Fantasy Football Projections: How to Read Them and When to Trust Them
Fantasy football projections are widely misunderstood and frequently misused. A projection like “Josh Allen: 26.4 points” is a mathematical expectation — the average outcome across many simulations of that game, not a prediction of what will happen. If you ran that game 1,000 times, Allen might average 26.4 — but individual outcomes would range from 8 to 55 points.
Projections Quick Reference
| Use case | How to use | When projections are reliable | When to override | Mistake to avoid |
|---|---|---|---|---|
| Start/sit decisions | Compare players in the same decision; small gaps (within 2 pts) are noise; large gaps (5+ pts) are meaningful | Players with stable, high-volume roles — RB1 with 18 carries/game, WR1 with 10 targets/game | When you have current information the model doesn’t (a practice report, a depth chart change that just dropped) | Making a start/sit switch because a projection moved 1.2 points — within any model’s margin of error |
| Waiver wire decisions | Rank pickups against your current bench players; use the projection to decide add priority, not to predict exact outcome | High-volume, clearly defined roles (injury replacement with immediate starter status) | When a player’s role just changed and projections haven’t updated yet — the model lags the news | Waiting for projections to update before acting on a clear role change — the news is more valuable than the model in this case |
| Trade evaluation | Use season-long projections, NOT weekly — a weekly projection inflated by one good game is recency bias baked into a number | Players with 8+ game samples in their current role | When a player’s role, team, or health situation has materially changed since the sample was built | Using one week’s projection after a big game to value a player in a trade — use their 10-game average instead |
| Consensus vs. individual | Consensus (averaged across multiple models) beats any single model for baseline accuracy; individual models reveal where the market disagrees | Consensus is most reliable when all platforms agree within a narrow range | When there’s 5+ point spread between platforms — use the divergence as information: this player’s outcome is genuinely uncertain | Ignoring platform variance — when models disagree by 5+ points, the position is risky, and that’s itself useful |
| Checking timing | Check Wednesday/Thursday for baseline; final check Sunday morning pre-kickoff for injury confirmation | Sunday morning projections for questionable players who have since been ruled out or cleared | None — Sunday morning is the freshest data before lineup lock | Checking projections obsessively mid-week — models don’t change dramatically unless major news drops |
How Projections Are Built
The primary inputs:
- Expected game environment (Vegas total, spread)
- Player’s historical averages (stats per game, per target, per carry)
- Current season role data (snap counts, target share, carry share)
- Matchup adjustment (how the defense performs against this position)
- Volume expectations (projected touches, targets)
Where projections break down:
- Committee RBs where touch distribution is unclear
- Injury replacements whose new role isn’t fully established
- Players with volatile roles (touchdowns as their only path to big games)
- DST and kickers — enormous game-to-game variance that any model handles poorly
The Two Times to Check Projections
Wednesday or Thursday: Baseline for lineup decisions and waiver moves. The key data (injury reports, game O/U totals) has stabilized by midweek.
Sunday morning before kickoff: Confirm game-time statuses. A Wednesday projection for a “questionable” player changes meaningfully when they’re ruled out Sunday morning.
Between those two checks: Nothing. Refreshing projections daily for stable information adds anxiety, not insight.
When to Override a Projection
Override when: You have current information the model doesn’t (a practice limitation that just dropped); the player’s role changed and the model is lagging; there’s a clear modeling error (player projected for their season average despite a wildly different game environment this week).
Don’t override when: You feel like a player will do well; a TV analyst is confident; last week’s result is influencing you disproportionately.
Projections aggregate large sample sizes of information into one number. Your intuition rarely has more information than the model — only different information.
For using projections alongside matchup data in actual start/sit decisions, see: Fantasy Football Start/Sit Decisions: A Framework for Making the Right Call. For the role data that feeds the best projections, see: How to Read Fantasy Football Snap Counts and Why They Matter.