PATTERN+ PREDICTIONS
Updated

Study · re-run every night · updated

What the strategy search keeps, and what it threw away

On the night of Oct 6, 2026, the search tested 3,144 versions of 1,783 trading rules, each buying the coins at one end of a measure and shorting the other end. 55 passed the first cut on the older 60% of their history (with this many tests at once, a luck test of about 3.2 was needed); 8 also earned on the newer 40% they never saw. The book keeps every rule that ever passed: 8 still passing, 3 faded, each paper-traded from the night it was found (the first on Oct 6, 2026). Before fees. Past patterns, not financial advice.

Tests

3,144

1,783 rules

First cut

55

older 60%

Held

8

newer 40% too

In the book

8

still passing

First written · 1 revision

Share on Xcopy link

§ 01 · The night's search

The night's search

as of · the strategy search, Oct 6, 2026

The night's strategy search as a funnel: Tests run 3,144; Different rules 1,783; Passed the first cut 55; Held on the newer part too 8Tests runTests run · 3,144 · bars on a log scale3,144Different rulesDifferent rules · 1,783 · bars on a log scale1,783Passed the first cutPassed the first cut · 55 · bars on a log scale55Held on the newer part tooHeld on the newer part too · 8 · bars on a log scale8The night's strategy search as a funnel: Tests run 3,144; Different rules 1,783; Passed the first cut 55; Held on the newer part too 8Tests runTests run · 3,144 · bars on a log scale3,144Different rulesDifferent rules · 1,783 · bars on a log scale1,783Passed the first cutPassed the first cut · 55 · bars on a log scale55Held on the newer part tooHeld on the newer part too · 8 · bars on a log scale8

Each rule ranks the coins by one measure (a move, a volatility, a volume, a positioning reading), buys one end and shorts the other, and holds for a set number of days, in every market or only while Bitcoin is above or below its 200-day average. Its oldest periods start Mar 31, 2019; the newer part starts between Aug 28, 2023 and Aug 25, 2026, depending on how long each measure's history is.

§ 02 · What it keeps8 kept

What it keeps

as of · the survivors, counted

8 rules passed both cuts. Across them, the typical rule made +0.29% a period from the coins it bought over the ones it shorted (+0.05% from the bought side alone), with a luck test of 2.2 on the newer part.

  • 7 held 1 day
  • 1 held 3 days
  • 3 in every market
  • 5 only in bear markets
  • 6 among all the coins
  • 2 with the majors, the thinnest, the newest and the wildest left out

The book keeps every rule that ever passed, paper-traded from the night it was found: 8 still passing, 3 faded.

§ 03 · What it threw away

What it threw away: the closest misses

as of · neither proven nor kept

The rules that came closest, in full. Each missed a cut: the dot is its luck test on the older part (against the first cut's bar), the diamond on the newer part (against its own bar).

