In-Season vs. Preseason Fantasy Projections: Key Distinctions
Preseason and in-season fantasy projections are built on different data foundations, updated on different schedules, and answer fundamentally different questions. The distinction matters because treating them interchangeably — running a Week 9 waiver decision through a preseason model, for instance — introduces a category error that shows up as bad roster choices. This page breaks down how each type works, where each belongs, and how the gap between them shapes every projection-dependent decision from draft day through the playoff push.
Definition and scope
A preseason projection is a forward estimate of a player's full-season statistical output produced before meaningful game data exists for the current year. These projections lean on prior-season numbers, offseason personnel changes, training camp depth charts, and historical aging or injury patterns. A running back projection issued in July, for example, might weight three prior seasons of carry volume alongside any offensive line turnover in the offseason.
An in-season projection narrows the time horizon and sharpens the inputs. Rather than projecting a full season, in-season models typically target a single upcoming game, a remaining schedule slice, or a rest-of-season window — and they update as new information arrives. That "new information" is precisely the domain explored in Projection Update Schedule: injury reports, snap counts from the past two weeks, recent target-share shifts, and Vegas game totals all feed into models that may refresh multiple times between Sunday and the following Thursday.
The scope difference is not just temporal. Preseason projections operate under high uncertainty — they model a player's expected role, which may not match actual usage. In-season projections can observe actual role and adjust accordingly, though sample size constraints introduce their own noise (see Sample Size and Projection Reliability for how projectors handle small observed samples mid-season).
How it works
Preseason models build a baseline through a weighted blend of historical performance, typically applying heavier weight to the most recent season and diminishing weight to earlier years. A three-year weighted average might assign 50% weight to Year N-1, 30% to Year N-2, and 20% to Year N-3. These baselines are then adjusted for:
- Positional scarcity scaling relative to scoring format (Scoring Format Impact on Projections covers how PPR versus standard settings shift positional values)
In-season models layer current-year observations on top of a prior baseline, progressively replacing the preseason estimate as the sample grows. By Week 6 of an NFL season, a wide receiver's target share over the first five games carries substantial predictive weight — enough to shift projected catch totals significantly from the August estimate. This is the empirical backbone of Regression to the Mean in Fantasy: early-season outliers get pulled back toward the prior, but the prior itself gets updated.
The FantasyPros consensus model documentation and public work from researchers at the MIT Sloan Sports Analytics Conference have both demonstrated that blended models — those combining preseason priors with accumulating in-season evidence — outperform either approach used in isolation.
Common scenarios
Scenario 1: Draft preparation (preseason model territory)
A manager using projections to anchor a snake draft in late August is necessarily working with preseason estimates. The practical application is explored in Applying Projections to Draft Strategy, but the key point here is that preseason projections should be evaluated on their assumptions, not their point estimates. What role does the model assume for this running back? What happens to the projection if that assumed role doesn't materialize?
Scenario 2: Waiver wire decisions (in-season model territory)
A receiver who's seen 11 targets in each of the past 3 weeks following a starter's injury has generated observable role data that a preseason model couldn't anticipate. The decision to add or drop that player belongs to in-season projection logic — specifically the kind of Usage Rate Adjustments in Projections that models should implement in real time.
Scenario 3: Trade evaluation (both models, reconciled)
Trade valuation requires holding two time horizons simultaneously: what a player is worth for the rest of this season (in-season) versus what they project to do next year (preseason logic applied forward). This is why Trade Value and Projection Data treats rest-of-season and dynasty/keeper values as distinct outputs.
Decision boundaries
The clearest rule: use whichever model has more actual data for the decision being made. Before Week 1, there is no current-year game data — preseason projections are the only option. By Week 8, a player has 7 games of observed usage, and those observations should dominate a weekly lineup call.
A secondary boundary involves update frequency. Preseason projections are essentially static between publication and the season opener. In-season projections should, in principle, update whenever meaningful information changes — injury designation, a depth chart change, a weather report for a cold-weather dome game. The Fantasy Projection Lab home tracks projection freshness because a stale in-season number is, in practical terms, worse than no number at all.
The third boundary is position sensitivity. Quarterbacks show relatively stable in-season profiles — 8 games of data for a starting QB is quite reliable. Running backs behind uncertain offensive lines can shift 30% or more in projected carry volume based on a single injury designation. For position-specific sensitivities, Running Back Projection Methodology and Quarterback Projection Methodology detail how each position's projections respond differently to in-season signals.
Treating preseason projections as obsolete after Week 1 is the wrong move — they provide the prior that stabilizes noisy early-season samples. Treating them as still-valid by Week 10 is the equally wrong move in the other direction. The skill is knowing when the accumulated evidence has earned the right to override the original estimate.