Reference graders and the join that decides them

Join lab: L4.2 ended with four text assertions and five rows they cannot read, so this lesson builds the reference grader the module plan names next — and finds its accuracy is a property of the join, not of the comparison. A label-string join matches 0 of 47 released cases because the store labels failures with 9 open codes and the dataset with 7 axial categories; L2.2's alias table matches 47, and 11 cases sit in 2 categories that absorb two codes each. The reference stores an integer of cents and the trace a two-place string, so comparing as stored agrees on 0 of 30 recorded cases and reading money agrees on 30. Keyed on scenario, label and amount the grader decides all 34 failures at 100.0% accuracy — and 4 of those catches are another row's reference. Adding the day shares no key across 50 traces, borrows nothing, and decides 30 rows: 4 missed failures, 16 passes left alone, 92.0% accuracy. The 4 borrowed catches are exactly the 4 failures with no released case, and the reference set covers 0 of the 16 passes, so its 0 false alarms is a property of the reference set.

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