Tested production source: 6f1270d793217783889b5f22779f34ad93228247, parent 18a8e31441b39af26a997d1916ae6238fb2158bd. The source revision tested by the preserved September 6 isolated experiments remains 323c12f6c17753b0b408a3f0e34265939395d864. This directory is subsequent evidence integration; its own commit is not the tested source revision.
All 11 changed source files were read back at the committed SHA and matched the previously tested candidate byte-for-byte. All 75 materialized file Git blobs matched the complete commit tree. This materialization includes the full core-engine source/tests, all eight workspace package manifests and the frozen lockfile; it is not a claim that every repository package was tested.
The source-identical candidate had already passed pinned pnpm9.15.0 frozen installation, the 19-surface claim gate, evidence structure gate, 26 verifier regression tests, strict core typecheck and all65 discovered core tests across six files. prior-candidate-validation.json associates those preserved earlier runs with identical source; it does not relabel their old logs as new post-commit execution. An earlier candidate compiler run failed with TS2883; the final original-workspace compiler log passed after an explicit public Express type was added. The original compiler also lacked osc declarations. Both failures remain preserved with the later pass.
After the committed-source identity check, five fresh processes ran the canonical package benchmark with the actual commit SHA supplied. They preserved 1,250 measured calls and250 warmups across the five existing cohorts. Rounded outputs agreed across all five runs and with the historical baseline. Timings remain descriptive in-process compute measurements; there is no latency threshold.
The new full-dependency probes import actual installed TypeScript modules, including the runtime type barrel omitted by the historical isolated loader. All500 minority fixtures and the temporal, invalid-config, median and majority-tie observations match the historical record. Epoch-zero expiry now changes from active to expired. The17 historical aggregator assertions still pass; two added assertions confirm epoch-zero cleanup of the map and public state. Eight real OSC codec/event cases pass for scalar, empty and multiple arguments; only socket opening is stubbed, so no UDP-delivery result follows.
The one-line expiry repair, scalar OSC normalization, source-checked narrow declarations, explicit Express export type and IPC-free benchmark launcher have bounded evidence. Minority filtering, confidence calculation, direct invalid smoothing, temporal policy and median semantics were not redesigned. Their reproduced counterexamples remain open implementation/policy questions, not human outcomes or evidence of applicant independent mastery.
The initial pnpm11 attempt failed and appended placeholder build entries to the workspace YAML. Those bytes and the failed log remain preserved locally. The original Git-verified workspace YAML was restored, and pinned installation plus final gates and benchmarks were rerun before source integration. The final source commit contains no workspace or lockfile change. Earlier failed CLI benchmark launches and a source-untagged import-hook smoke output remain separate; none is counted in these five exact-commit benchmark runs.
Both exact-source GitHub Core Engine Evidence runs still failed before runner assignment: push34403004996/job102639174869 and PR34403010684/job102639195448 each have runner0, zero steps and zero artifacts. github-actions.json preserves the observations. The GitHub execution gate remains infrastructure_execution_blocked; local execution cannot close it.
Use the tested source revision in one checkout/materialization and keep these later harness files in a separate directory. Set OMNI_REPOSITORY_ROOT to that source root, OMNI_OUTPUT_DIRECTORY to a fresh output directory, and SOURCE_REVISION to the tested commit above. From the source core-engine package directory run each of full-dependency-probes.mjs, full-dependency-aggregation.mjs and full-dependency-osc.mjs via node --import tsx with its absolute harness path. The shared verifier checks all75 materialized Git blobs/SHA256s and refuses another source revision. Do not change the historical isolated loader guards to accept repaired source.
The benchmark command remains corepack pnpm --filter @omni-dromenon/core-engine test:bench -- --output <fresh-json-path> with the exact source SHA in SOURCE_REVISION. Inspect archive member names and regular-file types before extraction, then validate every size and SHA256 against raw-archive-receipt.json. Raw logs and results are unchanged. The prior September6 archive remains at its existing path and its recorded SHA256 was independently reverified.
The original new-run environment is local Linux x64, Node24.19.0, TypeScript6.0.3 and pnpm9.15.0. The GitHub workflow uses Node20 and has not run. No network, browser, capacity, security, participant, rehearsal or live-performance result is claimed. The legacy package engine declaration includes Node18 releases older than --import availability; aligning that declaration is separate compatibility work. No claim is made that all advertised Node18 versions were tested.
A separate compatibility run subsequently used an actual Node20.20.2 binary with the exact source SHA. Both preinstall gates, all26 verifier tests, strict typecheck, all65 core tests and one canonical benchmark passed. That benchmark preserved250 measured calls and50 warmups, with matching rounded outputs. It reused the existing locked dependency tree installed under Node24, so this is not a fresh Node20 installation claim. One Node20 benchmark does not establish repeated timing behavior in that environment. The GitHub runner and artifact-upload gates remain unobserved.