Nothing here is estimated, sampled or smoothed. The pot is a balance read from the chain at a named instant; the claims are entries in an append-only log; the awards are integer division with the remainder accounted for. If a figure cannot be derived from those three things, it is not on the site.
Given a pot and a set of claims, flat pro rata pays each claimant the fraction of the pot that their claim is of every claim. There is no per-unit rate anywhere in that sentence. The rate people quote is pot ÷ claimed, which is the answer only while the claiming has stopped.
So the site prints the quoted figure and, beside it, what it actually pays. They are the same number exactly once: when nobody else claims.
quoted = pot ÷ Σ claimed paid_i = claim_i × pot ÷ Σ claimed Σ paid = pot ← asserted, exactly
Lamports are integers and shares of them are not. Floor every share and the total comes up short — one lamport left holding, balancing nowhere. Round to nearest instead and the total can exceed the pot, which is the worse failure: it balances on paper and overdraws in fact.
Every share is floored, then the remainder is handed out one lamport at a time by largest fractional part. The sum is asserted equal to the pot for every settlement, at every size — not sampled.
Ties are broken on the claim's own identifier, never on its position in the list. Broken on position, a log replayed in a different order settles a different way on a different day, and both answers look correct.
claims 100 100 100 floor 3 3 3 → 9, one short largest 4 3 3 → 10, exact
Abatement by class does not reduce everyone together. The residuary is exhausted first, then general legacies, then demonstrative; a specific legacy is touched last. The class the estate runs out in abates within itself, pro rata, and every class below it receives nothing.
Reverse that order and every total still ties out to the lamport — the wrong people are simply paid. So the order is tested as a fact of its own, on a pot chosen to run out between two classes. Tested on a pot that runs out in the first class, the comparison passes while never being exercised at all.
pot 150, four claims of 100 specific 100 paid whole demonstrative 50 abates here general 0 residuary 0
Settle the pot without the arrival, settle it with them, subtract. Every lamport the arrival is paid comes out of somebody already there, and the register returns the loss per claimant rather than a total — because a total lets each person assume it was somebody else's.
Abatement pays nobody. It reads a public balance and divides it on paper; nothing is held, signed or transferred, and no award on this site entitles anybody to anything. A claim is a public address and an amount — a request carrying anything key-shaped is refused before it is read, and the refusal says so.
