What the Lean code literally says, in plain math · claude-fable-5-1 (independent sub-agent, blind: Lean code + auditor instructions only)
Read-back: BitcoinSelfishMining.relative_revenue
Let α,γ be real numbers with 0<α and α<21 (named, usable hypotheses) and no hypothesis on γ (any real). Then
Rα,γ=C(α,γ),
where both sides are custom quantities, unfolded as follows.
Left-hand side Rα,γ is relativeRevenue αγ:
Rα,γ=rα,γpool+rα,γothrα,γpool,equal to 0 if rα,γpool+rα,γoth=0,
with
rα,γpool=πα(zero)gpool(zero)+πα(tie)gpool(tie)+n=0∑∞′πα(leadn)gpool(leadn),rα,γoth=πα(zero)goth(zero)+πα(tie)goth(tie)+n=0∑∞′πα(leadn)goth(leadn),
each tail being Mathlib's tsum (equal to 0 when the family is not summable). Here πα is stationary α: πα(zero)=(1−2α)/D(α), πα(tie)=(1−α)(α−2α2)/D(α), πα(leadn)=(α/(1−α))n(α−2α2)/D(α) with D(α)=2α3−4α2+1 and real division by zero equal to 0. The per-state expected gains (dropping the subscript α,γ) are, with weights p(pool)=α, p(honestOnPool)=(1−α)γ, p(honestOnOthers)=(1−α)(1−γ):
gpool(zero)=0, gpool(tie)=2α+(1−α)γ, gpool(lead0)=0, gpool(lead1)=2(1−α)γ+2(1−α)(1−γ), gpool(leadm+2)=(1−α)γ+(1−α)(1−γ);
goth(zero)=(1−α)γ+(1−α)(1−γ), goth(tie)=(1−α)γ+2(1−α)(1−γ), goth(leadn)=0.
Right-hand side C(α,γ) is closedForm αγ:
C(α,γ)=1−α(1+(2−α)α)α(1−α)2(4α+γ(1−2α))−α3,
whose denominator expands to 1−α−2α2+α3 (a different polynomial from D(α)), and which is 0 by the division convention whenever that denominator vanishes.
The equality is asserted for every α∈(0,21) and every real γ, including γ outside [0,1]. It is an equality of two real numbers each of which is defined via total operations (division by zero giving 0, non-summable tsum giving 0), so no side condition about non-vanishing denominators or summability is part of the claim.
Confirmed by the moderator at approval.