rdmsm4x changelog — Local LLM Benchmark canonical-source promotion

Local time: 2026-08-30 22:30 America/New_York
Host: rdmsm4x
Scope: source promotion, packaging correction, and test-boundary evidence

What changed

• Created canonical prototype project /Users/richh/dev/ai/local-llm-benchmark from retained staging source /Users/richh/dev/_migration/fleet-ingest/rdmbair15m5/dev/local-llm-benchmark.
• Retained staging unchanged. Excluded .agents, .venv, caches, live Git metadata, .env*, auth.json, cookie-named files, PEM/key/id patterns, and platform debris.
• Copied 150 eligible files to 150 destination files with an empty post-copy differential under the same exclusions.
• Created local commits:
  • d7c1ee6b3e6e0eac507827cb8294b47112412298 — promoted source
  • 8f9d0d8bde08d3633829c0701eebbd6dd7e47c21 — safety contract/provenance/resume
  • 342e3c1ea029065d2329c53eb4809d558fe5e72c — declared missing reportlab dependency and recorded runtime test gate

Verification

• First canonical uv run pytest tests/ created an isolated project environment and revealed five collection errors: fixture code imports reportlab, but it was not declared.
• Added reportlab>=4.0 to pyproject.toml, refreshed uv.lock, and reran the complete suite.
• Current canonical result: 399 passed, 3 failed in 21.02 seconds.
  • One test requires optional MLX.
  • Two tests assume a live Ollama model/API.
• No model weights were downloaded, no inference service was started or reconfigured, and no user documents/Photos libraries were read.

Next owner action

Split fixture/unit coverage from opt-in live MLX/Ollama integration checks through injected adapters or explicit markers. A real benchmark requires a separate host-capacity, model-installation, data-retention, and runtime authorization.

Apple Notes publication

This session is Background rather than Aqua, so the required Apple Notes entry is pending. From Terminal.app on the rdmsm4x desktop session run:

zsh ~/scripts/notes_changelog.zsh ~/dev/LLM/Claude/changelogs/rdmsm4x-changelog-20260830-2230-local-llm-benchmark-promotion.md