Skip to content

The Tribunal

A public white paper for an autonomous trading architecture where deterministic engines trade, a probationary LLM court governs, and every seat is judged by measured outcomes.
Current doctrineThe Inversion
Trading layerPython Engines
Governing layerLLM Tribunal
Authority ruleMeasured Lift

Mobile white paper

Not a model chasing trades. A court that only earns authority above systems that can already be measured.

The Tribunal is an evidence-governed decision architecture for autonomous trading research. The live design separates deterministic trading engines from the LLM governing layer: code produces auditable targets; the court earns the right to allocate, veto, contextualize, or stand down.

The Inversion

Trading engines produce auditable outputs. The LLM court governs allocation, context, and stand-down decisions only where it proves lift over code.

01

Trading triumvirate

The Engines

Three deterministic seats define the trading layer: Reversal is active as daily target generation; Carry is the next build; Trend is gated at a longer horizon before construction.

02

Capital authority

The King

The King does not get authority by title. It must improve allocation, sizing, and stand-down decisions against a frozen Python control on the same rows.

03

Context and conscience

Prophet + Sentinel

The Prophet supplies verified context and counsel. The Sentinel watches regime and capital danger. Both remain advisory unless their contribution is measured.

You are not building a trader.

You are building a court that can replace its own assumptions.

Models are mortal.

The verified record is immortal.

Public architecture

Show the institution. Keep the recipes private.

This site explains the separation of powers, evaluation law, and operating philosophy without exposing thresholds, feature construction, prompts, model weights, or execution logic.

01State Packetsanitized inputs
02Engine Targetsauditable code
03Governing CourtKing · Prophet · Sentinel
04Result Ledgerverified feedback

read-only governance stream

Sanitized Tribunal Trace
The Inversion
t+00s[INGEST]market-state packet sealed; public surface redacted
t+07s[ENGINE-R]daily target weights generated; execution authority separate
t+14s[ENGINE-C]candidate book in development; no authority granted
t+21s[SENTINEL]regime mandate attached; capital posture checked
t+28s[PROPHET]computed brief attached; counsel remains probationary
t+35s[KING]allocation gate compared against deterministic control

The architecture is the answer

Flat systems aggregate bias. The Tribunal separates powers until evidence grants authority.

Every layer exists because the layer below has a job it should not be trusted to solve alone.

EnginesDeterministic trading seats produce auditable target outputs before any LLM layer receives authority.
CourtThe King, Prophet, and Sentinel govern allocation, context, and danger posture under paired measurement.
CanonVerified lessons survive model replacement. Rejected claims remain recorded so the system does not repeat old errors.
  • Earn your seat. No permanent authority. A component holds power only while measured contribution justifies it.
  • Separate trade from governance. Engines produce targets. The court allocates, contextualizes, or withholds only when the test record supports it.
  • Shadow before promotion. New candidates run silently first. Behavior change is not enough; paired measurement decides.
  • Retire with honor. The original LLM signal seats became the ordeal that clarified what should and should not own trading authority.
  • Compounding record. Each generation inherits the verified record, not the founder’s favorite hypothesis.

Built by Cedrick Baker — [email protected]

The Tribunal is an autonomous decision architecture where deterministic systems, LLM governors, and verification loops remain separate until evidence grants authority.

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