2026-04-27 - Phi-4-mini Four-Layer Fused Shard Intent

Intent: Probe whether a larger fused Phi-4-mini shard beyond the validated 3-layer/288M pattern remains ANE-resident before any broader scale-out, following validation-before-scale discipline and the project ANE-only gate policy.

Setup: Planned non-destructive build: one INT8 stateful CoreML shard for layers [0,4) under local artifacts, then compile and run strict MLComputePlan residency validation. No cleanup/deletion, no golden validation unless requested after residency, and no energy benchmarking.

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

Surprise / hurdle: The open question is whether fusing a fourth layer crosses a CoreML/ANE placement boundary even though smaller fused shards have stayed resident.

Lesson: Fused-shard scale-out should advance only after the next larger representative shard proves strict ANE residency.

Next: Build and compile only the layers [0,4) probe shard, then record strict residency results before considering golden validation or larger fused ranges.

Refs: research/ANE_CHAIN_SCHEMA.md