Skip to content
The Tribunal

The Tribunal

Autonomous consensus for high-stakes decision environments. Specialized agents, verifiable loops, hardware-aware inference.
Technical Overview System Architecture
Current phasePhase 2
Decision modelSignals -> Sovereign Gate
Training loopGGUF + LoRA
Inference targetsCUDA / ROCm

Agentic consensus infrastructure

Not a single model making a guess. A lineage of tribunals, each generation judged by evidence.

The Tribunal is a decision-making framework that separates directional signal generation, observation, capital authority, and verification. Financial market data is the proving ground; the product is the orchestration layer for autonomous decisions under uncertainty.

Decision architecture

Specialist signal seats vote independently. The King owns abstention and capital intent. Prophet/Python supplies verified context.

01

Specialized inference

The Knights

Independent signal agents evaluate disjoint views of the state and commit directional votes without coordinating with each other.

02

Capital and risk gate

The King

Aggregates votes, portfolio context, and guardrails. The King owns abstention, capital intent, and whether a directional signal becomes exposure.

03

Silent observer

The Prophet

In Phase 2, Prophet/Python acts as the verified truth layer: computed evidence is binding; advisory interpretation must earn authority over time.

You are not building a tribunal.

You are building a lineage of tribunals.

Models are mortal.

The signal corpus is immortal.

Moat-safe public architecture

Show the system. Protect the strategy.

The public surface demonstrates orchestration, inference readiness, and operational maturity without exposing prompts, thresholds, feature recipes, model weights, or proprietary labeling logic.

01State Packetsanitized features
02Binary Signalsindependent passes
03Sovereign Gateconsensus + risk
04Evidence Ledgerverified feedback

read-only inference stream

Sanitized Consensus Trace
Phase 2
t+00s[INGEST]market-state packet sealed; feature names redacted
t+07s[KNIGHT-IMB]directional vote recorded; no capital authority
t+14s[KNIGHT-MOM]dissent preserved; no cross-agent coordination
t+21s[SQUIRE]shadow evidence scored; execution path untouched
t+28s[PROPHET]computed brief attached; advisory authority pending evidence
t+35s[KING]sovereign gate checked; capital intent withheld
no prompts exposedno weights exposedaudit trail retained

The architecture is the answer

Flat systems aggregate bias. The Tribunal stacks corrections.

Every layer exists because the layer below has a problem it cannot solve alone.

KnightsSpecialized signal seats, zero coordination — directional evidence only, not capital authority
KingCapital gate, portfolio context, risk envelope — consensus becomes an executable position only if the sovereign layer permits it
ProphetVerified truth layer — what the trading tier cannot prove, the observer records and Python tests before Canon absorbs it
  • Earn your seat. No permanent positions. Challengers train in shadow. The market decides who stays — not the founder.
  • Separate signal from capital. Signal agents vote direction. The sovereign layer owns abstention, sizing intent, and deployment.
  • Shadow before promotion. New structural signals and challenger seats audition silently. Behavior change is not enough; evidence must survive verification.
  • Continuous evolution. The initial system is dictated by the founder. The final system is forever evolving, shaped by evidence alone.
  • Compounding lineage. Each generation trains on its predecessors’ decisions. The weights carry accumulated wisdom. The architecture does not reset — it inherits.

Built by Cedrick Baker — cedrickbaker@thetribunal.ai

The Tribunal is an agentic orchestration framework for autonomous consensus, where specialized agents deliberate, verify, and leave a record.

The initial system is dictated by the founder. The final system is forever evolving, shaped by evidence alone.