The closest misses' luck tests, older part then newer part: #1 5.2 then 1.5; #2 5.1 then 1.0; #3 3.1 then 2.6; #4 2.4 then 3.2; #5 2.8 then 2.7; #6 0.1 then 5.4; #7 1.9 then 3.4; #8 3.7 then 1.5first cut 3.2newer part 1.60#1#1 · not kept · older part: luck test 5.2 (bar 3.2) · newer part: +1.5 (bar 1.6) · whole history +0.35% a period, 1,112 periods#1 · not kept · older part: luck test 5.2 (bar 3.2) · newer part: +1.5 (bar 1.6) · whole history +0.35% a period, 1,112 periods#2#2 · not kept · older part: luck test 5.1 (bar 3.2) · newer part: +1.0 (bar 1.6) · whole history +0.48% a period, 1,112 periods#2 · not kept · older part: luck test 5.1 (bar 3.2) · newer part: +1.0 (bar 1.6) · whole history +0.48% a period, 1,112 periods#3#3 · not kept · older part: luck test 3.1 (bar 3.2) · newer part: +2.6 (bar 1.6) · whole history +2.16% a period, 206 periods#3 · not kept · older part: luck test 3.1 (bar 3.2) · newer part: +2.6 (bar 1.6) · whole history +2.16% a period, 206 periods#4#4 · not kept · older part: luck test 2.4 (bar 3.2) · newer part: +3.2 (bar 1.6) · whole history +0.48% a period, 397 periods#4 · not kept · older part: luck test 2.4 (bar 3.2) · newer part: +3.2 (bar 1.6) · whole history +0.48% a period, 397 periods#5#5 · not kept · older part: luck test 2.8 (bar 3.2) · newer part: +2.7 (bar 1.6) · whole history +2.44% a period, 206 periods#5 · not kept · older part: luck test 2.8 (bar 3.2) · newer part: +2.7 (bar 1.6) · whole history +2.44% a period, 206 periods#6#6 · not kept · older part: luck test 0.1 (bar 3.2) · newer part: +5.4 (bar 1.6) · whole history +2.01% a period, 111 periods#6 · not kept · older part: luck test 0.1 (bar 3.2) · newer part: +5.4 (bar 1.6) · whole history +2.01% a period, 111 periods#7#7 · not kept · older part: luck test 1.9 (bar 3.2) · newer part: +3.4 (bar 1.6) · whole history +0.15% a period, 2,746 periods#7 · not kept · older part: luck test 1.9 (bar 3.2) · newer part: +3.4 (bar 1.6) · whole history +0.15% a period, 2,746 periods#8#8 · not kept · older part: luck test 3.7 (bar 3.2) · newer part: +1.5 (bar 1.6) · whole history +0.29% a period, 1,112 periods#8 · not kept · older part: luck test 3.7 (bar 3.2) · newer part: +1.5 (bar 1.6) · whole history +0.29% a period, 1,112 periodsThe closest misses' luck tests, older part then newer part: #1 5.2 then 1.5; #2 5.1 then 1.0; #3 3.1 then 2.6; #4 2.4 then 3.2; #5 2.8 then 2.7; #6 0.1 then 5.4; #7 1.9 then 3.4; #8 3.7 then 1.5first cut 3.2newer part 1.60#1#1 · not kept · older part: luck test 5.2 (bar 3.2) · newer part: +1.5 (bar 1.6) · whole history +0.35% a period, 1,112 periods#1 · not kept · older part: luck test 5.2 (bar 3.2) · newer part: +1.5 (bar 1.6) · whole history +0.35% a period, 1,112 periods#2#2 · not kept · older part: luck test 5.1 (bar 3.2) · newer part: +1.0 (bar 1.6) · whole history +0.48% a period, 1,112 periods#2 · not kept · older part: luck test 5.1 (bar 3.2) · newer part: +1.0 (bar 1.6) · whole history +0.48% a period, 1,112 periods#3#3 · not kept · older part: luck test 3.1 (bar 3.2) · newer part: +2.6 (bar 1.6) · whole history +2.16% a period, 206 periods#3 · not kept · older part: luck test 3.1 (bar 3.2) · newer part: +2.6 (bar 1.6) · whole history +2.16% a period, 206 periods#4#4 · not kept · older part: luck test 2.4 (bar 3.2) · newer part: +3.2 (bar 1.6) · whole history +0.48% a period, 397 periods#4 · not kept · older part: luck test 2.4 (bar 3.2) · newer part: +3.2 (bar 1.6) · whole history +0.48% a period, 397 periods#5#5 · not kept · older part: luck test 2.8 (bar 3.2) · newer part: +2.7 (bar 1.6) · whole history +2.44% a period, 206 periods#5 · not kept · older part: luck test 2.8 (bar 3.2) · newer part: +2.7 (bar 1.6) · whole history +2.44% a period, 206 periods#6#6 · not kept · older part: luck test 0.1 (bar 3.2) · newer part: +5.4 (bar 1.6) · whole history +2.01% a period, 111 periods#6 · not kept · older part: luck test 0.1 (bar 3.2) · newer part: +5.4 (bar 1.6) · whole history +2.01% a period, 111 periods#7#7 · not kept · older part: luck test 1.9 (bar 3.2) · newer part: +3.4 (bar 1.6) · whole history +0.15% a period, 2,746 periods#7 · not kept · older part: luck test 1.9 (bar 3.2) · newer part: +3.4 (bar 1.6) · whole history +0.15% a period, 2,746 periods#8#8 · not kept · older part: luck test 3.7 (bar 3.2) · newer part: +1.5 (bar 1.6) · whole history +0.29% a period, 1,112 periods#8 · not kept · older part: luck test 3.7 (bar 3.2) · newer part: +1.5 (bar 1.6) · whole history +0.29% a period, 1,112 periods

older partnewer partthe bar each must pass

  1. Rank the coins by the last 3 days' move: buy the bottom 20% and short the top 20%, for 1 day, only while Bitcoin is below its 200-day average.

