Module 01 · Foundations · Interactive lesson
Precision
& Recall
A model scores every benefit application on how clearly the veteran qualifies. One threshold decides who gets fast-tracked. Move it, and you trade one kind of mistake for another — and every mistake is a real person. Drag the line and watch the confusion matrix breathe.
Live · 200 applications
98 fast-tracked
Score ≥ 50 → fast-tracked; below → held for manual review.
Precision–recall curve
The gold dot is your current threshold. Drag the line above and watch it ride the curve.
Confusion matrix
at threshold ≥ 50
Eligible veteran, correctly fast-tracked.
Not eligible, wrongly fast-tracked — a benefit goes to the wrong claim.
Genuinely eligible, wrongly denied — a veteran waits who shouldn't.
Not eligible, correctly held back.
Live metrics
Precision
0.79
Of those fast-tracked, the share truly eligible. Up = fewer false positives.
Recall
0.82
Of all eligible veterans, the share we reached. Up = fewer false negatives.
Accuracy
0.81
Share of all decisions that were correct.
F1 score
0.80
Harmonic mean of precision and recall — one balanced number.
The tradeoff, in human terms
Raise the threshold and you demand more confidence before fast-tracking. False positives fall (precision rises) — but you start denying genuinely eligible veterans, so false negatives climb (recall drops). Lower it and the trade reverses. There is no setting with zero of both: the clouds overlap. In a governed system you choose the point on this curve deliberately, document why, and send the borderline cases to a human — never let one silent number decide alone.
Govern, in practice
Precision and recall aren't just chart axes — they're the two ways an eligibility model can fail a veteran. Scholars pick the operating point on the record, the choice is audit-logged, and borderline scores route to human review before any claim is touched.
Mission first, people always.