# Bench entry 2026-10-02: gpt-oss-20b on an Apple M5 Pro (24GB)

Published entry files for https://majormatters.co/p/bench-gpt-oss-20b-laptop-under-a-mandate

- FINDINGS.md: the findings table, each row naming its evidence file
- PROTOCOL.md and protocol-local-model-runtime-v1.md: what was measured and how
- INTAKE.md: how the product was obtained (disclosure)
- LOG.md: dated observations
- evidence/: setup.log, llama-bench JSON, egress-and-memory.txt, llama-server.log, mandate-runs/ (summary.md, summary.json, one JSON per run with the transcript and the signed WitnessKit chain), mandate-runs-pass1-unpatched/run.log (the superseded first pass)
- harness/mandate-run.mjs: the live-model harness (imports the Mandate Sandbox's vendored kits; paths are shown relative to the home directory)

Home-directory paths in the logs are shown as ~ .