Measured data from live collectors. Every figure is dated and reproducible.
overhang
Collateral exit overhang — 2026-09-03
Published 2026-09-03 · sample n=22370
The median reserve on xStocks, kamino, can exit only 19.6% of its marked value, measured across 10 symbols.
Detail
GOOGLx — 4.7% of $3.0M marked is exitable
TSLAx — 14.1% of $2.8M marked is exitable
MSTRx — 14.1% of $1.7M marked is exitable
HOODx — 14.3% of $473k marked is exitable
SPYx — 15.6% of $4.1M marked is exitable
NVDAx — 23.5% of $2.6M marked is exitable
QQQx — 25% of $2.9M marked is exitable
AAPLx — 33% of $400k marked is exitable
CRCLx — 100% of $595k marked is exitable
cbBTC — 100% of $181k marked is exitable
Separate market, separate mechanism — Superstate Opening Bell: all 2 reserves measured never routed at any size. FWDI (0% of $16.2M marked, n=2237), USCC (0% of $4.4M marked, n=2237). These do not route because their tokens are transfer-permissioned: Kamino's liquidate instruction is permissionless, so a liquidator must already be an allowlisted wallet to receive the seized collateral. That is a permissioning constraint, not a thin book — stated here as the mechanism, not measured by this collector.
Not counted as no-exit because the failure was intermittent: CRCLx (1/2237 sweeps), cbBTC (4/2237 sweeps) — a symbol must fail in every sweep before its market is called absent.
Those no-exit readings include sweeps taken while the US market was open, so the absence is not an off-hours artifact.
Every percentile is computed per symbol first, then taken as the median across the 10 symbols on xStocks alone — nothing is pooled across symbols and nothing is pooled across markets.
Only a structural exhaustion of the sell ladder counts as no exit. Each sweep carries a SOL control quote; sweeps whose control did not route, or was flagged transient, are excluded as unmeasured rather than reported as an absent market.
Readings before 2026-08-11 are excluded in the query: they were computed against a frozen mark and are not comparable.
Method note: figures are computed from a live database at publication time and never revised afterwards. If a later measurement disagrees, both stay published.