How do you evaluate a DFS projection source?

Two metrics matter: mean absolute error (MAE) per player-slate, and rank-order correlation between projected and actual outcomes. A source with 5 fantasy points MAE and 0.75 rank correlation is above average. Below 0.65 rank correlation is a red flag; below 0.5 is guessing. Track both across 30+ slates before drawing conclusions — one bad slate is noise, not signal.

What is systematic bias in DFS projections?

Systematic bias is a projection source consistently over- or under-projecting a specific player class — always overprojecting rookies, always underprojecting bench-role players, always missing on rushing quarterbacks. Bias is worse than random error because it's exploitable in the wrong direction; sharp DFS players identify their source's bias and adjust manually.

Should I use multiple DFS projection sources?

Yes, if you know how to combine them without doubling the noise. Weighted-average combining reduces variance when sources have uncorrelated errors. Combining sources with correlated errors (multiple sources using the same base data) adds no information. Sharp DFS players use a primary source + one uncorrelated secondary source for cross-check.

How do you spot a bad DFS projection source?

Symptoms: projections that don't update after major news (a starter ruled out at noon but the source still projects him at 25 minutes at 5 PM), projections that ignore matchup effects (same player, same salary, same projection against every opponent), or projections whose rank order routinely diverges from public consensus in ways that don't backtest well.

Does DFS Degen publish its own projections?

The simulator ingests projections from configured sources and generates the correlated Monte Carlo output layered on top. Projection quality is upstream — the simulator is only as accurate as the projections it's fed. Elite tier lets you upload your own projections via CSV so you can override any source you don't trust.

DFS Projection Quality: How to Evaluate a Projection Source

By DFS Degen TeamPublished August 4, 202610 min read

DFS projections are the input every other decision depends on. Not all sources are equal — MAE ranges from 4 to 8 fantasy points per player-slate across the public projection landscape, and rank-order correlation varies from 0.65 to 0.85. Sharp DFS players evaluate sources empirically, identify systematic biases, and combine sources only when their errors are uncorrelated. Bad projections make good simulation useless.

A common failure mode in DFS: heavy investment in lineup construction and portfolio optimization on top of projections that are only slightly better than random. The compounding cuts both directions — great projections plus mediocre construction beats mediocre projections plus great construction, most weeks. This piece walks through how to evaluate a projection source empirically rather than anecdotally.

The two metrics that matter

Mean Absolute Error (MAE). Average absolute difference between projected and actual fantasy points per player-slate. Lower is better. Public DFS projection sources range from ~4 (top-tier) to ~8 (baseline) on NFL main-slate skill positions. MLB and NBA MAE ranges are wider because those sports have higher inherent variance.
Rank-order correlation (Spearman's ρ). How well the projected ranking of players matches the actual ranking. This matters more than MAE because DFS rewards relative ranking (best value per position), not absolute point accuracy. Above 0.75 is strong; below 0.65 is a red flag.

The backtest methodology

Collect 30+ slates of projections vs actual outcomes for the same sport / site.
Compute MAE by taking the mean of |projected - actual| across all player-slate observations.
Compute Spearman's ρ separately for each slate by ranking players by projection and by actual, then correlating the rank orders.
Segment the results by player price tier ($3k-5k, 5k-7k, 7k+) — a source with strong overall MAE but weak performance on top-tier players is problematic because your GPP anchors are almost always top-tier.

A source with 5-point MAE on average but 8-point MAE on $8k+ NBA players is systematically failing at the salary tier that decides GPPs. Overall MAE hides the tier- specific problem; segmented MAE surfaces it.

Common systematic biases

Rookie overprojection. Sources that lean heavily on preseason narrative or highlight-reel samples overproject rookies. Systematic; exploitable by fading rookies more heavily than the source suggests.
Bench-role underprojection. Sources that key off season-average snap counts underproject veterans promoted to expanded roles mid-season. Manual adjustment upward on news-driven promotions captures the delta.
Rushing-QB miss. Traditional projection models trained on QB passing stats underproject dual-threat quarterbacks (Josh Allen, Lamar Jackson, college dual-threats). Their rushing contribution is systematically underweighted.
Weather blindness. Sources that ignore weather forecasts overproject QB and WR in high-wind games and underproject RB in the same games. Manual weather adjustment corrects this.

Combining multiple sources

Weighted-average combining reduces variance IF the sources' errors are uncorrelated. Two sources using the same base data (both scrape the same fantasy site's projections) have correlated errors — combining them adds no information. Two sources using genuinely different methodologies (one statistical, one narrative-based) combine usefully.

Sharp DFS players use a primary source with proven backtest performance + one uncorrelated secondary for cross-check. When the two diverge substantially on a player, dig in — usually one of them missed a specific news beat and the delta is worth verifying manually.

Building your own projections

Elite-tier DFS Degen users can upload their own projections via CSV. This is the endpoint of projection quality — full control over the input, no dependence on a third-party source. Requires substantial modeling work (~50 hours of setup + 5 hours per slate) but produces the highest-quality projections for players who put the time in.

Monte Carlo simulation explained — simulation quality is upstream of projection quality; both compound
DFS injury news timing — news drives the largest per-slate projection updates
DFS correlation deep-dive — projections capture means; correlation captures joint distributions
DFS Degen plans — Elite unlocks CSV projection upload

DFS Projection Quality: How to Evaluate a Projection Source

Frequently asked questions

Keep reading

Put the theory into practice