    luck test 5.2 on the older part (trimmed 5.5), 1.5 on the newer · 1,112 periods · +0.35% a period (trimmed +0.31%, middle +0.36%) · 58% won · passed the first cut, failed the newer part

  2. Rank the coins by the last 3 days' move: buy the bottom 10% and short the top 10%, for 1 day, only while Bitcoin is below its 200-day average.

    luck test 5.1 on the older part (trimmed 5.4), 1.0 on the newer · 1,112 periods · +0.48% a period (trimmed +0.44%, middle +0.42%) · 57% won · passed the first cut, failed the newer part

  3. Rank the coins by the last 30 days' move: buy the top 20% and short the bottom 20%, for 7 days, only while Bitcoin is above its 200-day average, leaving out the majors, the thinnest, the newest and the wildest coins.

    luck test 3.1 on the older part (trimmed 3.1), 2.6 on the newer · 206 periods · +2.16% a period (trimmed +1.51%, middle +1.11%) · 61% won

  4. Rank the coins by their Binance funding over the last week: buy the top 10% and short the bottom 10%, for 1 day, leaving out the majors, the thinnest, the newest and the wildest coins.

    luck test 2.4 on the older part (trimmed 2.5), 3.2 on the newer · 397 periods · +0.48% a period (trimmed +0.46%, middle +0.50%) · 60% won

  5. Rank the coins by the last 30 days' move: buy the top 10% and short the bottom 10%, for 7 days, only while Bitcoin is above its 200-day average, leaving out the majors, the thinnest, the newest and the wildest coins.

    luck test 2.8 on the older part (trimmed 2.9), 2.7 on the newer · 206 periods · +2.44% a period (trimmed +1.87%, middle +1.69%) · 59% won

  6. Rank the coins by their own volatility over 60 days, Bitcoin's part taken out: buy the bottom 10% and short the top 10%, for 14 days, only while Bitcoin is above its 200-day average.

    luck test −0.1 on the older part (trimmed 0.8), 5.4 on the newer · 111 periods · +2.01% a period (trimmed +3.53%, middle +4.83%) · 68% won

  7. Rank the coins by how closely they moved with Bitcoin over 60 days: buy the top 20% and short the bottom 20%, for 1 day.

    luck test 1.9 on the older part (trimmed 2.8), 3.4 on the newer · 2,746 periods · +0.15% a period (trimmed +0.15%, middle +0.11%) · 52% won

  8. Rank the coins by their distance from the 50-day average: buy the bottom 20% and short the top 20%, for 1 day, only while Bitcoin is below its 200-day average.

    luck test 3.7 on the older part (trimmed 4.2), 1.5 on the newer · 1,112 periods · +0.29% a period (trimmed +0.25%, middle +0.18%) · 54% won · passed the first cut, failed the newer part

§ 04 · How it is tested

How it is tested

as of · methodology

A luck test (t) says how far a result sits from zero, in units of its own noise: about 2 is rarely luck for one test, but many tests at once need more, which is what the first cut is for.

Every rule, in every pool and by two detectors (the plain average and an average with the best and worst tenth of periods left out), goes through one 5% false-discovery cut on the older 60% of its own periods. Then, on the newer 40% it was never chosen on, it must earn on the same side with a luck test of 1.645 or more, every period counted: a trader cannot skip the worst ones after the fact.

Every result is before fees. Our price history holds today's coins: coins that died before we began collecting are missing, which flatters a long side and hurts a short side the further back it goes.

§ 05 · Revisions1 dated

Revisions

as of · every change to its headline numbers, dated

This study is re-run every night on the whole history. It was first written on Oct 6, 2026; its numbers last moved on Oct 6, 2026. Each night its headline numbers moved is listed here, newest first.

  1. 3,144 tests · 55 past the first cut · 8 held · 8 in the book

§ 06 · More

More research

as of