NBA ledger report -- latest pregame snapshot 2026-10-16T23:32:26Z; latest graded slate 2026-10-16 (native and preseason rows)
Model tags: fourfactors_hl25_b2b_carry25_phase_luck_talent_ft_v6 / fourfactors_hl25_b2b_carry25_phase_luck_talent_ft_avail_v6 / fourfactors_hl25_b2b_carry25_phase_v4
Sources: data/nba_ledger.csv (native, pregame-locked); data/nba_reconstructed.csv (reconstructed, hindsight); data/nba_preseason.csv (exhibitions, Kalshi).
Headline: flat 1u on the LEAN (model's side), at the price named. Value-side bets are graded only as the pre-registered hypotheses (the blanket value bet was retired 2026-10-06).
Read every ROI against its null (q x payout - 1: the ROI if the no-vig market is right, about minus the hold), NOT zero.
+/- is 1 SE; z = (ROI - null) / SE. fav = 1u on the market favourite of the same games at the same price (same-row baseline).
Bases, books and seasons are never pooled. Price bands are descriptive, not filters.
Market: Kalshi (last pre-tip minute; fee in the price) wherever a row has it, from 2025-26; earlier rows keep their sportsbook close. The hypotheses stay on the sportsbook prices they were registered on.

== NATIVE (pregame-locked, forward): 0 rows, 0 graded, 0 pending
  No graded native rows with a closing line yet.

== PRE-REGISTERED HYPOTHESES (frozen; analysis.HYPOTHESES)
  Native at the registered price beside the hindsight rule on the reconstructed rows. Never pooled; one season is not a verdict.
  H1    Games 1–9 · value side · model P − no-vig q ≥ 0.08
    native @ pregame           no bets
    hindsight @ close          n=115   56-59       +20.48u  ROI  +17.8% +/- 12.8  (null -4.0%, z +1.7)  win 48.7% vs q 43.1%  | fav -22.6%
  H2    Games 10+ · value side that is the favourite
    native @ pregame           no bets
    hindsight @ open           n=987   693-294      +6.25u  ROI   +0.6% +/-  2.2  (null -4.1%, z +2.2)  win 70.2% vs q 67.3%  | fav +0.6%
  H3    Games 10+ · value side · model P − no-vig q ≥ 0.12
    native @ pregame           no bets
    hindsight @ open           n=260   132-128     +49.47u  ROI  +19.0% +/-  8.7  (null -4.2%, z +2.7)  win 50.8% vs q 40.9%  | fav -4.2%
  H4    Games 10+ · value side · model P − no-vig q ≥ 0.12
    native @ first snapshot    no bets
    hindsight @ open           n=260   132-128     +49.47u  ROI  +19.0% +/-  8.7  (null -4.2%, z +2.7)  win 50.8% vs q 40.9%  | fav -4.2%
  H2·F  Games 10+ · value side that is the favourite
    native @ first snapshot    no bets
    hindsight @ open           n=987   693-294      +6.25u  ROI   +0.6% +/-  2.2  (null -4.1%, z +2.2)  win 70.2% vs q 67.3%  | fav +0.6%

== RECONSTRUCTED (leave-one-season-out, closing price: HINDSIGHT, never forward evidence)
  Season table (lean at the close):
    book · season                n        W-L    units     ROI   +/-    null  fav ROI   LL model-close
    ESPN BET 2024-25          1209    825-384   -42.74   -3.5%   2.0   -4.2%    -4.3%   +0.0118 +/- 0.0050
    Kalshi 2025-26            1209    834-375   -10.61   -0.9%   2.1   -3.2%    -3.3%   +0.0159 +/- 0.0055
-- ESPN BET close, 2024-25: 1209 games
  Lean — the model's favourite
    All picks                n=1209  825-384     -42.74u  ROI   -3.5% +/-  2.0  (null -4.2%, z +0.3)  win 68.2% vs q 67.8%  | fav -4.3%
    Games 10+                n=1071  732-339     -42.17u  ROI   -3.9% +/-  2.1  (null -4.2%, z +0.1)  win 68.3% vs q 68.0%  | fav -3.6%
    Games 1–9 (carryover)    n=138   93-45        -0.57u  ROI   -0.4% +/-  6.5  (null -4.2%, z +0.6)  win 67.4% vs q 66.6%  | fav -10.1%
  Log loss  model 0.5940  close 0.5821  model - close +0.0118 +/- 0.0050 (1 SE; negative = model better; z +2.4)
  Brier     model 0.2042  close 0.1995  model - close +0.0047 +/- 0.0022
  Lean accuracy 68.2%  vs close favourite 69.1%  (accuracy is not calibration)
-- Kalshi close, 2025-26: 1209 games
  Lean — the model's favourite
    All picks                n=1209  834-375     -10.61u  ROI   -0.9% +/-  2.1  (null -3.2%, z +1.1)  win 69.0% vs q 67.8%  | fav -3.3%
    Games 10+                n=1071  739-332     -19.81u  ROI   -1.9% +/-  2.2  (null -3.1%, z +0.6)  win 69.0% vs q 68.4%  | fav -3.7%
    Games 1–9 (carryover)    n=138   95-43        +9.20u  ROI   +6.7% +/-  6.9  (null -3.5%, z +1.5)  win 68.8% vs q 63.3%  | fav +0.0%
  Log loss  model 0.5837  close 0.5677  model - close +0.0159 +/- 0.0055 (1 SE; negative = model better; z +2.9)
  Brier     model 0.1995  close 0.1937  model - close +0.0058 +/- 0.0024
  Lean accuracy 69.0%  vs close favourite 69.7%  (accuracy is not calibration)

== PRESEASON (exhibitions, Kalshi prices; not regular-season skill): 66 rows
-- Kalshi close, 2026-27: 66 graded
  at the pregame:
    Lean — the model's favourite
      All picks                n=66    33-33       -12.09u  ROI  -18.3% +/- 10.6  (null -4.0%, z -1.3)  win 50.0% vs q 58.4%  | fav -17.0%
  at the close:
    Lean — the model's favourite
      All picks                n=66    33-33       -10.17u  ROI  -15.4% +/- 11.6  (null -4.0%, z -1.0)  win 50.0% vs q 57.7%  | fav -17.5%