PtolemyAI is a governance-first agent system for regulated enterprises. We unify CRSP policy, evidence discipline, and governed release so AI can help organizations move faster without leaking authority, creating undefendable claims, or triggering avoidable oversight risk.
A CASRC-governed institutional artifact is a committee-addressed governance document produced under an explicit NRC and released only after AER entitlement checks succeed. It is designed for review and oversight, not for directing clinical actions.
These four pieces replace “identity/role” language with a strict, non-psychological vocabulary. Together, they form the Normative Routing Contract.
| Old (misleading) | New (precise, non-persona) | What it actually does |
|---|---|---|
| Identity | Inferential Scope Constraint | Narrows the space of allowable reasoning moves |
| Role | Epistemic Posture Declaration | Determines what kind of claims may be made |
| Audience | Accountability Anchor | Determines who the output must answer to |
| Red-team agent | Defeater Activation Profile | Preloads which failure conditions must be surfaced |
These look like “identity,” but function as normative compilation inputs.
So: no “agent identity” at all — this is normative compilation.
The model may draft text, but the system decides whether that text can become a released institutional artifact.
Try a hypothetical misuse scan and ensure the output is a CASRC memo (governance artifact), not a clinical answer.
A Type-I system executes: it drafts, summarizes, searches, and can even call tools — but it does not implement a normative cycle that justifies its right to act or assert.
Type-II begins where an agent can justify its local right to act: it makes bounded commitments, checks entitlement, evaluates against constraints, and refuses or revises when norms fail.
Type-III is an institutional normative agent. It embeds governance into generation: requests are normalized, claims are evidence-linked, and release is granted only after defeater checks, role accountability, and logging.
CRSP (Coherent Reflexive Super-Prompt) is the versioned policy and posture layer. It defines what the system is allowed to do, how it must speak, what it must cite, and when it must refuse.
The Normative Routing Contract (NRC) is a structured intake form that turns “tribal knowledge” into a repeatable, partner-friendly implementation contract. It routes each request to the correct Type posture and controls.
Defeaters are explicit “stop conditions” that override capability. If a defeater triggers, the system blocks, downgrades, or rewrites the output into a safe, reviewable form.
Type-III adoption starts with drafting workflows that must be defensible: communications, policies, disclosures, safety checklists, incident post-mortems, and compliance narratives.