If grains.sql tw_lags composed the ten team windows the share tree and every other team_week reader consume (own plays_l3, pass_rate_l3, points_l3, points_trend, points_allowed_trend, plays_trend, pass_rate_trend; the opponent's plays_l3, pass_rate_l3, points_allowed_l3 via the same table) as geometric pools instead of flat l3 / l3-minus-l8, does the player line move end to end?

Owner lane: construction

Question

If grains.sql tw_lags composed the ten team windows the share tree and every other team_week reader consume (own plays_l3, pass_rate_l3, points_l3, points_trend, points_allowed_trend, plays_trend, pass_rate_trend; the opponent's plays_l3, pass_rate_l3, points_allowed_l3 via the same table) as geometric pools instead of flat l3 / l3-minus-l8, does the player line move end to end?

Prediction

G-A0 fantasy MAE within +-0.01 pts/row and the Bonf-3 interval spans zero (nil), targets and carries within +-0.005; G moves most forecasts by 0.05-0.3 pts. C-A0 is WORSE than A0 on fantasy by +0.005 to +0.04 with the 95% interval excluding zero on at least one endpoint, because the shuffled plays_l3/pass_rate_l3 also reach the volume anchor. If C is nil the block is inert end to end and G is uninformative.

Finding and verdict

MEASURED NOTHING. Geometric recomposition of the ten team windows (own + opponent copies, both sides of every game, every consumer of team_week at once) moves fantasy MAE +0.0048 [Bonf-3 -0.0088, +0.0187], targets -0.0030, carries +0.0023, all spanning zero, on 20,523 player-weeks 2023-25 with the replica exact. The whole block is near-inert end to end: a cross-team derangement of it costs +0.0088 carries [+0.0019,+0.0156] and +0.0074 fantasy (spans zero).

Reasoning

Closes the last untested route of H-021's team windows. The anchor route (0624Z) found a real rush gain at the team-game level (-0.041 carries/team-game) that dilutes to nil per player; the share-tree + opponent route is now nil too, and the block itself is worth about +0.01 carries/row through the model by derangement.

Next action

H-021's team windows are closed on both routes. Remaining items the row carries (conservation, latest_rows) are not composition questions of team_week; the only live H-021 lead is 0118Z's season-share term in the start/sit order, which is a decision question and belongs to its own row.