Is the board's RB-carries cell a cancellation produced by served non-starting QBs' carries draining the conserved team rush pool inside lines_for/_reconcile, so that scaling those QBs' carries by P(QB1 leaves) before _reconcile fixes the RB room?

Owner lane: allocation

Question

Is the board's RB-carries cell a cancellation produced by served non-starting QBs' carries draining the conserved team rush pool inside lines_for/_reconcile, so that scaling those QBs' carries by P(QB1 leaves) before _reconcile fixes the RB room?

Prediction

A: replayed shipped near 3; the shadow lowers it but does NOT reach the 1.5 bar, and the clean teams stay positive (pool level carries part). B: on the trained unconditional row the QB2 leak is smaller than the board's, the shadow moves team-game RB carry MAE by less than 0.2 and all-row fantasy MAE by |delta| < 0.01.

Finding and verdict

REFUTED as registered (bar 1.5 missed: 2.19), and the registered falsifier FIRED: the 10 clean teams sit at +2.885 before and after, so the pool level is part of the board's RB gap. The board shadow does lower the RB gap to prior-season actual (-0.75 [-1.38, -0.20]), but C1 (same mass off the QB1) gets two thirds of that (-0.50), so the board effect is mostly generic QB->RB mass, not QB2-specific. On actuals the fix is harmful: +0.011 [+0.004, +0.020] (2025) and +0.009 [+0.001, +0.018] (2024) fantasy pts per roster row. Do not scale backup QB carries in lines_for; the board's excess backup-QB carries are a served-row question.

Reasoning

The registered statistic compares a board to the previous season's actual, which is not a forecast target, so it cannot tell a right redistribution from a wrong one; the added actuals endpoint could, and it says the trained chain already allocates backup QB carries a little BELOW truth.

Next action

The live board serves 1.64 non-starter QB carries a team-game; the trained row serves 0.66 and the truth is ~0.7. The question that remains is which SERVED rows carry the extra ~0.9 (stale 2025 window H-246, promoted/buried QB2 H-240/H-258, P(dressed) H-151). It is a served-row question, and it overlaps H-246/H-240; not filed as a new row.