Research

An evidence library, not a blog. Every finding carries its method, its sample and its limits — and the ones we later refuted stay published.

• One leveraged wallet in five is misclassified when you ignore free collateral
  74 wallets holding open BTC positions. Naive margin maths says 96% could defend their position. The venue says 76%.…
• Twenty-five crypto assets are worth 2.65 independent bets
  60.4% of all variance in a basket of liquid perpetuals is a single common factor. Diversifying across coins mostly buys …
• Crypto order books lose 15% of their depth during large moves
  1,324 days of Binance order-book depth, 3,733,943 snapshots, zero missing days. In quiet markets the book is unchanged t…
• The spread was 0.00154 bps. The cost was 5.19.
  And a zero maker fee does not rescue it. We checked, because that is the first thing anyone assumes.…
• The number everyone would put on the front page turns out to be a constant
  433 published clusters over three days of Hyperliquid BTC. 71% sit at exactly 100% cannot-defend. Splitting at the media…
• Your backtest needs 13 years to prove what it claims
  We found this by discovering our own tests were underpowered by one to two orders of magnitude. The best result in one e…
• Liquidation maps show where forced selling is. We tested whether it moves price.
  228 independent episodes across 28 symbols. Forced flow returned +40.07 bps at 15 minutes and beat a permutation null. A…