# docs-refresh result Partial completion: the three documentation corrections are delivered. The requested real replica measurements could not be produced: the specified Linux runtime failed to execute. No fabricated measurements, changed pins, code changes, engine runs, or claims about other workers are included. ## Changes and reasons - `docs/contracts.md`: replace the foundation-only and planned status with the implementations present in this checkout; link implementation documentation; describe committed `gates384` (384 records) and `spots576` (576 records). Preserve every interface definition, signatures, bank schema, `.6f` rule and stop rules verbatim. Add `research/solved` to the existing credit header. - `docs/reference.md`: describe the existing pinned installer instead of a future installer; direct publication readers to README as the pinned URL table entry point without assigning a commit for this wave. Add `research/solved` to the header. See the verification limitation below concerning README. - `docs/score.md`: remove the broken copy-and-patch recipe and its obsolete `uname` assertion; give direct setup and materialization commands. Correct the real replica implementation status and link its evidence. Include both requested replica commands with an explicit failed-setup qualification instead of invented round lines. Label the existing engine measurements as previously recorded evidence. Retain the Taylor header. ## Commands and outcomes All commands were bounded below 470 seconds; this task did not start a simulator workload or a benchmark after runtime setup failed. No sleeps or long-running workload were introduced. Persistent setup state is under `scratch/`. Documentation editing and static checks were short commands. There is no performance benchmark result for this attempt. Read-only inspection commands (each completed with exit 0; some were grouped into a single shell invocation): ```sh git status --short git rev-parse HEAD rg --files ammfast baselines docs | head -80 cat docs/ammfast.md docs/replica-banks.md docs/setup.md cat docs/contracts.md docs/reference.md docs/score.md cat ammfast/__init__.py sed -n '1,240p' score.py sed -n '1,240p' setup_fast.sh sed -n '165,280p' docs/score.md cat ammfast/__init__.py rg -n 'gates384|spots576|complete|raw.githubusercontent' baselines/replica/validation.json README.md | head -25 sed -n '1,100p' baselines/replica/validate.py sed -n '1,75p' baselines/engine/validation.json sed -n '1,90p' README.md sed -n '280,330p' docs/score.md sed -n '240,295p' setup_fast.sh ``` Initial status was clean. HEAD matched the assignment's starting tree. Inspections established that the APIs, banks, live adapter and explicit-platform installer exist. Some combined inspection output was truncated; focused reads subsequently covered the replaced scorer block and implementation status. Runtime command: ```sh bash setup_fast.sh --platform linux-x86_64 --cache "$PWD/scratch/cache" --scratch "$PWD/scratch" --replica-only ``` Exit **3**. Output started with: ```text setup_fast: platform=linux-x86_64 mode=replica offline=0 files=4 deadline=420s ``` Failure at **setup_fast.sh:280**: the extracted pinned `python3` reported `cannot execute binary file`. The installer reported that the interpreter selected by `--platform linux-x86_64` does not run on this machine, child exit **126**, and that nothing was installed. No source correction or pin substitution was attempted. This is a runtime failure, not missing requester input; no `.imd-blocked.json` was written. Consequently `. scratch/env.sh`, materialization under that runtime, and these scorer commands were **not run**: ```sh python -B score.py --baseline baselines/replica/qt1-ref-64000-65023.json --replica --policy COMBO --bar 0.1 python -B score.py --baseline baselines/replica/TwoComp-64000-65023.json --replica --policy qt1-ref ``` For both commands: exit code not applicable; actual round lines **none**; stop reason **none**; seeds requested **none**; scorer seconds **not measured**. Their bank selection is 64000..65023 (1024 available), and source inspection establishes a first request of 64000..64127 when executed successfully. This is not execution evidence. The live-example portion of the assignment remains unmet. Editing command: `python3 -B - <<'PY' ... PY` (inline pathlib replacements restricted to the three documents), exit 0. Verification commands used `TMPDIR="$PWD/scratch/tmp"`: ```sh python3 -B baselines/replica/validate.py git diff --check git diff --stat python3 -B - <<'PY' # Inline static assertions and SHA256 calculation, described below. PY git status --short ``` All completed successfully (combined shell exit 0). The bank validator printed: ```json {"complete": true, "bound_files": 22, "banks": 9, "bank_rows": 12288, "gate_pairs": 384, "spot_pairs": 576, "distinct_comparisons": 576, "gate_max_abs_error": 1.1937117960769683e-11, "spot_max_abs_error": 1.1937117960769683e-11, "ok": true} ``` `git diff --check` printed nothing. Diff stat: three files, 57 insertions and 39 deletions. The inline check compared contracts against `git show HEAD:docs/contracts.md`: identical fenced signature/schema blocks; intact replica definitions, bank definitions and scorer rules. It checked the evidence array lengths and `research/solved` in all three credit headers, then calculated hashes. Output ended: ```text PASS: contract definitions unchanged; evidence counts and credit headers checked ``` Git/Python launchers emitted sandbox diagnostics about being unable to create an `xcrun_db` cache file (`Operation not permitted`); their requested reads, edits and static checks nevertheless completed. No host configuration was inspected or changed. These diagnostics are distinct from the fatal pinned-runtime failure above. Result creation: `mkdir -p artifacts/goal/batches/e04-seal/items/docs-refresh` and a shell heredoc writing this file, exit 0. Final result size, credit, allowed-path and whitespace checks are recorded by the final command output. This artifact is left untracked, not force-added. ## SHA256 of delivered documents ```text c5a5dfc378f8b214b17fa67bf4a89b99c9c096203df85015dc72c13a4a6334ec docs/contracts.md e49ea5b6b7f885f753d012999344befec3e051f840b8f1e6a248476859ea3cef docs/reference.md 2778c8995be356e02c9b38130743a6a423b57ce322f987f5d369a6e741e92836 docs/score.md ``` ## Statements not independently verified - Requested live replica round lines, stops and timings are unavailable. No benchmark, simulator parity rerun, or engine integration was executed. The offline validator verifies committed evidence and bindings, not its original execution. - The requested README publication wording points to the designated publication entry point. In this starting checkout README still says final URLs are pending, and does not yet carry the final kit URL table. Its eventual table cannot be verified here. README is outside this item's edit scope; no claim is made about another item's work or a final wave commit. - Previously recorded engine integration measurements and hashes in `docs/score.md` remain historical evidence, not measurements reproduced by this refresh. Broader Gate P, platform and runtime claims in unchanged paragraphs were not re-executed. ## git status --short Before result creation: ```text M docs/contracts.md M docs/reference.md M docs/score.md ``` The result artifact is under ignored `artifacts/`, remains untracked, and therefore does not appear in ordinary short status. Only the three permitted source documents were changed.