Risk grades
Each market carries a letter. It is a weighted read of several measured axes, calibrated against real liquidation outcomes, and there are things it is known not to catch.
The axes
- oracle
- What prices this market, and how far that price has historically fallen and recovered.
- leverage
- How much room the liquidation threshold leaves before a position is underwater.
- utilization
- How lent-out the loan reserve is, and how sharply its rate has spiked before.
- liquidity
- Whether the position can be exited at size.
- size
- How much is actually in the market.
How the weights were set
Not by taste. They were fitted against a dataset of real liquidation events gathered across every protocol we ingest, and validated on a separate, independent label the fit never saw: markets that ended in bad debt. A proposed re-weighting is only adopted when it improves on both, which has meant rejecting changes the optimiser preferred.
What a grade does not tell you
The liquidation dataset is uneven across protocols, and one major venue contributes no events at all, so the label the weights were fitted against is confounded by which protocols report. And on a specific family of markets the raw score is actively inverted against bad-debt outcomes: what catches those is a separate solvency cap, not the axes.
A grade is a summary of measured things. It is not a prediction, and it is not a substitute for reading what the market actually holds.