Roster Construction After the Draft: How to Allocate Your Bench

A Bench Is a Budget

Draft prep tends to run out before the bench does. The starters get tiers, ADP gaps, and a plan for the turn; the last picks go to whoever is left when the room starts guessing. The bench still absorbs most of a season's roster churn, which makes that improvisation expensive.

Each bench slot costs a pick and then costs a weekly decision, and the slots do different work. A slot spent on the wrong job never shows up as a loss until the week you reach for it and find nothing useful there. The mix of upside, insurance, and near-term production is a construction question that projection ranges can answer.

The Three Jobs a Bench Slot Can Hold

An upside swing has a median below startable and a P90 that reaches a starter's week. You hold it for the right tail and accept that the median never pays, so the stretch of the P90 above replacement is the entire asset. The house cutoff for a real swing is a P90 at least four points a week over replacement, a judgment call that no backtest set.

Insurance is a player whose value depends on an event you can name, usually a particular starter missing time. The board publishes each player's own median and range, and no surface projects the weeks a teammate sits, so an insurance piece is a scenario you price yourself. An insurance slot without a starter's name attached is a swing that forgot its reason.

A short-term starter is a player you expect to put in a lineup within a few weeks, on a bye or to cover a hole, whose median already clears replacement. Among players who pass that test, the job goes to the highest P10, because you are buying the floor.

Reading the Split Off Your Starters

Your starters set the budget. Positions where your starter's range sits well above replacement need little insurance and can spend on upside, while positions with a steep drop from your starter to the next available body earn the insurance slots. On many rosters that drop is steepest at running back, and your own roster is the only place to confirm it.

A sensible starting split for a six-slot bench in a twelve-team league is two swings, two insurance pieces with names attached, and two short-term starters for the first bye cluster. The split is house judgment, and it should move with the lineup it protects. A fragile lineup pushes it toward insurance, and a sturdy one buys more swings.

Deeper rooms change the arithmetic, as the fourteen-team page works through. At the draft itself, run the same three jobs on the season board, where each median and range sits against the twelve-team baselines of QB12, RB30, WR36, and TE12. The four-point cutoff is a weekly number with no season equivalent, so use season P90s to rank swing candidates against each other.

An In-Season Re-Check, With Invented Numbers

In season, the same budget runs on weekly numbers. Take six bench candidates from an invented set of weekly projections, with numbers made up to show the shape, tied to no current run, and measured against an invented flex-level replacement line of 9.0 points. Within one position and usage tier the site's ranges scale with the median, so the P90 test separates players across positions and roles and mostly reduces to the median inside one.

A, a receiver in a low usage tier, projects 6.2 with a P90 of 16.1, 7.1 points over the line and well past the cutoff. F comes from the same receiver tier at 4.0, so its P90 scales to 10.4 and clears the line by only 1.4; inside that tier the lower median decided it. Tight end B sits beside A at 6.0, but on this invented board its position's range runs narrower, and a 12.0 P90 clears the line by 3.0, short of the cutoff despite a median almost level with A's.

The backup running back, D, projects 4.1 with a 9.8 P90 and fails the swing test outright. D earns a slot only with a starter's name attached, because its value lives in the weeks that starter misses, a scenario no published range shows.

C and E both clear the line at the median. C, a receiver in the middle tier, projects 10.4 with a P10 of 4.2, while E, a running back in the middle tier, projects 9.6 with a P10 of 5.3. E takes the bye-week start. C's higher median buys points rather than a floor, which is a different job than the one this slot was held for.

D carries one more caveat worth knowing before its payout week. The weekly model does not split usage shares out as a separate step, so a backup's projection does not inherit a missing starter's workload the moment the starter is ruled out, and the first week an insurance piece starts is the week its number understates the job most.

Where This Sits in the Plan

The last four rounds are a fill against three named jobs, ordered by what the early rounds left uncovered. A roster that went heavy at running back early has already bought its own insurance and can spend late picks on range width, and the build comparison page carries the structural version of that trade. The bye count that decides between one short-term starter and two comes from bye-week planning.

During the season, the start/sit comparer puts weekly P10, P50, and P90 side by side, which covers the median, P10, and P90 tests; replacement level and the insurance scenario stay yours. The weekly model behind it reruns daily, and the weekly ledger keeps the record on those ranges. Read it, then decide how hard to lean on them, including this piece.