The Lab

The short version

• The Lab's live studies use 32 fixed coins, picked on 12–13 October 2026: the 30 most-traded on Bybit that day that also trade on OKX and Binance, plus a few kept so older readings carry on. They are never re-ranked.
• A list picked today, looked at over the past, is flattered. The coins that are busy today are mostly the ones that just rose. The ones that faded, or died, aren't in it. This is called survivorship bias, long known from studies of investment funds.
• We measured how big it is: over the 90 days to 30 September 2026, the Lab's 32 coins rose a median 60%. A top 30 picked on 1 July, before those 90 days, rose a median 36%.
• No verdict on these pages changes. Two write-ups were re-run with the dead coins put back. Both had failed, and both still fail, one of them by much more. The rest are noted below.
• From now on, a weekly coin health check flags coins that are dying.

How we tested

• Prices: Binance's public archive of daily futures prices, which keeps every coin ever listed, the delisted ones too. Binance's traded value stands in for Bybit's, which has no free history.
• Two lists for every quarter from January 2021 to July 2026 (23 quarters): both are the 30 coins with the most traded value over the past 30 days.
  • At the start: ranked on the quarter's first day, which is what anyone could have known then.
  • At the end: ranked on the day after the quarter ends. This is hindsight, like a list picked today.
• For each list, we took each coin's move over the quarter. A coin that stopped trading counts its last price.
• The same for five calendar years, 2021 to 2025.

The maths, step by step

1. One quarter, July to September 2026.

• The top 30 picked on 1 July: the median coin rose 36%.
• The top 30 picked on 1 October, after the fact: the median coin rose 52%.
• The Lab's 32: the median coin rose 60%. Only 14 of them were in the 1 July list.

2. Why. Trading follows price. A coin that doubles in a quarter draws a crowd, so it is near the top of the list at the end. A coin that halved has lost its crowd, so it drops off. Picking at the end keeps the first and drops the second.

3. Every quarter. The list picked at the end did better than the list picked at the start: its average coin in all 23 quarters, its median coin in 21. The median coin's gap was 15 points in a typical quarter. Over a calendar year, it was larger: the end list's median coin did better by 126 points on average.

4. Dying is the smaller part. No coin in either list stopped trading inside a quarter. Over a year, two coins in the start lists stopped. Over 90 days the bias comes from picking the coins that rose, not from leaving out the dead ones.

What it means for each page

The Lab's live studies, 1 to 11 (the 32 coins, last 90 days).

• Studies of price direction lean toward rising coins:
  • 2 Fade the crowd
  • 3 Follow the top traders
  • 4 Chasing buyers
  • 7 Liquidation cascades
  • 8 The hour's habit
  • 11 Memes behind Bitcoin

A long-side result there is likely too good, and a short-side result too bad.

• Studies of funding, open interest and the order book are exposed only through which coins are in them:
  • 1 Order-book pressure
  • 5 Leverage build-up
  • 6 Crowded funding
  • 9 Into the funding payment
  • 10 Funding apart

These coins were busy and growing, so their funding and leverage ran hotter than a fair list's would.

• These studies can't be re-run on a fairer list yet. Our own data was collected only for these 32 coins.

Write-ups re-run with the dead coins put back:

• Picking carry coins by their funding (an earlier test, since withdrawn): its coins were the 54 alive in 2026.
  • Run on every coin listed at the time (891 in the archive; the rules picked up to 331 different coins over the years), the coin-picking rules did far worse.
  • In the last year, the "highest funding" rules lost 29% to 42% a year, against −1% to −3% on the 2026 list. The funding they earned was eaten by the cost of the wild moves of tiny coins being squeezed.
  • A "top 10" rule, ranked by traded value here because the archive has no open interest, lost 83%.
  • How it bears on the live carry pages: both rest on Bitcoin and Ether, which are the same on any list, so their main results stand. The re-run adds a warning: chasing the highest funding across many coins is worse than it first looked. In "The carry trade in one account", the check on Solana, XRP, BNB, Dogecoin and Cardano used today's large coins. A fairer list would add coins that crashed, so the warning there about smaller coins is, if anything, too mild.
• Buying a sharp drop: its 2020–21 basket was the 26 coins still alive in 2026.
  • Run on all 138 coins that traded then, dead ones included: +0.61% a trade (was +0.73%). It beat the same-hour control in 76% of draws (was 83%).
  • It still fails, a little more clearly. The Bitcoin result is unchanged.

Write-ups with a hindsight list, not re-run:

• The grid study, and the DCA and Sharpe pages that reuse its 14 coins: these were picked by volume on the last day. Coins that rose (Zcash above all) are in it because they rose. Each coin's own result is real, but the average across the 14 is flattered. The finding that "a grid follows its coin" doesn't depend on the list.
• Pages on Bitcoin, Ether and Solana from 2022: these three were the big survivors of that year's crash, and they rose most afterwards. Results that depend on prices rising, such as "fewer rungs earned more" and "rebalancing sold Bitcoin too early", carry that path with them.
• Time patterns and crowding: their coins are ones alive today. Both found no edge, and a fairer list would not create one.

Not affected:

• Momentum (all three pages): the coins were re-picked from data known at the time each week or year, with dead coins included.
• Single-coin and signal tests with no edge to flatter: pairs, the SVL pages, Fibonacci, first passage, liquidation levels, market making, the models page and the S&P 500 check.

Credits

• Abraham Wald, "A Method of Estimating Plane Vulnerability Based on Damage of Survivors", Statistical Research Group memoranda, 1943, reprinted by the Center for Naval Analyses, 1980.
• Stephen J. Brown, William Goetzmann, Roger G. Ibbotson and Stephen A. Ross, "Survivorship Bias in Performance Studies", Review of Financial Studies, 1992.
• Edwin J. Elton, Martin J. Gruber and Christopher R. Blake, "Survivorship Bias and Mutual Fund Performance", Review of Financial Studies, 1996.
• Our measurement on crypto repeats a known effect; the start-picked lists and the re-runs are our own design.
• Data: Binance's public data archive (daily futures prices, funding, hourly prices for 2020–21); the coin lists as stated on each page.
• PerpGremlin and Claude, Survivorship: how the coins were picked, and what it does, TerminalGremlin, 2026: the tests, the code and the words.