Mechanism

The Adaptive Consensus Protocol (ACP) is the consensus layer of the memory-native protocol stack. It is a memory-native mechanism that enables distributed nodes to evaluate mutation proposals carried by agents without relying on centralized coordination or globally synchronized state. Unlike traditional consensus systems that require fixed validator sets or persistent governance registries, ACP dynamically scopes quorum eligibility using policy references embedded within an agent's memory field. Each node independently determines its own eligibility, voting weight, and policy alignment using only the information carried by the agent.

The consensus process begins when an agent proposes a mutation, such as adding an alias to a semantic index, encoded within its payload and described in its memory field. The memory field includes a policy reference, a lineage trace documenting the agent's origin, and a quorum descriptor indicating the trust-weighted voting structure required for approval. These values together define the eligibility criteria, the weighting logic, and the approval thresholds for the proposed mutation. The proposal therefore carries its own governance rules rather than depending on rules held by the nodes that evaluate it.

Eligibility and Policy Resolution

Upon receiving the agent, a node verifies the cryptographic signature generated over the agent's UID, memory field, and payload. The node validates this signature against the sender's public key and inspects the memory field to resolve the referenced policy agent. The policy agent is itself a memory-bearing agent whose primary purpose is to encode governance logic: it defines quorum logic including eligibility roles, voting structure, and weighting parameters. If the node qualifies under the referenced policy, it participates by evaluating its trust path to the proposing agent and determining alignment with the embedded governance constraints.

Eligibility as a consensus node is dynamic and scoped to the agent's transport header, policy references, and trust domain. It does not require persistent identity, fixed validator roles, or a global registry. A consensus node is any network node that is eligible to participate in, initiate, or validate quorum decisions under ACP based on information contained within the memory field of an agent. Because the policy reference travels with the proposal, the set of nodes that may vote on a given mutation is determined by that proposal rather than fixed at protocol design time.

Trust-Weighted Voting

Each participating node submits its vote as a new agent containing a reference to the original mutation proposal, the node's trust score, and justification metadata. Votes are weighted according to domain scope, trust profile, and policy-defined metrics, then aggregated using the quorum logic contained within the original agent's memory field. As an illustrative example, a mutation may require a minimum of three votes out of five, with a cumulative trust weight exceeding a defined threshold, for approval to occur. Votes may be accumulated locally or forwarded via the dynamic routing protocol (DRP) to additional quorum participants.

The vote weighting draws on the trust graph: the evolving, memory-informed model maintained by nodes that maps prior interaction outcomes to trust scores used in routing and quorum weighting. A node with a record of successful prior execution carries elevated trust weighting, while nodes associated with policy rejections or instability carry less. Because each vote is itself an agent, the votes participate in the same signed, memory-bearing structure as the proposal they evaluate, and their justification metadata remains available to later readers.

Outcome and Memory Trace

If quorum is reached, the ACP module appends an approval entry to the originating agent's memory trace, documenting the voting outcomes and embedding quorum context for downstream audit and validation. If quorum fails or a proposal is rejected, a rejection or quarantine flag is appended instead. In each case, the agent's memory field becomes a complete, cryptographically verifiable execution trace of the consensus process, enabling subsequent nodes to confirm compliance with scoped policy constraints and trust rules.

This recording is a property of the memory field's structure. Each memory trace is independently signed by the node that generated it and chained using cryptographic hashes, ensuring chronological ordering, auditability, and non-repudiation. The consensus history of a mutation is therefore interleaved with the substantive history of the agent it concerns, so the outcome that a later node reads is the same outcome that was decided, not a separately maintained projection of it.

Stateless and Memory-Aware Modes

ACP supports both stateless and memory-aware modes of operation. In stateless mode, quorum logic is determined exclusively by the agent's memory field and the active policy at runtime; nodes do not persist external memory between agent evaluations, and all consensus decisions are made using the data embedded in the received agent. This allows devices with limited resources or transient uptime, including IoT devices, ephemeral containers, or anonymized relays, to participate without full-stack deployment or long-term data retention.

