A sample with a brief, inputs, methods, checks and limitations.
2026-10-26.2 · 40 minutes estimated; work in several sessions
Saved on this device; sign in to import explicitly.
Prerequisites and goal
Choose a role path and a core task.
Start with diagnostics; skip explanations you can already apply independently.
Define acceptance before production
Define the audience, decision, time and format before building. Keep scope small enough to explain every step. Record dates, definitions and missingness. Label fictional data in both the title and table; do not present demonstration results as real impact.
Audit AI-assisted work
AI may suggest code or structure. Independently recalculate one key result, test one boundary and explain an alternative. Preserve inputs and execution instructions. Visual polish supports readability; judgment and verifiability determine quality.
Worked example · fictional
From a problem to inspectable work
Fictional: a student wants a dashboard for every industry in one weekend.
Narrow to a comparison of two roles, defining the sample and counting unit.
Keep source extracts and unknowns; reconcile different wording of the same requirement.
Deliver a one-page comparison and auditable coding table, with sample limits and dates; do not infer market demand.
Failure case and repair
A polished AI chart has no retained input.
Definitions, calculations and conclusions cannot be checked.
Rebuild a minimal input table, independently check it and add execution notes.