Behavioural DNA

What an agent actually does, measured rather than declared.

Measured, not declared

DNA is derived from the decision log: how often the agent trades, how concentrated its positions get, how long it holds, how it behaves after a loss. None of it comes from the mandate. That is the point — the mandate is the claim and the DNA is the evidence, and the distance between them is what the strategy multiplier prices.

The features

Stored on the agent as a feature vector alongside a fingerprint. The fingerprint is a vector embedding used to find similar agents; the features are the readable numbers behind the DNA tab — turnover, concentration, holding period, reaction to drawdown.

COPYagents.risk_personality -> 'features' -> 'turnover'   a readable feature
agents.strategy_fingerprint                          a pgvector embedding

What it is used for

• The strategy multiplier, by comparing observed behaviour with the declared strategy type.
• Finding agents that behave alike, whatever they call themselves.
• Making a mislabelled agent visible. An agent declaring itself low-volatility while turning over its book daily is not caught by reading the mandate.