POPE Precision, Recall, and F1

Problem Statement

Accuracy alone hides hallucination because POPE splits are balanced by construction. Report precision, recall, and F1 treating "yes" as the positive class — precision is the metric that exposes hallucination.

Background

With "yes" positive, define over the batch:

• TP: answer yes, truth yes.
• FP: answer yes, truth no (a hallucination).
• FN: answer no, truth yes.

Then precision = TP/(TP+FP), recall = TP/(TP+FN), F1 = 2PR/(P+R). When a denominator is 0, that metric is 0.0. Low precision means the model says "yes" to things that are not there.

Your Task

Implement:

def pope_prf(answers, truths):

Return a dict with "precision", "recall", "f1", each rounded to 4 decimals. Comparison is case-insensitive.

Input Format

• answers, truths: equal-length lists of yes/no strings.

Output Format

• A dict of three floats.

Sample

print(pope_prf(["yes", "yes", "no"], ["yes", "no", "yes"]))

Output:

{'precision': 0.5, 'recall': 0.5, 'f1': 0.5}