mrothroc/sc35-s4d-lin-mixlab

docs: link the mixlab narrative writeup

docs: say where the example clip comes from

Add deployment scope (not streaming, no unknown class), trim overclaims

Editorial pass: tighter opening, US spelling, section renames

Correct the randn guidance: noise has no keyword, and it is not a confident answer

State the PyTorch port's serialization and training limits precisely

Real wav example, measured preprocessing warnings, file inventory, inference:false

Add verify_labels.py fixture, note CPU-only port, correct loader docstring

Add labels.json (index order is not alphabetical), warn that transformers/widget do not apply, correct loader error and input prep

Fix loading example, ship mixlab_ckpt loader, scope parity + portability claims

S4D-Lin reproduction on Speech Commands 35-way raw waveform (96.14% test)

initial commit