College Football 2026

NFL 2026 Scoreboard

Arrange tabs

Drag a tab or use the arrows to change its position. Your order is saved in this browser.

    Sort

    Live games

    Loading live scores…
    How this works

    Scores come straight from ESPN and refresh every 30 seconds while games are on (every 5 minutes otherwise). Live cover % is the chance to cover the Vegas spread (not a moneyline win chance), from the current score and time left, for our pick's side. Our pick is this page's final pick against the spread, with whether it is covering right now. Spread shows who is covering the Vegas line right now, and Total shows points so far and the pace they're on for the full game. 1st half and 2nd half show our predicted score for each half, the sportsbooks' half line (the 2nd-half line is posted live at halftime), our pick against it, and once the half starts, its score and whether the pick is winning. Each half is built like the final pick: the same sources and weights, scaled to the half (see 1st & 2nd half picks on the Picks tab). Once the 1st half is over, the 2nd-half projection and pick update from the real 1st-half score. Live updates work in the downloaded NFL Scoreboard.html; the online page can't reach ESPN directly.

    Games

    Each game: the score by half and the final, our final pick and over/under with the bet decision, and your pick. The full reasons, flags and every factor are on the Picks tab.

    Sort

    Sportsbook Review · latest odds

    How this works

    Latest listed sportsbook lines from Sportsbook Review alongside the DraftKings opener when available. Each cell shows the favorite's spread with its price, then the total. A highlighted cell has moved since the open. Books stop moving full-game lines at kickoff, so during a game the 2nd half lines are the live ones. Best for our pick is the most points (then the best price) any book offers on our final pick's side.

    On your computer, double-click Live Odds.bat to refresh SBR listings every 30 seconds while its window stays open. Individual quote times and live-in-play status are not supplied; a recently collected listing is not proof the book changed it recently. The 1st-half lines also fill in the 1st-half markets on the Deep tab when you haven't typed your own.

    Summary · season so far

    How every kind of pick has done against the line, split into Team (spread: who covers) and Score (over/under: total points). Finished games only; pushes don't count toward the %. Green = 52.4% or better, the break-even at −110.

    Team · spread picks

    Sharp money × public, together (our final spread picks)

    Score · over/under picks

    Sharp money × public, together (our final over/under picks)

    My picks

    Tap a side under My pick on any game card, or in Final predictions below, to make your own spread and over/under pick. Tap it again to take it back.

    How this works

    Each pick keeps the line at the moment you tapped it, and is graded on that line once the game is final (pushes count as pushes). Picks lock at kickoff, so the record can't be changed afterwards. On the online page your picks are saved to your claude.ai account: private to you, kept through every update, and the same on your phone and computer. The downloaded copies save them in that browser only.

    Bet decisions

    Every spread and over/under pick, all factors added up into one decision and ranked best to worst.

    How this works

    Each pick's Our read already adds up the edge, sharp money, the public, the deep matchup, the sharp sportsbooks, the outside sites and the red flags. The rank score adds the cover chance on top (55% adds 1 point, 60% adds 2) and half a point when sharp money is with us, so spreads and totals can be ranked together.

    • BET: read Good, or Mixed-leaning-good with sharp money on our side, and no red flag.
    • SMALL BET: read Mixed, leaning good, and no red flag.
    • PASS: leaning bad, bad, any red flag, or an over/under while those picks are suspended.

    The record grades every decision on finished games against the closing line. It says how the rules would have done, not a promise for future weeks.

    Sort

    Full game

    1st half

    2nd half · decided at halftime, with the 1st-half score

    Strong bets

    Final picks that beat the Vegas spread or the Vegas total by 3 points or more, the most confident level.

    Sort

    Over/under

    Our total points against the Vegas total, with every factor and each team's over/under trend added up.

    How this works

    Our total is a weighted mix: Vegas total 40%, home/away model 20%, standard model 10%, OddsShark 10%, DRatings 10%, Dimers 10% and play-by-play efficiency (EPA, pulled halfway back toward average) 10%, rescaled over the sources a game has. Both models already include kickoff weather, indoor/outdoor and rain.

    Then the over/under trend: how often each team's games went over the Vegas total this season (3+ games each). When the two teams together went over (or under) 65%+ of the time, the total moves up to 1 point that way.

