Our Track Record
Every grade we publish is recorded the moment we publish it and graded later against what the price did next, compared with the S&P 500 over the same days. This page shows that record as it stands, wins and losses alike. Nothing here is a backtest.
Over 73,294 Buy calls spanning about 9 separate 7-day windows, our Buy grade ran about 1 point behind the S&P 500, give or take 0.6 points over the following 7 days, and level with the average stock we rate, give or take 0.8 points.
What each grade returned
| We said | Calls | Typical return | Won | vs S&P 500 | vs Avg Stock |
|---|---|---|---|---|---|
| Strong Buy | 5,079 | -0.3% | 47% | level (±1.0) | level (±1.3) |
| Buy | 73,294 | -0.9% | 41% | behind -1.0% (±0.6) | level (±0.8) |
| Hold | 104,579 | -0.7% | 42% | behind -0.9% (±0.8) | level (±0.8) |
| Sell | 4,919 | -0.5% | 44% | level (±1.0) | level (±0.9) |
| Strong Sellmeasuring | 75 | – | – | measuring | measuring |
“Typical return” is the median call, what usually happened. “Won” is the share of calls that ended positive over the window. The two comparison columns answer different questions: “vs S&P 500” is whether you would have done better just buying the index, while “vs Avg Stock” is whether our ranking picked better than the average name we rate. The index is dominated by the largest companies, so in a stretch where those lead, every grade can trail it even when the ranking is working. A figure is printed only when it is larger than twice its own margin of error; inside that margin the cell reads “level”. A row marked “measuring” has not closed enough separate windows yet; hover it for the reason.
Our model's top fifth
Beside the grades, a learned model ranks every name each day. Its record is measured on the fifth it ranks highest, the names it shows as its strongest, over the following 30 days.
Measuring 0 graded picks so far across 0 of the 5 separate windows a verdict needs. So far: still gathering its first measured windows.
Against the S&P 500: not yet measured against the S&P 500.
Measurement rules version 2026-10. How this is measured
How this is measured
Recorded when published
A call is stamped the moment the grade goes out. Nothing is reconstructed after the fact, and the record underneath is append-only: a past run is never edited.
Graded only after the window closes
A grade is measured against prices that came after it, over a fixed window, and only once that whole window has passed. An unfinished window is absent, never padded.
One call per name per day
A name rated several times in a day counts once, so a busy day cannot weigh more than a quiet one.
Against the index, over the same days
Each call's return is compared with the S&P 500 over the identical window, and separately with the average stock we rate. The first asks whether you would have done better in an index fund; the second asks whether our ranking helped.
Medians, not averages
The typical call is the median. An average is carried by a handful of volatile names and reads better than what usually happened.
Error bars across separate windows
Thousands of calls made in the same week are one observation of one market. The margin of error is computed across non-overlapping windows, and a cell needs at least five of them before it prints at all.
Nothing printed inside its own margin
A difference against the index is shown only when it is larger than twice its margin of error. Inside that margin the cell says “level”, whichever way the raw number leans.
What is not here
No backtests, no simulated books, no “would have returned” figures, and no chosen examples. Those exist in the product, behind a login, labelled hypothetical, and kept apart from this page on purpose.
Past results do not predict future ones. Nothing on this page is investment advice. See the disclaimer.