2026-04-27 - Phi-4-mini Layers 8–11 Run-Batches Intent

Intent: Use the new run-batches stage for the next actual bounded Phi-4-mini build+test batch, layers 8–11 only, following validation-before-scale discipline and the project ANE-only gate policy.

Setup: Planned command: the bounded batch stage of the Phi orchestration script; preflight before the batch, then existing per-layer convert, compile, strict MLComputePlan residency, and numerical smoke gates.

Result: Intent recorded before execution; no placement, cosine, latency, energy, or artifact-count results yet.

Surprise / hurdle: The key risk is proving batch orchestration can run one real four-layer batch without expanding into full 32-layer conversion or bypassing per-layer gates.

Lesson: A stopped-after-one batch is the safe next step for batch automation when every layer still passes residency and numerical smoke before scale-out.

Next: Run only layers 8–11 through the guarded batch; do not run full 32-layer conversion, performance, energy, cleanup, or deletion.

Refs: research/ANE_CHAIN_SCHEMA.md