What a Questionable Tag Does and Does Not Change About a Projection

What the weekly page does with the tag

The weekly model treats an injury designation as a cutoff. A player whose current designation is Doubtful, Out, IR, PUP or suspended drops off the page. A Questionable player stays projected, and his median and band print exactly as they would with a clean report.

That rule is narrow on purpose. The weekly model blends this season's scoring, with recent weeks weighted up, with the last two seasons and a position mean, scales the blend by the game's implied team total, and publishes a median. It has no input for practice reports, snaps, routes or depth charts, so a limited session midweek has no way into the number.

The implied team total is the only way the game gets into a projection. If the market marks a team's total down because a starter is in doubt, the rerun that picks up the new total scales every player on that team with it, the tagged player included. That is the market's read arriving through the only channel the model has, and it moves the whole offense together.

The page reports a tagged player's projection from his history and his team's total. Whether he suits up is a separate question, and the answer belongs to you.

Why the band keeps its shape

A P10-P90 band's shape comes from history. At each position, the weekly model places a player in one of three brackets by the size of his own weekly projection, and every player in a bracket prints the same proportional band, fitted per scoring format. That proportion is how far projections in that bracket actually finished from their outcomes.

A tag moves no projection, so it pushes no player across a bracket line, and he prints the same proportional band he carried before the report. That band is a minimum on risk: the spread every player in that bracket carries before anything specific to this week gets stacked on top.

That minimum is where a Questionable player's real risk hides. A tagged player can carry genuine weight at zero, and none of that weight shows up in the printed P10. Two medians a point apart already sit deep inside each other's bands; reading overlapping projection bands covers when that makes a call a coin-flip, and a tag on one side tilts the flip before any arithmetic starts.

An invented start-sit pair, priced by hand

Take an invented pair of half-PPR receivers — numbers made up to show the shape, with band proportions standing in for the fitted ones. Receiver A prints a median of 12.0 and a band of 5.4 to 19.8. Receiver B prints a median of 11.0 and a band of 5.0 to 18.2. Both land in the same bracket, so both carry the same proportional band: 45% of the median at P10, 165% at P90.

Receiver A carries the Questionable tag. The page prints him unchanged, and on the page he is the start by a point.

Say your read of the week — practice reports, beat coverage, how this staff has handled the tag before — puts a 20% chance on A sitting. A player who sits scores zero, so A's real distribution stacks 20% of its weight at zero and leaves the rest on the printed distribution. That mixture is what you compare against B.

Its P10 falls inside the stack at zero, so A's floor is zero. Its median is the point where the zero stack plus a slice of the printed distribution reaches half: 0.2 plus 0.8 times p equals 0.5, p comes out to 0.375, and A's real median is the 37.5th percentile of his printed band. Interpolating between the printed P10 and P50 puts that near 10 on this invented shape, about a point under B.

The top barely moves. The mixture's P90 is the printed distribution's 87.5th percentile, a shade under 19.8. A 20% inactive chance erases the floor, costs about two points of median and shaves a fraction off the ceiling.

Drop the inactive chance to 10% and the damage keeps the same shape. The zero stack still reaches the P10, and the median slides only to about the 44th percentile of the printed band — roughly 11.0 on this shape, a dead heat with B. The tag reaches the floor at a smaller inactive chance than it takes to reach the median.

This arithmetic flatters the tagged player. A receiver who plays through the tag can leave early or run a reduced route tree, which thins the middle of his distribution as well, and the two-outcome mix above ignores that. Treat the numbers as the least the tag can cost.

What that means for a lineup call

The tag costs you in whichever part of the distribution your matchup depends on. Favored by a wide margin and protecting a floor, the lost P10 is the whole cost, and the healthy receiver with the lower median can be the right start. Needing a ceiling as an underdog, the tag costs little, because the top of the distribution barely moved. Floor-versus-ceiling lineup logic is worked through in the piece on variance players; here it only has to point the right direction.

The tag is a floor problem long before it is a median problem. Price it at the floor first.

The weekly start-sit board reruns daily, so a tag that upgrades to Out removes the player on the next rerun, and a tag that clears leaves his line where it was. Between those two events, the median on the page is a healthy-player median, and the zero stack is yours to add.

Season-long availability is a different mechanism on a different model. The draft board comes from the season model, rerun by hand, and it folds missed time into expected games weighted across the last two seasons; how missed games reach a draft projection is laid out there. None of it touches a weekly designation.

Where the record comes in

The weekly model's calls are graded on the public accuracy ledger, which scores each week's first posting, frozen before its first kickoff. A later rerun can move the week's page and never moves the grade. How the brackets and their bands are fitted is on the methodology page, with the misses stated as precisely as the hits.

This piece covers one slice of a weekly decision: the part a tag touches and the part it leaves alone. Pair it with the start-sit board for the medians, add your own inactive chance to anyone tagged, and read the record, then decide how hard to lean on any printed median with a tag beside it, including the arithmetic here.