Cross-Mesh Divergence Detector

Mechanism

Each participating mesh M_i maintains its own credentialed-observation chain in the five-property form described in the parent provisional. The detector subscribes, for each mesh, to a designated comparison stream: a projection of observations that the mesh asserts can be cross-referenced against peers. Comparison streams are typed: a "physical-event" stream carries observations whose payloads encode the same observed phenomenon (e.g., a measured RF emission at a given location and time); a "reconciled-time" stream carries observations binding mesh-local clocks to a shared reference; an "authority-assertion" stream carries observations of the form "authority A asserts X" where A is recognized in multiple meshes.

The detector pairs observations across streams using a correlation key — typically a tuple of (event-type, declared-time-window, declared-spatial-window, authority-class). For a correlation set S = {o_1, …, o_k} of paired observations from meshes {M_1, …, M_k}, the detector applies a stream-typed divergence function δ(S) returning a scalar (or vector) deviation. For physical-event streams, δ may be the maximum pairwise residual of the measured quantity. For time streams, δ is typically the maximum pairwise clock skew. For authority-assertion streams, δ may be a discrete indicator (assertions agree or disagree on a categorical attribute).

When δ(S) exceeds a declared threshold τ — itself a credentialed observation issued by the monitoring authority and bound to the comparison stream — the detector emits a divergence event. The event is a credentialed observation referencing every contributing observation in S as predecessor, carrying the computed δ, the active τ, the inferred divergence pattern (e.g., monotonic drift, abrupt step, intermittent disagreement), and a monitoring-authority signature. Downstream reconciliation workflows admit the event against their admissibility predicates and proceed under lineage-bound merge.

Operating Parameters

A reference deployment monitoring two to eight meshes operates with comparison-stream throughput in the range 1,000–50,000 paired correlation sets per second per stream, depending on stream type and event rate. Detection latency from the arrival of the last contributing observation in a correlation set to event emission is bounded at 50–500 ms wall-clock for 95th-percentile traffic, dominated by correlation-key indexing rather than by δ computation.

Threshold parameters τ are declared per stream and per authority. Physical-event thresholds are typically expressed in units native to the observed quantity (e.g., ±1.5 dB for RF power, ±50 m for position). Time thresholds are typically in the range 100 µs – 10 ms depending on whether the consuming workflow is real-time control or post-hoc audit. Authority-assertion thresholds are typically binary (zero tolerance for categorical disagreement) but may admit weighted disagreement when authorities assert with declared confidence.

Divergence-pattern classification operates on a sliding window of recent divergence events for the same correlation key. Default window sizes are 32–256 events. Classifier outputs include "noise" (within statistical envelope of expected jitter), "drift" (monotonic trend), "step" (abrupt level change), "intermittent" (sporadic exceedances), and "structural" (sustained disagreement). The classification is itself recorded in the divergence event so that downstream reconciliation can dispatch on pattern.

Alternative Embodiments

The detector may be deployed as a centralized service consuming all comparison streams, as a peer-distributed service in which each mesh runs a local detector against its peers' streams, or as a hierarchical detector in which leaf detectors handle bilateral comparisons and a root detector aggregates higher-order divergences. The disclosed mechanism is independent of deployment topology so long as every emitted divergence event is itself a credentialed observation referencing its contributing observations as predecessors.

Correlation-key construction may be exact (observations are paired only when correlation keys are bit-identical) or fuzzy (observations are paired when correlation keys fall within a declared similarity neighborhood). Fuzzy correlation enables comparison of observations made under independently quantized spatial or temporal windows, with the similarity neighborhood itself being a credentialed parameter.

Divergence functions δ may be configured per-stream from a library of canonical functions (max residual, root-mean-square residual, Kullback–Leibler divergence over discrete distributions, edit distance over structured payloads) or supplied as monitoring-authority-signed custom functions. Custom δ functions enter the chain as credentialed observations and are subject to the same predecessor referencing as any other parameter.

Adversarial-resilient embodiments employ byzantine-robust pairing in which a divergence event requires corroboration from a quorum of monitoring authorities before it is admitted as a reconciliation trigger. Such embodiments protect against false-divergence injection at the cost of increased detection latency.

Composition With Other Primitives

Divergence detection composes natively with no-consensus federation: the parent provisional describes federations in which meshes do not seek a single agreed state, and divergence events are precisely the structural surface across which such federations expose their disagreements. Composition with lineage-bound merge enables reconciliation workflows to merge divergent state along the lineage path of contributing observations, preserving authority-attributed history rather than collapsing to a single resolved value.

