DK / Rays Odds

rays odds.

Monte Carlo
Live model · 2026 season

How do the Rays get in?

Loading the season…
Fetching standings and the full 2026 schedule…
Reach the playoffs

Win the AL East

First-round bye

Win the World Series

Magic number

Where the season lands.

Every simulated season ends with a final win total. The spread below is the model's honest uncertainty — not a single projection, but the full distribution of outcomes across every simulated remainder.

The leverage board.

Swing per game

The feature nobody builds. For every upcoming game, the model measures the odds conditional on each outcome — then reports the gap. That difference is exactly what the game is worth. Green means a home win helps the Rays; orange means it hurts.

Tonight's rooting guide.

Same math, pointed at your evening. Of the games on the next slate, here's who to pull for and what each result is worth in odds points.

Skill, luck, and the standings.

American League

Run differential predicts the future better than win-loss does. Pyth is each team's Pythagenpat expectation; Luck is how many wins they're running ahead of it. Rating is the blended talent estimate the simulation actually uses.

TeamW–LPythLuckRatingProjPlayoffsDivision

Odds through the season.

computing…

The model replayed from opening day — re-estimating talent from only the games played to that point, then re-simulating the rest of the way. This is the squiggle: what the model would have said, week by week.

Turn the knobs.

Re-simulates live

Every assumption in the model is exposed. Change one and the whole board re-runs.

Games of league-average play blended into each team's run differential. Team win% needs about 70; run differential is less noisy, so ~45 is the honest default.

How much of the rating comes from game-by-game Elo (which tracks recent form) versus season-long Pythagenpat (which is steadier).

Elo points of posterior uncertainty, redrawn once per simulated season and held fixed. Set this to zero and the intervals collapse — that's the bug most homemade models ship with.

MLB home advantage is small and shrinking — about 53%, worth roughly 24 Elo points.

How hard each game moves a rating. Baseball is mostly noise game to game, so K is tiny — about 4, against roughly 20 for the NFL.

Seasons played out in full. More sims means smoother tails and steadier leverage numbers.

How the model works.

Five layers
  1. 1Talent. Each team's runs scored and allowed become a Pythagenpat expectation, k = ((RS+RA)/G)^0.287, regressed toward .500. In parallel, an Elo rating walks the entire completed game log with a margin-of-victory damper. The two are blended and re-centered on 1500.
  2. 2Game probability. Two ratings become one number the usual way — 1/(1+10^(−Δ/400)) — with home field folded into Δ. This is Elo's form of Bill James's Log5.
  3. 3Correlated uncertainty. Each simulated season redraws every team's true talent from its posterior and holds it fixed for all their remaining games. If the Rays are really a 90-win roster, that's true for all of September at once. Skip this and your intervals come out far too narrow.
  4. 4Standings and tiebreakers. All remaining remaining games are played out, then seeds resolve under the real post-2022 rules — no Game 163. Head-to-head record, then intradivision win%, then intraleague win%.
  5. 5October. Three division winners and three wild cards per league. Seeds 1–2 bye; 3v6 and 4v5 play a best-of-three entirely at the higher seed; then a best-of-five and two best-of-sevens with the real home patterns.
Why the title odds look low. A team that is a 55% favorite in any single game still wins a best-of-seven only about 61% of the time. Stack four rounds of that and even the best roster in baseball rarely clears 20%. A model that tells you a club is a lock for October and a longshot in November isn't contradicting itself — that's the actual shape of the sport.
Source · MLB StatsAPI Method · Pythagenpat, Elo, Log5 No key, no backend, all client-side