Cluster hit rate
A public, after-the-fact check of the liquidation clusters: of all big clusters recorded at least N hours ago, how many did the Hyperliquid mark price touch within N hours.
Definition
A public, after-the-fact check of the liquidation clusters: of all big clusters recorded at least N hours ago, how many did the Hyperliquid mark price touch within N hours. Published for every coin and for all coins together.
In one sentence: Share of large liquidation clusters that price actually reached within the time window after we recorded them.
Current data · all markets
no data yet| horizon | clusters | reached | hit rate |
|---|---|---|---|
| 4 h | 375 | 17 | 5% |
| 24 h | 252 | 23 | 9% |
From the data shown above: First row: 4 h, clusters 375, reached 17, hit rate 5%.
How we compute it
Formula
hit rate (N h) = clusters reached within N hours ÷ clusters recorded at least N hours ago
reached (long) = mark ≤ top edge of the bucket · reached (short) = mark ≥ lower edge- Every 15 minutes the 3 largest buckets within ±15 % of the mark holding at least $5M each are recorded; the same bucket is recorded at most once per 24 hours.
- A cluster counts as reached when the mark touches the bucket (longs: mark at or below the top edge; shorts: mark at or above the lower edge), checked against 30-second mark samples.
- An observation counts for a horizon only once it is at least that old; hit rate = reached ÷ clusters.
How to read it
- It measures how often clusters were reached in the past, with the sample size next to it.
- It is a description of past data, not a probability for the next cluster.
Common mistakes
- Reading the hit rate as a probability for the next cluster. It counts past observations only.
- Ignoring the sample size. A rate from a few dozen clusters moves a lot with every new observation; the scoreboard shows the count next to every rate.
Honest limits
- Wicks between two 30-second samples are not seen, so the hit rate is a lower bound.
- Early weeks have few observations; small samples swing a lot.
Questions and answers
Why can the hit rate be lower than what really happened?
The mark price is sampled every 30 seconds. A wick that touches a cluster and returns between two samples is not seen, so the published rate is a lower bound.
Use it live
Step by step
API
Not in the REST API yet.
Every key uses the data of its plan; see the developer docs for parameters and examples. Developers · API docs
Plan availability
- Free: every coin, current values and 30 days of history.
- Pro and above: history up to 1095 days, CSV export.
Terms defined here
Cluster hit rate
Methodology. Facts from public data. Not investment advice. · Report incorrect data