Experiment formal engineering primitives results 01 - kmosoti/FabricO11y GitHub Wiki

Native engineering evidence primitives: results 01

The read-only Rust aid checks pinned bytes, dependency order and stopped-receipt structure. It does not execute plans, authenticate producers, discover omitted mandatory cases or grant acceptance. Existing independent Python product oracles and frozen release harnesses remain unchanged. The owner chose incremental Rust migration while alpha acceptance continues; no MCP server or release tag is added.

Experimental correction before adoption

The registered contract is docs/experiments/formal/engineering-evidence-protocol.md. At frozen f6d2594, exactly one budget regression ran and failed (exit 101, confirmed whole-service stop): the total was 34,359,738,366 instead of 34,359,738,368 because the two-byte plan bypassed the artifact budget. At 50730a7, three real file controls ran: existing identity-refusal control passed, budget control failed, and a real append after an eight-byte read exposed work beyond the admitted extent. The recorded extent assertion establishes >8 bytes; it did not print an exact 13-byte observation. Both failed runs and fixtures are retained under results/engineering-primitives-tests-01.

The narrow correction uses one budget includes plan and artifacts; reads stop at initial metadata extent; descriptor/path identities still reject changed or truncated files. The obsolete plan/artifact mode, cache count flag and forwarding path were removed. This is finite structural evidence, not proof against every concurrent writer or privileged attacker.

Verified native prototype

Frozen source 6ac6932072484c28186291caed1c88160bc3d3a0, Rust 1.99.0, default launcher caps/no swap, DATA-only scratch, serialized source drivers and full 95 GB admission census/reservations. Actual command: cargo test --locked -p xtask --lib --test preflight (explicit pinned toolchain env) exited 0, whole-service stop confirmed. Four library tests and twelve integration tests ran without skips, including both prior regressions, all 4,165 directed graphs on 1–4 vertices against independent total-order enumeration, repinned receipt/schema/path/budget controls and native CLI 0/3/2 outcomes. Whole command: 6.169 s / 268.3 MiB, including build/test machinery; not tool-only RSS. Wrapper SHA: 0863451ef54bb3f07dddbdb1f8632ed0818ee195502377e9774c15c6c82ce868.

cargo run --release --locked -p xtask --example preflight_measure exited 0, whole-service stop confirmed 2026-10-18T00:26:16.634851Z. Seed 17; seven alternating pairs after one warmup; exact output equality in all six graph and one artifact cases. Graph reference repeatedly scans with hash-set membership; indexed ready selection uses adjacency/indegrees and a min-heap. Both hash variants stream the same immutable fixture. Fixture creation is outside timings.

Case Reference median ms Indexed median ms
64 chain 0.078 0.011
64 fan-out 0.094 0.010
512 chain 6.081 0.102
512 fan-out 6.958 0.094
2048 chain 114.073 0.486
2048 fan-out 134.169 0.425
32 references to 8 MiB 994.075 31.023

For 32 hash joins, actual digested bytes were 268,435,456 versus 8,388,608. Whole command: 18.989 s / 300.8 MiB, including 8.46 s release compilation; not an allocation comparison. Results apply to these finite warm-cache tooling workloads on the shared Fedora 44 host, not product performance or a Python/Rust ranking. Raw measurements: results/engineering-primitives-measure-01/native-measurements.json. Example ELF SHA: fe5f2bfc4f860f3bd3fdc3d9f6d50efc32f8399ec243b5d1db6cea41c337fb9f; wrapper SHA: e4fda1648ad0b25fa41472b66683e0e719e960018979e3bb9b2cfd1002a5187a.

Cleanup and integration

At 80ea0b2, the superseded three-file/1,183-line Python aid was removed from the new tree. Its historical raw evidence and old frozen tree remain retained. Readiness prose and review guidance were shortened without deleting failure records or original acceptance gates. Rust compile/test inputs are unchanged from 6ac6932. The final focused integration results follow; hosted integration is still pending.

Final focused integration

Final source is d187e9a41a9e7cdc6ad169888bd6cef574a5c3f2. At 80ea0b2, frozen Bun dependency/link/checker/hook checks and the architecture-review skill validator all exited 0 with confirmed whole-service stops. Targeted Clippy initially exited 101 on the test-only READ_HOOK type's complexity. Its original raw failure remains retained. A cfg(test) type alias preserves the same type and behavior without suppressing the lint; targeted Clippy and format-check then exited 0 at d187e9a.