Composition with cross-jurisdictional governance allows divergence thresholds to be jurisdiction-specific; the same nominal disagreement may be admissible in one jurisdiction's mesh and a reportable event in another. Composition with the dispute primitive allows any contributing mesh to challenge a divergence event by filing a credentialed counter-observation; challenges enter the chain and may suspend or reverse the triggered reconciliation.

Composition with composite licensing intersection enables divergence events themselves to be subject to license envelopes — for example, when a divergence event's payload contains classified observation residuals, downstream readers must satisfy the relevant scope envelope before the event is admitted to their workflow. Composition with cross-marketplace composition enables divergence detection to operate over comparison streams that span commodity-class marketplaces: a discrepancy in reported settlement state between two participating marketplaces is itself a divergence event under the same mechanism, enabling reconciliation of marketplace state under lineage-bound merge.

Distinction From Prior Art

Consensus-divergence detection — the family of techniques used in distributed databases and blockchain networks to identify when a quorum has failed to converge — presupposes a consensus-seeking architecture. The disclosed federation, by design, does not seek consensus; meshes are autonomous and divergence is an expected operating regime, not a failure mode. Consensus-divergence detectors emit alerts to recovery procedures aimed at restoring a single agreed state; the disclosed detector emits credentialed events that are themselves first-class artifacts in a chain that preserves disagreement.

CRDT (conflict-free replicated data type) conflict detection silently merges divergent state under commutative operators; conflicts, where they arise at all, are resolved by data-type semantics without producing an evidentiary record. The disclosed mechanism is, by contrast, evidence-producing by construction: every divergence event is a credentialed observation whose chain position and predecessor references are auditable independently of any merge outcome. Furthermore, the disclosed mechanism applies to observation streams that are not commutative in any meaningful sense — RF measurements, authority assertions, time references — and would not benefit from CRDT semantics even if they could be coerced.

Conventional cross-system anomaly detectors typically consume unsigned telemetry and produce alert records with no cryptographic chain to the underlying observations. The disclosed mechanism is distinguished by its operation strictly over signed observations, by emitting only credentialed events, and by the explicit composition with no-consensus federation and lineage-bound merge.

Worked Example

Consider two coalition-partner meshes, M_alpha operated by a national defense ministry and M_beta operated by a civilian aviation regulator, both observing the RF environment over a shared border region. Each mesh independently captures observations of detected emissions: a measured power, a measured center frequency, a measured time of arrival, and a measured bearing. The detector subscribes to a "physical-event" comparison stream from each mesh and pairs observations whose declared spatial windows overlap and whose declared time windows lie within 200 ms. For a pair (o_alpha, o_beta) the detector computes δ as the maximum residual across power (dB), frequency (kHz), and bearing (degrees). The active threshold τ for this stream is a credentialed observation declaring (1.5 dB, 5 kHz, 2°). When a paired set produces δ exceeding τ on the bearing component while remaining within τ on power and frequency, the detector emits a divergence event tagged "intermittent" pattern and "bearing-only" classification, signed by the joint monitoring authority. A reconciliation workflow downstream of the event determines that one mesh's antenna calibration drifted and triggers a lineage-bound merge in which both observations are retained, with bearing fields annotated by the calibration-correction credential rather than overwritten.

Crucially, the operational outcome differs from a consensus-divergence handler, which would treat the disagreement as a fault to be resolved by selecting one mesh's reading and discarding the other, and from a CRDT handler, which has no semantically meaningful merge for bearing-of-emission observations. Under the disclosed mechanism, the historical record retains both readings, the credentialed calibration correction explains their difference, and any downstream consumer (a coalition operations cell, an air-traffic safety review) can re-derive the corrected bearing or interrogate the original readings independently of any single mesh operator.

Disclosure Scope

This disclosure supports claims directed to: a method for detecting state divergence among a plurality of federated meshes by comparing signed observations across typed comparison streams under correlation-key pairing; a system in which divergence events are emitted as credentialed observations referencing contributing observations as predecessors and carrying a monitoring-authority signature; a method for classifying divergence patterns over sliding windows of recent events and for triggering reconciliation workflows under lineage-bound merge; and corresponding non-transitory computer-readable media bearing instructions implementing the foregoing. The disclosure further supports dependent claims directed to centralized, peer-distributed, and hierarchical detector topologies; exact and fuzzy correlation-key construction; library and custom divergence functions; and byzantine-robust quorum-corroborated divergence admission. The disclosed mechanism is independent of any specific consensus protocol, mesh implementation, or observation modality, and is intended to read on any embodiment in which divergence among federated meshes is detected and recorded as described.