Method
The research process, in eight checks.
Eight checks every idea passes before it is allowed near capital — each with a concrete test we can verify, and each test written down.
01
Clean data
Point-in-time, survivorship-free, adjusted for corporate actions. A model can only see what its data lets it see.
testevery field stamped as_of ≤ t
02
Stated hypothesis
The mechanism, the expected sign and the kill criterion are written down before any test runs.
testhypothesis + kill criterion committed before any backtest
03
Honest reporting
Every variant tried is reported, the trial count is shown, and results are deflated for multiple trials.
testN trials reported with every result
04
Real costs
Spread, market impact, fees and borrow are charged in every test, at realistic size.
testnet of costs, or not shown
05
Clean sources
Only lawfully obtained data, and no edge that depends on manipulation.
testevery data source and venue reviewed before use
06
Steady iteration
Abandon what fails, strengthen what holds, and keep the record of both.
testevery rejection logged with its reason
07
Live monitoring
Live behaviour is compared with backtested behaviour continuously; drift is a signal, not an anomaly to explain away.
testlive vs. backtest drift checked daily
08
Focused capital
Capital goes where the evidence is strongest, inside a risk budget no single strategy may exceed.
testrisk budget per strategy, enforced