Descriptive record — not a forecast. This page records what our liquidation map showed before this cascade (snapshot timestamps shown), what then happened on our multi-venue tape, and the model's measured historical record as it stood before the event. Every accuracy figure carries its sample size and a Wilson 95% lower bound. It does not predict future price moves and does not claim the model forecasts when a cascade will start.
Cascade Autopsy: XAU — 2026-07-16
2026-07-16T13:45:32Z → 13:50:31Z UTC · generated 2026-07-20T17:39:02Z · artifact 2026-07-16T1345-XAU
What happened
Venues
binance, bybit, okx
Direction
predominantly LONG liquidations (79% of events) — price fell
| Symbol | Liquidated | Events | Share |
|---|
| XAUUSDT | $320.0M | 1 | 99.5% |
| SNDKUSDT | $742,837 | 114 | 0.2% |
| ZECUSDT | $263,912 | 2 | 0.1% |
| SOXLUSDT | $227,569 | 34 | 0.1% |
| BTCUSDT | $124,511 | 7 | 0.0% |
Timeline (UTC, minute by minute)
| Minute | Events | Liquidated | Top symbol | Venues |
|---|
| 13:44 | 31 | $112,961 | SNDKUSDT | binance, okx |
| 13:45 | 57 | $202,054 | SNDKUSDT | binance, bybit, okx |
| 13:46 | 111 | $320.7M | XAUUSDT | binance, bybit, okx |
| 13:47 | 11 | $42,597 | SNDKUSDT | binance, okx |
| 13:48 | 83 | $519,859 | SNDKUSDT | binance, bybit, okx |
| 13:49 | 27 | $142,638 | SOXLUSDT | binance, bybit, okx |
| 13:50 | 22 | $23,552 | SNDKUSDT | binance, okx |
| 13:51 | 17 | $33,995 | SNDKUSDT | binance, okx |
Venue breakdown
| Venue | Liquidated | Events |
|---|
| okx | $320.5M | 111 |
| binance | $949,687 | 160 |
| bybit | $177,013 | 26 |
Independent re-read of our own captured tape (verify: PASS). Cross-venue capture is the product moat.
What the map showed BEFORE the cascade
No pre-cascade map snapshot exists for this symbol (level capture not active for it at that time, or outside the retention window). We do not reconstruct or backfill a before-map — that would be look-ahead.
The measured record, as it stood before this cascade
As of 2026-07-16 13:22 UTC (the last calibration record before this cascade, n=15 day-clusters), the model's walk-forward calibration was below its own validation gate (the Wilson 95% lower bound on out-of-sample precision@N did not clear the published target). Per our honesty rules we do not quote the figure itself while it sits below gate — only the status. The level-by-level record above stands on its own as the descriptive evidence for this event.