Preserve unfavorable trained rank0 derivative evidence

[CompleteDotTech]

Evidence

Preserve the completed actual-trained rank0 FP32 derivative failure for issue #3. Corrected inventory is 294 tensors / 494,090,176 scalars. Original incorrect-inventory failure remains preserved separately.

Only batched/serial passes. HF backbone fails 132/290 gradients; HF task fails 194/294; native serial/full and batched/full each fail 22/294. Frozen tolerances and acceptance flags are unchanged. This probe constructs no optimizer and establishes no CPU-state optimizer arithmetic, quality, or issue closure result.

Portable receipt hashes and all five comparison counts were checked against the original worker/supervisor artifacts. Historical checkpoint birth and reviewed runtime provenance remain distinct. Later diagnostic interventions are external and have not changed production source.

Validation

• Locked development environment restored with uv sync --locked --extra dev --extra native --python 3.13.
• Full pytest: 875 passed, 62 skipped.
• Required compileall: passed.
• Source snapshot audit and feature-matrix verifier: passed.
• Commit SSH signature verified locally.

Refs #3. Keep issue open; remaining acceptance work is not complete.