Irregular sampling and prospective regularization v0.1

The pipeline now supports a session-adaptive regularization declaration:

ResampleOperation(rate_hz="median", max_gap_factor=1.5)

Placed before BaselineDFFOperation("asls"), this converts timestamp jitter from a structural incompatibility into an explicit, auditable workflow choice. It does not alter the completed IBL v0.3.2 analysis.

Every resampling operation records its requested and resolved rate, source interval CV, median and maximum source intervals, target/source rate ratio, resolved gap threshold, fraction of target samples protected as gaps, and maximum and 95th percentile distance to a source sample. Original timestamps and values remain on the source_time dimension.

The frozen smooth-signal benchmark is irregular-resampling-v0.1.json. Across 20 and 50 Hz sampling, three signal frequencies, and 0.5% timestamp jitter, all six scenarios passed the prospectively stated 1% normalized-RMSE threshold. The worst result was 0.248% at a 3 Hz signal sampled at 20 Hz.

This is an engineering acceptance test, not proof that interpolation preserves all photometry features. Sharp event transients, discontinuities, missing runs, event coverage near protected gaps, and a new held-out real-data comparison remain required before regularized AsLS is promoted beyond experimental status.

The subsequent sharp-transient and missing-run study is reported in transient-gap results v0.1.1. It supports median-rate linear jitter regularization but rejects silent interpolation across missing event-window samples.

The governing rationale is SDR-0011.