Theorem 4.1 — Tight four-thirds loss of full transparency
ProvedTransparencyTeamContracts.Submodular.theorem_4_1For every finite multi-agent contract instance with a finite nonempty state space, arbitrary prior , nonnegative state-independent costs, and monotone submodular state technologies , full revelation satisfies
Here optimizes the public information and signal-dependent contract jointly with principal-favorable pure Nash equilibrium selection. No zero-baseline or positive-cost restriction is imposed. The factor is sharp even for additive technologies: for every , there exists a finite additive instance with
The two clauses form one goal. The multiplied universal inequality also covers zero optimal payoff and does not require division by the full-information value.
Differences from the source:
- encoding: finite agents and states, real probabilities and contracts, optimal utilities → Fin indices, a simplex prior, and suprema of operational contract and public-scheme payoffs; reason: finite relabeling preserves all instances; admissible payoff sets are nonempty and bounded, and suprema give the same optimal values without asserting attainment
- encoding: monotone submodularity with marginals tested for every agent → monotone IsSuccess plus the published diminishing-returns predicate for agents outside the larger set; reason: the omitted membership cases follow from monotonicity, so the function class is identical
- encoding: full revelation is a 4/3 approximation and that factor is tight even for additive rewards → a multiplied 3/4 utility bound conjoined with a positive-payoff additive counterexample to each beta in [1,4/3); reason: multiplication preserves the approximation guarantee at zero payoff; the quantified sharpness is precisely impossibility of any better approximation factor
- encoding: public signals on compact spaces and expected payoffs → compact Borel signal spaces, measurable posteriors and strategies, and explicit payoff integrability; reason: these are the usual meaningful-expectation conventions; finite-state measurable contract payoffs are bounded and hence integrable, so no source instance or support size is excluded
import Mathlib import Definitions.Def_TransparencyTeamContracts_Submodular_Information
namespace TransparencyTeamContracts.Submodular
/-- Theorem 4.1, printed p.9 (PDF 11). Both the universal guarantee and the
additive sharpness clause are part of this single goal. -/
theorem theorem_4_1 :
(∀ (n m : ℕ) (M : Model n m) (D : Belief m),
(∀ ω, SubsamplingSM.Offline.IsSubmodular (M.f ω)) →
(3 / 4 : ℝ) * optimalInformation M D ≤ fullInformation M D) ∧
(∀ β : ℝ, 1 ≤ β → β < (4 / 3 : ℝ) →
∃ (n m : ℕ) (M : Model n m) (D : Belief m),
(∀ ω, FairTeamContracts.IsAdditive (M.f ω)) ∧
0 < fullInformation M D ∧ β * fullInformation M D < optimalInformation M D) := by sorry
end TransparencyTeamContracts.Submodular
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.