Five gates, run in cost order.

A study runs its cheapest killers first. Any gate can end the engagement on the spot; only a format that clears all five earns a go.

The gates

1 · The category’s record, and the cost to prove. Before format work, before machines, before any site: the category’s own record is read in filings, dated press, and the operators’ own statements, and the total capital required to prove the model is priced against the return available at demand somebody has actually measured. This gate is first because it costs nothing and the evidence is already public. A model that cannot pay back at any measured demand is not rescued by a better site, a better machine, or a better menu. Project 01 built before it priced this, and the amendment is permanent.

2 · Regulatory posture. Statutes are read in their own text, not in summaries. Then the enforcement record is read in the cities’ own open data, because how a rule is policed is a separate fact from what it says. A format that works only while enforcement lags is excluded.

3 · The machine. Buy or build, priced landed: freight, duty, tax, brokerage, and the electrical field evaluation that overseas sticker prices omit. Two binary filters ride along: the operator must own the payment flow, and the parts must carry certifications an inspector will recognize.

4 · The site. Only a format that has cleared the first three gates earns a site assessment: exposure, zoning, utilities, and the site constraints that bound what a given structure can hold, circulation among them. A site can cap a use, and that cap is worth knowing before a buildout is drawn. What a site cannot do is rescue a format the arithmetic has already closed, which is why this gate runs fourth and not first.

5 · Kill criteria, then a verdict. Numbered, numeric, dated criteria with the exit action written in advance. The study closes with one of three words: go, pivot, or kill, plus the named conditions under which the question would reopen; absent those, a kill is permanent.

Standing rules

A short list of invariants that recommendations cannot violate: no human staff in store operations; the operator is the merchant of record; no format whose ticket hands the payment rails more than about eight percent of revenue; cheap and reversible before fixed and final.

Everything is receipted

Regulatory text is quoted from the statute. Financial figures come from filings. A vendor’s marketing claim is labelled as one. Load-bearing claims are adversarially checked against primary sources before a document is trusted, every task ends in a written report, and the reports are the record.

What that looks like in practice

Project 01 ran this method against a real corner: eight concepts entered, and each died against a named criterion or a falsified premise. The verdict was kill, on the category’s published record and the cost of proving otherwise. The misses, and the method change they forced, are published in the study.