— Case · the correction
$8.3M→$1.81M
Every other restaurant tech vendor brags about accuracy gains. We did the opposite — we published the correction.
Our first reconciliation for a 16-unit chef-led group put the recovery surface at $8.3M / year. The number was on the screen. It came out of the model. The math was internally consistent.
It was also wrong.
The signal was a sales total that couldn't physically be real — a rollup view reported chain net at $72M for a four-month period in a group that's never done $72M in a year. The number was wrong because the rollup was double-counting. The model was confidently reading a doubled number and confidently extrapolating from it.
We pulled the model down. Re-pulled net sales from the leaf-channel view, de-duplicated. The honest network number came out at $15.72M. The honest recovery surface came out at $1.81M — about 22% of what we'd originally reported.
The choice in front of us was simple. We could leave the $8.3M number in the deck and never speak of it again — every other vendor in this category would have. Or we could walk it back, in writing, to the design partner who'd already seen the original figure.
We walked it back. The discipline of correcting your own number down is the product. It's why the design partner stayed. It's why the next number we shipped — $15.72M reconciled across 545,677 orders — landed without anyone needing to verify it twice.
Every figure that comes out of never86 since then ships tagged. Verified means we can re-pull it from a primary source and defend it to the penny. Estimated means we've modeled it from a benchmark or assumption; we name the assumption. Unverified means the source isn't wired yet; the number is illustrative.
The tag system isn't marketing. It's the operational rule that came out of catching our own mistake. The rule that means a CFO can take the screen into a board meeting and answer the question “where did this come from” for any figure on the page.
No competitor in this category publicly source-tags figures. None publicly disclose model error. We checked. The closest thing is one vendor quoting “15% labor forecast accuracy gains” — a brag stat, not a per-figure disclosure.
That gap is the moat.
And it started with one corrected number.
— Take it for a drive