The scan statistic never drops below `1`.
ProvedLogic.JFeature.one_le_maxRatioStataether-cataloglogic
The scan statistic never drops below 1. Pure pigeonhole: some cell
always carries at least a proportional share of the hits.
theorem Logic.JFeature.one_le_maxRatioStat(u : ι → κ) (Hs : Ω → Finset ι)
(hκ : (univ : Finset κ).Nonempty) (hne : ∀ w, (Hs w).Nonempty) (w : Ω) :
1 ≤ maxRatioStat u Hs hκ w := by sorry
Formalization Note Transplanted verbatim from the Aether Catalog source Logic/JFeatureSweepSynthesis.lean; the statement is byte-identical to the source declaration, elaborated with autoImplicit disabled in the platform environment.
Preamble
-- Thm stub generated from Logic/JFeatureSweepSynthesis.lean
import Mathlib
import Definitions.Def_Logic_JFeatureMarginalBlindness
import Definitions.Def_Logic_JFeatureMaxStatistic
import Definitions.Def_Logic_JFeatureSweepSynthesis
import Definitions.Def_Logic_PhaseRouteAlignment
/-
# Synthesis: what a flat j-feature sweep does and does not establish
This file ties together the three strands of the paper-248 analysis.
* `pval_maxRatio_eq_one` : the **uncalibrated** scan statistic
`max_k rate(cell k) / globalRate` is `≥ 1` on *every* draw of the null
ensemble (pigeonhole, `Logic.JFeature.exists_fiber_rate_ge_globalRate`), so the
naive test "the best cell exceeds the global rate" has permutation p-value
exactly `1`. Only a max-statistic calibration can say anything.
* `sweep_blindness_two_views` : a carrier living in the *joint* structure is
invisible to marginal features in **two independent statistical views** — the
contingency view (enrichment ratio exactly `1`, this development) and the
regression view (degree-1 `R² ≤ 0`, `Logic.PhaseRoute.Rsq_additive_nonpos`).
Flatness of a marginal sweep is therefore a property of the *test*, not
evidence about the carrier.
-/
open Logic.JFeature
open Finset
variable {ι : Type*} [Fintype ι] [DecidableEq ι] [Nonempty ι]
variable {κ : Type*} [Fintype κ] [DecidableEq κ]
variable {Ω : Type*} [Fintype Ω] [Nonempty Ω]
omit [Fintype Ω] [Nonempty Ω] inFormal statement
theorem Logic.JFeature.one_le_maxRatioStat(u : ι → κ) (Hs : Ω → Finset ι)
(hκ : (univ : Finset κ).Nonempty) (hne : ∀ w, (Hs w).Nonempty) (w : Ω) :
1 ≤ maxRatioStat u Hs hκ w := by sorrySource