We are recruiting the founding cohort of the measurement-assurance retainer. A small roster of companies whose evaluation numbers we keep true, quarter after quarter.
Not a dashboard and not a subscription to a tool. A quarterly cycle that ends in a number somebody outside your company stands behind.
Each quarter we take the numbers your buyers, board, or press will actually quote, and we verify the instruments they came from. Not a dashboard. A verified number with a receipt.
A signed document stating which instruments were read, what was found, what was fixed, and what the verified configuration is. Built to be handed to procurement, an auditor, or an insurer.
Datasets drift, judges get swapped, graders get patched. We re-run the verification each cycle so the letter stays true after the code changes.
We run standing probes against the judge models your evaluations depend on, so a silent behavior change upstream shows up in your inbox before it shows up in your numbers.
Before a number goes on a model card, a blog post, or a fundraise deck, we check the claim against the instrument. 48-hour turnaround.
The work behind this is public: the ledger of landed fixes, the map of the audited surface, and the scanner.
One verified number per category per cycle. When a seat is taken, it is taken.
Send the number you are about to publish. We check it against the instrument, free, in 48h.
If the instrument needs work, that is the Level-2 audit.
The retainer keeps it fixed: quarterly re-verification, drift watch, updated letter.