Evaluations and datasets · Lesson 13 of 18 · 5 min

Detect evaluation leakage

Evaluation leakage happens when held-out information influences tuning or selection. A contaminated test set can exaggerate generalization performance.

Open the lab Course map

Mastery 0/3Run itExplain itApply it

learner@clairos:~ $

↑ history · Tab complete · Ctrl+C cancel · Esc, Tab to leaveExit 0

Your objectives

Hints are always available. Wrong turns never cost XP.

1
Read the case evidence
Read /opt/ai-lab/cases/aiops-evals-datasets-013.json. Separate the recorded facts from your assumptions.
Hint & worked example
Evaluation leakage happens when held-out information influences tuning or selection. A contaminated test set can exaggerate generalization performance.
The complete case JSON is read successfully.
```
cat /opt/ai-lab/cases/aiops-evals-datasets-013.json
```
Common mistake to avoid: The test remains independent because the answers were copied only once.
2
Isolate the deciding signal
Select the independent_test field from the case JSON. What does it establish—and what does it not establish?
Hint & worked example
Remove the leaked examples and evaluate on genuinely held-out cases.
A successful JSON query isolates independent_test.
```
jq .independent_test /opt/ai-lab/cases/aiops-evals-datasets-013.json
```
Common mistake to avoid: Using a different case or an unrelated signal as the justification.
3
Record your operational decision
Choose the justified code: replace-contaminated-test, change-without-evidence, or ignore-the-signal. Save that one code followed by a newline in /home/learner/lab/aiops-evals-datasets-013-decision.txt.
Hint & worked example
Remove the leaked examples and evaluate on genuinely held-out cases.
The decision artifact contains the correct code, not just a matching command in history.
```
printf '%s\n' 'replace-contaminated-test' > /home/learner/lab/aiops-evals-datasets-013-decision.txt
```
Common mistake to avoid: The test remains independent because the answers were copied only once.

Understand the pattern

Evaluation leakage happens when held-out information influences tuning or selection. A contaminated test set can exaggerate generalization performance.

Interpret the evidence

Remove the leaked examples and evaluate on genuinely held-out cases.

A tempting mistake

The test remains independent because the answers were copied only once.

All names, measurements, prices, versions, and thresholds in this exercise are fictional training data. The evidence supports this bounded scenario, not a claim about a live provider.

Worked example and expected result

$ cat /opt/ai-lab/cases/aiops-evals-datasets-013.json

{
  "fixture": "Fictional training data; not a live measurement.",
  "test_answer_in_prompt_examples": true,
  "independent_test": false
}

Exit status: 0. The command completed successfully in the training fixture.

$ jq .independent_test /opt/ai-lab/cases/aiops-evals-datasets-013.json

false

Exit status: 0. The command completed successfully in the training fixture.

$ printf '%s\n' 'replace-contaminated-test' > /home/learner/lab/aiops-evals-datasets-013-decision.txt

(no output; exit 0)

Exit status: 0. The command completed successfully in the training fixture.

Try a variation

Name the assumption that would have to change for “replace-contaminated-test” to stop being the justified action. Do not silently replace missing evidence with a guess.

Explain it

What does the evidence in “Detect evaluation leakage” justify?Make the change now and gather the missing evidence afterward.Remove the leaked examples and evaluate on genuinely held-out cases.The test remains independent because the answers were copied only once.

Apply it elsewhere

A colleague concludes: “The test remains independent because the answers were copied only once.” What should you do?Reject that conclusion; return to the case evidence and its stated limits.Accept it because the training command completed.Treat this fixture as a live measurement.

Keep the momentum.

Complete the lab, then explain and apply the pattern.

Next: Review dataset provenance Try an incident drill

Anonymous lesson and mastery progress lasts for this session. Sign in to keep synced lesson XP between sessions; mastery stays device-local. No speed bonus, lost lives, or hint penalties.

Keep learning

Case methodology: evidence, limits, decision