Does the start/sit decision the board hands a manager -- the SERVED row, after the consensus blend -- order two startable men worse than the TRAINED row every decision gate grades, and by how much against the FantasyPros consensus?

Owner lane: serving

Question

Does the start/sit decision the board hands a manager -- the SERVED row, after the consensus blend -- order two startable men worse than the TRAINED row every decision gate grades, and by how much against the FantasyPros consensus?

Prediction

Frozen in the runner docstring before the run, informed by H-190's unpaired levels (disclosed): product S-T accuracy -0.6 to -1.6pp with a Bonf-4 interval excluding 0; raw -0.4 to -1.2pp; regret +0.05 to +0.20 pts per pair on the product; SEL_S cost no smaller than SEL_T; the noise-matched rival N costs less than S; cross-season rows cost the most per pair. H-089's own prediction: a FURTHER 2 to 5pp on flex_sym. H-089's refute rule: product S-T within 1pp of zero with an interval spanning zero.

Finding and verdict

TESTED. The premise holds and its size was over-called. On H-089's own identical pairs, the board's served number orders two startable men worse than the trained number by -0.82pp [Bonf-4 -1.63, -0.01] and +0.12 [+0.03, +0.22] pts regret per pair; H-017's raw arm -0.51pp, interval spans 0. H-089 predicted a further 2-5pp: FAILED on magnitude, not refuted by its own rule. Stronger than registered: that pair set is selected on the TRAINED scale; on the close calls the SERVED board actually presents the cost is -1.52pp [-2.38, -0.64], +0.18 pts regret (raw -2.22pp, +0.26), and on a consensus-only selection -1.27pp. The consensus's margin over the served product is +3.1pp against +1.7 to +2.3pp over the trained product. RB and WR carry it; QB and TE indistinguishable from zero. Almost all of it sits on age-1 rows (last week's game), and 2025 alone is -0.75pp with an interval spanning zero.

Reasoning

Every decision number in the project (H-017, CP-H17, H-190, H-204) grades the trained row, and H-190's served/trained pair of levels was never differenced, never given regret and never re-selected on the scale the board shows. This is the paired, interval-bearing version, with the selection the consumer faces.

Next action

(1) The repair is H-222's: serve a row whose windows include the last played game (CP-H49 generalised past the share columns); this runner's SEL_S contrast is its decision bar -- S - T within 0.3pp. (2) Every decision gate (decision_stats callers, CP-H17) should score the served row AND select pairs on the served scale; SEL_T under-states the served cost by half. (3) Prospective: 2026 has fpapi_live (1,045 rows) and aar_forecasts player fantasy_points weeks 1-5; grade issued flex_sym vs fpapi_live once 4+ weeks settle -- 3 clusters now is not an interval. (4) 2025 is the smallest season on every selection; if it persists in 2026 the gap may be shrinking with the fit's history, which is a different claim.