Who should I start? Week 4 start 'em, sit 'em

Each player gets a projected stat line — targets, catches, yards, carries, touchdowns — built from five named factors, then scored in your league’s format with the same rules as the fantasy points calculator:

• Volume: his share of his team’s targets, carries or pass attempts, times how much the team usually throws and runs.
• Efficiency: catch rate and yards per target, carry or attempt, pulled toward the league average until he has built a sample.
• Scoring chances: touchdowns from his red-zone and goal-line work, not from last month’s touchdown luck.
• Vegas: this week’s implied team total against the totals in the games behind his baseline. A higher total means a few more plays and a lot more touchdowns.
• Market prior: where drafters took him in the preseason, counted as 2 games of evidence, so it fades as his season goes on.

On top of that sits availability: a Questionable tag multiplies the projection by 0.85, a Doubtful tag by 0.70, and a teammate ruled Out frees up some of his share. Defense against the position is shown for context but isn’t in the number when the game has a Vegas line: in testing it added nothing once the line was in. It stands in only for a game with no line yet.

The call comes from the gap. The two highest projections are compared at the precision we print them (one decimal each), and the gap picks a band: under 1 point is a coin flip, then a slight lean, a lean, a start and, at 5 points or more, a clear start. Each band prints how often gaps that size went the higher projection’s way in two seasons of testing (the table below). With three or four players, each card also says how far it sits behind the one above it.

The two biggest differences. Each projection splits into the parts above, and the tool names the two that move the gap most: more targets, a better Vegas spot, an injury tag, a higher touchdown rate, the preseason market. Each reason quotes the actual inputs, so you can check it against what you know. A difference worth less than 0.1 point is never offered as a reason.

Coin flips don’t get invented tiebreakers. When the gap is under a point, the tool names the higher floor and the better Vegas total, and only if they really differ.

The range bar runs from each player’s floor to his ceiling: the 10th and 90th percentile of how players at his position finished against their projection in 2024–25. Weekly scoring is lopsided (a few big games pull the average up), so the ceiling sits further above the projection than the floor sits below it.

We tested the projections walk-forward: each week projected only from the games before it, with the model fit on 2025 Weeks 4–11 and graded on weeks it never saw (2025 Weeks 12–18 and all of 2024 Weeks 4–18). For every pair of starter-level players at the same position in the same week (the top 18 QBs and TEs, the top 36 RBs and WRs by volume), we checked whether the higher projection outscored the lower one.

What that means. The bands held within 1–3 points across both seasons, which is why the tool can print a rate next to a call. It is also why a call is never a promise: a coin flip is about 52%, and even a clear start went the other way about 1 time in 5.

The edge is real but modest. Across all of those starter pairs, the higher projection won 63.9% of the time in late 2025, against 60.6% for each player’s plain season average; in 2024 it was 62.1% against 61.6%. Weekly fantasy scoring is mostly noise. The value here is as much the reasons and the range as the call.

Other scoring. The bands were measured on Half-PPR gaps. In Standard, PPR or custom scoring the tool applies the same bands to the gap in your scoring; we haven’t measured those separately, and the tool says so under the call.

Quarterbacks, especially early in the season. QB calls are the least reliable part of the model. In 2024 Weeks 4–7, a quarterback’s plain season average picked the better of two starters more often than the projection did (68.5% against 63.6%). Treat a QB lean as a lean.

Role changes. A backup promoted this week has no share history yet, so he projects low. A starter back from injured reserve is projected on his old games, and his teammates’ shares aren’t cut back to make room. A traded player carries his old team’s share until he builds a new one.

Who can play. A player on injured reserve (or released, or otherwise off the active roster on nflverse’s weekly rosters) can’t start. A player who missed his team’s last game is “status unknown” until the week’s report settles it: his card keeps his projection if he plays, and the call puts availability first instead of treating the pair as a coin flip.

Late scratches. A game-day inactive counts only if nflverse’s weekly rosters list him before the site’s last refresh. Injury designations apply once that week’s official report is out (the Week 4 report is applied now), and the last refresh before the early games reflects Friday’s report. Check the injury report before kickoff.

Who gets projected. Players need a real 2026 role: on average at least 15 pass attempts a game for a QB, 6 carries + targets a game for an RB, 3 targets a game for a WR, 2 targets a game for a TE. Anyone below that, or without a 2026 game yet, shows as “not enough 2026 data to project”. A player on a bye or ruled Out shows as “can’t start”, and a player whose game has kicked off drops out until next week.

Scoring the projections can’t follow. 100-yard and 300-yard bonuses, long-touchdown bonuses and return touchdowns aren’t projected: a bonus pays on one big game, and a projection is an average. Kickers and team defenses aren’t in this tool; see kicker streamers and DST streaming rankings.

A fantasy tool, not a guarantee. FantasyTools analysis is opinion for entertainment purposes only — not betting, financial, or professional advice.