ModelsRisk
A quiet price can sit on
a crowded book.
Leverage Stress reads how much borrowed exposure the market is carrying, which side is carrying it, and what moved to change that.
The price barely moved.
24 hours, inside a fifth of a percent
What it
actually reads.
- Perpetual futuresOpen interest, and how fast it is changing
- Funding and basisWhat one side is paying the other to stay in
- Forced closingWhether positions are already being closed out
Observation classes, not vendors. Which venues sit underneath is documented in the API.See what each source covers
Every reading
carries its context.
- "model":
- "leverage_stress",
- "market":
- "BTC",
- "state":
- "elevated", qualitative, never a probability
- "pressure_side":
- "long",
- "scored_against":
- "prior_30d",
- "observed_at":
- "2026-09-19T14:32:00Z", when it was read
- "coverage":
- "monitored_derivatives_venues", what it saw
- "limits":
- "Not a price forecast", what it will not claim
Published fields are in the API documentation.
What it will
not tell you.
- 01It does not forecast a price, a direction or a liquidation.
- 02It reads the venues we record, not the whole market.
- 03Elevated is a comparison against this market's own history, not a universal threshold.
Reach it the way
you work.
- MCP
This reading becomes a tool inside your AI assistant.
Connect an agent - API
The same reading over HTTPS, in your own application.
API reference
About this
model.
Is an elevated reading a signal to sell?
No. It says exposure is concentrated and expensive to hold, which is a description of the state of the book. What that is worth depends on everything the model cannot see — your position, your horizon, and the rest of the market.
What is the score compared against?
The same market's own recent history, not a fixed number. A level that is ordinary for one instrument is extreme for another, and a threshold that ignores that produces alerts that are only ever about which market is busiest.
Which venues are behind it?
The derivatives venues we record continuously. Coverage travels with the reading, so a market outside it comes back marked rather than estimated, and the API documents which venues those are.
Has it been tested?
There is frozen-protocol research behind the current version, and it is research: the first historical run did not clear the bar for a deployed forecast, and that is written down rather than rounded up. What ships today is a reading of the present state, not a track record.
The other
five.
Is someone buying or selling steadily while the price barely reacts?
PositioningPositioning DivergenceWhich groups of participants are starting to take opposite sides?
RiskIn developmentCascade RiskWhere could forced selling or buying outrun the liquidity available to absorb it?
ExpectationsIn developmentExpectations GapDo options, prediction markets and the underlying price the same scenario?
PositioningIn developmentTrader ConvictionDoes this bet actually matter to the trader making it?
One question.
One answer.
With what it read, when it read it, and what it will not claim.