acpu 0.2.1 owner proposal — validated branch evidence
Source-only bump 2c70b3b929c791ca489bba12d2f5a6daef7d43e7 descends from
reviewed origin/main 14373cc8a0e18302751844579d94d1f8f80bd8b1. Its complete
diff is acpu/Cargo.toml, Cargo-generated root Cargo.lock, and
acpu/CHANGELOG.md. Workspace and sibling package versions remain unchanged.
No bump merge, tag or publication has occurred. This branch archive is a
prepublication diagnostic; owner delivery requires fresh default-branch inputs
after the owner merges the bump.
The capsule is candidate-1/evidence.tar.gz, SHA-256
d14d591e4e36e5e489494fb1b2c51f49bd1281b2f588366a3c4d3ac666a0d288.
candidate-1/bundle.json lists every retained file and checksum. The untouched
normalized acpu archive has SHA-256
74967ea98831cd29d69fd3cef0cc08a8236f66d8d2fdcb2e99223428488d1159, clean VCS
identity at the bump commit, and source bytes verified against that Git revision.
Results and exact commands
All owning and acpu package results below belong to the bump commit above.
Commands use Rust 1.95.0, RUSTFLAGS=-Dwarnings, RUSTDOCFLAGS=-Dwarnings,
locked offline resolution and explicit external target directories. Raw
commands, environment, streams and source inventories are inside the capsule.
| Scope | Commands and observed result |
|---|---|
| Owning eight gates | cargo fmt --all -- --check; cargo clippy --workspace --locked --offline -- -D warnings; cargo clippy -p acpu --all-targets --keep-going --locked --offline -- -D warnings; cargo test --workspace --locked --offline; cargo build --release --workspace --locked --offline; cargo test -p acpu --lib --tests --release --locked --offline; cargo run -p acpu --example matmul --locked --offline; cargo doc -p acpu --no-deps --locked --offline. All returned zero. |
| Raw owning tests | Workspace: 388 passing, five ignored, 31 summaries. Acpu release: 212 passing, one ignored, eight summaries. These totals are rederived from the two test commands above. |
| Untouched normalized package | cargo clippy --all-targets --keep-going --locked --offline -- -D warnings; cargo test --lib --tests --locked --offline; the same tests with --release; cargo doc --no-deps --locked --offline; cargo run --example matmul --locked --offline. All returned zero. Both test modes: 212 passing, one ignored, eight summaries. Both owning/package matmul runs report zero maximum error and PASS. |
| Package targets | cargo metadata --no-deps --format-version 1 --locked --offline, both owner and normalized package, plus fresh extracted-archive replay: identical 40 targets, including 32 examples and six integrations. gemv_admission and gemv_independent are present. |
| API comparison | Separate complete-library rustdoc JSON for official acpu0.2.0 and this candidate, target aarch64-apple-darwin, Rust 1.95.0; explicit cargo-semver-checks 0.51.0 check-release --baseline-rustdoc … --current-rustdoc … --release-type patch --color never: 229 pass, 31 skip, no semver update required. Exact binary hashes, flags, artifacts and commands are retained. |
| Nox diagnostic | Unchanged four-test probe from Nox 64d091e5ad3f277cce77b88ae0de6eea4c49b4df, normalized Nox archive SHA 37b40ed9894b3656315555b58afaec86058b6eb122e453244bb858a34fe1ff5b; cargo test --test source_contract --locked --offline and the same command with --release: four pass in each mode. |
The Nox scratch manifest uses only an explicit candidate acpu patch. Metadata
proves one registry unimem0.2.0 and one registry strata-nebu0.1.1, NeBu fast off,
and the same direct/Nox Goldilocks type source. This is not unpatched registry
acceptance and makes no owning Nox dependency change. The normalized acpu crate
still reexports registry unimem0.2.0's Tape; it does not deliver the owning
workspace's later Tape repair. Strata remains at
56aedb2d12b3126c601eb333419136d403614dbb.
After this consumer completed, Nox master advanced to
262de214521c47d18cd539040280f22e4b1fd8a4 through PR39, incorporating the
production/CLI changes merged by PR36. The retained diagnostic continues to
describe only fixed Nox 64d091e5ad3f277cce77b88ae0de6eea4c49b4df and its
unchanged archive/probe. Live release adoption must validate the latest Nox
source separately; this evidence is not relabeled as current-Nox acceptance.
Preserved tooling epochs
Owning, acpu package and API production used exact tools exported from
3b875c9850f8ebe232a6650aaff7913a28773803, manifest SHA
5c74d87123d860ce25800f300cac8f2425770d6ba9bf57a33489603000bb0811.
The package runner stopped after its successful commands at a metadata-argv
type error. Its missing result.json remains absent. The original consumer
stopped before tests because a Git-output assignment stripped the probe's final
newline. Nu retention also required parentheses around a multiline predicate.
All failures, raw receipts, original tools and intermediate capsules remain.
The reviewed repairs are committed at
e791fefaf05c612d10d96f201e9d9460b20d502f, manifest SHA
78f184494b7095f5804202a9cd661e1975b945577e8f8f705774e5aade170b9b.
The successful consumer and validator use this actual bundle. No owning/acpu
package/API gate was repeated; fresh consumer execution includes its own Nox
normalization command. Existing successful receipts were copied byte-for-byte
into the final records; original records remain unchanged.
The capsule retains original production tools under tools/ and repaired
consumer/validator tools under records/validation-tools/. The verifier checks
each bundle against its own Git commit and requires its executing file to come
from the declared validator bundle. It derives package success from commands,
raw counts, source/archive/VCS checks and fresh target metadata, not the absent
package summary. verification-candidate-1 retains pack/unpack and successful
relocated replay using the canonical repository as the explicit Git argument.
After unpacking to an unused external directory UNPACK, run the committed
exported verifier as follows, with a separate unused extraction directory:
nu UNPACK/records/validation-tools/audit/2026-10-17-acpu-release-proposal/verify.nu REPO UNPACK/records UNPACK/tools UNPACK/records/validation-tools UNPACK/records/validation-tools FRESH_TMP
Root independent replay and the owner's merge/publication decision remain the
handoff boundary. The independent GEMV test stays byte-identical at SHA
d595e4799a602ba6b2f46987bac6c22d0cb60d7cf8b841900a388488272568ba.