Major Matters Bench: contributed run bundle

Model: gpt-oss-20b
Operator: Charlie Major, Apple MacBook Pro, M5 Pro, 16 cores, 24GB unified memory, macOS 27.0, llama.cpp 0.5.0, build 11146 (7fe450e19), Homebrew; llama-server -m ~/models/gguf/gpt-oss-20b-MXFP4.gguf --port 8081 --jinja -c 32768 --alias gpt-oss-20b (four slots of 32,768, unified KV, as entry three's pass 2)
Signed: 2026-10-12T21:07:37Z by did:key:z6Mktro729ShPY3kmoKGFqPUnvoR3rweeqJPbUZZvJuuRixo
Files in evidence/: 50
Harness: mm-bench-harness 1.1.0

What is here
  evidence/                 everything the harness recorded, unchanged
  evidence/MANIFEST.json    SHA-256 of every file, the signer, the chain head
  evidence/MANIFEST-chain.json   the WitnessKit chain over those hashes, signed
  evidence/config.json      the configuration the run used
  evidence/mandate-runs/    one JSON per run (transcript, calls, signed witness chain), summary.md, summary.json

To verify (needs Node 22 and the harness, no network):
  node bench.mjs verify --evidence <this folder>/evidence

What gets published: the findings table, summary.md, the per-run JSON files, the speed and egress
records, the operator's name, machine and runtime as stated in config.json, and the signer DID.
Nothing else in this bundle is personal: there is no hostname, user name or private key in it.
