Track record

What this record actually shows

The figures below are recomputed on every publish, from the tracker's own evaluation log. Last recomputed: 2026-09-18.

The ideas were all saved. The real track record does not exist yet, and the paper one shows no demonstrable edge.

• Every evaluation is dated and stored, so the record is complete from 2026-04-06 onwards: 1191 recommendations over 440 companies. Nothing has been back-dated or removed.
• Real money: 1 round trip. AFYA, opened on an algorithm candidate and closed 36 days later for +0.48%. A second position was transferred in, not chosen by the method. One trade at roughly zero proves nothing, in either direction. (Last checked 2026-09-04.)
• Paper test on 1123 recommendations with an entry price, bought at the close of the recommendation date, benchmarked against the median of the whole tracked universe over the same window: the positive verdicts (397) show a median excess of +0.75%, the rest (726) +0.02%. That puts the recommended names ahead of the ones we did not recommend — by a margin far too small, on a sample this young, to be called an edge.
• Rank correlation between the score and the excess return is -0.019 over the full sample and -0.018 at a 30-day horizon. On the samples with real size, the score does not order returns.
• The reason is the window, not the method: the median holding period is 17 days, and 792 of 1123 observations are under 30. A Graham/Buffett/Lynch method is built for years. Measured at 17 days you are measuring market noise.
• One thing survives, at the bottom end. On the 24 observations held more than 90 days the ordering is correct and the rank correlation turns positive (+0.293) — driven by the AVOID end, which loses 23.5 points against the benchmark. So far the method has been right about what to avoid. That subset holds no BUY verdicts at all, so about what to buy there is still no evidence.
• Known biases, stated: repeated evaluations of the same ticker a few days apart are counted more than once; the universe is what the author chose to analyse, not a random sample; prices are closes from the local daily cache and are not dividend-adjusted; there are no transaction costs and no position sizing. The table measures a signal, not a strategy.

Real positions: 1 round trip (AFYA, 0.48% over 36 days).

This is not investment advice. Past performance, hypothetical or real, does not predict future results. The history below is published because EU Market Abuse Regulation art. 20 requires the record of past investment recommendations to be disclosed alongside the methodology.