How to read a winners' list without being recruited by it: separate the sample from the population, label the return driver, and decide in advance which of these outcomes your own process would have refused.
1. Never read a "best performers" list without asking what the whole sample did
The repeatable method
- Establish the population the winners were drawn from, and its size — a top ten out of forty is a different claim from a top ten out of four thousand.
- Get the average of the whole set, not just the survivors, and compare that with the benchmark.
- Ask for the losers' list from the same exercise. If it is not published, treat the winners' list as advertising.
- Weight the conclusion by whether anyone actually held the picks in the sizes implied.
Here: the whole-sample number is given — readers' picks averaged
+17.3% against the S&P 500's
+16% — which is a 1.3-point edge, not the +107% to +349% the page displays. The
8 January companion reports the same year at +19.0% vs +17.1% on the ten most-picked names, so even the "whole sample" figure moves with the cut chosen.
Watch for
- An equal-weighted average of picks presented as a portfolio return. Nobody held them equally, and nobody rebalanced.
- The benchmark quoted at two different values in two issues three days apart (+16% and +17.1%). Check the basis before building on either.
2. Attach the return driver to every winner, and check whether it is repeatable
The repeatable method
- For each big move, name the single event or condition that caused it — a contract, an approval, an index addition, a price cycle, a financing.
- Classify it: recurring (structural demand, pricing power) or one-off (a signed deal, a trial readout, an index event).
- Discard the one-offs from your own idea generation, however large the return.
- Where the driver is a single counterparty, size the concentration risk that produced the upside.
Here: NBIS re-rated on "a multi-year AI infrastructure deal worth over $17 billion" from Microsoft; IREN on "about $9.7 billion", also Microsoft; RKLB on an $816m Space Development Agency award; KTOS partly on joining the S&P MidCap 400; OLMA on early trial data for palazestrant. Only MU's driver — pricing and utilisation in memory — is a cycle rather than an event, and that is the one Slegers later argues against.
Watch for
- "Pick-and-shovel" used to describe a business with one customer. The phrase implies diversified demand; two of these do not have it.
- Index inclusion listed beside fundamentals as though it were one. It is a flow, and it does not recur.
3. When results contradict your style, publish them and say so in the same issue
The repeatable method
- Report outcomes your framework would not have produced, rather than omitting them.
- State the mismatch explicitly and in terms of business characteristics, not of taste.
- Do not adjust the framework in response to a single year of a different style working.
- Keep the record, so the same names can be revisited a full cycle later.
Here: "
These are not the type of companies we usually cover at Compounding Quality. Most businesses on this list are cyclical by nature. They are exposed to commodity prices, capital cycles, technological shifts, or regulatory changes." The warning is also placed
before the list — "on average, (way) more risk was involved" — so a reader meets it first. Contrast the
August growth interview, where a contradictory framework is printed at length with no house rebuttal at all.
Watch for
- Style drift arriving as "we're broadening the universe." Fundsmith's mid-2026 turn to momentum is the archive's cautionary case.
- A disclaimer that is only in the conclusion, where fewer readers reach.
4. Treat record margins as a question, not a confirmation
The repeatable method
- When a cyclical business posts its best-ever margin, locate the previous peak and what followed it.
- Ask whether the pricing came from scarcity (temporary) or from a structural advantage (durable).
- Check whether the industry is currently adding capacity — the surest sign the pricing is borrowed.
- Decide whether the current price requires the peak to persist. If it does, the position is a bet on no mean reversion.
Here: MU is reported at +226.8% because "pricing improved, utilization rose, and Micron's earnings power surprised even optimistic investors." Six months later the same facts are the argument against it (
July 2026): margins above the 2018 peak in a business with no pricing power, Grantham's "if you make obscene profits, you'll get ferocious competition," and the verdict "that's not a bet that I want to make." The same evidence, read forward instead of backward.
Watch for
- Commodity producers described in secular-demand language — UUUU here is uranium and rare earths framed as a technology trend.
- A capital raise reported as a strength (EOSE). Under the same author's own low-capital-intensity test, it is the opposite.
Methods distilled from the archived Compounding Quality post for personal study. Not investment advice.