Evidence first
Every signal is grounded in visible market data and portfolio context, never in a model’s confidence alone.
A trade is only as good as the reason it can give
AUMERA turns market movement into decisions that have to defend themselves: challenged by other agents, checked by an independent risk kernel, and reconciled in a ledger before anything learns from the outcome.
Decision lifecycle
From a market observation to a sealed record in five steps
This is the path one decision takes through layers L01 to L05 of the system. No step can be skipped, and no model acts alone.
- 01Observe
Point-in-time evidence
Public order book depth, trades, and bars are captured with their venue and age, so the decision can later be judged against what was actually visible.
- 02Challenge
Adversarial cross-examination
Specialists publish independent views, and others attack the premise: false breakouts, thin liquidity, crowded positioning. Weak ideas die here.
- 03Synthesize
A proposal with its own exit
What survives becomes one explicit proposal with an entry, a size range, and the exact market condition that would prove it wrong.
- 04Authorize
Independent pre-flight veto
The risk kernel checks exposure, leverage, liquidation distance, and data health. It can refuse or narrow the proposal. Confidence does not count.
- 05Record
Paper execution and its outcome
Approved proposals become paper orders. The rationale, risk decision, and resulting trade remain available for review.
Principles
Three rules every decision follows
Separate authority
Research confidence cannot authorize capital. Risk enforcement is an outside supervisor, not a colleague.
Explicit invalidation
Before an order is staged, the proposal states the exact condition that proves it wrong.
Invariants
Limits the runtime enforces, whatever the models say
Volatility-scaled leverage
Allowed leverage shrinks as volatility and portfolio stress rise. Limits can only narrow a position, never widen it.
Decision history
Research, risk decisions, and paper trades can be inspected together.
No unchecked authority
Agents run inside research boundaries. No model has a direct path to any market, and live capital is not approved.
Follow one decision from thesis to verdict
Members can inspect available proposals and paper outcomes in the workspace. Missing specialist evidence is recorded as an abstention.