​
The one idea

The best trader in a pool of 30,000 is usually the luckiest, not the most skilled. If you rank a large population by Sharpe ratio and take the top name, you have run a multiple-comparisons experiment and selected its most extreme outcome. Some of that extremity is skill; most of it, in a large enough pool, is variance. A scorecard that reports raw Sharpe alone is reporting the experiment’s noise as if it were its finding.

​
The correction chain

1

Probabilistic Sharpe Ratio (PSR)

The probability that the true Sharpe exceeds a benchmark, given the observed track record’s length, skew, and kurtosis. Corrects for the fact that a short record supports a weak claim regardless of how good it looks.
2

Expected maximum Sharpe

Given N independent trials, what is the highest Sharpe you would expect from luck alone? This is the bar a genuinely skilled trader has to clear.
3

Deflated Sharpe Ratio (DSR)

PSR measured against the expected maximum instead of against zero. This is the headline number, and it is deliberately much less flattering than the raw figure sitting next to it.
4

False discovery rate

Across the whole scanned population, what fraction of the names clearing the bar are expected to be false positives? Reported alongside the board so a shortlist is read as a shortlist, not a list of certainties.

​
var_sr decides whether any of this works

var_sr is the variance of Sharpe ratios across the trial population. It is the input that makes the deflation meaningful.If var_sr is set to 1.0, the expected-maximum term collapses and every Deflated Sharpe Ratio silently becomes zero. There is no error and no warning — the board renders, every name scores 0, and the output looks like a population with no skill in it rather than a misconfigured statistic.If every DSR on a board is exactly zero, check this parameter before concluding anything about the traders.

​
What the scorecard does not claim

  • Not a prediction. A high DSR is a statement about a past record surviving a multiple-testing correction. It is not a forecast.
  • Not a recommendation. No position sizing, no entry, no target.
  • Not survivorship-free. The correction handles multiple comparisons across the scanned population. It cannot recover traders who blew up and left the venue before the scan.
  • Not fee-aware by default. See docs/BACKTEST_REALISM.md for the cost model applied to backtests; a raw scorecard is gross of trading costs unless stated.

​
Verdicts

The scorecard reduces to a small set of verdicts rather than encouraging readers to compare fourth decimal places. The composite score is noir_score in the codebase — an identifier the Scienter rename deliberately left alone, since renaming a persisted field is a migration, not a rebrand.

​
Access

Full specification: docs/SCORECARD_METHODOLOGY.md. Section 7, “What this does not claim”, is the one to read before quoting a number externally.
Scienter publishes impersonal market signals of general and regular circulation. Nothing here is investment, legal, or tax advice, or a recommendation to buy, sell, or hold any asset. Scienter is not a registered investment adviser or broker-dealer. Trading digital assets can result in the total loss of your funds — see Disclaimers for the full text.