wavelab
wavelab counts Elliott waves on BTCUSDT and, for every count, states the level that would prove it wrong. It is a research engine, not a signal service — and over nine years, its counts could not be told apart from a random twin's.
What nine years said
Every count the engine produced was replayed causally over BTCUSDT 4-hour bars from 17 August 2017 to 13 September 2026, with the entry, the stop and the target it would have drawn at the time. 1,453 trades resolved. Each was paired with a twin entered at random one to twenty bars later, with the same geometry — which is what separates "does the wave count contribute anything?" from "did BTC go up?". Both arms pay the same cost: a 7.5 bp taker fee on each side, charged in R.
| Measure | Elliott | Random twin |
|---|---|---|
| Expectancy per trade, both arms finished (n = 1,451 pairs) | +0.007R | +0.024R |
| Difference, Elliott minus twin, with its 95% interval | −0.017R [−0.092R, +0.054R] | |
| Fibonacci fit vs outcome (n = 1,453 trades) | rho −0.0011, p 0.966 | |
The wave counts did no better than the twin. The difference per paired trade cannot be told apart from zero: its interval, from a stationary block bootstrap, runs from below to above it. Nor is the twin shown to be better. How well a count fitted Fibonacci ratios said nothing about how its trade ended.
Why it still runs
A backtest stops at the last candle in the file. The engine still runs because new bars are the only data that could still change the result: they are out of sample by construction. Its counts are not yet recorded or scored the way the ledger below is, so for now the nine-year result above is the only measured evidence.
The ledger
Beside the waves, the engine runs 88 pre-registered strategies and writes what each one says about every 4-hour bar into a public, sealed record: promised before the bar closes, paired with a coin-flip control, scored from the exchange's public archive, and recomputable by anyone. Since 17 October 2026 each plan is sealed within minutes of its bar's close and published once its eight-day horizon has passed; the first record, kept from 24 September, published each plan soon after writing it and is now closed. The expectation sealed with it is no edge: nine years of backtesting the catalogue found none, as the test above found none for the waves.