Module 13 · Responsible systems

Module 13: construction feedback

Try practice before reading the worked notebook. Download solutions or open notebooks/practice/module13-solutions.ipynb in the complete source tree. If a download link does not open, use the offline access route.

recall_denominator

Recall conditions on actual positives, not predictions. For recall_denominator(8, 2, 11, 30), the expected result is 19.

def recall_denominator(tp, fp, fn, tn):
    return tp+fn

safe_rate

Unavailable is not evidence of parity. For safe_rate(8, 19), the expected result is 0.42105263157894735.

def safe_rate(numerator, denominator):
    return numerator/denominator if denominator else None

wilson

State the conditional independent-binomial assumption before interpreting the interval. For wilson(8, 10), the expected result is [0.49016247153664183, 0.9433178485456247].

def wilson(successes, observations):
    from practical_ml_ai.learning import wilson_interval
    return list(wilson_interval(successes,observations))

Complete-study results feedback

In the verified CI run 40697234797 (October 28, 2026), the group has TP 8 and FN 11: recall 8/19 = 0.42, with an illustrative Wilson interval of about 0.23–0.64 (flagged as a small denominator). The female group's interval is about 0.57–0.64 on 590 actual positives. The small group's interval is wide; overlap is not a test of equal treatment, and the intervals assume a sampling model the Adult extract does not establish.

Reasoning task feedback

A group with TP = 0, FN = 0, FP = 0 and TN = 5 has selection rate 0/5 = 0, but recall's denominator is 0, so safe_rate(0, 0) is None. A group with TP = 3, FN = 0 has recall 1.0. Nothing about the first group's misses was measured, so comparing the two recalls is impossible, not a finding of parity.

Judge the explanation

Compare your explanation with the criteria in the study workflow: passing the visible checks alone is developing; a changed case that you predict, implement and explain meets the task.

Transfer feedback

Compare two groups with equal rates and unequal counts. Explain why this does not establish equitable outcomes or justify a group ranking. A defensible response reports counts and intervals with each rate: equal rates on 10 and 100 records carry very different uncertainty. It explains that neither a group ranking nor equitable outcomes follows without the decision context and the people affected.