Architecture Overview — Quantum Inquiry

DDRP + CAAP

A two-layer accountability architecture. DDRP extracts what obligations exist. CAAP proves who acted on them and under what authority. Neither layer modifies the other.

Two-Layer Architecture Cryptographic Separation Domain-Agnostic Open Protocol

The Architecture

Most compliance systems answer one question: what is the current state of this obligation? DDRP and CAAP answer two different questions, at two different points in the lifecycle of a professional obligation, using two architecturally isolated layers.

DDRP answers

What obligation-creating language exists, was it detectable at production, and was each obligation structurally resolved or left open?

CAAP answers

Who acted on the obligations in this artifact, when, under what authority, and against which cryptographically verified version of the record?

Layer B
CAAP — Contextual Accountability Attribution Protocol
Attribution — append-only event log with required actor, timestamp, and authority basis per event. Hash-bound to specific artifact versions. No modification. No deletion.
Interface
DDRP Artifact Interface Contract v1.0
SHA-256 hash handoff only · CAAP has zero write access to DDRP · Mandatory schema validation · No partial acceptance · One-way data flow
Layer A
DDRP — Deterministic Document Review Protocol
Extraction — deterministic, hash-stable, immutable artifact. Lexical detection only. No AI in the verification layer. Unaware of CAAP. Produces artifacts and stops.

The Interface Contract

The interface contract is the only formal connection between the two layers. It defines the JSON schema a DDRP artifact must satisfy before CAAP will accept it. An artifact that fails validation is rejected entirely — no partial acceptance.

FieldPurpose
artifact_schema_versionStructural compatibility. CAAP rejects unsupported versions without coercion.
artifact_hashSHA-256 of canonical artifact payload. CAAP recomputes and must match.
document_idStable identifier of the source document.
extraction_timestampUTC timestamp of when the DDRP extraction occurred.
canon_rules_versionVersion of the lexical extraction rules used.
obligations[]Array of obligation objects produced by DDRP. May be empty.
absent_fields[]Explicitly recorded structural elements not found. May be empty.
A hash mismatch is treated as evidence of mutation, not a recoverable error. Any change to artifact content produces a different hash and therefore a different artifact identity. Events logged against the previous artifact remain permanently associated with that version.

Why the Separation Is Enforced

When extraction and attribution share one mutable system, a structural vulnerability emerges: the record of what obligations existed and the record of what was done about them can be retroactively aligned. The distinction between observation and interpretation collapses. The audit trail becomes a narrative document rather than an evidentiary one.

The commercial stack is built for workflow. DDRP + CAAP is built for proof.

DDRP produces the record. CAAP references it by hash only. Neither can modify the other. The boundary between what was found and what was done is cryptographically enforced — not by convention, not by policy.

The Policy Connection

FedRAMP RFC-0024 (January 2026) mandates that federal cloud providers transition from human-written compliance narratives to machine-generated deterministic telemetry. The RFC explicitly prohibits generative AI outputs as compliance evidence.

The architectural parallel: FedRAMP/OSCAL implements deterministic telemetry for federal cloud security. DDRP + CAAP implements the same principle — deterministic extraction, immutable artifact, hash-bound action attribution — as an open protocol applicable to any professional document environment.

The epistemological position is identical: compliance must be demonstrated by a verifiable, reproducible record, not a written assertion. The difference is scope. OSCAL applies to one domain. DDRP + CAAP applies wherever professional obligations under document require an auditable evidentiary chain.

Where the Stack Sits

Three commercial categories address portions of this problem space. None implements both layers with architectural separation.

Property DDRP + CAAP CLM Platforms GRC Platforms FedRAMP / OSCAL
Deterministic obligation extraction YESNO (AI/ML)NO (manual)YES (telemetry)
Hash-stable immutable artifact YESNONOPARTIAL
Layer separation (extraction / attribution) YESNONONO
Append-only event log (structural) YESNONONO
Authority basis required per action YESNONONO
Hash-bound action to artifact version YESNONOPARTIAL
No probabilistic inference in verification YESNONOYES (mandated)
Domain agnostic YESYESYESNO
Open non-commercial protocol YESNONONO

Every commercial system in this analysis can produce an audit log. None can produce an audit log in which every entry is cryptographically bound to the specific version of the document from which obligations were deterministically extracted, with a required authority declaration per action, in a structurally append-only form. That is the gap this stack fills.

The EU AI Act Connection

The documentation infrastructure the EU AI Act requires — without AI.

Neither DDRP nor CAAP is an AI system. Neither is subject to the EU AI Act. That is the point. Organizations deploying high-risk AI systems under the Act need an accountability layer that is deterministic, human-reviewable, and not itself subject to the probabilistic failure modes the Act is designed to govern.

The stack maps directly to the Act's documentary obligations:

DDRP covers

  • Article 11 — Technical documentation
  • Article 12 — Record-keeping and logging
  • Article 13 — Transparency and traceability
  • Article 17 — Quality management systems

CAAP covers

  • Article 17 — Accountability and authority chains
  • Article 19 — Post-market monitoring
  • Article 20 — Serious incident reporting basis
  • Article 21 — Cooperation with authorities

The Documentary Accountability Substrate (DAS) is the governance framework that formalizes this mapping — introduced in the Routledge volume on agentic AI governance and published on Zenodo (DOI: 10.5281/zenodo.19369623).

Open Source

All three protocols are published as open-source repositories. Non-commercial. Freely available.