-
SIDECAR1 rule 3: measured evidence that the answerability gate must run at inference, not as a training filter
-
SIDECAR1: measured survival table (HLS, TS, CMAF, copy, transcode, HEVC) and fingerprint numbers
-
SIDECAR1: encoder-coupling claim scoped to the measured ROI result (no benefit above the Johnson line)
-
nanolab 0.4.0: Sidecar/1 envelope (answerability record per measurement, refusal as row, per-row fingerprint, monotone-alignment rebinding); priority statement with prior-art credits; 21 tests pass
