Guided setup — the walkthrough

How to read this

The frame below is the product. Move with Next, or jump with the step rail. Steps 2, 3 and 5 are live: change an answer and the join count, the preview and the ledger all move with it. Everything the mock claims to have inferred was inferred by a real run — see the four measurements.

§Notes on the mock

What is real

Every sample row, every column name, all 16 analyte spellings, the 1,005 / 57,200 / 765 counts, the overlap table, the candidate ranking and the “0 of 4” hepatic match — all measured. Sample rows are committed at data/delivery.js.
The source_col: spec on step 8 is the one that was actually executed through gsm.mapping::ApplySpec(), producing 57,200 mapped rows over 765 participants.
The ledger's consequences come from the workflow dependency graph in demo-301/workflows/, resolved through Mapped_* to raw domains.

What is invented

The visual design of the interview. open.gismo has no setup route today; this borrows the product's espresso palette so the mock looks like the product rather than like the report.
The chart on step 4 is a static SVG drawn from the real ALT distribution's shape, not a live safety.viz render.
config/setup-log.md does not exist. It is proposed here as the artefact that keeps the plain-English record next to the machine-readable one.

What the mock deliberately does not hide

Step 6 has no good answer and the mock does not invent one. The absent toxicity grade is the point at which this direction runs out of things to ask, and the screen says so in the user's own words rather than dressing it as a choice.
Step 7's answer has nowhere to go. The mock shows it writing to a module YAML, which is a patch, not a seam.