7.1.7
- Defer percentile-threshold materialization until a bootstrap path actually needs the threshold field, reducing setup overhead for dask-backed day-of-year percentile count indices.
- Keep percentile bootstrap cases using
threshold_min_valueon the exact safe tiled path after Kraken validation showed the experimental fast path was not field-identical on real precipitation data. - Keep
cftimepercentile bootstrap cases on the exact safe tiled path after Kraken validation found a remaining one-cell mismatch in the experimental fast path. - Add maintainer notes for bootstrap architecture, readable implementation boundaries and the validated fast-path support matrix.
Validation
Kraken validation on Wednesday, July 29, 2026 confirmed:
- exact fallback agreement for wet-day percentile bootstrap reruns;
- exact safe fallback routing for
cftimepercentile bootstrap in this release; - local bootstrap regression slice passing before merge.