Research · 2026-08-05
Backtests are hypotheses. Live capital is the peer review
Nothing we deploy is trusted because it is clever. Evidence under duress, then a budget that cannot be argued with.
A backtest is a claim about the past, written with the benefit of hindsight. It is useful the way a laboratory is useful: it can falsify. It cannot certify.
We treat the simulated track as a hypothesis to be attacked — Monte Carlo resampling of the trade sequence, parameter neighbourhoods, costs at their worst hour. Only then does a risk budget exist, and the budget is not advisory.
When a model approaches its boundary, the system does not negotiate with conviction. Live capital is the peer review. The backtest is the abstract.