Week 26 — Capstone IV — Ship, Measure, Defend

Capstone · Mock review #7: the final panel

1. Ship (~2.5 hrs)

• Production environment separate from dev and preview, deployed from CI only
• Staged rollout: you → 2–3 friendly users → full pilot group, watching SLOs and errors at each step
• Runbook covering the top 5 failure modes, kill switch, rollback and contacts
• The experiment from week 25 running

2. Measure (~2 hrs)

At the end of the week (or as late as the sample allows):

• Experiment readout with a confidence interval and a decision (ship, kill or iterate)
• North-star and guardrail metrics against the targets set in week 23
• AI-feature production quality: a sample of real outputs graded against your eval rubric. Did production match the offline evals?
• Actual cost vs the estimate

3. Launch memo (~1 hr)

Write one page for an exec audience covering:

• what you built and for whom
• the result, with numbers and uncertainty
• the risks you’ve handled and the ones that remain
• what it would cost to scale
• the recommendation and the next three bets

No jargon.

4. Capstone defence (~1.5 hrs)

Tell Claude: “Run the final panel: a staff SWE, a senior data engineer, an SRE and an exec. 60 minutes. Don’t go easy.” Present:

• the design doc
• a demo
• eval results
• the experiment readout
• the cost
• the postmortem from any real incident

Pass bar:

• You defend each major decision with its trade-offs.
• You identify your system’s weakest point before the panel does.
• You say “I don’t know, here’s how I’d find out” where that’s the honest answer.
• The exec understands the memo without asking for translation.

5. Retrospective (~1 hr)

Write retro.md:

• your week 1 vs week 26 self-assessment on CLI, git, Python, TypeScript, SQL, Databricks, AI engineering, production, system design and stats (1–5 each)
• where estimates were most wrong, across the whole course
• what you’d now do differently as a PM, and what you’ll stop doing
• the next 3 skills to deepen

6. Publish (~1 hr)

• Clean up the repo: README, architecture diagram, how to run it, and screenshots.
• Turn your 26 teach-backs into an index page.
• Write a public post: “26 weeks from PM to product engineer: what I built and what I learned.”

Done when

• Capstone live with pilot users
• Experiment readout and metrics reviewed
• Launch memo written
• Final panel passed
• Retrospective written
• Repo and teach-backs published