did not reach the 64 MiB sealing threshold. The full R2 trial and unflagged crash-cut tests provide the sealing evidence.

The runner output-path guard was then corrected in commit c8a0c1b. Its 66 related tests passed in launcher unit 056b12832b1f43bfb1f9410cb418980c (exit 0, 5.209 seconds, 46.2 MiB). This guard change does not alter the preceding smoke outcome.

The complete replacement freeze was archived and all 466 members verified before removal in unit fabric-work-5adaf862e45648cfb4dea2daa5e837da (exit 0; 3.724 seconds; 40.4 MiB peak). default-smoke-freeze.tar.gz is 16,794,345 bytes, SHA-256 a94164eb590010a2677bd824c0aff4380625087b612686b8a3ae06f32ec87cfc. The private archive includes source, binaries, provenance and run output. The stopped cgroup and absent tagged processes were checked; default-smoke-cleanup.json confirms removal of exactly the owned 58,287,608-byte freeze. Compact summaries and launcher/runner receipts remain beside the archive.

Final checks

The unflagged cargo xtask checks --profile fast --receipts $FABRIC_DATA/results/readiness-continuation-fast-03/checks passed all 17 checks, exit 0, under resource-launcher unit fabric-work-37b7e81db5c54e72ba7c998140e4bebe. The launcher command explicitly removed all five experiment selectors listed in build.json. Elapsed time was 329.399 seconds; CPU was 728.407 seconds; whole-job cgroup peak was 15,354,585,088 bytes, with zero memory-pressure/OOM events and no swap. The group stopped and its private scratch was removed.

Manual documentation verification used python3 -B tools/bench/labs/cross_system/sweep_documentation.py --out $FABRIC_DATA/results/readiness-continuation-docs-01 through the resource launcher. Unit fabric-work-502bf34625e04cde9a575eeaf23a3ddb exited 0: documentation, checker negative probes and hook tests all passed. The authenticated private Bun copy was removed; hooks and CI remain disabled. These receipts complete the default-adoption checks without changing historical benchmark outcomes.