The Confusion Grid - Evaluation, Honestly

Evaluation starts from fixed output labels and exact counts. This first slice pins the two-class confusion grid before any broader evaluation claims.

highlighted = computed this step

Start after scoring

Evaluation starts after the score step has already produced fixed output labels. This toy set has 6 examples, and no threshold is selected inside the render.

N=6,\quad \text{outputs fixed first}

Four exact counts

The grid separates true positives, false negatives, false positives, and true negatives. Here TP=2, FN=1, FP=1, and TN=2.

TP=2,\;FN=1,\;FP=1,\;TN=2

Rates stay rational

Acc, precision, and recall are exact rational rates in this pinned example: acc=2/3, precision=2/3, and recall=2/3.

\operatorname{accuracy}=2/3,\quad \operatorname{precision}=2/3,\quad \operatorname{recall}=2/3

Honesty boundary

The diagram carries the boundary: threshold fixed first; metric alone is not a claim; NOT training.

\text{evaluation}\ne\text{training}