Markets whose oracles are privately operated, watched continuously.
Hyperliquid's builder-run HIP-3 perps trade against oracles their deployers control. The Terminal reads every number, spread, depth, mark-to-oracle divergence, activity, as a percentile of that market's own recorded distribution for the current session phase, and keeps a ledger of every sustained excursion past the 99th. A quiet grid is the normal state.
How to read this
Percentiles. Every number is ranked inside that market's own last 28 days for the current session phase. A spread at p67 is wider than 67% of recorded readings; p99+ is beyond nearly all of them. Depth reads the other way: a low percentile means a thinner book than usual.
Gauges. All gauges point the same way: the further right the needle, the more unusual the reading in the direction that worsens market quality (wider spread, thinner book, larger divergence). The two ticks mark p95 and p99.
Chips. A card only gets a chip when a reading stays unusual: stretched after 60 seconds beyond p95, anomalous after 90 seconds beyond p99.5. A quiet grid is the normal state, by design.
Episodes. An episode is a bigger event: past p99.5 sustained for five minutes or more. Each market's detail view lists its episodes with duration, peak and how much traded during each one.
Mark-oracle. The gap between the exchange's mark price and the deployer-operated oracle, in basis points, signed. It is only judged when the underlying market is open; a frozen weekend oracle is context, not an event.
Digging in. The table ranks markets by their most unusual reading right now. Click any row or card for bands, oracle behaviour, the typical day and the episode ledger.
| Market | Most unusual now | Percentile | Episodes 7d | Last episode |
|---|---|---|---|---|
| HYPE | – | – | 1 | 1d ago |
| xyz:AAPL | – | – | 0 | 10d ago |
| xyz:AMZN | – | – | 1 | 4d ago |
| xyz:BRENTOIL | – | – | 4 | 2d ago |
| xyz:CL | – | – | 6 | 2d ago |
| xyz:GOLD | – | – | 1 | 5d ago |
| xyz:GOOGL | – | – | 1 | 3d ago |
| xyz:META | – | – | 0 | 8d ago |
| xyz:NVDA | – | – | 0 | 12d ago |
| xyz:SILVER | – | – | 2 | 2d ago |
| xyz:SP500 | – | – | 3 | 6d ago |
| xyz:TSLA | – | – | 0 | 8d ago |
| xyz:XYZ100 | – | – | 2 | 6d ago |
These are market-quality observations, not recommendations. Live values come from Hyperliquid's public API, fetched by your browser and smoothed over the last three 30-second readings. Percentiles are against pooled row-level distributions from our own continuous recording: 28 completed days, 13 Jul 2026 to 9 Aug 2026, per session phase (ET). Flags are one-sided and sustained: stretched means beyond the 95th percentile of the risk tail for 60 seconds or more (an attention marker, expected on a card or two most of the time), anomalous means beyond the 99.5th for 90 seconds or more, and divergence is judged only in phases where the underlying market is normally open (a frozen oracle is context, not an event). The episode ledger requires more: past the 99.5th sustained for five minutes or longer, nearby runs merged, applied to every recorded row since 10 May 2026 (89 days). Baselines computed 2026-08-10; young markets drift, so bands use a trailing window rather than the full history. Data is "as is" and may be wrong or delayed; nothing here is investment advice or an invitation to trade.