The dataset review lab and the refresh cadence

Review lab: L3.3's released 47 cases go under review and the frozen gate turns out to cover 7 of 18 cells (38.9%) — the 11 it misses are the cells L3.1 called starved. Two coders agreed on 42 of 50 traces (84.0%), but those 8 disputed traces expand through parentage into 13 of 47 graded cases (72.3%), and 0 of the 13 sit in the gate. Three twelve-case samples find 4, 5 and 7 of the 12 actionable verdicts. Graduating one case per starved cell closes the coverage gap and drops the gate's fresh share from 36.4% to 18.2%. An incident-only cadence reads 0 of 47 arrivals.

This Evals Engineering lesson is part of a paid plan. Sign in with a plan that includes this course to read it.