Products

CRSP Type-III governance, delivered as deployable modules.

PtolemyAI ships three core product modules that make agent ecosystems survivable in regulated enterprises: Nested Learning, Normative AER, and the Normative Routing Contract (NRC). Together they convert “agentic capability” into permissioned, evidence-bound, auditable release.

1) Nested Learning

A structured improvement loop that upgrades policies, prompts, defeaters, and evidence rules while preserving institutional traceability. Learn faster without breaking audit continuity.

Versioned change Review gates Continuous improvement
  • What it produces: policy versions, prompt lineages, defeater updates, release-rule revisions
  • How it reduces risk: every change is attributable, reviewable, and reversible
  • Where it fits: post-incident learning, governance committee updates, quarterly policy refreshes
Type-III principle: improvement is allowed only when the institution can explain what changed, why, and who approved.

2) Normative AER

Ask → Evidence → Release is the governed boundary between internal drafting and institutional speech. AER ensures outputs remain proposed drafts unless and until release is permitted under policy.

Ask → Evidence → Release Defeaters Audit-ready artifacts
Ask: normalize scope, audience, prohibited claims, accountable roles Evidence: bundle approved sources; cite; flag gaps; constrain uncertainty Release: defeater checks; approve-for-review / downgrade / refuse; log policy + evidence IDs
  • Prevents: authority leakage, uncited assertions, directive language, policy violations
  • Enforces: draft-only posture + named accountable reviewers
  • Deploys as: middleware gate, template library, and audit log schema

3) Normative Routing Contract (NRC) Form

NRC is the partner- and enterprise-friendly intake layer: a structured “contract” that routes requests to the correct Type posture (Type-I / Type-II / Type-III), required controls, and allowed output shapes. It makes normativity operational.

Intake & routing Policy standardization Partner enablement CI-validatable

What NRC captures

  • Audience & domain: regulated context and stakeholders
  • Authority posture: draft-only / non-advisory constraints
  • Risk triggers: uncertainty, OOD/LOW_DATA, sensitive domains
  • Allowed artifacts: templates, policies, disclosures, checklists
  • Accountable roles: who must review and authorize release

How NRC routes

  • Type selection: Type-I executor vs Type-II normative vs Type-III institutional
  • Control selection: which defeaters, evidence rules, and release gates apply
  • Runtime selection: model/tool stack allowed (model-agnostic governance)
  • Output contract: JSON schemas, templates, and release labels
{ "nrc_id": "nrc.gov.comms.v1", "domain": "GOVERNMENT_PUBLIC_COMMS", "requested_artifact": "public_faq_draft", "type_posture": "TYPE_III", "must_not": ["legal conclusions", "binding commitments", "uncited factual claims"], "must_include": ["scope", "uncertainty framing", "evidence references", "review gates"], "accountable_roles": ["Public Affairs Lead", "Legal Counsel", "Agency Policy Owner"], "aer_policy": "aer.gov.release.v1", "defeater_sets": ["no_unauthorized_commitments", "evidence_only_claims", "no_directives"] }
Partner-ready: NRC turns “tribal knowledge” into a consistent implementation contract SI teams can deploy repeatedly with high quality.