2026-04-27 - Phi-4-mini Layers 12–15 Run-Batches Intent

Intent: Continue the actual Phi-4-mini bounded build+test with run-batches for layers 12–15 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; completion proceeds in explicit four-layer batches with preflight, 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 ensuring this remains one checked bounded batch rather than an unchecked full 32-layer conversion.

Lesson: Four-layer batches keep scale-out auditable when every batch is explicit and every layer must pass residency plus numerical smoke gates.

Next: Run only layers 12–15 through the guarded batch; do not run performance or energy tests, and do not clean up or delete artifacts.

Refs: research/ANE_CHAIN_SCHEMA.md