No numbers without provenance
The board pack on this desk has an empty figures section. That was a choice, and it is the one decision on this desk most worth explaining, because every finance document faces the same pressure and most of them give in to it.
The rule
A number appears on this desk only if it can be traced to a fact id in company/product-facts.md, or derived from one by arithmetic that is written down.
That is the whole rule. It is short because a longer one would have exceptions, and an exception is how the first untraceable number gets in.
What the rule cost this time
product-facts.md records the numbers section as open (N-1): no figures were supplied for this toolchain and none reproduce from any repository or report in scope. So section 3 of the board pack lists the figure names, the source they will come from, and the word "open" six times.
It would have been easy to do otherwise. Hosting and source control at this scale have public list prices. A monthly cost could have been assembled from them, a burn figure stated, a runway divided out. Every step of that chain is defensible on its own, and the result would have read like a finished document.
It would also have been a number with no owner. Nobody approved it, nobody could be asked where it came from, and the first person to check would have found a figure assembled from list prices rather than from what was actually billed.
Why one bad number is expensive
A document's credibility is not the average of its claims. It is close to the minimum.
A reader who finds one figure that does not reconcile does not discount that figure. They discount the document, and reasonably so, because they now have no way to tell which of the other numbers were assembled the same way. Everything true in the pack gets cheaper because of the one thing that was not.
The asymmetry runs the other way too. A pack that says "open" where it does not know makes every figure that is present more believable, because the reader has evidence that this author will leave a gap rather than fill it.
Why not just wait to publish
Because the shape is useful now and the shape is checkable now.
Publishing the structure lets the figure names, their sources and the method be reviewed before the numbers arrive, which is the right order. A method argued about in the abstract is argued about on its merits. The same method argued about once it has produced a number people like is argued about differently.
It also puts a date on the gap. Section 3 names what is missing, who closes it, and when: owner review of product-facts.md by 2026-10-03 (T-4), with wave 2 gated behind it (T-5). A missing number with a date on it is a tracked item. A missing number with no date is a thing everyone has quietly agreed not to mention.
The same rule, applied to failures
It works in the other direction too. The one recorded failure, a host serving a 404 for its Open Graph image after an icon set merged, appears in the pack with its fact ids (F-1, T-2) because it is in the record. The softer finding from the same review is reported as noted rather than fixed (F-2), because that is what the record says and not because it is the flattering version.
A pack that reports its failures with the same discipline as its figures is making a claim about how the figures were produced, and that claim is the only thing the reader actually has.