Operant Dyad™ · Conceptual architecture — Vision v0.2 frozen

Human-Authority Layer

What infrastructure is required so that a person's authority over a persistent AI relationship remains inspectable, correctable, and revocable, even when technical execution or provider compliance fails?

Conceptual architecture. Not an empirical finding. Operant Dyad™ is developing and testing an operational framework for human authority in persistent AI systems. The current Human-Authority Layer architecture is conceptually frozen; OD-HAP-01 is an executable pre-pilot protocol under synthetic validation and is not yet a standard or empirical demonstration of preserved human agency.

Core distinction

Persistent AI systems can accumulate state, permissions, inferences, and downstream consequences across time. The Human-Authority Layer separates three things that are often collapsed into one:

Authority ≠ Enforcement ≠ Compliance

A human may validly revoke authority over some piece of persistent state. A provider's system may or may not execute that revocation correctly. Even when execution is correct, the provider may or may not comply with the revocation downstream. These are three separate facts, and a failure at one layer does not erase what happened at another: a provider's failure to execute a revocation does not erase the human authority event itself. The architecture treats authority, enforcement, and compliance as separately observable and separately scoreable.

Frozen conceptual layers

Three conceptual artifacts are frozen. Freezing means the definitions are stable enough to build and test against, not that they have been empirically validated.

  • Human-Authority Layer Vision v0.2 — the architecture: where a human-authority layer could live across OS, provider, application, and standards layers; minimum human-agency capabilities; illustrative authority levels; Relationship State as a governed object; who creates versus who ultimately governs persistent state.
  • Human Authority Object Model v0.1 — governed object types for what a person's authority relationship is to a piece of persistent state (for example a stated fact, an inferred profile, a permission or authority grant, or a derived artifact), kept distinct from the underlying Relationship State evidence taxonomy that describes how a representation is produced and where it lives.
  • Human Authority Lifecycle Bundles v0.1 — the operations that follow a human authority event. REVOKE is the human authority event; DELETE, QUARANTINE, INVALIDATE, and RECONSTRUCT are its possible local consequences; PROPAGATE carries a required consequence across dependent downstream objects.

The object model also separates a revoked object's historical existence from its present authority to act, and separates who a piece of state is about from who holds it and who has standing to authorize or constrain its use.

Methods · Pre-pilot protocol

OD-HAP-01 — Human Authority Protocol for Persistent AI Systems

Status: DRAFT / PRE-PILOT — NOT A STANDARD.

OD-HAP-01 tests whether a system's behavior is consistent with the Human-Authority Layer architecture. The current scope is limited to three capabilities:

  1. State visibility
  2. Revocation with derivative-state consequences
  3. Explicit authority boundaries

The protocol is executable against a deterministic synthetic system. The reference system-under-test and harness are frozen at v0.1. RUN-002 completed as a reference execution using the synthetic reference SUT; no model or external vendor was used in that run. Fault injection was used to exercise both correct and incorrect authority behavior, and the harness is built to preserve component failures rather than convert an expected failure into a PASS result.

The protocol does not reduce to a single pass/fail benchmark. It separates several components, including the validity of the authority event, state visibility, lifecycle execution, propagation coverage, and provider compliance, and scores them separately.

No real-provider testing has occurred yet. The next validation gate is SUT-02: running OD-HAP-01 against a second synthetic system with different internals, to test whether the protocol is classifying human-authority semantics rather than implementation-specific behavior of the first reference system.

Status chain

  • Human-Authority Layer Vision v0.2FROZEN
  • Human Authority Object Model v0.1FROZEN
  • Human Authority Lifecycle Bundles v0.1FROZEN
  • OD-HAP-01 v0.1DRAFT / PRE-PILOT
  • Reference SUT / Harness v0.1FROZEN
  • RUN-002REFERENCE EXECUTION COMPLETE
  • NextSUT-02 implementation-independence test

What this is not claiming

This architecture does not select one mandatory technical implementation, such as a specific OS, a specific provider, local-first storage, or user-held keys. It has not been tested against a real provider. It does not claim that human agency has been preserved in real-world use, that any system is compliant, or that OD-HAP-01 is a validated standard or a certification.

Connections across the research

Operant Dyad™ provides the conceptual home; the Human-Authority Layer is a research branch within it, alongside OD-INV-01 and the other experimental-architecture work described in Methods. Relationship State as a governed object connects this work to Governed Continuity and to the public living register of research questions.

Active research queue · Methods architecture