We tested it, it did not survive the evidence, and the projections did not change. Here is the whole result, including what it does not prove.
_This is one of a series on hypotheses that failed. Every projection change here has to beat a held-out test before it ships, and most candidates do not. Publishing only the ones that worked would misrepresent how the model is built._
_Not recorded._
14-day forward window, points per appearance, matched same-date controls, bootstrap clustered on pid.
| arm | calls | **distinct players** | flagged Δ | control Δ | DiD | 95% confidence interval | verdict |
|---|---|---|---|---|---|---|---|
| Batters | 54 | 8 | +0.162 | −0.100 | +0.261 ± 0.190 | [−0.085, +0.654] | no information |
| Pitchers | 186 | 32 | −0.723 | −0.373 | −0.350 ± 0.852 | [−1.983, +1.392] | no information |
Neither interval excludes zero. Falsifier 1 fired in both arms: the flagged group does not decline further than comparable players who were equally hot and were not flagged.
The batter arm points the wrong way. Flagged batters improved (+0.16 pts/game) while their matched controls declined (−0.10). On 8 players that is not a finding, but it is not the direction the surface claims either.
The pitcher arm has the right sign and no precision. Flagged pitchers fell (−0.72) and controls fell too (−0.37); the −0.35 gap sits inside an interval five times its own width.
Rejected. Nothing in the projections changed as a result of this study.
_Written up from the internal audit record. The numbers are the ones the study produced; nothing was re-run to make the story cleaner._
Curated picks where the model has the highest decision conviction. Updated every render.
Run rankings to populate Market Edge.
Run rankings first.
Mark a player as drafted to get a recommendation.
Studies and awards - the research-side surfaces of the managr platform.
Live snapshot of the projection system.
Backtest, parsimony, ablation, and benchmark - one synthesized report. Output renders in Validation.
Final Value = Raw Projection × MasterConfidence × TeamContext. Each layer answers one question - no overlap.
Recency weighting: prior seasons contribute by recency (more recent = more weight) and sample size (more PA = more weight).
Shrinkage: observed rates regress toward position-specific population means by an amount inversely proportional to sample size. Catches small-sample outliers.
Quality-of-contact adjustments: xwOBA, barrel rate, and bat-tracking metrics replace luck-driven outcome stats with skill-driven ones.
PT modeling: projected PA is anchored to workload tier (full-time, regular, platoon, backup catcher) using historical role distributions.
Monte Carlo: N simulated seasons per player using projected mean + uncertainty, producing P10 / P50 / P90 distribution and bust%.
Confidence layers: four orthogonal multipliers: Roster (depth chart), Role (PT certainty), Sample (career PA), Market (consensus alignment). The Audit tab shows all four for every player.
VAR: Value Above Replacement at the player's primary position. Replacement levels are floored per position so SS scarcity doesn't artificially inflate stable middle infielders.
Tune component weights against historical seasons. Stored in session.
-
-
-
Walk-forward backtests, calibration quality, and benchmark comparisons. Run walk-forward first - most tools depend on its pair pool.
Projects every historical batter-season from prior data only. Reports RMSE, MAE, Spearman ρ, hit rates. Populates the pair pool other validators use.
Walk-forward over historical seasons. Reports RMSE, MAE, Spearman ρ, top-N hit rates.
Distribution, uncertainty, per-archetype performance, calibrated tiers, draft sim.
Load any rival projection or ADP source. Auto-detects player_name + rank / adp / projected.
Where the model finds value the market is missing.
Curated picks from current rankings.
Where the model has measurable advantage.
The safety layer. Realism, false-confidence, forensics - what to remove.
The most important governance tools. Output renders in Validation.
Per-era ablation, error clusters, bias, correlations, removals. Each tells you what to remove. Run walk-forward first.
The smallest model that retains predictive power. Every feature beats the burden of proof.
Live pipeline status and data freshness.
-
Last known status of each underlying data source.
Source status will appear once data has been refreshed.
Issues from the most recent data refresh, if any.
-
MVP, Cy Young, Rookie of the Year - historical winners with their stats, plus model-predicted current-season winners.
Predictions from the current 2026 projection model. Top candidates per award based on projected fantasy points + advanced-stat underlying. League assignment from current team.
Coaching rosters for every MLB team plus a multi-year Fantasy Points Above Average ranking of every MLB coach.
Findings article + all-time top-100 performer leaderboard from every World Baseball Classic (2006-2026).
A growing collection of fantasy baseball studies. Each card opens a detail page with the methodology, the data sources, and (where the data exists today) live computed findings.
Active player injuries, refreshed from upstream sources.
| Player | Team | Pos | Status | Explanation | Replacement |
|---|
Enter players for each side, one per line. Values use your scoring weights and injury-adjusted projections.
| Player | Pos | ProjPts | Status |
|---|
| Player | Pos | ProjPts | Status |
|---|
Every signal, formula, and data source the model uses for player evaluation - organized by product.
Configure your draft session. This stays on this device only.