'Order-book walls are support': 91-94% just get cancelled

Source: Order-book / DOM trading, widespread

What's left is 0.3-0.8 bps against a 6 bps spread.

The few that get eaten carry an effect too small and too fast for anyone but high-frequency firms.

This verdict is about the strategy as published, not about the person who shared it.

The claim

Here’s what the strategy promises, in its own terms.

Large resting bids and offers act as support and resistance; trade against them.

How we tested it

Same six checks as everything else here: random entries, coin flips, a look-ahead audit, unseen years, double costs and a luck check. How we test →

What we found

38 days of Hyperliquid order books (659,000 snapshots on BTC, ETH, SOL, SUI). When a big wall vanished at the touch, it was cancelled 91-94% of the time and consumed 3-5%. Pulled walls carry no information (profit factor 0.96, 0 of 5 walk-forward folds). Consumed walls do (1.26, 5 of 5), but the edge is 0.3-0.8 basis points, dead at a 6 basis-point spread and gone within 5 seconds. Only high-frequency firms can touch it.

The numbers

The headline numbers from our test.

Days	38
Snapshots	659,000
Walls cancelled	91-94
Walls consumed	3-5
Pulled wall PF	0.956
Consumed wall PF	1.259
Edge	0.3-0.8
Half life	<5

• Profit factor: money won divided by money lost. Below 1.0 loses money. Around 1.0 is break-even.
• R-multiple: profit or loss measured in units of the amount risked per trade. +1R means you made what you risked.
• vs random: how the strategy did against the same trades with random entries.