How They Compose

Independence

Each protocol works alone. You can deploy ACAP without Phala, NERVE without PACE, or any single protocol in isolation. The primitives are self-contained and the capabilities.extensions entries are independent.

Composition

When multiple protocols are deployed on the same AgentCard, they produce compound guarantees:

PACE + ACAP (Accessibility + Consent)

Consent becomes accessibility-conditional. Before creating a ConsentRecord (ACAP), the system performs a ConsentCapacityCheck (PACE) to verify the principal can meaningfully consent right now. Consent given during sundowning hours or without the principal’s language support is flagged as non-compliant.

PACE + Phala (Accessibility + Welfare)

Satisfaction measurement becomes capability-aware. Phala’s implicit signals (response time, engagement) are normalized against the principal’s declared capabilities. A 200-second response from a principal with response_timeout: 300s is fast, not slow.

NERVE + ACAP (Integrity + Consent)

Trust-triggered re-consent. When NERVE’s MicroglialObserver drops an agent’s trust_score below 0.4, any active consent records for that agent are invalidated. The agent must re-establish consent before acting on the principal’s behalf.

NERVE + Phala (Integrity + Welfare)

Outcome valence drives myelination. Phala’s positive/negative valence signals strengthen or weaken NERVE’s SynapticChannel myelination, creating a feedback loop where well-performing pathways become faster and poorly-performing ones are attenuated.

All Phases Together

The full phase-protocol lifecycle:

1. PACE checks: can this principal participate?
2. ACAP checks: has the caller honored the callee’s terms?
3. NERVE checks: is this agent still trustworthy?
4. Phala measures: did the action serve the principal?

Each phase feeds the next. Accessibility gates consent. Consent gates action. Integrity monitors the action. Welfare evaluates the result. The result feeds back into trust (NERVE myelination) and future routing (Phala BeliefUpdate), closing the loop.

ACAP Extensions

ACAP carries reference extensions that layer specific deployment concerns onto the Core chain without modifying it: governance tiering, category preferences, regulatory context, and audit projection. The extensions compose in turn: category preferences defines the (category, dimension) grid, regulatory context enforces jurisdictional floors on the same grid, governance tiering consults the effective floor when classifying a re-consent diff, and audit projection renders the full trail into a regulator-facing report.

Pramana across all phases

Pramana is cross-cutting rather than phase-bound. Every consequential output produced at any phase (a PACE accessibility decision, an ACAP consent finding, a NERVE trust assessment, a Phala satisfaction event) can itself be wrapped in a typed ClaimAttestation so the source and procedure that warranted the output survive as a re-verifiable record. The phase protocols are not changed; the attestation is layered on each emitted claim.

Abhyasa under all phases

Abhyasa is the delivery layer, cross-cutting like Pramana. The governance obligations the phase protocols produce — an ACAP consent revocation, a Phala corrective BeliefUpdate whose loss would leave an agent acting on withdrawn authority or a rejected preference — are carried under custody with a deliver-or-report guarantee: each is applied, explicitly declined, or escalated to the principal, never silently lost over a lossy or partitioned channel. The phase protocols are unchanged; Abhyasa wraps the propagation of their asymmetric-cost obligations and adds a principal-side fail-safe for when the channel defeats delivery.

On the AgentCard

All these protocols declare themselves in the same capabilities.extensions array. See For A2A Developers for the full AgentCard example.