The published backtest re-measured after the 13 November data repairs (2026-11-14)
Public summary of audit 328. No strategy rule, threshold or selection logic changed.
What changed
On 13 November the stored returns were rebuilt under one entry rule, German purchases received their returns, and Swedish filings hidden by an over-broad duplicate-symbol check were re-admitted. The same backtest as on 12 November was run again on production, read-only, with the same settings.
The numbers, before and after
2022-01 to 2026 Q1 decisions, EUR portfolio, total returns, 0.6% round-trip cost. Sharpe and gaps on each stock's real daily prices.
| Figure | Published until now (12 November) | 14 November |
|---|---|---|
| Ten rows served first, annual return | 2.5% | 5.3% |
| STOXX Europe 600 on the same dates | 9.6% a year | 9.0% a year |
| Ten rows served first minus the index, points a year (95% interval) | -6.3 (-12.8 to -0.3) | -3.0 (-9.5 to +3.8) |
| Annual return of the proof basket | 8.1% | 7.1% |
| STOXX Europe 600 on the basket's dates | 8.1% a year | 7.3% a year |
| Basket minus the index, points a year (95% interval) | +0.4 (-6.4 to +6.6) | +0.1 (-5.4 to +6.2) |
| Research cell, annual return | 2.1% | 2.3% |
How to read it
- The rows shown first still trail the index, by about 3 points a year, and the interval now contains zero. The proof basket earns what the index earns. No edge is demonstrated.
- The change comes mostly from filings the backtest did not have before (German purchases) and from returns now computed under one rule. It is a correction of the data, and the size of the move is well inside the uncertainty.
- A backtest is not a live record and does not establish an investment advantage.