GAMMA Mind Meld Ops

Use for reproducible Mind Meld runs and focused diagnostics.

References

• src/mind_meld/README.md
• docs/BENCHMARKING.md
• tools/README.md

Workflow

1. Choose surface: tools/run_mind_meld_cli.py or gamma.py mind-meld.
2. Use fixed prompt, fixed --steps, and --summary-only for comparable runs.
3. Add --meld-diagnostics when debugging swaps, logits, or templates.
4. For quality instability, test --shared-chat-template, --order-neutral, and --soft-swap.
5. Prefer same-family model pairs before cross-architecture experiments.

Commands