What FantasyPros weight minimises fantasy-point MAE per position beside our number, and is the shipped pooled consensus.WEIGHT = 0.40 the wrong constant or the wrong shape?

Owner lane: q-language

Question

What FantasyPros weight minimises fantasy-point MAE per position beside our number, and is the shipped pooled consensus.WEIGHT = 0.40 the wrong constant or the wrong shape?

Prediction

Registered on H-235 at 2026-09-15T1843Z, before the script existed: trained pooled optimum 0.55-0.75 (H-190's 0.70), QB highest. H-190 CONFIRMED iff trained pooled optimum in [0.60, 0.80], QB optimum >= 0.80 and the T walk-forward-vs-0.40 contrast's Bonferroni-8 interval excludes 0 below; REFUTED iff pooled optimum within 0.40 +/- 0.10 and that contrast spans 0; else PARTIAL. Follow-ons (written before each ran): the archive's edge does not grow on unannounced snap collapses (honest vintage); the clamp binds on >= 15% of covered rows at w=1, (w=1, no clamp) is worse than (w=1, 0.6), and the 2-D argmin is at w < 0.95 with MAX_MOVE < 0.6.

Finding and verdict

PARTIAL by the registered rule, and the claim's SHAPE is wrong. Direction confirmed and stronger than H-190 said: the MAE-optimal FantasyPros weight at the shipped clamp is 0.95 pooled [0.85, 1.00], QB 1.0, RB 0.95, WR 0.95, TE 0.85; walk-forward per-position weights (0.85-1.0) beat 0.40 by -0.0488 [-0.0700, -0.0227] Bonferroni-8 on 2024-25 trained rows and -0.1076 [-0.1308, -0.0821] on the served row; flex_sym +2.73pp [+1.74, +3.54] served. REFUTED on 'per-position': one pooled weight is equal (+0.0008). And the weight is confounded with MAX_MOVE: unclamped, the optimum is 0.60 pooled and the consensus alone barely beats us (-0.029); the clamp is doing a third of the work. Archive vintage: the falsifier found no outcome leakage (consensus is WORSE than us on unannounced collapses, edge -0.386, bias +4.38 vs +4.14), which does not prove a pre-kickoff vintage. DISCOVERY; not a change proposal.

Reasoning

H-190's number (0.70) came from a curve that crossed the clamp; the weight a lane fits is whatever the clamp leaves.