Three point six cents of live cover for every dollar already stolen [1] is where any underwriting decision on this book starts, and the more useful version of the same arithmetic is the temporal one: at a realised loss run of roughly $191 million a month, the whole outstanding capacity of the on-chain market amounts to about twenty-one days of theft [2]. That prices the product honestly, and it means the outstanding cover is a rounding item in a treasury's loss distribution, not a transfer of risk.
The exits are the informative part. CoinGecko's stated reasons run to high premiums and an inability to find capital providers willing to underwrite [4], which is an underwriter saying the tail cannot be fitted, and the concentration data shows why: ten incidents carried 72.5% of the value [8], about $2.63 billion, or $263 million each, roughly double every dollar of cover outstanding anywhere on-chain [3]. No underwriter writes that kind of exposure against $130 million of capital.
The tail is a different business, or rather the more interesting version of this story sits in the tail: take the top ten out and 235 incidents share about $1.0 billion, averaging $4.25 million apiece [4], which is an ordinary small-limit book that a $130 million pool could carry many times over. Five of nine protocols left anyway [3], which says the premium required to cover adverse selection on that book exceeded what buyers would pay [4]. Price and value diverged, and the sellers walked away from the table.
One coincidence worth keeping in view: capacity fell by $33.0 million between the two readings, and cumulative payouts across the market's entire life sit near the same $33 million [5]. The market has withdrawn cover equal to every dollar it has ever paid a claimant. What remains excludes phishing, private key theft, employee error, market volatility and losses on unsupported chains [10], so the vectors that an operator's own controls actually govern sit outside the policy whether or not the policy is bought.
Binance's self-funded reserve of about $1.16 billion [11] is 32% of the industry's nineteen-month loss total [6], a real number backed by an actual balance sheet rather than a marketing figure, and it is also a reserve funded by the balance sheet it protects, correlated with the event it exists to survive, which is precisely the property an insurance policy is bought to break. For everyone below that size [12], the available risk transfer is the design of custody itself, paid for in engineering rather than in premium.
One caveat is worth flagging here: CoinGecko is counting on-chain protocols [3], so if traditional carriers or captive structures grew their crypto limits over the same window, the story is migration and this report does not measure it. The second way to be wrong is arithmetical. If the coming nineteen months produce a fraction of $3.63 billion [5], the same $130 million [1] becomes a materially different ratio with no new capital raised at all. Until one of those shows up, the honest planning assumption is that no policy exists at size, and the reserve is the product.
