Every metric,
in plain English.

A single rolling number that captures team strength. Starts at 1500; goes up after wins, down after losses — more for an upset.

Every team starts at 1500 (the league baseline). Each completed game updates both teams: the winner gains points (more if they beat a higher-rated opponent), the loser loses points. The K-factor and home advantage are tuned per sport. NHL is tighter than NBA because hockey scores are sparser and OT/SO outcomes carry more variance.

The model's pre-game estimate of the home team's chance to win, expressed as a 0–100% number.

Computed from the two teams' Elo ratings + the home advantage. Logistic curve: ±400 Elo points = ~91%/9%. We also blend the prior with score margin once a game is in progress (the LiveWinProbability card) for an honest in-game estimate.

How honest the model is about its own confidence. A perfectly calibrated model that says 70% wins exactly 70% of the time.

We bucket every prediction by the model's stated probability (50–55%, 55–60%, …, 90%+) and compare to the actual win rate at each bucket. A perfectly calibrated model rides the diagonal on /model/accuracy's calibration plot. Above the diagonal = underconfident; below = overconfident. Within ±3pp of the diagonal we call calibrated.

Bucketed label for a pick: Lock (highest), Edge, Lean, Tossup. Tier thresholds tuned per sport.

Lock = highest model conviction (margin from 50% ≥ 18pp for NBA/MLB/WNBA, ≥ 14pp for NHL). Edge = strong lean. Lean = modest. Tossup = near coin-flip. NHL thresholds are tighter because the league has narrower probability ranges; a 65% NHL favorite is genuinely a strong play.

Constant rating bonus added to the home team's Elo before computing win probability.

Each league has a different home-court / home-ice / home-field edge. NBA ~75 Elo points (home wins ~60%), WNBA ~70 (~58%), NHL ~50 (~55% — lowest of the four leagues), MLB smaller. Home advantage is added to the home team's pre-game rating in the win-prob formula.

Edge size in "percentage points" — the gap between our model's win probability and the market's implied probability.

If the model says 60% home and the closing moneyline implies 50% home, the magnitude is 10pp ('ten percentage points'). Larger magnitude = sharper edge. Edges under ~3pp are at the edge of model noise; meaningful sharps live at 5+ on the same side. The /best-bets leaderboard ranks by magnitude.

The arithmetic difference between two percentages. Distinct from a percent change.

If a team's win probability moves from 50% to 55%, that's a 5 percentage-point move — not a 10% increase. We use pp because relative changes between probabilities ('14% increase') are misleading.

The win probability baked into a moneyline. Negative odds → favorite, positive → underdog.

American moneyline of −150 implies the favorite wins 60% of the time (150 / (150+100)); +200 implies the underdog wins 33.3%. Bookmakers add juice / vig so implied probabilities sum above 100%; magnitude calculations use the de-juiced version where possible.

A bet on which team wins, priced in American odds.

Negative numbers mean a favorite (wager that much to win $100); positive means an underdog (win that much on a $100 wager). −150 means risk $150 to win $100; +200 means risk $100 to win $200.

A bet on the winning margin. The favorite must win by MORE than the spread.

NBA Lakers −7.5 means the Lakers must win by 8+ for a "spread bet on the Lakers" to cash. Hockey uses puck line (typically ±1.5 goals). Baseball uses run line (typically ±1.5 runs).

A bet on the COMBINED final score. Over wins if the two teams' total exceeds the line.

NBA over/under of 224.5 means an over bet wins if both teams' combined points are 225+. Pushed totals (exact match — only on whole-number lines) refund the wager.

A team's record covering the point spread, regardless of who actually won.

A team that's −7.5 favored and wins by 5 still LOSES against the spread. ATS records reveal whether a team is overrated or underrated by the market. Sharp bettors track ATS more than straight win-loss.

The bookmaker's built-in margin. Why both sides of a coinflip are usually priced at −110.

On a 50/50 outcome, a fair price would be +100 / +100. Books charge ~10% vig: −110 / −110 means you risk $110 to win $100. The book makes money over volume even if both sides hit 50%. Translates to ~52.4% break-even win rate at standard vig.

Polar starfield where each star is a team or player. Distance from center = magnitude (rating gap, edge size, etc.).

Used on /standings (rating constellation), /best-bets (edge constellation), /constellation (cross-sport master), and the players directory (leaders). Star size + halo scale with the underlying metric. Click any star to drop into its detail page.

Heat grid of recent games on team pages. Color = win/loss, intensity = score margin, slash = model called the wrong side.

Each cell is one game in the team's last 30. Green = won, red = lost. Intensity scales with margin: a 1-pt win is faint; an 8+ pt win saturates. A diagonal yellow slash means the model picked the OTHER side and lost — surfaces 'upset wins' and 'favorite losses' at a glance.

