Trust-Weighted Voting in ACP: Domain-Scoped Votes Accumulated Against Agent Memory

Mechanism

Trust-weighted voting under ACP proceeds in four interlocked phases that operate as a single structural primitive at the transport layer. The first phase is domain scoping. Every proposal that requires consensus is annotated, at the moment of construction, with the domain of authority to which it belongs. Domains are not free-form labels; they are addresses within a hierarchical authority namespace that travels with the agent memory of each participant. A proposal addressing the routing of cryptographic key material is scoped to a different domain than a proposal addressing the admission of a new peer, and a participant's vote in one domain carries no implicit weight in the other. Domain scoping prevents the leakage of authority across functional boundaries that pervades reputation systems built on a single global score.

The second phase is weight derivation. When a participant casts a vote, that vote is not counted as a unit. It is multiplied by the trust weight that the participant holds within the relevant domain at the moment of the vote. The weight is computed deterministically from the participant's accumulated history of decisions in that domain, where each historical decision contributes a signed quantity reflecting the outcome of the decision as later verified against the lineage record. A participant whose past decisions in a domain have produced verifiable, accepted outcomes accrues weight in that domain. A participant whose past decisions have been overridden, contradicted by subsequent evidence, or implicated in a tamper event loses weight in that domain. The accumulation function is monotonic in good outcomes and non-monotonic in adverse outcomes, with the precise schedule encoded in the protocol rather than left to local interpretation.

The third phase is accumulation against quorum logic. Each domain carries, in the agent memory of every participant, a quorum function that defines the threshold of accumulated weight required for a proposal in that domain to be considered ratified. The quorum function is itself a structural element of the protocol; it is signed, lineage-tracked, and cannot be altered except through a meta-proposal that itself reaches quorum under the prior quorum function. As votes arrive, each participant independently accumulates the weighted sum and evaluates it against the quorum function. Because the inputs (the vote, the weight at time of vote, the quorum function) are all carried in agent memory and cryptographically committed, every participant arrives at the same accumulation result without needing to consult a central tabulator.

The fourth phase is explicit authority rotation. ACP does not permit indefinite accumulation of weight by any single participant. Each domain encodes a rotation schedule that decays the weight of long-standing participants, redistributes a portion of accumulated weight to newer participants whose recent decisions have been verified, and forces periodic reconfirmation of authority through fresh interaction. Rotation is explicit in two senses: it is recorded in the lineage as a discrete event with its own timestamp and signature, and it is bounded by parameters that any participant can audit. There is no implicit drift of authority and no hidden incumbency advantage.

Tamper evidence threads through all four phases. Every vote, every weight computation, every quorum evaluation, and every rotation event is appended to a hash-chained lineage structure that travels with the agents themselves. Any attempt to alter a historical vote, retroactively adjust a weight, or rewrite a rotation event invalidates the chain at the point of tampering and is detected by any participant who replays the lineage from a prior known-good state. The protocol guarantees not merely that tampering is difficult but that tampering, if attempted, leaves a permanent visible scar on the record.

Operating Parameters

The protocol exposes a small set of parameters that govern its quantitative behavior in deployment. The domain namespace depth controls how finely authority can be subdivided; shallow namespaces produce coarse-grained weight aggregation while deep namespaces permit precise scoping at the cost of larger memory footprints in each agent. The weight accumulation rate, expressed as the marginal weight contribution per verified successful decision, governs how quickly a new participant can earn meaningful authority in a domain. The decay coefficient applied during rotation determines the rate at which accumulated weight is redistributed. The quorum threshold, expressed as a fraction of the total accumulated weight in the domain, determines how broad an agreement must be to ratify a proposal.

Each parameter has a defined operating range and a default within that range. Domain depth is typically bounded between three and seven levels in production deployments; deeper hierarchies introduce coordination overhead that outweighs the benefit of finer scoping. Accumulation rates are calibrated so that a participant acting in good faith reaches meaningful weight after on the order of tens of verified decisions, not hundreds; the calibration reflects the practical cadence of decision-making in agent populations. Decay coefficients are typically set so that a participant who ceases to act in a domain loses material weight within a small number of rotation cycles, ensuring that dormant authority does not accumulate. Quorum thresholds are domain-specific; high-stakes domains carry thresholds approaching unanimity, while low-stakes domains operate at simple majority. The protocol enforces consistency of these parameters across all participants in a domain through the same lineage mechanism that enforces vote integrity.

A further parameter governs the verification window during which a decision's outcome is treated as still subject to revision. Within the window, an outcome is provisional and contributes a reduced increment to the deciding participant's weight; after the window closes without contradicting evidence, the contribution is promoted to its full value. The window length is calibrated against the characteristic latency at which contradicting evidence might surface in the relevant domain, and it is the parameter most often tuned during deployment as operators learn the empirical distribution of late-arriving evidence in their environment. The calibration is itself recorded in the lineage, so that historical weights remain auditable against the parameters that were in force at the time of accumulation.

