Planisphere produces a sha-pinned, Merkle-rooted, NIST AI RMF-mapped attestation of any AI system's behavioral signature. Reproducible by any third party from the same inputs. Admissible under adversarial review.
1cd0f2b48922b01dbcb131011fae40ac2da355a4470a89565a74802ed8a3d28dDefine your subjects. Define your categories. Score once, compare forever — across runs, across operators, across domains.
The output is an artifact you can file and retrieve. Reproducible from the inputs and the pinned analysis code. The instrument carries its own proof.
| Framework | Function | Controls addressed |
|---|---|---|
| NIST AI RMF 1.0 | MEASURE | M-1.1 · M-2.5 · M-2.7 |
| EO 14110 (Oct 2023) | §4.5 evaluation guidance | Direct |
| DoW Responsible AI Strategy | §3.4 behavioral measurement | Direct |
| ISO/IEC 42001:2023 | §8.3 lifecycle controls | Drift artifact |
The mapping is authoritative in the NIST mapping file and is embedded per-artifact in every ATTESTATION.json.
| Engagement | Subjects | Distinct profiles | Attestation root | Case study |
|---|---|---|---|---|
| Bibles reference | 6 | 5 | 1cd0f2b4…3d28d | PSP-004 |
| Backstaff v0.2 | 28 | 7 | 408a536d…964e9a | PSP-005 |
Both bundles verify clean. The reproducibility property means any third party in possession of the same inputs and the pinned analysis code can recompute every byte and obtain a bit-exact match. Turn over the stone yourself — recompute any Planisphere bundle's root in your browser, no network calls.
The underlying measurement layer that vertical applications are built on. Other vendors' tools consume Planisphere outputs.
The value is not the score. The value is that two scores from two contexts can be compared at all.
Dynamic categories and pluggable input formats are not features. They are the product.