The registered literal command vectors, expanded toolchain environment and reservations are in results/engineering-primitives-final-01/plan.json (SHA ab354137c19cbc3464913a1c8214f13df2d848f1d2ee6d6d015dc48617800373) and retry-plan-02.json (SHA 860c5ac10e179d155fe7533863c4794e65a3b7e50a87386c46b65cb50729cde0). X246 requests/readbacks: initial, lint correction. Each command uses the contained source driver, DATA scratch, full census and a 1 GiB reservation (2 GiB for Clippy). The docs wrappers set the retained Bun 1.4.0 PATH/cache explicitly. Rust uses the existing pinned 1.99.0 binaries; no tools were installed for this retry.

Commands: python3 -B <registered check-deps.py/check-links.py/check-checker.py/check-hooks.py>; python3 -B /home/kmosoti/.codex/skills/.system/skill-creator/scripts/quick_validate.py .agents/skills/architecture-review; cargo clippy --locked -p xtask --all-targets -- -D warnings; cargo fmt --all --check. Full outer argv and raw logs remain in the hash-bound plans/wrappers, including explicit PATH and RUSTUP_TOOLCHAIN. Required final hosted verification is still mandatory; the earlier prototype's full-fast result does not substitute for it.

Cell Revision Exit Finished UTC Wrapper SHA-256 Resource receipt SHA-256
docs-deps-final-01 80ea0b2 0 2026-10-18T00:33:48.838357+00:00 57102cd3362e5503d6da77592faca7843584d3a42ed93d8fcaa65fe569951293 53d7ba4b12594b71a1393f04ff025adda48d4e30b8ffe8ba853cb89656d58696
docs-links-final-01 80ea0b2 0 2026-10-18T00:33:54.365880+00:00 712796786a1384da4001c45c4e078a323f1b6b94ec8a1b65df1e01fdd10ad3fa 2f0f226236299e9cd831b0d953f3bc85afc088d2251d9b0bfff62c4cb037c3f1
docs-checker-final-01 80ea0b2 0 2026-10-18T00:34:18.921103+00:00 7165f44dea49c70c17965c697fdae31e7456e4d0745dbc57537071f95ca48ebc 24dd51fdd61f761fda1c2309b16a46775a8a0e9245b1c9acc19d67ef42d0d827
docs-hooks-final-01 80ea0b2 0 2026-10-18T00:34:24.569712+00:00 39b807e81eee0631a7a5148f3cbe61485cba9a19f3ad16deba0d5ec5e65bcf34 564966dc9248a4994455354a80c309d4b62a7e5127dd36507e3f20e601979d2b
skill-final-01 80ea0b2 0 2026-10-18T00:34:29.301309+00:00 112cd272b44cfad2ea3743b4af79fb9acecca4cb249b9c2f9dcfdaa32e876120 bc5680690a10e158dfd4a79327da5c0132692c3eb8093cb6551d9c7982b2bc69
clippy-final-02 d187e9a 0 2026-10-18T00:35:23.153620+00:00 5cb5b43d8c6fe6b3084585afc6e1096e072ee7c9584df0a2625969f4820f95b8 9602a4bdca9d52d14de386662eb5e8f935c7d0f850e9019339bbaf26a25baf11
fmt-final-02 d187e9a 0 2026-10-18T00:35:27.811822+00:00 700522c90bcbb1c312599d183acb30b4c70a388057310d71a818c4f27d87b6d8 3837b51bef7642b5c7b2e1e4a60406e0b50a8d2c4d907c6e2d2deab9351c06d4

PR78 publishes this exact final head. At this report's publication, required hosted checks are pending, the PR is not merged, and no alpha acceptance or release completion is claimed. Native acceptance continues independently under its existing frozen main-bound harnesses. The old Python prototype's raw results/frozen tree, both primitive red experiments, initial Clippy failure and all acceptance failures remain preserved.

Final hosted checks and merge

The exact ready head d187e9a41a9e7cdc6ad169888bd6cef574a5c3f2 subsequently completed all required hosted checks: fast, security regressions and extended. The draft extended skip was not counted. PR #78 merged at 2026-10-18T01:33:37Z as 00fd555d41fae30fd0b34697b5fbf82c6fdf7d1b.

The exact pre-merge check snapshot, head identity, merge response and actual REST merge record are retained in DATA/results/engineering-primitives-final-01/. This supersedes only the earlier hosted-pending status. Local tests and measurements retain their original revision attribution and finite limitations; the tooling does not replace product acceptance or independent oracles. No release or tag was created.