Internal — do not ship to learners. This note is for the author/recorder.
Every example-output.md and everything under exercise/solution/ is an AI-generated stand-in. It was produced by actually running the author's prompts/assets (reasoning modules) or by computing on the real datasets (data modules) — not hand-waved — so a reviewer can see the intended shape and the folders aren't empty. When you record, replace these with your actual on-screen outputs. The walkthrough.md is your script/reasoning; the example-output.md is the artifact to end on.
Reasoning modules — SP1–SP6 and M6 (Claude Code, generative)
Outputs are genuine Claude runs, but generative output varies run to run. When you demo live (or solve the exercise), capture your actual output and replace example-output.md if it differs materially. Content/structure will match; exact wording won't.
Data modules — SP7, SP8, SP9 (computed — accurate) Outputs (CSVs, charts, numbers) are computed from the real provided datasets and are correct: CPA ≈ $51 / +5.9% systematic bias / CVR on-target / LTV $96 (discount) vs $447 (full-price) / India-as-confound. Keep as-is, or re-run the analysis live and replace. If you ever regenerate the datasets, re-run the analysis so the numbers stay in sync.
M8 (AI creative) — direction-only
example-output.md has the real Claude direction + self-contained image prompt + copy, but no images (no image tool in the authoring environment). At recording, generate the actual images in your image tool and add them to the solution.
M9 (background agent) — live web sources, time-sensitive The brief + audit use real web sources captured at authoring time (competitor pricing; market-size estimates). Prices and market figures change — re-verify every cited source at record time and update. The teaching point should still hold (pricing is trustworthy and consistent; the market-size "CAGR" disagrees wildly across firms — 6.75%–19.5%), but confirm the specific numbers and links.
| Module | Output provenance | Action at recording |
|---|---|---|
| SP1–SP6, M6 | genuine Claude run (varies run to run) | re-run live; replace if materially different |
| SP7, SP8, SP9 | computed from real datasets (accurate) | keep, or re-run analysis |
| M8 | Claude direction only, no images | generate real images, add to solution |
| M9 | live web sources (time-sensitive) | re-verify sources + update numbers/links |
- Exercise solutions are one strong example — student submissions will legitimately differ; grade against the requirements in the exercise
INSTRUCTIONS.md, not against an exact match. - SP5 demo has no provided senior-dog persona panel (personas are seeded live) — see the open decisions in the README.