Reviewing how the engagement documents would be verified, the question came up of what standard they had to meet before shipping to a marketplace where corporate buyers screen consultants.
Assume a frontier model audits every page for logic and arithmetic, and build the deliverable to survive that rather than to read well.
The existing verification asked whether a claim was permitted. It never asked whether the analysis held. Those are different failures and the second one is more expensive, because a buyer's model finds a total that does not sum in seconds and then discounts everything shipped beside it. The cheapest defense is to run the hostile check first and publish the result.
A reasoning pass was added separately from the compliance pass: arithmetic must recompute, a term must hold its meaning end to end, every load-bearing assumption must be stated with its basis, the conclusion must follow, and the first hostile question must be answered inside the document. The house standard became his own merchandise plan, which reconciles a headline number three independent ways and prints the variance, and flags every input as documented, benchmarked or placeholder.
Publishing your own uncertainty converts an audit from a risk into a selling point. Most consultants hand over a model that looks certain; handing over one that says exactly where it is solid is the rarer and more valuable act.
He set the bar by describing the reader, not the rule. A buyer with a model open is a specific person, and that is easier to write for than a standard.