Companies with at least two of: top-quartile R&D intensity, top-quartile stock-comp intensity, top-quartile share volatility (revenue under $5B). Measured 2015-2019 WITH delisted names included: 38% produced a +50% year within four years — 2-3× the market's base rate — and 19% delisted in collapse. Membership marks capability, both directions. Which members win was not predictable in our own study, so this is a cohort lens, not a pick list.
That paragraph is the whole claim. Everything below explains how it is measured and why it is framed the way it is.
Three traits, each a capacity to try: how much of its revenue a company spends on R&D, how much it pays in stock, and how much its own share price moves for reasons the market as a whole does not explain. A company in the top quartile of at least two, while still small enough that revenue sits under $5B, is a member.
Membership is a description of a company's shape, not an opinion about its price, its management or its prospects. We publish it because the shape turned out to concentrate large outcomes at both ends — and because a cohort that concentrates the good outcomes concentrates the bad ones too.
The evidence behind this cohort is a study built specifically to include the companies that didn't make it — names that were delisted, acquired or wound up are usually missing from financial databases, and a cohort measured without them looks far better than it was. With those names restored, the cohort's members produced a +50% year within four years about two to three times as often as the market at large, and delisted in collapse about a fifth of the time.
We then asked the obvious next question: within the cohort, can we tell the winners from the casualties in advance? We pre-registered two candidate discriminators and tested them properly. Neither one separated the two groups at better than coin-flip odds. A post-hoc look at what the "failures" actually were is instructive — a large share of them are not corpses at all, but companies that were acquired.
So we do not publish a capacity score, a capacity rank, or a "strength of membership" grade. There is no such number anywhere in the product, and a screen on this cohort sorts by market cap. Naming a best member would assert precisely what our own study says we cannot see.
Each trait is then placed as a percentile against the entire universe — which is what the "top-quartile" test in the rule refers to. The percentile columns in the screener are those ranks on a 0–100 scale.
If our data can't measure one of the three traits for a company — a filer that never tags an R&D line, say — that trait is null. It is not treated as a zero, and it is not treated as a failed test. Membership then needs two of the traits we can observe, and we publish a count of how many that was, so a blank never has to be guessed at.
This matters more than it sounds. Treating an untagged line as zero would quietly place every such company at the bottom of the distribution — a claim the data does not support — and would shift every other company's percentile at the same time. Roughly a third of the universe has fewer than two measurable traits; for those names, membership is simply not established, which is different from not a member.
A company whose revenue we cannot convert to dollars is never asserted into the cohort either: the size gate is a dollar threshold, and we would rather show nothing than gate on a number we can't compare.
The screener carries an optional Capacity column group — membership, the three component percentiles, and how many were measurable — plus a built-in "Capacity cohort" screen. On a member's security page, one line under the snapshot strip names the membership and how many traits it met. There is nothing else: no badge on non-members, no ordering, no score.