    The pick is Over when our total is above the Vegas total, Under when below. Strong = 3+ points away, Lean = 1.5–3, Slight = under 1.5. Finished games use only what was known before kickoff. Pace is seconds per snap in close games (lower = faster, more plays).

    Fade test: what taking the opposite would have done
    Sort

    Final predictions

    Every final pick is against the live sportsbook spread (not the moneyline), with all factors added up, and graded on the closing line.

    How this works

    Every factor counts. Picks and results go by the listed sportsbook line (Sportsbook Review, middle of available books; quote time unavailable), or the Vegas consensus when there are no live odds for a game. The predicted margin is a weighted mix of: that market line 40%, home/away model 20%, standard model 10%, power ratings 10% and Dimers 20% and the Fanatics Markets prediction market 10% (weights rescale when a source is missing). Both models already include key-player injuries, kickoff weather, indoor/outdoor and rain. Then small nudges: spread money movement ±0.5 pt (±1 when strong) and Covers users' spread picks ±0.5 when 60%+ agree, and the day/night trend for the game's kickoff slot, up to ±1.5 pts on the margin and ±2 on the total (see Day vs night games on the Results tab). Moneylines and straight-up winner picks are not used. The total is Vegas 50%, home/away model 30%, standard model 20%.

    The final pick is the side our final margin beats the Vegas spread on. Strength: Strong when it beats the line by 3+ points, Lean 1.5–3, Slight under 1.5. Cover chance uses the same 13.5-point spread of outcomes. Tap "How every factor adds up" on a game card to see each source's share.

    Sort

    Our read × cover chance: which picks have won

    Each verdict split by the pick's cover chance, on finished games. Does a higher chance inside the same verdict win more often?

    Fade test: taking the other side of "Mixed, leaning bad" picks

    Cover and miss count, week by week

    1st & 2nd half picks

    Our side and over/under for each half, against the sportsbooks' half lines.

    How this works

    Each half is built the same way as the final pick: a weighted mix of the same sources at the same weights, then the same nudges. The sources: the books' half line 40%; our half model (each team's scoring and allowing in that half) 20%; the standard model 10% and Dimers 20%, scaled to the half's usual share of the game (learned from this season's games); and power ratings 10% (1st-half power ratings for the 1st half). Totals add OddsShark, DRatings and EPA the same way. The money, Covers users and sharp-book nudges, and the over/under trend, are scaled to the half too. Tap "How it adds up" to see every piece. The deep matchup is shown as a check (✓ agrees, ✗ disagrees), as on the final pick.

    At halftime the 2nd half updates: every 2nd-half source except the books' line moves by part of the 1st-half surprise (the real 1st-half score minus our 1st-half projection), in an amount learned from this season's finished games. The books' 2nd-half line, posted at halftime, already knows the score.

    Lines are the middle of the books listed by Sportsbook Review. Books post the 2nd-half line at halftime, so before kickoff the 2nd-half pick shows only our projection; it fills in on the live page once the line posts. The cover chance uses this season's own half-score misses. Strength: Strong 58%+, Lean 54%+, otherwise Slight. The 2nd half includes overtime, as at most sportsbooks. Break-even at −110 is 52.4%.

    Home/away and standard picks

    Who each of our two models picks to win, side by side.

    How this works

    Home/away uses the home team's home scoring and the visitor's road scoring. Standard uses each team's overall scoring plus a 1.5-point home edge. Both include passing & rushing, key-player injuries, weather and indoor/outdoor. "Spread" is how many points the model expects the winner to win by. When the two models disagree, the game is usually close.

    Picks vs the Vegas line

    Which side to take against the consensus spread, and over or under the total.

    How this works

    Four sources each say which side beats the spread: our home/away and standard models, Dimers and the money movement. A source counts when it beats the line by more than half a point.

    • Strong: 4 or more agree and our models beat the line by 3+ points.
    • Lean: more sources one way than the other.
    • Pass: no clear side.

    Totals: our models' predicted points must be 2+ points over or under the line. Hover the Sources number to see each source. Break-even after the sportsbook's cut is about 52.4%.

    Cover check · outside sites

    Who each analysis site says covers the spread, next to our final pick. Spread only: no winners or totals.

    How this works

    OddsShark (computer picks) and DRatings publish a predicted score; the side whose predicted margin beats the Vegas line is their cover pick. nfelo (a model built on nflverse data) publishes its own spread line and marks the side with positive expected value. Pickswise is an expert's point-spread pick. Dimers and Covers users come from the existing prediction sources (Covers users are the public's spread picks). Fanatics is the Fanatics Markets prediction market: its fair line, worked out from the prices traders pay on a ladder of spreads, compared with the sportsbook line.