Two teams' Elo arcs overlaid on one chart. Pulsing dots are current ratings; 'if X hosts' splits show model lean.

Used on /{league}/teams/[abbr]/vs/[opp]. Each line is the team's rating arc; the gap between the lines IS the model's lean for the matchup. The 'if home' splits use the same Elo + home-advantage formula the in-game model uses.

The thin bar at the top of every page rotating between live count and top-edge spotlight.

Auto-refreshes every 60 seconds. Live mode shows '🔴 N live · 3 MLB · 2 NHL'. Edge mode shows tonight's sharpest model-vs-market gap. Slate mode (off-nights) shows 'N games on tonight's slate.' Dismissible — click the × to hide on this device.

How much a play improves the team's expected score.

A 6-yard gain on 3rd and 5 produces far more EPA than a 6-yard gain on 1st and 10. EPA converts every play into a common currency: scoreboard impact. Positive EPA means the offense gained value; negative means it lost value.

Per-touch efficiency. The gold standard for skill-position value.

Total EPA divided by plays. Filters out volume, shows you how valuable a player is when they're involved. A player with 0.25 EPA/play on 200 touches is more impactful per touch than one with 0.15 EPA/play on 300 touches, even when their counting stats look comparable.

Share of plays that gain positive EPA.

A play is 'successful' if it meets down-and-distance expectations: 40% of needed yards on 1st down, 60% on 2nd, 100% on 3rd/4th. High success rate = consistent. Low success rate with high EPA/play = boom-or-bust profile.

Team or player efficiency vs. league average on every play.

Defense-adjusted Value Over Average. Grades every play against the league-average outcome for that situation (down, distance, field position), then adjusts for opponent strength. Positive = above average, negative = below. Developed by Football Outsiders.

Percentage of team's passing targets a receiver gets.

Receiver targets ÷ team total targets. 25%+ is elite-tier usage. Target share is sticky week-to-week and a better future-projection input than raw receptions.

Single metric combining target share + air yards share.

1.5 × target share + 0.7 × air yards share. Captures both how often a receiver gets the ball thrown their way AND how far downfield those targets travel. 0.70+ is WR1 territory.

Percentage of offensive snaps a player is on the field for.

The floor of opportunity. A skill player needs 65% or higher snap share to be reliable for fantasy. Watch committee backfields and injury returns; snap counts often lead the box-score numbers by a week.

Share of team pass plays the player runs a route on.

More granular than snap share. Isolates receiver-specific usage by excluding run plays. RBs with 60%+ route participation are real pass-catching threats; 80%+ WR route participation is elite.

How much better a QB completes passes than the average passer would.

Each pass attempt has an expected completion probability based on depth, throw location, and pressure. CPOE subtracts expected from actual. +5% CPOE = elite accuracy; -3% = below average.

Average air yards of a QB's throws (or a receiver's targets).

Short pass-game QBs run 6–7 yard ADOT; deep-ball QBs push 9–10. For receivers, high ADOT correlates with boom-or-bust weekly variance.

Share of dropbacks where the QB is pressured.

Offensive-line quality proxy. QBs lose ~0.3 EPA/play under pressure. If pressure rate is trending up, project passing stats down regardless of matchup.

Yards gained after the ball is caught.

Total receiving yards minus air yards at the catch point. Splits receivers into route runners (low YAC, high ADOT) vs. YAC monsters (slot receivers, RBs).

Actual YAC minus the average YAC for that catch location + defender proximity.

Isolates receiver skill from scheme. A receiver with +1.5 YACOE is breaking tackles and winning after the catch at a higher rate than average.

Yards between receiver and nearest defender at the moment of the pass arrival.

Tracked by Next Gen Stats. 3.0+ yards of separation on average = open more often than not; 2.0– is contested-catch profile.

Rushing yards earned beyond what the average back would on the same carry.

Each carry has an expected yardage based on box count, defenders in the gap, and the blocking scheme. RYOE isolates the running back from the offensive line. Positive = the back creates yards; negative = the line creates yards.

Percentage of carries with 8+ defenders in the box.

Tough-matchup proxy. Stacked boxes crush rushing EPA. RBs with 30%+ stacked box rate carry an outsized positive signal when they still produce; they're earning every yard.

Expected points a team will score based on the Vegas spread + total.

(Total + Spread) / 2. A team in a 48-point game favored by 3 has a 25.5 implied total. Shapes game script: high implied totals favor pass-catchers; low totals favor defenses.

Expected flow of the game: pass-heavy, run-heavy, or neutral.

Favored teams with big leads run more to close out games; underdogs down 10-plus pass more to catch up. Projected game script shifts RB carries versus WR targets materially.

How fast a team runs plays when scoring is competitive.

Seconds per play in neutral game state. High-pace teams (25s/play) create more total plays and more total fantasy points per game for everyone involved.