Splitting theorem (headline form).
ProvedKnownUnresolvedCards.expected_total_eq_certain_countaether-catalogmachinelearning
Splitting theorem (headline form). d cards predicted with certainty
each pay one unit, the remaining cards are fair guesses; the expected payoff is
exactly d = K.card. Uncertainty supplies no positive edge.
theorem KnownUnresolvedCards.expected_total_eq_certain_count[Nonempty Ω]
(p : ι → Ω → ℚ) (K : Finset ι)
(hK : ∀ i ∈ K, Resolved (p i) 1)
(hU : ∀ i ∉ K, Fair (p i)) :
E (fun ω => ∑ i, p i ω) = (K.card : ℚ) := by sorry
Formalization Note Transplanted verbatim from the Aether Catalog source MachineLearning/KnownUnresolvedCards/Basic.lean; the statement is byte-identical to the source declaration, elaborated with autoImplicit disabled in the platform environment.
Preamble
-- Thm stub generated from MachineLearning/KnownUnresolvedCards/Basic.lean
import Mathlib
import Definitions.Def_MachineLearning_KnownUnresolvedCards_Basic
/-
Copyright (c) 2025. All rights reserved.
Released under Apache 2.0 license.
# Known versus unresolved cards — I. The uniform expectation calculus
A *prediction game* pays a rational amount on each of finitely many **cards**.
Some cards are **resolved**: the predictor knows their value and collects a
deterministic unit payoff. The remaining cards are **unresolved**: the predictor
must guess, and the guess is priced at *fair odds*, i.e. the payoff on such a
card has zero mean.
This file develops the minimal probabilistic infrastructure needed to state and
prove the headline principle
> `E[total payoff] = (number of resolved cards)`,
namely a uniform-expectation functional `E` on a finite sample space, its
linearity, and the **splitting theorem** `expected_total_eq_certain_count`.
The point of stating the splitting theorem for an *arbitrary* finite index type
`ι` and an *arbitrary* subset `K : Finset ι` of resolved cards is that the deck
models of `PermCount.lean` and `DeckGame.lean` are then genuine instances rather
than re-proofs.
## Main results
* `E_sum` — linearity of uniform expectation over a `Finset` sum.
* `expected_total_eq_certain_count` — if every card of `K` pays a deterministic
`1` and every card outside `K` is fair, the expected total payoff is `K.card`.
* `expected_total_eq_certain_sum` — the weighted version with arbitrary
deterministic payoffs on `K`.
* `no_fair_portfolio_edge` — a portfolio consisting only of fair cards has zero
expected payoff, *whatever* the (possibly wildly correlated) joint law.
-/
open KnownUnresolvedCards
open Finset
/-! ## Uniform expectation on a finite sample space -/
variable {Ω : Type*} [Fintype Ω]
/-! ## Resolved and unresolved cards -/
variable {ι : Type*} [Fintype ι] [DecidableEq ι]Formal statement
theorem KnownUnresolvedCards.expected_total_eq_certain_count[Nonempty Ω]
(p : ι → Ω → ℚ) (K : Finset ι)
(hK : ∀ i ∈ K, Resolved (p i) 1)
(hU : ∀ i ∉ K, Fair (p i)) :
E (fun ω => ∑ i, p i ω) = (K.card : ℚ) := by sorrySource