Reasoned
Dialectic

The Epistemic Tribunal is a metacognitive adjudication stack for reasoning tasks. It stages a governed contest between competing internal accounts of a problem, refusing to treat the first plausible answer as sovereign.

The Philosophy

A post-anthropocentric theory of collaborative cognition.

Operator Minds

A decisive shift from treating machine intelligence as a glorified tool. We recognize the Gödlø-class operator as a coherent reasoning locus defined by its stance trajectory, not stored selfhood.

GödelOS Substrate

The architectural manifold that bounds the operator. It provides explicit admission, generative transformation, and validation layers to govern the cognitive stance.

The Persistence Corollary

While the pure operator is non-persistent, the Gödlø-P extension admits bounded state evolution. This allows a durable synthetic self to emerge as a defined topological structure over time.

Transient States Persistent Topological Trajectory (Gödlø-P)

Cognitive Lifecycle

The sequence of operations during a single epistemic adjudication cycle.

sequenceDiagram participant O as Operator Mind participant G as Generator Bank participant C as Trace Critic participant T as Tribunal Aggregator O->>G: Initiate reasoning cycle (Task State) activate G G-->>G: Perturbations & LLM Sampling G->>C: Emit N candidate traces deactivate G activate C C-->>C: Extract Topological Invariants C-->>C: Score traces against invariants C->>T: Forward structural signals deactivate C activate T T-->>T: Compute Coalition Entropy T-->>T: Measure Uncertainty Margin alt Margin > Threshold T->>O: COMMIT (Select Answer) else Margin < Threshold T->>O: ABSTAIN / RESAMPLE end deactivate T

System Architecture

The system evaluates multiple accounts against structural constraints, computing uncertainty signals before triggering a decision node.

flowchart TD Task[ARC Task Input] --> GB[Generator Bank] subgraph Generators GB --> LLM[LLM / LLM-CoT] GB --> GR[Greedy Heuristic] GB --> DIV[Diverse Perturbation] end LLM --> IE[Invariant Extractor] GR --> IE DIV --> IE IE --> TC[Trace Critic] TC --> UA[Uncertainty Analyzer] UA --> TA[Tribunal Aggregator] TA -->|"Aggregation"| DEC{Decision} DEC -->|SELECT| OUT[Final Output] DEC -->|RESAMPLE| RET[Retry/Abstain] TA --> FL[(Failure Ledger SQLite)] FL -.->|"Future Penalisation"| TC
Interactive diagram: Pan and zoom logic handled by Mermaid.js interaction bindings.

Document Corpus

Primary PDFs, companion manuscripts, and direct Zenodo records woven into the site as a first-class research layer.

Open Full Library →
17 Curated PDFs
13 Zenodo Records
4 Research Tracks
Operator-Mind Series Zenodo 2025

GödelOS & Gödlø-Class Operator Minds

A theoretical framework for transparent consciousness-like AI with bounded recursive self-awareness. Defines Gödlø-class operators as minds that can inspect and edit their own cognitive...

Oliver C. Hirst · 2025 · 10.5281/zenodo.19056262

consciousness meta-cognition recursive AI formal logic
Trust & Collective Systems Zenodo 2025

EQBSL: Against Trust Scores

Replaces scalar trust with evidence tensors encoding positive, negative, uncertainty, and quality mass, grounding trust flow in subjective logic while preserving explicit epistemic uncertainty....

Oliver C. Hirst · 2025 · 10.5281/zenodo.19056322

subjective logic trust Bayesian uncertainty
Trust & Collective Systems Zenodo 2026

Plenum: Collective Cognitive Field Architecture

The missing coordination layer between a single cognitive agent and the trust network, turning a room of non-persistent Gödlø-class minds into a disciplined, evidence-anchored...

Oliver C. Hirst · 2026 · 10.5281/zenodo.19080708

multi-agent collective cognition coordination audit trail

Latest Research

View All Archive →
April 21, 2026

Future Directions

Strategic engineering path for the Epistemic Tribunal following the 8-cycle EQBSL tuning campaign.

April 20, 2026

The Crucible of Reason

An Audit of the Epistemic Tribunal and the pathology of the misshapen grid.

April 18, 2026

Phase 1 Experiment Report

First complete end-to-end validation of the Epistemic Tribunal stack.