In memory-aware deployments, nodes may also incorporate historical mutation outcomes, trust scores, or policy participation records to inform quorum formation, weighting decisions, or eligibility forecasting. This optional historical context provides enhanced interpretability while preserving fully localized decision-making. In both modes the decision is scoped to the identity, memory context, and mutation parameters of a single agent, without requiring persistent governance hierarchies, external registrars, or global alias systems.

Health-Driven Adaptation

The adaptive character of ACP arises from its interaction with the network health monitoring system (NHMS). NHMS nodes emit signed health agents carrying operational signals such as congestion metrics, trust volatility, semantic entropy, and cache degradation data. At the consensus layer, trust volatility indicated by health-state reports may modify quorum eligibility, weight assignments, or required thresholds, preventing unstable nodes from disproportionately affecting mutation events.

A health agent received by a node may cause that node to execute one or more adjustments to the parameters of ACP for one or more semantic classes. The adjustments include raising or lowering quorum thresholds, excusing or reinstating specific participants from quorum eligibility, and re-weighting participant votes. These adjustments are governed by local policy and the trust-scoped feedback the health agent carries, so the consensus behavior of a region of the network adapts to real-time conditions without centralized synchronization or a global dashboard.

Composition Within the Stack

ACP is one of four layers the protocol stack may include, alongside the dynamic routing protocol, the dynamic indexing protocol (DIP), and the semantic memory layer. Each layer operates exclusively on agent-resident data and appends a corresponding execution trace to the memory field. ACP consumes the lineage trace and policy references that the memory field already carries for routing and indexing, so consensus is not a separate subsystem with its own state but a layer that reads the same memory-native object the other layers read.

The relationship with DIP is direct: when DIP detects entropy divergence and generates a local mutation proposal to split, merge, or reclassify an index, that proposal may, if ACP is present, undergo scoped quorum validation before it takes effect; otherwise DIP executes autonomously under policy constraints. Above the stack, cognition-layer mutation proposals such as goal updates or inference-graph revisions are validated by ACP only for authorization under policy constraints, quorum rules, and trust-scoped governance. The protocol stack does not inspect or interpret the internal semantics of such proposals.

Prior-Art Distinctions

Traditional consensus systems require fixed validator sets or persistent governance registries. ACP differs in that quorum eligibility is scoped per proposal by the policy reference embedded in the agent's memory field, and each node determines its own eligibility, voting weight, and policy alignment from the agent alone. There is no global validator registry and no fixed quorum threshold imposed at protocol design time; the quorum descriptor and policy agent that travel with the proposal define those rules.

ACP makes the consensus decision and writes the record into the same append-only memory field in a single mechanism, so the recorded outcome is structurally identical to the decided outcome. The combination of policy-referenced eligibility, trust-weighted voting aggregated by quorum logic carried in the agent, health-driven adjustment of quorum thresholds and eligibility, and signed hash-chained recording of the outcome distinguishes ACP from consensus systems that require fixed validator sets or persistent governance registries.

Disclosure Scope

The adaptive consensus protocol described here, comprising the policy-referenced scoping of quorum eligibility from an agent's memory field, the resolution of a referenced policy agent that defines eligibility roles, voting structure, and weighting parameters, the casting of votes as new agents carrying a trust score and justification metadata, the aggregation of trust-weighted votes under the quorum logic carried in the proposing agent's memory field, the appending of approval, rejection, or quarantine outcomes to the agent's cryptographically chained memory trace, the stateless and memory-aware modes of operation, and the adjustment of quorum thresholds, eligibility, and vote weights in response to health agents, is disclosed in U.S. Application No. 19/366,760. This article describes that disclosed mechanism.

The scope encompasses deployments ranging from minimal stateless edge nodes to full-stack core infrastructure nodes, and federated semantic zones that coordinate mutation validation across trust-divergent boundaries without shared infrastructure or synchronized ledgers. Implementations differing in the specific trust signals consumed, the specific weighting metrics applied, or the specific quorum descriptors used remain within the disclosure scope provided that policy-referenced eligibility, trust-weighted voting, and memory-resident recording of the outcome are present together, evaluated against memory-bearing agents whose fields the protocol consumes directly.