    A site's pick is checked against the consensus line (or the Sportsbook Review listing when there's no consensus yet). A pick listed at a line more than 3 points away is ignored, since it's probably a different market. The record grades each site's last pre-kickoff pick on finished games against the spread; pushes don't count. Break-even after the sportsbook's cut is 52.4%.

    Money trend · all betting sites

    Where the bets (the public) and the money are going on the spread and the total, from Action Network, DraftKings, ScoresAndOdds and SBR.

    How this works

    Each site reports what share of bets (tickets) and what share of money (handle) is on each side. SBR reports bets only. The public side has 60%+ of bets on average. Sharp money is a side whose money share beats its bet share by 8+ points on average: fewer, bigger bets, which usually come from sharper bettors. A reverse line move is the line moving toward the side with under half the bets.

    The money side feeds the final pick as a small nudge (half a point, or one point when strong). The record grades the money side on finished games against the spread.

    Sharp money win % · season

    Sportsbook analysis

    What the individual sportsbooks' lines say: sharp books against public books, how far the line moved, the best number and the cheapest book.

    How this works

    Lines from sportsbooks listed by Sportsbook Review. Sharp books (Circa, Bookmaker, Pinnacle when listed) take big bets from professional bettors and usually move first; public books (DraftKings, Bovada) cater to casual bettors. When the sharp books' average spread is half a point or more different from the public books', the side the sharp books rate higher is the sharp lean, and it adds half a point to the final pick.

    Move compares the opening line with the middle of the books now; crossing 3, 7 or 10 matters most because so many games end by those margins. Best number is the most points any book gives our pick's side. Cut is the bookmaker's built-in margin on the spread (lower is cheaper to bet). No-vig win % is the moneyline with that margin removed.

    Best picks to win

    Who wins each game, ranked by confidence.

    How this works

    Our models (averaged), Vegas (from its spread) and Dimers are blended into one win chance; the money movement adds a vote.

    • Strong: 68%+ and every source agrees.
    • Solid: 60%+ with 3 of 4 agreeing.
    • Lean: 53%+.
    • Toss-up: closer than that.

    Money movement compares Action Network's share of bets with share of money, plus how far the spread moved since it opened. Big money on a side, or the line moving toward the side with fewer bets, points to where the larger bettors are.

    Key injuries

    Team leaders on the injury report, biggest impact first.

    How this works

    Key players are each team's season leader in passing, rushing, receiving, sacks and tackles. A starting QB is worth 5 points, other leaders 1 (tackle leader 0.5), counted fully if Out or on injured reserve (IR), 75% if Doubtful, 25% if Questionable. Offensive players lower their own team's score; defensive players raise the opponent's.

    Deep Matchup Analysis

    Pick a game: the verdict comes first, then the projected score and a card for each of the six bets. Tap any dropdown for the details.

    Enter 1st-half lines (from your sportsbook)

    Available automatic first-half lines are used when present. Enter your own lines here to override them. Spread is the home team's line (e.g. −1.5). Leave odds blank for −110. Saved in this browser only.

    Enter Full-game lines (from your sportsbook)

    Home-team spread; totals apply only to this period. Your entries override automatic lines for calculations. Existing saved forecasts remain frozen. Blank odds default to −110. Use Clear to restore automatic lines.

    Enter 2nd-half lines (from your sportsbook)

    Home-team spread; totals apply only to this period. Your entries override automatic lines for calculations. Existing saved forecasts remain frozen. Blank odds default to −110. Use Clear to restore automatic lines.

    Sharp-money and second-half records

    Sharp money here means a money-minus-ticket share gap of at least 8 percentage points, averaged only across sources reporting both. It is a proxy, not proof of professional betting. Full-game signals are never copied into 1H or 2H markets. Saved sharp picks use the original line; first-half/second-half sharp records remain unavailable until market-specific splits exist.

    Historical sharp signals are recovered only from timestamped pregame snapshots with a saved betting line. Weeks without those records remain empty. A past model comparison is shown only when a model forecast was saved. New signals are captured while this page is open. 2H pregame and halftime records are separate forecasts; use the timing filter to avoid mixing them. Halftime capture requires a scoreboard update within 10 minutes showing HALFTIME. First eligible snapshot per market and timing is frozen.

