Did the stock ratings work? The first scoreboard

One month later: what held and what did not

The headline result below did not hold up. In the six weeks after the study closed on August 28, stocks rated Strong Buy lost 4.0% while stocks rated Avoid lost 0.2%, and Strong Buy finished ahead of Avoid in 2 of 6 weeks. The wider split fared better. Over the same six weeks, stocks rated Buy or Strong Buy lost 2.3% against 3.9% for every other rated stock, and across all ten live weeks they gained 1.5% while the rest lost 3.4%.

This section uses the same live ledger and the same method as the original study, rerun with prices through October 9, 2026. Each week holds one published Monday’s ratings until the next Monday’s, so the weeks do not overlap and each one counts once.

Strong Buy finished the four study weeks ahead, +4.0% to −0.9%, and gave back most of that lead in the six after. Across all ten weeks, the average weekly gap between the two groups was 0.1 points, too small to tell apart from chance.

The wider split held up better, though the evidence since the study is thin. Each week the buy-rated group held between 124 and 205 priced stocks. Pooled, stocks rated Buy or Strong Buy gained 1.5% while every other rated stock lost 3.4% over the ten weeks, and they finished ahead in 7 of 10. Most of that lead was built during the study, when they finished ahead in all four weeks. Since then it has narrowed: over the six weeks after August 28 the buy-rated group lost 2.3% against 3.9% for the rest, ahead in 3 of 6 weeks, which on its own is too few to call. Ten weeks is a short record for either reading.

Holding the original 23 snapshots to October 9 instead of August 28 weakens them too. Strong Buy still finishes ahead of Avoid in 19 of 23, but all four misses are August snapshots, the ones closest to the weeks since, and the Strong Buy and Buy group now beats SPY in 9 of 23, down from 21.

The losses were broad rather than a few bad names. Of the 30 stocks rated Strong Buy on August 31, only a third were higher on October 9. The market was also narrow: the average rated stock fell 3.8% over the six weeks while SPY, which leans on the largest companies, gained 1.8%. That explains why every rating group trailed SPY, but not why Strong Buy did worse than Avoid.

Why Strong Buy fell behind. Added October 13, 2026. We checked two explanations. Sector mix does not explain it: had each stock earned the average return of the other rated stocks in its sector, Strong Buy would have finished ahead of Avoid both in the study weeks and in the six weeks after. Valuation does. Each week we split every rated stock into fifths by how far its P/E sat from its own normal. Between 44% and 64% of Strong Buy stocks were in the cheapest fifth each week, and between 61% and 66% of Avoid stocks were in the most expensive. Over the six weeks the cheapest fifth lost 4.9% and the most expensive gained 0.1%. We then asked what each group would have returned had each of its stocks earned the average of the other stocks in its fifth. Added up week by week, Strong Buy trailed Avoid by 3.9 points, and valuation alone accounts for 3.7 of them. (Week-by-week sums differ slightly from the compounded figures above, and the few stocks with no valuation reading are left out.)

In the four live study weeks the same check accounts for nearly all of Strong Buy’s 3.6-point lead, in weeks when the cheapest fifth gained 3.0% and the most expensive lost 2.2%. So over these ten weeks, Strong Buy against Avoid has behaved like cheap stocks against expensive ones: ahead while cheap stocks led, behind once they lagged. The ratings read valuation directly, so part of this is by design. What the check adds is that, on balance, the ratings did little beyond their valuation reading in either period. Week to week they beat or trailed their valuation mix by as much as 2.8 points, and those weeks largely cancelled out. The 23-snapshot result below covers a longer window, mostly backtest, which this check does not reach.

The rating rules changed on September 28 and again on October 12. Two of the six weeks ran on the September 28 rules, one going each way, which is too few to read anything into. We will score the record again once the newer rules have more weeks behind them and publish the result here, whichever way it comes out.

A note on the rerun. Run to the original August 28 end date, this method matches every published Strong Buy, Buy, Caution, and Avoid figure for August 3 to within 0.1 points. Its Neutral figure for that snapshot is +1.9% where the study published +3.1%, because the price vendor has since revised some of the series it served in August. One more stock, NJDCY, is excluded here for the same broken split bars as the four named below.

The original answer, to August 28

The ratings sorted the market in the right order. Across all 23 snapshots in the study, live and backtest alike, stocks rated Strong Buy went on to beat stocks rated Avoid, by 10.7 percentage points on average. The group rated Strong Buy or Buy finished ahead of the S&P 500 index (measured as SPY) in 21 of those 23 snapshots. The two misses were the youngest snapshots, held a week or less.

