Methodology

How each number is produced, and which tier of evidence it sits on.

Every figure on this site is produced by Genesis, a research program that spent months testing whether leveraged-market signals survive contact with evidence. Most did not, and those refutations are published alongside the findings that held. Isobath is the instrument that program built; the record of what it got wrong is on the record page, and it is not curated.

The provenance ladder

Every figure on this site carries one of four tiers. There is deliberately no predicted tier.

Defensibility

The share of a cluster's notional held by wallets whose withdrawable is exactly zero. They cannot move their liquidation price without depositing from outside. This requires the venue's own figure — the obvious arithmetic misclassifies one wallet in five. The measurement →

Coverage

Scanned position notional divided by exchange open interest, reported for this scan rather than the best scan we have ever run. The current map states 34%; a full universe reaches an estimated 53.3%. Publishing the better number on a narrower scan would be exactly the dishonesty this site exists to avoid.

Which of these figures survive our own sampling

The position scan runs in two tiers: the whole frozen universe every few hours, and the 300 largest positions every hour. We paired them and measured what changes between the two, rather than assuming the faster one is a smaller version of the slower one.

Coverage and dollar totals survive. Paired scans agree on coverage to within a point and a half, and on forced notional to within about a tenth — because the notional sits in a handful of very large accounts that both tiers see.

Anything counted in positions does not. The hourly tier holds 28% of its positions within 10% of liquidation; the full scan holds 47%. Being a large account correlates with being a safe one, so ranking by size selects for distance from liquidation. Every figure on this site that counts positions states which tier produced it, and figures from different tiers are not comparable. The measurement →

Scale — the denominator

A cluster's size means nothing on its own. What decides whether it can move the market is its size against the liquidity standing in front of it, and no liquidation map publishes that comparison — including, until recently, this one.

We read the venue's own order book at three significant figures, which aggregates it into hundred-dollar buckets and reaches roughly ±2.5% of mid. Within ±1% that currently stands at $255.5M. Every cluster on the map is divided by it. Beyond the book's reach the figure is a proxy — we use the book at spot, because the book that will exist at a price several percent away is not something anyone can observe today, and the badge says estimated rather than pretending otherwise.

This scale comparison is consistent with CASCADE-1's result: its liquidation episodes did not add continuation beyond matched volatile minutes. It is not by itself a causal proof. The median measured cluster is under half a percent of the book (F-0014).

Map age

The position scan runs on a slower cadence than the order book updates. Every figure carries how old the underlying scan is, because a map presented as current when it is an hour old is precision that does not exist.

What we do not publish