Day 8. The board meeting is tomorrow morning at 9am and Mateo's slide deck is open on the second monitor. Slide 4 says 'Sentinel scribe — pilot expansion, 12 → 40 clinics'. The new prompt is good. Accuracy is up, the dose-mention rule survives the metformin case, the structured output parses cleanly, the four lies are named and defended, and the injection ladder from yesterday didn't crack the system prompt. Kai's been celebrating since lunch. Then Mateo walks over, looks at the deck, and asks the one question that puts the room back in its seat: 'How do we know it won't regress the next time someone tweaks it?' That's today. By 6pm you will ship the prompt, its 5-case eval, and the last-passing-run JSON as ONE versioned artifact that Mateo can replay on his laptop in 30 seconds. Everything you learned in Chapters 1 through 7 collapses into this one shape: prompt-as-versioned-artifact. If you can't replay the eval and watch it pass, you didn't ship.
You'll walk out able to
- Treat the prompt + eval + last-passing-run as one inseparable artifact, not three files in a folder
- Drive a multi-turn conversation to a schema-clean versioned prompt manifest
- Pick regression-on-replay over human review as the only deploy signal you trust under pressure
- Tell a real PR description from a vibey one — and name what makes it real
- Write 3 deterministic rollback-trigger eval cases that fire without a human in the loop
- Dial reasoning-effort for nightly regression runs without burning the cost budget
Your first task
Daria's old layout vs your new one.
Mateo opens the Notion folder Daria left behind: it's just prompt files. `v3_final_FINAL_works_pls_no_touch.md`, `v3_final_FINAL_v2.md`, `v3_final_FINAL_pleasework.md`. You're about to commit the new layout — one folder per prompt, three files in lockstep. Mateo wants to know which line of the change is actually load-bearing.
What you'll answer
Read the diff. Which change is the load-bearing one — the move that makes 'regression on replay' actually work?
Free · no card · we save your progress
Save your progress & earn the certificate
You've seen the opening — keep going free. 7 tasks · auto-graded · finish in 90 min and the certificate is yours. Free forever · no card.
No password. To sign in on your phone, request an emailed code on the sign-in page.
Already started? Sign in with an emailed code instead.