Playoff Probabilities
Season — — 0 seasons simulated — pp at 95% Starting
In Out2026 Schedule
Show
no games picked
Checking the live feed
Season — — 0 seasons simulated — pp at 95% Starting
In OutHow to read it. Every figure is washed by how likely it is: quiet grey towards nothing, deepening through blue towards certain. The heavier rule falls after the seventh club in each conference — everything above it is currently making the field. Club colours appear only in the marks, so a colour in the table always means a probability and never a team. A cell reading 0 keeps that same low grey but adds diagonal ruling: the outcome did not happen in a single simulated season. That is different from <1, a live chance too small to round up. Click any row for its full seed distribution.
Playoffs is the chance of earning any seed. Division is the chance of winning the division, #1 Seed of earning the conference's bye, Conf of winning the conference, and Title of winning the championship. In an expanded row, Seed 1 through Seed 7 and Out are mutually exclusive and together describe every simulated finish.
6k, 25k, and 60k select how many complete seasons are sampled. Larger samples reduce Monte Carlo noise; they do not make the underlying football model more accurate. The ± figure above the table is the worst-case 95% sampling margin near a 50% probability—not a confidence interval for whether the model itself is right.
After you force one or more games, a small corner arrow remains on every affected probability. It is the cumulative percentage-point change from an otherwise identical no-pick forecast using the same model and sample count—not merely the most recent game's movement. Clearing every pick removes the arrows.
Picking games. Choose a winner — or a tie — for any game still to be played and the whole season is simulated again from scratch. It does not filter a pre-computed set, so you can pick as many games as you like without the numbers falling apart. The count above the table is how many games you have picked; the rest are still being simulated. A record set in italic over a dotted rule is a projection — it counts your picks as well as the games actually played.
The slim figures below each unplayed game are this preview model's pregame win, tie, and win estimates. Picks do not rewrite these figures, and live games hide them because this forecast does not calculate in-game win probability. Final rows show the real score instead of reconstructing a historical estimate the model did not save.
Strength is rebuilt from 7,276 completed regular-season and playoff games from 1999 through 2025. Recent games carry more weight through game-by-game updates and annual regression toward average. Every franchise begins from the same neutral 1500 rating and follows the same rules; there are no team-specific adjustments.
This is not a live sportsbook consensus. Until fitted line-derived inputs arrive, Market expresses the same historical club strengths in points and passes them through a market-shaped probability curve. It is an alternate preview, not independent evidence.
@ marks the home club. International and other neutral-site games use vs instead.
Live scores appear above the forecast and on the corresponding schedule row. While a game is in progress, its picker remains hypothetical because this preview does not model live win probability. A validated final locks the row and enters the standings. When a new final is accepted, an open page reloads and replays all accepted 2026 finals into Elo before simulating the remaining season.
Pick the rest of the season and watch the odds move.
Monte Carlo preview
Background worker · progressive estimates · recent scenarios reused
The browser simulates every unpicked game off the main thread, shows an early estimate after 4,000 seasons, then refines to the selected sample count. Repeating the same model, picks, and sample count reuses a small in-memory result instead of running again. If a picked scenario needs a no-pick comparison for a newly selected model/sample combination, a second background worker computes that baseline once and reuses it for the rest of the session. The schedule is official and Elo now starts from real historical results; roster/QB effects and tiebreakers remain preview approximations. The ± figure above is the worst-case 95% sampling margin, in percentage points, near a 50% probability.
During active game windows, an open page checks a small same-origin scoreboard snapshot once a minute. Live scores never enter the model; only validated finals do.
Picks are encoded in the part of the URL after # and simulated in your browser; that URL fragment is not sent with ordinary web requests. Copy Link intentionally includes it so someone opening the shared link sees the same forced-game scenario.
Elo uses the history-based ratings directly. Market is not using current betting lines: for now it translates the same club strengths into points and applies a market-shaped curve. The two buttons explore different probability shapes, not two independent sources of information.
A chronological 2006–2025 online comparison covers 5,431 games. Lower is better: Elo's Brier score is 0.220517, versus 0.249356 for a neutral 50/50 forecast—an 11.57% relative improvement. That shows the historical ratings contain predictive signal; it does not prove the 2026 probabilities are perfectly calibrated. A held-out calibration and log-loss study are still future work.
The released schedule, accepted scores, and history-based Elo inputs are real. The ratings do not yet know about quarterbacks, roster changes, coaches, injuries, transactions, or betting markets. The tie estimate and winning-percentage/random tiebreak are also approximations rather than the NFL's complete procedure, so the forecast remains an experimental preview rather than a finished predictive model.
Matchups, dates, announced kickoff times, networks, and venues come from the official 2026 NFL schedule. Games marked TBD remain subject to the league's flexible scheduling procedure. Final scores, when present, are validated from ESPN's scoreboard. An open page checks the small same-origin snapshot once a minute only around active games; validated finals remain on their schedule rows and enter the standings.
The preseason Elo prior is rebuilt from completed games in nflverse's historical game archive using the score-margin update and annual regression described by FiveThirtyEight's published NFL Elo implementation. The exact input file is hash-pinned in this release so upstream changes cannot silently alter the forecast.
An independent rebuild in the spirit of FiveThirtyEight's retired NFL forecast. Not affiliated with FiveThirtyEight, ABC News, or the NFL. Club and conference logos are trademarks of their respective owners and are shown only for identification.