The protocol also exposes parameters governing the introduction of new participants. A bootstrap weight floor establishes the minimum trust weight assigned to a participant on entry to a domain, ensuring that newcomers are not entirely silenced by the accumulated weight of incumbents. A bootstrap horizon defines the period during which the floor is in effect; after the horizon, the participant's weight is governed exclusively by accumulated decision history. The bootstrap parameters are bounded so that the floor cannot be set high enough to permit a coordinated influx of new identities to overwhelm the accumulated authority of the existing population, defeating Sybil-style attacks against open domains.

Alternative Embodiments

The disclosed mechanism admits several alternative embodiments that preserve its structural properties while adapting to different deployment contexts. In one embodiment, the weight accumulation function is replaced with a piecewise linear schedule that introduces step changes at defined thresholds, producing a discretized authority structure suited to environments where graduated trust is preferred over continuous trust. In a second embodiment, the rotation mechanism operates on a sliding window rather than discrete cycles, smoothing weight redistribution over time. In a third embodiment, the quorum function is replaced with a multi-quorum function that requires simultaneous satisfaction of distinct thresholds drawn from non-overlapping participant subsets, defeating Sybil attacks that might saturate a single subset. In a fourth embodiment, domain scoping is extended with cross-domain coupling coefficients that allow weight in one domain to contribute, at a defined fractional rate, to votes in a related domain, supporting hierarchical authority structures without the brittleness of strict separation. Each embodiment is consistent with the disclosed mechanism and falls within the scope of the disclosure.

Composition With Other Mechanisms

Trust-weighted voting under ACP composes with the broader memory-native protocol stack in well-defined ways. It interoperates with the lineage-bound message format described elsewhere in the disclosure, using the same hash-chained structure to record vote events that the message format uses to record exchanges. It interoperates with the substrate-agnostic execution layer by ensuring that vote and weight state migrate with the agent across substrate boundaries, preserving authority continuity through host transitions. It interoperates with the negative-capability declaration mechanism by treating a participant's declaration that it cannot act in a domain as a structural exclusion from voting in that domain rather than as a mere preference. The composition is closed under the operations of the protocol; the output of any combination of these mechanisms remains a valid memory-native protocol state.

Distinction From Prior Art

Prior consensus mechanisms can be partitioned into three broad classes, each of which differs structurally from the disclosed mechanism. Stake-weighted protocols such as those used in proof-of-stake blockchains weight votes by externally posted economic collateral; authority is purchased rather than earned, and domain scoping is absent or vestigial. Reputation systems such as those used in peer review and recommendation networks compute a single global reputation score and apply it uniformly across all decisions; domain scoping is absent and authority can be transferred from low-stakes to high-stakes contexts without friction. Permissioned consensus systems such as Raft and Paxos assume a fixed roster of equally weighted participants and address only the problem of agreement, not the problem of governance.

The disclosed mechanism differs from all three classes in that weights are earned through verified decisions, scoped to domains, decayed through explicit rotation, and tamper-evident through lineage. No prior mechanism combines these four properties into a single transport-layer primitive. The combination is what enables the protocol to provide structural governance for autonomous agent populations that are not bounded by a fixed roster, not committed to a single global authority, and not protected by external policy enforcement.

Beyond the broad classes above, several specific prior systems warrant individual distinction. Federated identity protocols such as those derived from SAML or OAuth carry attestations of authority but do not accumulate weight from the outcome history of decisions made under those attestations; an authority once issued is exercised at full force until revoked, with no graduated track record. Distributed ledger consensus mechanisms such as Tendermint and HotStuff achieve agreement among a known validator set but do not internalize the concept of domain-scoped authority, treating all blocks under all topics as governed by a single weighted vote. Web-of-trust constructions such as the PGP keyring permit participants to express graded trust in one another but do not bind those expressions to verified outcomes and do not provide a structural primitive for rotation. The disclosed mechanism is distinguishable from each by reference to its specific structural commitments.

Disclosure Scope

The disclosure encompasses the protocol logic described above, the data structures that carry vote, weight, quorum, and rotation state through agent memory, the computational methods for deriving weights from lineage, the methods for evaluating quorum against accumulated weight, and the methods for executing rotation. The scope extends to embodiments in which any of these elements is implemented in hardware, in software, or in a combination, and to embodiments deployed in centralized, federated, decentralized, and edge topologies. The scope does not depend on a particular cryptographic primitive, a particular network transport, or a particular agent runtime; the structural properties of the mechanism are preserved across substitutions of these underlying components, provided that the substitutions preserve the integrity guarantees on which the lineage chain relies.