Descending triangle
Bottoms on the same floor with falling tops show ever more aggressive selling: the floor gives way.
Measured in crypto — Binance spot · 540 pairs, delisted ones included · 0.2% per round trip
- ✓invariance
- ✓costs
- ✓placebo
- ✓benchmark
- ?out of sample
- ✓multiple testing
What was measured
How many of those trades actually count
What it paid, before any deductions
Where it stalled · does it come back in the pieces?
The effect sits at the start of the period, not the end
What this result does NOT say
- One market, one universe
- Measured on 507 spot cryptocurrency pairs, delisted ones included. It says nothing about futures, equities or indices, nor about how the same technique behaves in another market.
- One window of time, not every window
- The measured period runs from 2017-08-17 to 2026-07-28. A market moves through regimes, and a technique can work in one and fail in another — the card measures the regimes that fit inside this window, not the ones still to come.
- One cost structure
- The cost charged is 0.10% per leg, in and out. Anyone paying more than that gets a worse result, and anyone paying less gets a better one — the verdict holds for this fee.
- One exit rule
- The trade was closed by: the auditor's fixed horizon (10 bars). The same entry measured with a different exit is a different strategy, and can earn a different verdict — it happens in this archive.
- The number of trades is not the sample size
- There are 3,543 trades, but only 247 independent market episodes: a single move fires the technique across dozens of assets at once, and counting those as separate observations inflates any result. It is the smaller number that governs the arithmetic. With 247 episodes, what the data supports is a range from +0.97% to +3.63% per trade — the published average is the centre of it, not the exact measurement.
- Daily bars
- Measured at the daily close. Nothing here measures what happens inside the day, and an intraday technique is not auditable with this data.
Numbers and reproducibility
The six controls
| control | status | t | threshold | episodes |
|---|---|---|---|---|
| invariance | passed | — | — | — |
| costs | passed | 3.12 | 1.97 | 247 |
| placebo | passed | 2.56 | 1.97 | 247 |
| benchmark | passed | — | — | — |
| out of sample | inconclusive | — | — | — |
| multiple testing | passed | — | — | — |
- invariance3543 signals across 750934 bars
- costsgross +2.303% · cost 0.200% · net +2.103% (t=3.12)
- placeboactual +2.103% · placebo -0.285% · excess +2.389% ± 0.932% (t=2.56 against a threshold of 1.97, 247 real groups, 177,150 sham dates, draw error ±0.044%)
- benchmarktechnique +2.10% · buy and hold (same horizon) +0.08% · excess +2.03%
- out of sampleasset half A: +1.650% (t=2.09, 220 episodes) · asset half B: +2.549% (t=3.43, 204 episodes) · liquid half (>= US$ 2,059,212/day): +1.842% (t=2.66, 214 episodes) · illiquid half: +2.368% (t=2.76, 198 episodes) · period 1/4 (2018-01-16 a 2022-09-15): +3.952% (t=3.69, 117 episodes) · period 2/4 (2022-09-16 a 2023-10-04): +3.861% (t=3.23, 36 episodes) · period 3/4 (2023-10-05 a 2025-06-20): -0.221% (t=-0.15, 58 episodes) · period 4/4 (2025-06-21 a 2026-07-18): +0.823% (t=0.56, 39 episodes)The edge decayed: +3.95% (t=3.69) in the first quarter of the period, +0.82% (t=0.56) in the last. The partitions that replicate above are across assets, not across time.
- multiple testing1 variation(s) tested · t=3.12 across 247 episodes (equivalent to t=3.10) · p≈0.0019 · false positives expected by chance ≈ 0.00
Reproducibility
- period
- 2017-08-17 to 2026-07-28
- assets that traded
- 507
- variations tested before this one
- 1
- gross per trade
- +2.30%
- net per trade
- +2.10%
- exit rule
- the auditor's fixed horizon (10 bars)
- horizon bars
- 10
- fee per leg
- 0.001
- episode days
- 10
- confirmation bars
- 5
- tolerance
- 0.03
- max window
- 120
Binance spot klines (delisted pairs included) · collected from 2026-07-27 23:31 to 2026-07-28 14:29 · 540 assets · 750,934 bars · 2017-08-17 to 2026-07-28
Seed not recorded: audits before 2026-07-29 used 20260728 by convention in the code, and the value is not in the record.
Hypothesis, filed before the result
Chart patterns are the best-known class in technical analysis and the one most damaged by the pivot error: almost every published backtest marks the signal on the pivot's own date, when on that date nobody knew it was one. Measured with a CAUSAL pivot — five bars of confirmation lag, which is what a real trader faces — I expect them to come out WORSE than the candlestick patterns, and for a legitimate reason: by the time the break is recognisable, the price has already moved. Specific, falsifiable prediction: none survives the Benjamini-Hochberg correction across the 52 patterns, and the control that kills the most will be the benchmark, not costs — unlike the candlesticks, because these trade far less often.
filed on 2026-07-28, before the number existed
The original, as it was filed
Padrões gráficos são a classe mais famosa da análise técnica e a que mais sofre com o erro de pivô: quase todo backtest publicado marca o sinal na data do pivô, quando naquela data ninguém sabia dele. Medidos com pivô CAUSAL — 5 velas de atraso de confirmação, que é o que o operador real enfrenta — espero que saiam PIORES que os de vela, e por um motivo legítimo: quando o rompimento é reconhecível, o preço já andou. Previsão específica e falsificável: nenhum sobrevive ao Benjamini-Hochberg de 52 padrões, e o controle que mais mata será o benchmark, não o custo — diferente dos de vela, porque estes operam bem menos vezes.
Pre-registration exists to keep prediction apart from rationalisation: written after the number, every hypothesis is right.
Earlier audits of the same technique
Each variation an author teaches enters as its own test, so that whatever might work in the strategy gets covered. The verdict held in all of them.