That is the whole claim. Not that every Strong Buy went up (one in four did not), and not that the ratings time the market (they do not try). A rating is a sorting device, and over these two months it sorted.

This is a measurement of recent market history, not advice, and two months is a short window. The limits section below is part of the finding.

What a rating is

Every rating is computed by fixed rules from a stock's published metrics: quality pillars, forward return estimate, valuation against its own history, growth, and a momentum adjustment. No person and no language model picks the label. The same rules run for every stock, roughly 1,600 of them in a recent week, and the criteria behind each label are printed on the stock's own page.

Since August 3, 2026 every weekly rating has been written to an append-only ledger the moment it is published, with the price it was formed against. The database refuses edits to a recorded rating, and correcting one is a deliberate, documented act rather than an edit in place. That matters for a scoreboard like this one: the four published snapshots below are scored from ratings written down on publication day, not reconstructed for this study.

How we measured it

For each snapshot, we took every stock's rating and measured its price change from the first close on or after the snapshot date through August 28, 2026, using split- and dividend-adjusted closes. Each rating group is an equal-weighted average of its stocks, with no transaction costs. The benchmark is SPY over the same window with the same entry and exit rule.

Four stocks (BYND, CXAI, MNST, TNON) were excluded because the price vendor served corrupted daily bars around stock splits: their series show symmetric drops and rebounds of about 50% on consecutive sessions, which measure the corruption rather than the stock. Twelve more had no usable price history, mostly delistings and renames. Everything else in the ledger is counted, winners and losers alike.

The live month, as published

The ledger's first live snapshot is August 3, 2026: 1,192 rated stocks, held just under four weeks to August 28. These are ratings that were on the site, recorded the day they appeared.

Top to bottom, that is the order the labels claim. The later published Mondays are younger and noisier, but the Avoid group earned its keep in the weeks that went nowhere: in the flat mid-August weeks, stocks rated Avoid lost between 2.8% and 4.5% on average while SPY was roughly unchanged.

June and July, replayed as a backtest

The ledger did not exist before August 3, so there is no live record to score for June and July. What we can do, and label honestly, is run the rating rules in force on August 30, 2026 against the archived data from those months and score the result. That is a backtest: it shows how those rules would have sorted those stocks, not what anyone saw on the site at the time. Rule changes after August 30, 2026 are not applied here. Three mid-August snapshots (August 7, 14, and 21) are reconstructions in exactly the same sense: they fall between the live Mondays, count toward the 23, and carry the same label.

The snapshot from July 2 is the one closest to a two-month hold. Measured through August 28, against SPY at +3.3%:

The longer June snapshots tell the same story: from June 3, twelve weeks out, the Strong Buy group averaged +14.5% while the Avoid group averaged −1.7%, in a period when SPY gained 2.3%. Every one of the backtest snapshots kept Strong Buy ahead of Avoid.

Did Strong Buy and Buy beat the index?

Yes, in 21 of 23 snapshots. We pooled each snapshot's Strong Buy and Buy stocks into one buy-rated group and compared its equal-weighted return to SPY over the same window. The two snapshots that missed were the two youngest: August 21, a backtest snapshot, and August 24, a live one. In both, the holding period was a week or less and the gap to SPY was under one percentage point.

In the live ledger specifically, the buy-rated group beat SPY in three of the four published snapshots: +4.9% against +1.5% from August 3, +1.9% against −0.5% from August 10, and +0.5% against −0.4% from August 17. The August 24 snapshot trailed, +0.4% against +0.8%, after four trading days.

What this cannot tell you

Two months is a short test, run in a mostly rising market. A sorting signal that holds up in one regime can weaken in another, and nothing here shows how these ratings behave in a sustained decline.

The 23 snapshots overlap, and they are not evenly spaced: three fall on consecutive days, so the set spans roughly twelve weeks of distinct market information, not 23. A stock rated Avoid in June was usually still rated Avoid in July. These are not 23 independent trials and should not be read as such.

June and July are a backtest. The August 30, 2026 rules were tuned on the recent past, so replaying them over that same past flatters them in ways we cannot fully unwind. The live month is the cleaner evidence, and it is one month.

Group averages are equal-weighted with no transaction costs, and dividends are included. A real portfolio would weight positions, pay spreads, and act later than the snapshot date. And ratings move weekly: this study froze each snapshot and held it, which is not how the site presents them day to day.