Capstone: The brief

The capstone is the Harbor Lane playbook run on your territory: a real business question from your job (or a realistic one from a domain you know), taken from vague sponsor ask to defended recommendation. Everything is fair game from the course; nothing is provided — finding and shaping the raw material is part of what's being assessed.

Choosing the question

• Pick a 'why' or a 'should', not a 'what'. 'Why did support volume spike in Q2?' or 'Should we keep the premium tier?' — questions needing investigation and judgment. 'What were Q2 sales?' is a lookup, and lookups can't earn the certificate.
• Pick something with reachable evidence. You need at least one data source you can actually access and 2-3 people you can actually interview (30 minutes each). A great question you can't get evidence for is a worse capstone than a modest question you can nail.
• Pick something someone wants answered. A real stakeholder — even just your manager — transforms the exercise: real questions in the readout, real stakes in the framing, and often, a real decision at the end. Capstones with live sponsors are consistently the strongest submissions we see.
• Size it to ~12 hours across two weeks: roughly 3 on discovery, 4 on analysis, 3 on the story, 2 on process/recommendation. If your scoping doc suggests triple that, shrink the question, not the quality. (That budget assumes you can free the hours — ideally from the recurring-report time Module 4 helps you reclaim. If your real job can't spare them, run the capstone on a realistic fictional version of your domain instead; it grades the same.)

The scoping doc (submit before you start)

1. 1Write the question in one sentence, plus the decision it informs and who makes it.
2. 2Write the metric definition at the heart of it — the one sentence you'd get a sponsor to sign (Module 1's discipline; most capstone drift traces to skipping this).
3. 3List evidence sources: interviews planned (role, what they know, what they might be motivated to shade) and data available (source, grain — say what one row means, time range).
4. 4State your starting hypothesis, labeled as one, with what would confirm and what would kill it.
5. 5Set up the workspace: one AI project holding the brief and everything to come; your numbers log started; the data-policy check done for whatever you'll paste (mask real customer PII — the habit matters more in the capstone than anywhere, because this data is real).