Credit engine
How to read a score. And what it cannot tell you.
One number between 0 and 1000 for every address we have scanned, recomputed daily. This page is what is behind it — the inputs, a worked example, the limits, and who runs it today.
01The scale
0 – 1000, in three bands.
Something in the record does not hold together. The report says which of the three readings pulled it down, and why.
Where most addresses sit. Deliberately rendered without a hue — a 310 and a 590 are different, but neither is a verdict.
A record that holds up across market phases, a consistent identity, and standing beyond trading. Only these addresses can be followed on Copy Trading.
02Inputs
Three readings of the same address.
300+ on-chain features, grouped into three sub-models that never see each other's output.
Gravity Index
How an address behaves under risk, across market phases rather than in a single run.
- Profit quality
- Risk control
- Win matrix
- Market phase
- Leverage discipline
- On-chain footprint
Anything off-chain. Centralised exchange activity, OTC, and private venues are invisible to it.
Social Credit
Who an address is connected to, and whether that identity holds its shape over time.
- Verified connections
- Invitation graph
- Risk sharing
- Identity consistency
Reputation that never touched chain. A well-known trader with a fresh address starts with nothing here.
Eco Engagement
How far an address participates beyond trading — and what it holds while it does.
- Cross-chain activity
- Protocol engagement
- Governance behaviour
- Asset health
Chains and protocols we do not index yet. Absence of activity here is absence of evidence, not evidence of absence.
A composite model turns the three readings into one score.
0x7c4a…9f2e
G 941 · S 872 · E 903
0x5a3c…1f60
G 978 · S 551 · E 842
0x2e91…07bd
G 612 · S 903 · E 889
0xb44f…d150
G 917 · S 566 · E 704
The last two addresses score the same. They are not the same address — and no weighted average could produce both. What is published instead is this page: the inputs, the boundaries, the limits, and who operates it.
03Worked example
One address, read three ways.
A real address with the identity redacted. The data is not redacted.
04Limits
What a score cannot tell you.
Written because the questions get asked anyway, and a page that answers them first is worth more than one that waits.
Every input is something that already happened. A 900 describes how an address has behaved; it does not say what it will do tomorrow, and it is not a recommendation to follow anyone.
One person can run many addresses, and one address can be run by many people. Nothing here identifies a human being, and no score should be read as a judgement about one.
Hyperliquid, recomputed daily. Other venues, centralised exchanges and anything off-chain are outside what the model can see.
An address with no history returns n/a, never 0. A new address is unknown, not bad — and we would rather say so than guess.
05Who runs it today
This is not yet a protocol.
The credit model is run by the HyperTrend team. There is no third party computing or verifying scores.
Ingestion, storage and computation are centralised on infrastructure we control.
The model is not deployed as an on-chain protocol. Scores are served from our systems, not read from a contract.
06Use it
Three ways in.
Look up an address
A full report for any address we have scanned. Sign in to the app; a lookup spends credits.
Launch app →Credit API
Scores and filters for your own product — sybil screening, risk gates, address lists.
BuildingRead the docs
Data sources, refresh behaviour, anomaly detection, and the full write-up of everything on this page.
Docs →