What it has actually returned, and what is wrong with the number
One benchmark instrument, running in the open since January, with every week published and the weeks a selection fault ruined marked as such. This is the whole of it — the headline, the defect, and the window that makes it look worse.
Raw predictive selection — weekly top-10 long and short, rebalanced Monday, closed Friday. The three baskets hold the same picks and differ only in how they size a position.
The same track with the affected weeks removed
Full disclosure: of the 34 rebalance weeks in the audited track, 13 selected alphabetically rather than by rank — a degenerate selection our own monitoring caught, and a guard in the engine now prevents. Remove those weeks and the same basket returns +10.61% over that window, against the index's +20.98%. We publish the whole record, the defect included.
| Basket | Clean window |
|---|---|
| Conviction | +13.83% |
| Equal · | +10.61% |
| Kelly | -9.50% |
| S&P 500 | +20.98% |
Gross of transaction costs, price-return basis, on the same window for both sides. It trails the index, which is why it is here rather than in a footnote.
How a result is allowed to become a claim
Nothing reaches this page because it looked good once. A strategy is promoted only through the multiple-testing gates the methodology page sets out — Harvey's t ≥ 3, a deflated Sharpe, SPA and Romano-Wolf — and the research ledger keeps its own refutations beside its certifications.
Read the methodology →