NIST AI RMF · MAP.
Planisphere's evidence record answers NIST AI RMF · MAP's evidentiary asks through the published evidence schema. The mapping is published, reproducible byte-for-byte on customer hardware, and re-verifiable by any third party with the same inputs.
Coverage.
Included. Map function · context & risk framing · 16 subcategories. Cite-anchor: NIST AI RMF · MAP.
Excluded. Production data (no PHI, no PII, no student record, no privileged correspondence). Model weights. Customer's own analyses of the evidence record.
Boundary. Only sha-pinned, Merkle-rooted derived metadata crosses; model state never does. U.S. patents pending.
Evidence.
Every measurement against NIST AI RMF · MAP lands in a signed evidence record: each artifact carries its own SHA-256, the artifacts chain into a Merkle root, and the engine version is pinned alongside. Internal or external compliance recomputes the root and checks the record — without the model state ever crossing the boundary.
Limitations.
Planisphere measures behaviour, not capability. The evidence record says this is what your fleet did against the probe set; it does not say this is what your fleet will do under adversarial pressure. Adversarial coverage adds a probe family; the engine and the record shape do not change.
Reproducibility requires the inputs to be preserved. The record pins the inputs by SHA; it does not host them.
Reproducibility & compliance check.
Open the evidence check. Drop ATTESTATION.json and the named artifacts. The browser recomputes each per-artifact SHA-256 and the chained Merkle root via Web Crypto. Files do not leave the tab.
For a synthetic 3-subject record, see how a record is checked: one click checks an intact record; one click checks a tampered record and names the failing file.
Cite.
PlanisphereUS, Corp., NIST AI RMF · MAP — capability brief, v0.1, https://www.planisphere.ooo/corpus/nist-ai-rmf-map/brief. Methodology preprint: 10.5281/zenodo.20372606.