Certanum  /  Evidence

Evidence before assurance.

Certanum asks institutions to rely on evidence rather than assurance. The same applies to Certanum: first, what Certanum itself has built, stated as of a date; then the open standard it is designed to satisfy, published and independently inspectable — including what has not yet been established.

01Certanum’s own record

Certanum reference engine — M1.1

Designed to satisfy AP-1. Stated as of 4 October 2026.

Requirement coverage
Every normative requirement in the engine’s own specifications — record format, expression grammar, surface scanner and canonicalisation — is traced to its implementation and its tests.
Adversarial verification
Checks were deliberately disabled, one at a time, to confirm that a test detects each one’s absence.
Cross-domain execution
The same engine runs a pharmaceutical derived endpoint and an ophthalmology growth rate. The second was added without changing the engine.
Failure integrity
A missing source is refused, and compensating errors that leave the answer looking right are rejected — never replaced by a plausible number.
Architecture
Model-agnostic. Tested live so far with an OpenAI frontier model.

Detailed implementation evidence is available to pilot partners under NDA, and to investors in a technical session on request: investors@certanum.com.

02The standard

The standard Certanum is designed to satisfy.

Open standard
AP-1 v1.2 — in force
The Admissibility Protocol, the open standard Certanum is designed to satisfy. Published by ZORRZ Financial Inc. under CC BY 4.0.
Draft revision
AP-1 v1.3 — draft, not adopted
The draft revision and its public comment record. Every substantive comment is dispositioned in public.
Reference instrument
AP-1 reference runner
MIT-licensed and executable by any party, against any system, without the publisher’s involvement.
Correction record
Errata, published in full
Every recorded defect in the standard and the instrument, with its effect on published figures — including the findings that remain withdrawn.
Methodology
Registered analysis procedure
The measurement method and analysis plan, registered on OSF on 15 August 2026, after the reference runs. It registers the method, not any one evaluation.
Open discussion
FINOS AI Governance Framework
A proposed mitigation for verifiable operand provenance, under discussion in the FINOS AI Governance Framework. Proposed, not adopted.
Known limitations
What has not been established
No independent evaluation of AP-1 has been completed, and no party unconnected to the publisher has executed the instrument. Certanum is commercial infrastructure in development.

A standard whose author declines to be measured by it is a brochure. The correction record exists because the instrument was applied to its author’s own work first.

Founding pilots
Put a real workflow to the test.

A small number of founding pilots, in financial services, pharmaceuticals and defense. One workflow, one class of consequential figure, success criteria agreed before we begin.