Mode B · Evidence tier
Reproducible enforcement behaviour and a per-output conformance record an auditor or regulator can rely on, bound to the governing policy version, tamper-evident, replayable.
What the evidence layer adds
Every accept/reject decision is bit-identical on re-execution: certifiable, not statistical. Re-run the input, get the same verdict, always.
Every output carries a signed record naming the policy version, verdict and timestamp, evidence written by the system, not about it afterward. Rule 4 →
Regulatory context
We make these checkable. We don't replace them.
Its record-keeping and high-risk provisions in force from August 2026 demand evidence a system behaved as declared. The conformance record is that evidence, per output.
Operational-risk controls over AI-driven processes require you to show the control operated. We add a replayable record beneath the control, not a new control.
The AI management-system standard asks for documented evidence of control. We supply the machine-written artefact that documentation points to.
Measure and Manage functions call for verifiable evidence over time. The record is the verifiable measurement, bound to the policy version that produced it.
The signature asset
The record is the product's hero object: it names what governed, what was decided, and when, bound so any later tampering is detectable.
Request a sample record →In practice
The firm's permitted disclosure and suitability language is configured as grammar. Each customer communication ships with a conformance record. Provable conformance, not a policy statement about conformance, is what clears the compliance gate when procurement asks.
The honest boundary
Adds Rules 5 and part of 6 to the software baseline; Rules 3 and 7 in full arrive at the hardware rung. The eight rules →