Risk score

One number for how much can go wrong, built from risk alone and never from return. Higher is worse.Note 13

What this score is not

It is not a rating of how good a strategy is. Return is not an input — not as a component, not as a tie-break, not inverted. A number that mixed performance in would rank strategies by quality, and a quality ranking is the leaderboard this site exists to counterbalance. Sorting this page does not tell you what to buy; it tells you what could hurt.

It is not the provenance grade. How well evidenced a strategy is lives in evidence, deliberately apart: averaged together, a thoroughly documented risky vault would score like an undocumented safe one and you could not tell which you were holding.

The weights are a judgement, not a finding. No data set could give the right ones — that would need many observed failures with known causes, and the missing record of failures is this site’s whole premise. They are published below so you can disagree, and every component is shown separately so you can ignore the total and read the parts.

How it is built

Six components, each already published on its own page here — the score adds no measurement of its own, so anything it says can be checked against the page that says it. Each is read onto a 0–1 scale where 1 is the worst, multiplied by its weight, and the total renormalised over whatever could be measured.

Worst fall suffered · 25%: Peak-to-trough on the holder-return curve, so a large withdrawal is not counted as a loss. A 50% fall is the worst reading: it needs a double to recover, and past that the difference between bad and catastrophic stops changing what anybody would do. It carries the most weight because it is the only component that has already happened — the rest are exposures that may never cost anything.
Shared exposure · 20%: How one-sided the crowd is around the instruments this vault holds, from crowding. Everybody holding the same instrument in opposite directions is not this risk; everybody holding it the same way is.
Distance to liquidation · 17.5%: The nearest adverse move that would breach maintenance margin, from liquidation stress. A cushion of 25% or more reads as no risk from this component; nothing left reads as the worst.
Could you get out · 17.5%: What share of the vault could be withdrawn at the last reading — for a lending vault, whatever the markets it lends into have not lent on. A quarter free or more reads as no risk from this component; nothing free reads as the worst. It carries the same weight as distance to liquidation because it is the same risk in another venue’s vocabulary: a leveraged book is closed out by its margin, a lending vault is held by its utilisation, and almost no vault is exposed to both. Null where the venue does not say, which is not the same as none.
Largest depositor · 12%: The share one wallet holds. Below a quarter it stops deciding whether a single withdrawal breaks the strategy; at everything, the vault is one person’s mind.
Stopped operating · 8%: The share of observed days the strategy actually traded. A vault whose operator switched the bot off has a flat return and no risk showing anywhere else.

A component we cannot measure does not vote — it is neither counted as safe nor as dangerous, and the share of Scoring covers 77% of the capital on this page but only 48% of the vaults, and the gap is not an accident: a score needs a return history, so the vaults that go unscored are disproportionately the new and the small. That makes this page a fair description of where the money is and a poor census of how many vaults exist. The score that was measurable is shown beside every row. Below 50% measurable no score is published at all, because a number built mostly from absent components would rank a vault we know nothing about above one we do. That is why 699 vaults here have no score.

How it is built

Worst fall suffered · 25%

Peak-to-trough on the holder-return curve, so a large withdrawal is not counted as a loss. A 50% fall is the worst reading: it needs a double to recover, and past that the difference between bad and catastrophic stops changing what anybody would do. It carries the most weight because it is the only component that has already happened — the rest are exposures that may never cost anything.

Shared exposure · 20%

How one-sided the crowd is around the instruments this vault holds, from crowding. Everybody holding the same instrument in opposite directions is not this risk; everybody holding it the same way is.

Distance to liquidation · 17.5%

The nearest adverse move that would breach maintenance margin, from liquidation stress. A cushion of 25% or more reads as no risk from this component; nothing left reads as the worst.

Could you get out · 17.5%

What share of the vault could be withdrawn at the last reading — for a lending vault, whatever the markets it lends into have not lent on. A quarter free or more reads as no risk from this component; nothing free reads as the worst. It carries the same weight as distance to liquidation because it is the same risk in another venue’s vocabulary: a leveraged book is closed out by its margin, a lending vault is held by its utilisation, and almost no vault is exposed to both. Null where the venue does not say, which is not the same as none.

Largest depositor · 12%

The share one wallet holds. Below a quarter it stops deciding whether a single withdrawal breaks the strategy; at everything, the vault is one person’s mind.

Stopped operating · 8%

The share of observed days the strategy actually traded. A vault whose operator switched the bot off has a flat return and no risk showing anywhere else.