DELAWARE · C-CORPORATION
COMMERCIAL INFRASTRUCTURE IN DEVELOPMENT
Some numbers cannot be guessed.
Certanum is building verification infrastructure for AI systems that produce consequential numbers. Every figure computed from authoritative source data, traceable, reproducible — or withheld.
A tool call is not proof.
A computed figure and a generated figure are indistinguishable in text. A tool-call record establishes that a tool executed; it does not establish where the values entering it came from.
Three of the four links are probabilistic. A deterministic component inside a probabilistic pipeline does not make the pipeline deterministic. The wrong input, a skipped calculation or a mistranscribed result still produces a number that looks right.
Output inspection cannot separate them
The model interprets. Code computes. Evidence travels with the figure.
Certanum is designed to partition the generative component out of the numerical path. What remains in the path is conventional deterministic software, developable to the assurance bar the domain requires.
An operand that cannot be resolved does not degrade the answer. It withholds it. Refusal is a structural property of the architecture, not a policy applied to a model. Certanum does not certify the truth of the source data or the suitability of the domain logic; it establishes the derivation.
Don’t take the number’s word for it.
One consequential figure, three outcomes. The same arithmetic can be verified, rejected or withheld — depending on what the evidence shows.
Illustrative record with synthetic data, shown to explain the evidence model. It is not a product screenshot or a customer result.
Where an unevidenced number carries consequence.
The failure is structural rather than sectoral. It matters most where a figure is acted upon, recorded, submitted or defended.
AI is moving from generating language to executing workflows.
As AI systems invoke tools and supply inputs to consequential calculations, evidencing the numerical path stops being a research question and becomes an infrastructure one.
Sources as linked. Certanum is not endorsed by any company or authority named above. Descriptions are indicative and are not legal advice.
Certanum is not.
The measurement is open so the claim can be falsified. The system built to satisfy it is proprietary.
AP-1, the Admissibility Protocol, is an open standard for measuring whether a consequential figure produced by an AI system was computed, traceable, reproducible and refusable. It is authored by Marcus Rupp and published by ZORRZ Financial Inc. under CC BY 4.0, with a reference instrument under MIT — applicable by any party to any system, including Certanum’s, without the publisher’s involvement or consent.
Certanum does not control the standard and holds no certification, registry or accreditation role.
Entrepreneurship, AI assurance, mission-scale verification and financial technology.
Advised by Tim Moss — thirty years in engineering leadership. Sony Santa Monica founding member; Google Stadia ML and Meta Oculus Studios. Production machine-learning systems serving over 100 million users.
Affiliations are shown for identification purposes only. Both Co-Chief Technology Officer roles are held alongside primary employment under approved outside activity. No endorsement by NASA, JPL or Caltech is implied, and none is sought.
One workflow. One class of consequential figure. Success criteria agreed before we begin. Certanum funds the initial pilot work.