Lemma 2, p. 11 — the posterior beliefs , , and the closed forms of , ,
ProvedReviewBlackmail.Central.lemma_2Consider the review-blackmail model under centralized dispute resolution with the standing assumptions (, , nonnegative, convex and strictly increasing on , ), and let the timeliness and accuracy lie in . Let be the market's Bayes posterior that quality is high after observing review when it believes the seller responds to a malicious review with .
Then for every ,
and, writing ,
A positive review can only come from a high-quality product, the absence of a review carries no information about quality, and a negative review lowers the belief by an amount that depends on how the market expects the seller to handle blackmail. The closed forms are what Propositions 1–2 and Corollary 1 compare.
Formalization Note is M.post (keep (γ * δ) i) R, defined by Bayes' rule from the likelihood table, so the closed forms are a computation, not a definition. The closed forms are transcribed as printed, with written out.
import Mathlib import Definitions.Def_ReviewBlackmail_Central_Setting
namespace ReviewBlackmail.Central
theorem lemma_2 (M : Model) (hM : M.Standing) (γ δ : ℝ) (hγ : γ ∈ Set.Icc (0 : ℝ) 1)
(hδ : δ ∈ Set.Icc (0 : ℝ) 1) :
(∀ i : Resp, M.post (keep (γ * δ) i) .P = 1 ∧ M.post (keep (γ * δ) i) .Z = M.a ∧
M.post (keep (γ * δ) i) .N ∈ Set.Icc 0 M.a) ∧
M.post (keep (γ * δ) .n) .N =
M.a / (M.a + (1 - M.a) * ((M.β + (1 - M.β) * (1 - M.p / (M.a * M.θ))) /
(M.β + (1 - M.β) * (1 - M.p / (M.a * M.θ)) * (1 - M.θ)))) ∧
M.post (keep (γ * δ) .c) .N =
M.a / (M.a + (1 - M.a) * ((M.β * (1 - γ * δ) + (1 - M.β) * (1 - M.p / (M.a * M.θ))) /
(M.β * (1 - γ * δ) + (1 - M.β) * (1 - M.p / (M.a * M.θ)) * (1 - M.θ)))) ∧
M.post (keep (γ * δ) .s) .N = M.a / (M.a + (1 - M.a) * (1 / (1 - M.θ))) := by sorry
end ReviewBlackmail.Central
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.