A fully autonomous trading bot built by Radius Labs.

Autonomous trading bot / Completed

Built by: Radius Labs
System focus: Risk, execution & learning
Review posture: Explicit, modelled & bounded

Asset universe / 01

Multi-asset decision loop

Rather than evaluate a coin or symbol in isolation, ERGODIC resolves volatility-scaled sizing, covariance-aware exposure, liquidity-adjusted execution, and walk-forward recalibration into one constrained decision across its configured asset universe.

ERGODIC interface showing position sizing, execution-cost modelling, exit calibration, walk-forward windows, and learning-loop journal.

Methods atlas / Explicit assumptions

The theory is part of the interface.

ERGODIC makes its decision language legible: sizing, range, cost, covariance, and out-of-sample review remain visible where a system would otherwise become a single opaque score.

Kelly / information rate
Wilder / range analysis
Markowitz / portfolio theory
Almgren–Chriss / execution

Fractional Kelly sizing

f* = (bp − q) / bf applied = λ · f*J. L. Kelly / 1956

The raw fraction is scaled before it reaches a position-size decision, keeping the sizing rule explicitly bounded.

Volatility-aware exits

ATRₜ = ((n−1)ATRₜ₋₁ + TRₜ) / nTRₜ = max(H−L, |H−Cₜ₋₁|, |L−Cₜ₋₁|)J. Welles Wilder / 1978

A range-based volatility measure lets exit calibration adapt to current movement rather than a static threshold.

Cost-aware execution

C exec = fees + slippage + I perm + I tempmin E[C] + λ · Var(C)Almgren–Chriss / 2000

Fees, slippage, and impact remain visible before a theoretical decision is treated as an executable action.

Walk-forward validation

train → test → rollw* ∝ Σ⁻¹μMarkowitz / 1952

Rolling, non-overlapping windows keep calibration separate from out-of-sample review while portfolio context stays explicit.

Interface sequence / From context to action

Four views, one visible decision chain.

ERGODIC Command Deck interface showing market regime, system health, portfolio posture, activity pulse, and decision queue.

01 / Command Deck

A bounded operating view

A single surface makes the system’s posture, health, and queued decisions legible before an action is handed off.

ERGODIC Regime Intelligence interface with regime map, rolling probabilities, correlation matrix, volatility structure, and model consensus.

02 / Regime Intelligence

Classify before committing

Regime state, volatility structure, correlation, and model consensus appear together so a decision has a visible context.

ERGODIC Signal Swarm interface showing ranked strategy agents, agreement and dispersion map, factor attribution, and bounded handoff.

03 / Signal Swarm

Show the disagreement

Independent signals are not flattened into a black box: agreement, dispersion, attribution, and handoff state remain inspectable.