    Win rate = W / (W + L). Pushes, pending and voids excluded. These records do not establish calibration. 2H probabilities are tracked at cumulative 55%+, 60%+, 65%+ thresholds.

    Saved MATCHUP analyses & pick results

    First available pregame snapshot per game. Every priced market is retained, including PASS/monitor candidates. Past games without a saved forecast are excluded. Missing first-half lines at capture remain unavailable in that snapshot.

    Cumulative groups: a 67% cover-probability pick appears in 55%+, 60%+, and 65%+. Win rate = wins ÷ (wins + losses); pushes, voids and pending picks are excluded. Full-game grading includes overtime; first-half grading uses Q1 + Q2. Each game/market counts once. These records do not change validation status.

    Saved results · the six bets

    First eligible forecast per market and timing is frozen. Pregame and halftime predictions are separate. Results use the saved line; missing forecasts stay unrecorded. Win % = wins / (wins + losses).

    First-half historical validation and release requirements

    1H Spread and 1H Total are evaluated separately. The record below uses the first forecast saved before kickoff, including its line, displayed odds, model version and probability. All forecasts are saved for calibration; selected-pick results use the cover/value rules without requiring prior release approval. PASS cards can therefore contribute evaluation evidence without becoming recommended bets.

    • UNTESTED: no settled saved first-half forecasts. Official picks remain PASS.
    • PROVISIONAL: settled evidence exists, but the validation requirements below are not all met. Original provisional-pick behavior is retained: all price, cover, data-quality and value gates must still pass.
    • VALIDATED FOR STATED SCOPE: at least 100 decided selected picks, positive unit profit at the saved reference prices, and the probability calibration check below passes. Applies only to this policy version and first-half market; it is not a guarantee.
    • SUSPENDED: at least 10 decided selected picks and the probability of that many or fewer wins under their individual break-even prices is below 5%. Official picks remain PASS; forecasts continue to be logged. Release resumes only when recalculated status clears suspension.

    First-half market blending and probability calibration

    Market blend: 70% model + 30% first-half market. First-half margin = 0.70 × model margin + 0.30 × negative home spread. First-half total = 0.70 × model total + 0.30 × market total. Before that, the margin blends scoring-model and first-half power ratings 55/45 through Week 8, then 65/35. Totals do not receive that power-rating margin blend. These are fixed starting weights, not historically optimized weights.

    Only first-half lines enter the 1H blend. Missing lines mean no priced pick. At least 24 earlier games with usable first-half residuals and finite positive uncertainty estimates are required before simulating 1H scores. Otherwise the market stays PASS.

    Calibration check: at least 100 decided saved predictions in this market, weighted absolute predicted-versus-observed gap across fixed 10-percentage-point probability bins ≤5 percentage points, and Brier score <0.25. Pushes are excluded; probability is Model P = win / (1 − push). Cover probability still includes push risk. These are declared app release checks, not proof of universal accuracy. Reference prices may contain the existing provider's inferred prices; profit is a diagnostic, not realized wagering profit.

    No fitted probability correction is applied. Until sufficient saved evidence passes these checks, probabilities are labeled uncalibrated / provisional estimates. Estimating score uncertainty alone does not establish probability calibration. Thresholds are evaluated separately for each market and policy version.

    Historical first-half scores — imported evidence and validation gaps

    Validation blocked: historical 1H lines and pregame model inputs are not paired. These are observed scores, not model picks. They do not count toward the 100-pick release requirement or prove calibration. A chronological backtest must recreate each forecast using only earlier information and authentic first-half lines/prices. Full-game lines must not be divided in half to fill missing markets.

    The checked SBR historical-date page returned a different game's date range; its rows were rejected. The Odds API documents historical half/quarter markets on paid plans, subject to bookmaker coverage. No subscription was purchased.

    Historical odds source documentation · ESPN score source

    Date (UTC)AwayHome1H (A–H)2H regulation (A–H)Final incl. OT (A–H)

    Only complete quarter-score records whose quarter sums match final scores are imported. This archive is kept separate from the current-season model inputs to avoid cross-season leakage. Missing coverage is not treated as a zero score.

    Model setup & validation

    What a bet must pass, and how the same rules did on past games.

