Newest eligible ECR
The board uses the newest ready, unretracted expert-consensus observation published by the cutoff. Position rank is used only when an overall rank is unavailable.
How League Read ranks players
Public rankings keep attributed expert rank, consensus market ADP, and scoring-specific projections separate. League-specific recommendations then apply your scoring, roster, lineup, player availability, and acquisition context.
- Ranked players
- 276
- Attributed sources
- 3
- Consensus ADP coverage
- 97%
- Latest update
- Aug 17, 2026
Inputs, sources, and freshness
Every displayed input retains its source and observation time. The selected season, scoring format, team count, and quarterback format determine which evidence belongs on a public board.
Matching ADP only
Consensus ADP combines matching human drafts, human mocks, and provider consensus. Recency, sample size, confidence, and source independence affect the blend.
Scoring-specific points
Expected points are the model's average outcome. Floor is the calibrated 10th-percentile outcome and ceiling is the calibrated 90th-percentile outcome, so a modest average can still retain meaningful upside.
Sources stay visible
Provider names, links, formats, update times, rank ranges, draft samples, and market movement remain attached when the source supplies them.
Value Rank and tier boundaries
Market behavior, expert opinion, and League Read’s projection-derived value answer different questions, so the site labels them separately.
How experts order players
The retained ranking source supplies the board’s expert ordering and, when available, its best-to-worst rank range.
Where matching drafts select players
This is observed draft behavior. It does not use League Read projections and is not presented as a value prediction.
Projected value above replacement
Scoring-specific points are compared with positional replacement and adjusted for roster demand and scarcity.
Value Rank
rank((projected points − position replacement) × demand + scarcity)
Default 12-team, 1-QB public baseline
Tier boundary
next best plausible rank > current worst plausible rank + 0.5
Position bands are used when ranking uncertainty is unavailable.
League-specific recommendations
A connected league uses the same projection authority, then adds its rules, roster construction, available players, and timing. Each component is normalized from 0 to 1 before weighting.
60% VORP + 20% position + 12% scarcity + 8% confidence
VORP is the player’s projection minus the replacement projection at that position. League size and required starters determine replacement. ADP does not enter league value.
65% league + 12% lineup + 8% bench + 15% need
The lineup optimizer compares the roster before and after a player is added. Roster need cannot replace the underlying league value.
45% league + 32% lineup + 13% bench + 10% need
Waiver value gives more weight to immediate lineup improvement and the effect of the likely drop.
75% personal value + 25% acquisition urgency
ADP, availability at the next pick, waiver claim pressure, opponent need, and recorded owner behavior affect urgency without rewriting player value.
Chronological holdout evaluation
League Read evaluates a model only on outcomes that were not available during training. The window advances through history so later information cannot influence an earlier prediction.
Train earlier weeks
→
Predict the next untouched week
→
Score against actual results
→
Repeat with a larger window
MAE and RMSE
MAE measures the average miss. RMSE penalizes large misses more heavily. Decision-weighted MAE emphasizes players near lineup cutoffs.
Rank correlation
Spearman correlation measures whether the model orders players correctly, not merely whether point totals are close.
Top-player hit rate
The predicted starter-caliber group is compared with the players who actually finished in that group.
10th to 90th percentile
Floor and ceiling target the middle 80% of outcomes. Calibration is learned on earlier replay weeks and verified on later untouched weeks; width and large tail misses are scored together.
A testable probability
Confidence estimates the chance that actual points finish within that position's historical median projection miss. Earlier replay weeks calibrate the probability and later weeks test whether, for example, 70% confidence happens about 70% of the time.
Inactive and active outcomes
Expected points account for the chance a player is inactive. Floor and ceiling keep inactive and active outcomes separate, so availability risk can lower the average without erasing a playable ceiling.
Model promotion requirements
- Compare only models with the same scoring, roster, lineup, team-count, and quarterback settings.
- Beat the current champion or named baseline by the configured margin.
- Stay below the configured maximum holdout error.
- Beat the prior-average baseline across expanding windows.
- Target 80% later-week coverage for the 10th-to-90th-percentile range, with 60% retained as a minimum safety floor during model replacement.
- Reject confidence calibration that makes later-week probability accuracy materially worse.
Limitations
Consensus is not a prediction
Expert rank and ADP summarize other people’s rankings and draft behavior. They remain separate from League Read model output.
Coverage varies by source
Rank ranges, projection ranges, samples, and market movement appear only when the retained source provides them.
League context changes decisions
A public board cannot know private scoring rules, rosters, available players, waiver timing, or owner behavior.
Current validation status
League Read publishes a predictive-accuracy result only when it comes from a promoted chronological holdout report. If no qualifying report is available, the site shows no public accuracy claim and continues to label consensus evidence and model output separately.
View current rankings