EspañolWorking paper · in preparation
HA-FUGA
A polyphonic adversarial validation protocol for multi-agent AI systems.
Six dimensional agents collaborate across analytical boundaries. A council of three frontier models submits findings to adversarial debate. An interaction matrix M₆ₓ₆ formalizes coherence; an integrity chain makes every finding traceable to its source.
The problem
Multi-agent AI without an internal mechanism for productive disagreement
Multi-agent AI systems are increasingly used to produce findings on complex, high-stakes questions: system design at scale, scientific synthesis, multi-stakeholder coordination, decisions whose answers no single body of expertise can produce alone.
As these systems scale, two structural gaps appear. First, agents are typically organized by technical function (collect, classify, summarize) rather than by the dimensions of complexity that the underlying problem presents. Second, validation is added as a downstream layer rather than emerging from the structure of agent interaction itself.
The result is systems that produce confident outputs without auditable reasoning and without an internal mechanism for productive disagreement. HA-FUGA closes both gaps.
The protocol
Six dimensions, one orchestration
HA-FUGA organizes agents by the six dimensions of Horizons Architecture. Each dimension asks a question the others cannot answer; the protocol holds them in productive tension through the M₆ₓ₆ interaction matrix.
A fugue, polyphonically
Coherence emerges from counterpoint, not from consensus
The name invokes the musical fugue: a compositional form in which independent voices enter successively with the same subject, developing it through imitation, inversion, augmentation and counterpoint. The quality of the work resides not in any single voice but in the productive tension between them.
The six dimensions are the voices. Each enters with its own question, holds its own logic, and develops the same subject — the problem at hand — from a perspective the others cannot reach. Above them, a council of three frontier models forms a second polyphonic layer, submitting findings to debate in parallel, iterative, and anti-conformity modes.
Disagreement is information, not failure. When the three frontier voices converge too fast, the protocol activates anti-conformity: a deliberate adversarial mode that searches for the failure modes consensus tends to hide. When they diverge persistently, the dissent itself becomes a documented finding.
The cycle
What unfolds when HA-FUGA runs
Six phases compose a single cycle. Each phase is observable, each transition is traced, every finding is sealed.
- F1
Ingestion
Documents are processed, hashed and classified. Each text becomes a node in a knowledge graph with traceable provenance.
- F2
Dimensional reading
Six agents read the corpus from their respective questions and produce findings, each finding sealed with an integrity chain.
- F3
Cross-dimensional audit
Each dimension reviews the others' findings from its own perspective. Anomalies trigger detection, containment and eradication cycles.
- F4
Polyphonic debate
The council submits findings to debate in three modes: parallel (MORE), iterative (SAMRE), anti-conformity when convergence is too rapid to trust.
- F5
Synthesis
Validated findings flow into the final outputs. Every claim carries its evidence and its confidence. Dissent is documented, not suppressed.
- F6
Learning
The knowledge base updates. New dimensional rules accumulate. Patterns persist into the next cycle, sharpening the protocol over time.
Five principles
Validation as an emergent property of structure
- 01
Adversarial by Architecture
Validation emerges from the structure of agent interaction, not from a downstream verification step.
- 02
Incident Response Dimensional
Anomalies detected by any dimension trigger detection, containment and eradication cycles that span the system.
- 03
Integrity Chain
Every finding carries a verifiable chain of dimensional identity, evidence hash, access manifest, and confidence score.
- 04
Polyphonic Debate
A council of three frontier models debates in parallel (MORE), iteratively (SAMRE), and anti-conformity modes. Consensus is not forced.
- 05
Cross-Dimensional Audit
Each dimension audits the others from its own question, generating continuous mutual scrutiny rather than after-the-fact review.
Five metrics
Quantifying coherence, integrity and convergence
DSS
Dimensional Synergy Score
- The quality of interaction between dimensions across the M₆ₓ₆ matrix.
ACR
Adversarial Convergence Rate
- The rate at which the council reaches consensus through debate; values too high signal conformity, values too low signal incoherence.
EIS
Evidence Integrity Score
- The traceability of every finding back to source documents via its evidence hash.
DCS
Dimensional Coverage Score
- Whether every dimension contributed substantively to the analysis.
The kernel
The M₆ₓ₆ interaction matrix
The interaction matrix M₆ₓ₆ records the influence weights between every pair of dimensions. It is the formal kernel of coherence in the system: where each pair of voices reinforces, contradicts, or simply ignores the other becomes a measurable property of the structure rather than an after-the-fact intuition. Reading the matrix exposes synergies and tensions that no single-axis decomposition can recover.
The matrix is not static. As the cycle runs, the cross-dimensional audits update each cell with the quality of interaction observed between the two dimensions on this specific question. The matrix you read at the end is a compressed picture of how the dimensions actually engaged each other on the proposal — not a generic schema of how they could.
Four contributions
What HA-FUGA adds to the literature
- 01
Theoretical
First formal application of the Horizons Architecture notation to multi-agent AI validation. Organizing agents by dimensions of complexity, rather than by technical function, yields a tractable framework for problems that resist single-axis decomposition.
- 02
Methodological
A multi-agent system whose principle of organization is dimensional. Coherence between agents emerges as a property of the system, not as an after-the-fact verification layer.
- 03
Protocol
HA-FUGA itself: an integrating contribution that brings together AIR, AIP, TRiSM, D3, FREE-MAD and AuditBench / Petri over a dimensional ontology — a configuration no published framework offers.
- 04
Practical
Original end-to-end implementation under institutional confidentiality, demonstrating execution at scale with full integrity-chain auditability across heterogeneous data and stakeholder boundaries.
See it in action
Three runs, replayable instantly.
Each card replays a complete HA-FUGA cycle — the same protocol the live demo runs, but pre-recorded so it loads in seconds at zero cost. Every event, finding, audit, and council vote is preserved exactly as it happened.
Postal Innovation
Cross-border parcel triage · multi-agent
A multi-agent AI workflow coordinating the postal operator, customs authority, e-commerce platform and end-user. Six dimensions evaluate coherence with the Universal Postal Convention and destination customs regimes (EU ICS2, US ACAS).
Replay →
Postal Innovation
Last-mile rural reach · postal modernization
Extending postal coverage into underserved rural areas by combining the designated postal operator's network with informal couriers and digital addressing. The protocol surfaces which empirical claims rest on measured data versus inherited optimism.
Replay →
Postal Innovation
Postal remittances · cross-border financial inclusion
Using the postal network as the primary cross-border remittance channel for low-income migrant workers. Six dimensions weigh universal-service obligations, FATF compliance, capability gaps relative to specialized fintechs, and unit-economics assumptions.
Replay →