# Practitioner review and revision guide

Five tasks are worth 20 points each. Run the learner's code, trace conclusions to raw records, and require critical rule corrections even when the score exceeds 14. Actual practitioner allocation remains a delivery decision; do not pre-fill a practitioner's identity or claim that a review has already happened.

## Review conversation

Ask the learner to open the output and give a conclusion in three minutes, trace one piece of evidence in five, recalculate a boundary in five, and record one revision plus its acceptance check. Retain the original, revised output and test record.

## Questions by task

- Task one: move a conflicting ID to the start of the input. The accepted result must not change. Replay across feeds should compare canonical content, not source labels.
- Task two: move a future delivery before the cutoff and explain the cohort change. Later feed recovery must not rewrite the old historical review. Why is cancellation not on-time delivery?
- Task three: authorize op-7's role but retain its high frequency. The export conclusion must change. A denied request changed to 200 needs session and scope checks. Approved bulk archival is a necessary counterexample.
- Task four: compare a 25% rate at n=4 with n=20. The former needs more data first. Exactly 15 percentage points is Review needed. Explain a changed service weight.
- Task five: identify where production R5 tests will run. Frontend visibility cannot establish identity integration. Reconcile 140 hours to four 35-hour weeks.

## Revision record

Record finding → evidence → affected conclusion → correction → re-test → learner explanation. Complete at least one revision per task. Different visualizations, wording and architectures can pass when their rules, evidence, assumptions and acceptance are sound. A reference answer is not a script to memorize. Do not invent employment, client endorsement or outcome claims.