Skip to lesson
Exit
Capstone: Mock Loop & Self-Exam1 / 2

1 min lesson

Structure the write-up - decision first, evidence in the appendix

Work through the cases in "Structure the write-up - decision first, evidence in the appendix", pairing each signal with the move that fits.

Step 1 of 2

Structure the write-up - decision first, evidence in the appendix

Lead with the answer

Top of page: the cause, the blast radiusHow much breaks if a change goes wrong; the scope of potential damage. Press Enter for the full definition. in one number and the recommendation.

Example: “p99 rose 2.1s for the ~8% of users on client 2.14 in eu-west; recommend rollback. Evidence below.”

Evidence, then limits

Segmented charts and the confound checks you ran, in the appendix.

Close with limitations and next steps and keep the analysis reproducible (query + notebook).

A correct diagnosis buried under EDA still fails the round

Reviewers are senior and time-boxed. If page one is a histogram of latency and the recommendation is on page nine, you've demonstrated you can analyze but not communicate to a decision-maker - which is half the job here. The recommendation, the blast radiusHow much breaks if a change goes wrong; the scope of potential damage. Press Enter for the full definition. and the confidence go in the first three sentences. Everything that justifies them goes underneath.

Learn more

Full explanation

Self-score with a hiring-bar rubric

Self-score with a hiring-bar rubric

Axis
Clarity
Below bar
Decision buried; reader hunts for the point.
At hiring bar
Cause, blast radiusHow much breaks if a change goes wrong; the scope of potential damage. Press Enter for the full definition., recommendation in the first three sentences.
Axis
Correctness
Below bar
Blamed the global trend; missed a confound.
At hiring bar
Localized to a segment, ruled out traffic-mix, change-point confirmed.
Axis
Pragmatism
Below bar
Explored everything; recommended nothing.
At hiring bar
Made a call (roll back / fix forward) with the evidence that decided it.
Axis
Communication
Below bar
Notebook dump, not reproducible.
At hiring bar
Tight write-up, reproducible query/notebook, limitations stated.

Score the version you'd submit today, not the one you imagine after another pass.

QYour global p99 jumped after release N. You segment by client_version and no single version shows a jump, yet the aggregate clearly rose. What's the most likely explanation and your next move?