Our Methods

Transparency is core to TLDR Fantasy. Here's exactly how we generate projections, identify value, and build stacks.

Data Inputs

Our algorithms ingest real-time data from multiple sources:

  • •Salary Data: FanDuel and DraftKings slates (updated hourly)
  • •Vegas Lines: Over/under totals, spreads, moneylines (sharp consensus)
  • •Weather Data: Wind, temperature, precipitation forecasts
  • •Injury Reports: Real-time status updates and practice participation
  • •Snap Counts: Offensive and defensive snap percentages
  • •Historical Stats: Player performance, trends, seasonal adjustments
  • •Expert Consensus: Aggregate projections from multiple DFS publications

Projection Methodology

Our projection model uses a combination of machine learning and expert rules:

  1. Baseline Projection: Historical average for the player/position, adjusted for game script and matchup difficulty
  2. Game Script Adjustment: Is the team expected to be trailing (pass-heavy) or leading (run-heavy)? Vegas lines tell us expected game flow
  3. Snap Count Weighting: Snap percentage drives opportunity. High snap count + good matchup = higher projection
  4. Target/Touch Share: For pass-catchers, target share is the #1 driver. For backs, share of offensive snaps and red zone touches matter most
  5. Matchup Adjustment: Defense rank, secondary depth, and recent performance vs. position
  6. Weather Impact: Wind reduces passing yards and accuracy. Cold weather favors rushing. Rain/snow reduces offensive efficiency
  7. Expert Consensus Blending: We weight independent expert projections (DFS publications, professional handicappers)

Floor & Ceiling Calculations

We calculate floor and ceiling ranges based on historical volatility:

  • •Floor: 20th percentile of outcomes (conservative upside estimate). Used to assess lineup downside risk
  • •Ceiling: 80th percentile of outcomes (optimistic upside). Used for tournament/GPP upside analysis
  • •Volatility Scoring: Players with high variance (QBs, WRs) get wider floor/ceiling ranges than consistent players (RBs)

Value & Leverage Identification

Value is straightforward: projection (in points) divided by salary (in $1K). We flag value plays as above 0.25x.

Leverage is more nuanced. A leverage play is:

  • •High ceiling upside (80th percentile projection)
  • •Low ownership (typically <5% in GPP)
  • •Positive expected value over consensus

Stack Generation

Our algorithm generates optimal stacks by:

  1. Identifying Game Totals: Which games have the highest over/under? High-scoring games = better stacking opportunities
  2. QB-Pass Catcher Correlation: QBs score when their receivers score (positive correlation). We pair QBs with their top projected targets
  3. Bring-Back Selection: For GPP, we add a player from the opposing team to create "bring-back" stacks (hedge correlation)
  4. Salary Optimization: Ensure each stack fits within lineup salary constraints while maintaining quality
  5. Game Script Weighting: Prioritize stacks from games expected to be high-scoring and competitive

Human Curation

Our algorithms are powerful, but we don't blindly trust them. Every day, a human expert reviews:

  • •Late-breaking injury news and practice reports
  • •Coaching changes or formation shifts
  • •Outlier projections (if our model says 35 pts but expert consensus says 20, we investigate)
  • •Non-quantifiable factors (QB/HC public statements, roster moves)

This human layer ensures we catch things algorithms miss and reduce false positives.

Backtesting & Validation

We continuously validate our model performance:

  • •Projection Accuracy: We measure actual points vs. projected points weekly. Our target is mean absolute error (MAE) under 3 points per player
  • •Value Performance: We track which value plays hit (projected 3.0x value, actual ≥ 2.5x)
  • •Stack Correlation: We measure how often our stacks correlate as expected

Continuous Improvement

DFS is dynamic. Teams change, injuries happen, and game flow shifts. We update our models weekly based on:

  • •Backtest performance metrics
  • •User feedback and edge requests
  • •NFL coaching/roster changes
  • •Market sentiment (ownership trends)

Transparency Commitment

We believe in showing our work. TLDR Fantasy users can see:

  • •Why a player was recommended
  • •Full projection breakdowns (floor, ceiling, vegas adjustment)
  • •Stack reasoning and correlation logic
  • •Previous week's accuracy metrics