Validation
Validation is not the phrase "backtested". The bar is published—and so is everything that has not cleared it.
The bar
Four checks. All four, or nothing.
For a factor or hypothesis to enter a formal view, it must clear all four. Fail one and it stays in observation.
Check 01
Backtest
Direction and stability are tested on historical samples. Return without stability does not reach the decision layer.
Check 02
Forward behaviour
After the backtest, observation continues on new data. A factor that only fits the past and cannot persist is sent back.
Check 03
Directional stability
Direction must hold across samples and horizons. A factor that flips sign between windows is judged unstable.
Check 04
Cost & turnover
Returns are reported net of trading cost and turnover. No gross figures, and no marketing built on them.
Factor library
Factor library status
Scale figures for the factor library, which change as research progresses. Scale is not proof of capability—clearing the bar is.
Factors in research
202
Total factors currently maintained, including those under observation and validation.
Factor categories
9
Grouped by construction logic to prevent factors from being highly redundant.
Backtest window
2yr
The sample window covered by historical testing, published together with the methodology.
Validation results
Validation dashboard
This is where passed, observing and retired counts—plus per-category progress—will appear. Until the disclosure bar is met, we leave the numbers out.
Disclosable validation sample: accumulating
We do not publish pass rates, hit ratios or annualised figures while the sample is thin. When the bar is met we disclose, together: the sample window, the calculation methodology, passed / observing / retired counts, and a breakdown of primary failure causes.
System performance
System performance
System-level comparison—against broad indices, maximum drawdown, and the share of views held, revised or retired—needs a sufficient sample to mean anything.
Sample accumulating · no performance curve yet
Published views are being tested against subsequent facts on an ongoing basis. Once the sample clears the disclosure bar, we publish it with the number of independent trading days, the methodology, and the failure cases—rather than shipping a flattering curve first.
Read the bar before the view
If you read one page on this site, read this one. It explains what earns a viewpoint the right to enter a formal judgement.