Kayak-share rebalance scores are still 200s, but dispatchers say predicted needs van is wrong all week. Captured predictions sit in S3; the true van-needed labels land a day later in another prefix and were never joined. What should they run?

Short Explanation

Dispatchers say predicted needs van is wrong all week. Join captured predictions with the delayed ground-truth labels, then run a model-quality monitor. HTTP 200s rule out a 5XX problem, and a data-quality-only job skips that join.

Full Explanation

A model-quality monitor needs captured predictions joined to delayed ground truth, then a matching metric such as accuracy, precision, or recall. HTTP 200s rule out a 5XX problem. Comprehend and a data-quality-only job miss that join.

×