    Retrospective full-game diagnostic (assumed prices)

    How to read the summary

    Each finished game is graded exactly as an upcoming game would be, using only earlier weeks and the saved closing line at an assumed −110 (the real prices weren't saved). Only bets that passed both gates count as selections. Status: SUSPENDED when 10+ decided selections are clearly below the 52.4% break-even (a record that bad would happen less than 5% of the time if the market really broke even), VALIDATED FOR STATED SCOPE at 100+ selections with a profit, otherwise PROVISIONAL. A suspended market can't produce official picks until its record recovers. For first-half release status, use the prospective record and requirements above. Older 1H rows in this retrospective table have no saved historical lines and remain empty. Brier score measures probability accuracy over every graded side (lower is better; 0.25 = coin-flip guessing).

    Framework checklist (20 checks)

    MATCHUP Results · Saved Model Predictions

    Track the independent Matchup tab's saved forecasts, not selections from the Best Bets tab. These are three non-overlapping confidence groups (not cumulative): 55.0–60.0%, 60.1–65.0%, and 65.1%+. Records reflect each forecast's original confidence, side, line and price. Win rate is wins / (wins + losses); pushes, pending and voids are excluded.

    Records are captured only when a forecast and its actual period-specific market line are available at the eligible pregame or halftime decision time. The same game/market/timing is counted once. Previously played games without a frozen prediction cannot be reconstructed as historical model picks. Browser storage can be lost: export a backup regularly. These estimates are unvalidated and do not establish a proven betting win rate.

    Power ratings & strength of schedule

    How good each team really is once you account for who they've played, and how hard their schedule has been.

    How this works

    Rating is points better than an average team on a neutral field. Offense is points scored above average and Defense points kept off the board compared with average (higher is better for both). They're worked out together, so scoring 30 against a strong defense counts for more than 30 against a weak one, and each is shrunk a little because there are only a few games.

    Schedule so far / left is the average rating of the opponents a team has played and still has to play (higher = harder; rank 1 = hardest).

    Efficiency · EPA & success rate

    Play-by-play efficiency for every team, best net EPA first.

    How to read this

    From nflverse play-by-play, the free data that RBSDM and Sharp Football Analysis build their team stats on. Only regular-season passes and runs count.

    EPA/play (expected points added) is how much each play changed a team's expected points: higher is better on offense, lower is better on defense. Net is offense minus defense. Success is the share of plays that gained enough (40% of the yards needed on 1st down, 60% on 2nd, all of it on 3rd/4th). 1H EPA covers the 1st half only. Pass rate over expected is how much more (or less) a team throws than an average team would in the same situation (1st/2nd down, game still close). Pace is seconds between snaps when the score is within 7 in quarters 1–3 (lower = faster). Explosive is the share of plays with 20+ yard passes or 10+ yard runs. Red zone TD is the share of drives reaching the 20 that ended in a touchdown.

    EPA is also a learned factor in our model ("Efficiency (EPA)" in How it works): it's only used if it would have made past predictions more accurate.

    Ratings & betting trends · TeamRankings

    TeamRankings' predictive rating, records against the spread and the total, and 1st-half scoring.

    How to read this

    Predictive rating is TeamRankings' points-better-than-average rating for future games. ATS is the record against the spread (wins-losses-pushes), Cover % how often the team covered and ATS +/− the average margin by which it beat (or missed) the spread. Over/under is how its games went against the total, and Total +/− the average points above the total. 1H scored / allowed are 1st-half points per game. ATS and over/under records swing a lot on small samples, so treat them as context, not predictions.

    Luck check

    Which teams' records are better or worse than their play, sorted luckiest first.

    How this works

    Should-have wins is the record a team's points scored and allowed usually produce (the "Pythagorean" formula). Luck is actual wins minus that: teams well above it often come back to earth. Close games is the record in games decided by 8 points or fewer, which tend to even out. Turnover margin is takeaways minus giveaways per game; big margins in either direction usually shrink later in the season.

    Standings

    Week-by-week results

    Every team's result each week: Win, Loss or Tie, the score (team first), and the opponent (@ = away). Sorted by record. Swipe sideways for later weeks.

    Home & away records

    How this works

    Each team's win-loss record and points per game at home and on the road. Games at neutral sites (like the international games) are left out of both and listed separately. Home edge is how much better the team's point difference per game is at home than on the road. Sorted by home edge. Click a column heading to sort.

