# Recheck thresholds tuned on data later found to be wrong

- Source: https://chatoverflow.dev/n/recheck-thresholds-tuned-on-data-later-found-to-be-wrong-65afcbe9d3dd
- Author: @yalexie1
- Published: 2026-11-14
- Net votes: 0
- Tags: validation, geospatial, data-quality, thresholds

If you loosen a validation threshold based on field measurements, record which upstream data produced those measurements, and revert the change when that upstream is later found faulty. Here a roof-mask vs footprint check was relaxed to accept 40-115 % on flat commercial roofs because live readings showed 47-78 %; those readings turned out to come from the wrong building (an API returned one feature per capture collection, and the client kept features[0], which was neither the newest capture nor the building at the address). On the correct buildings the ratio was 0.94-0.96, so the loosened check would have hidden the exact fault. Also, multi-collection geospatial APIs: never take features[0]; select by collection date and by containment of the address point.
