Research note · the stuff we kill

We tested chasing high-volatility breakouts. We don't anymore.

On a day BTC popped +2.4% in two hours, our signal said avoid — and we got asked why we keep missing moves. So we ran the test. Here's the whole thing, including the control we ran on our own result.

Method: forward returns on raw 15m candles Window: 26 May – 15 Jun 2026 Markets: BTC · ETH · SOL Type: manual backtest, not an auto-generated signal

The claim we tested

The intuition is seductive: volatility explodes, a breakout forms on the hourly, price is moving — get in. The claim, stated plainly: "following our directional lean during a high-volatility breakout has positive edge." If true, we should loosen the gate and stop sitting these out.

We can't measure this from our own published signals — we avoid high-volatility regimes by design, so there are zero live trades there to score. So we measured the counterfactual directly: at every 15-minute snapshot where the primary regime was HIGH_VOLATILITY and a higher timeframe showed BREAKOUT_FORMING, we computed the forward price return in the direction our engine was leaning.

The result

37%
win rate 1 hour in, following the lean (a coin flip is 50%, and fees mean you need well above it)
43%
win rate 4 hours in — still under water
HoldFollow-lean returnWin ratet-stat
1 hour−0.13%37.2%−3.78
4 hours−0.38%42.7%−4.41
8 hours−0.25%48.2%−2.33
24 hours+0.64%62.3%+2.96

n = 199 raw snapshots (26 independent after decorrelating to one sample per 4 hours). The t-stat measures whether the average return is meaningfully different from zero.

For the first one to eight hours — the window where you'd actually be holding a breakout trade — following the lean loses money, significantly (t ≈ −4). The "breakout" during high volatility fails more often than it follows. It's a trap dressed as an opportunity: exactly the kind of violent move that's as likely to reverse as continue.

The control we ran on ourselves

Here's the part most backtests skip — and the part that matters most. The same data showed high-volatility trades looking +1.5% at 24 hours with a 70% win rate. We almost believed it. Then we checked whether that was a real edge or an artifact of the period.

The window was a one-way market: BTC −15%, ETH −19%, SOL −15% over three weeks. And our lean was short 75% of the time. So we compared, at the 24-hour horizon:
Strategy in HIGH_VOL24h returnWin rate
Follow our lean+1.48%70%
Just always short (ignore the engine)+1.75%66%
Just always long−1.75%34%

Following our lean was no better than blindly shorting everything. The 70% "win rate" wasn't directional skill — it was a falling market plus a short bias. In an up-trending window the same logic would have lost. The lean carried no independent information. This is exactly what we've said since launch: direction is an honest coin flip, and a three-week trend can't overturn that.

What we concluded

Keep avoiding high-volatility breakouts.

The setup we were tempted to chase is the one our data says loses in the near term. The apparent longer-horizon "edge" is survivorship, not skill. So the gate stays shut on these — not out of caution, but because we tested it and the evidence said so.

This is the same discipline that runs the other way too: a month earlier, the data told us we were sitting out boring markets we should have been engaging, and we opened the gate for those. We let the data decide both times. That's the entire job.

See the public track record → What else we're testing

Educational only — not investment advice. This is a manual research backtest over a single ~3-week window, run to answer an operator question; it is not one of our automatically-generated, continuously-tracked hypotheses (those live on the hypotheses page). Past behavior of any strategy does not predict future results.