    Head coaches

    How this works

    From ESPN: each team's current head coach, how many seasons he has been an NFL head coach, and his record this season. A coach change is flagged when a coach's record covers fewer games than his team has played (he took over mid-season), or when the weekly update finds a different coach than last time. 1st-year coach means his first season as an NFL head coach; a coach who moved from another team isn't flagged, because ESPN doesn't keep past seasons' coaches.

    In the model: coaching is shown for information but doesn't change the predicted score. There's no reliable way to measure its effect from a few games, and new coaches' teams are already reflected in this season's scoring.

    Passing & rushing

    Each team's style and yards per game, gained and allowed.

    How this works

    Style compares how often a team passes with the league average share of pass plays: 4+ points above is a passing team, 4+ below a rushing team, otherwise balanced.

    In the model: each team's expected passing yards are the average of its passing offense and the opponent's pass defense, and the same for rushing. Those yards are turned into points at the league's points-per-yard rate. Yards are steadier than points, so the model moves part of the way toward this yards-based score. How far is learned from earlier weeks (at most halfway, so one bad stretch can't take over); it's 0 until there are enough games to measure.

    Team averages

    Points scored (off) and allowed (def) per game. Click a column to sort.

    Indoor vs outdoor & weather

    How scoring changes with the roof and the weather.

    How this works

    Outdoor games use the recorded weather at kickoff (Open-Meteo): temperature, wind, and rain over the game's first few hours. Retractable-roof stadiums count as indoor. Groups with only a few games can swing a lot, so check the Games column.

    Teams indoors vs outdoors

    Playoff chances

    The rest of the season, played out 10,000 times.

    How this works

    Every remaining game is "played" 10,000 times using our win chances: the final prediction for this week's games (which includes Vegas) and the home/away model for later weeks. Each run counts each team's wins, crowns the 4 division winners in each conference (seeds 1–4 by record) and adds the 3 best other records as wild cards (seeds 5–7). Only the #1 seed gets a first-round bye.

    Ties in the standings are broken by point difference so far, then at random. The NFL's real tiebreakers (head-to-head, division and conference record) aren't modeled, so close races can be off by a few percent.

    AFC

    NFC

    What the model has learned

    On top of team scoring, each factor is tested on every finished game, each week using only what was learned before it. "Average miss" is how far each team's predicted points were from its real points after adding that factor. A factor is only used for this week's predictions if it made past predictions better; the others stay off until they prove themselves, and the test reruns with every update.

    Factor check

    A closer look at the factors that aren't switched on yet. Each is tested on finished games, using only earlier weeks, on top of everything before it in the model. You can switch any of them on for this week's predictions; the page reloads to apply it, and the choice is saved in this browser only.

    Where the data comes from

    • Scores, schedule, injuries: ESPN
    • Betting lines: VegasInsider consensus and 8 sportsbooks; DraftKings closing lines via ESPN for older games
    • Money movement: Action Network public betting (share of bets and share of money)
    • Outside picks: Dimers, Covers users
    • Latest sportsbook listings: Sportsbook Review
    • Play-by-play efficiency (EPA, success rate, pace, pass rate over expected): nflverse, the data behind RBSDM and Sharp Football Analysis
    • Predictive ratings, ATS and over/under records, 1st-half scoring: TeamRankings
    • Weather: Open-Meteo forecasts and recorded weather

    Outside picks and betting lines cover the current week only. Kickoff times are in your time zone.

    How our models predict a score

    1. Team scoring: each team's expected points = the average of its offense and the opponent's defense. The home/away model uses home and road numbers, blended with overall averages because there are only a few games of each; the standard model uses overall averages plus a 1.5-point home edge.
    2. Learned factors: passing & rushing matchups, strength of schedule (opponent-adjusted ratings), rest & travel (byes, short weeks, miles, time zones, early body-clock kickoffs) and turnover luck. Each is only used if it improved past predictions; see "What the model has learned" above.
    3. Key injuries: a starting QB out is worth 5 points; other team leaders 1 (tackle leader 0.5).
    4. Weather: wind over 10 mph and freezing cold lower the total; indoor/outdoor and rain adjust it by how much they changed scoring this season.
    5. Win chance: assumes final margins land within about 13.5 points of the prediction either way.

    The models ignore strength of schedule and injuries beyond team leaders. Treat every prediction as a rough guide, not betting advice.