Methodology

What a rate is

Every finding compares a situation with a baseline in the same league, for the same stat and the same kind of player. The baseline is the complement inside a stated universe: rain games are compared with other outdoor games where rain status is known, and the first game of a road trip is compared with other road games. Each finding page states its baseline.

pp means percentage points: the difference between two rates. If a rate is 40% in the spot and 62% in the baseline, the difference is −22 percentage points. A minus means the rate was lower than the baseline; a plus means it was higher. This is a description of the rates, not a betting instruction.

“X+” means the value reached X or more (the same as more than X−0.5), so a push can’t happen and is never counted as a hit. Player rows use the same usage filter on both sides: NFL QBs with 10+ pass attempts, RBs and receivers with 30%+ of snaps; NBA players with 20+ minutes; NHL skaters with 10+ minutes on ice. Those filters are in-game, so they describe what happened rather than something known before the game.

When a rate counts as an edge

1. Sample floor. Team and game rows need 200+ distinct games. League-wide player rows need 100+ player-games from 30+ distinct games, so one holiday slate can’t qualify. Play-level rows need 200+ plays.
2. Clear of the baseline. The 95% interval (the wider of a Wilson interval and a game-clustered interval) must exclude the baseline.
3. Multiple testing. About 2,800 comparisons were run, so a Benjamini–Hochberg false-discovery check (q ≤ 0.10) is applied.
4. Quality adjustment. Team rows are compared with a same-season expectation built from both teams’ other games plus home edge. Player rows are compared with each player’s own rate in his other games that season. A row is an edge only if this adjusted gap also excludes zero in the same direction. Game-level and play-level rows don’t need this step.

Rows that pass the sample floor but not the interval or q-value check are labeled Myth checked. Rows that pass both but fail the quality adjustment are Explained by quality. Team scoring and home-win rows without an adjustment are Context only. Rows below the sample floor aren’t published as findings; inside a season window they’re shown as Too small with their sample.

Model Tracker and This week

A model on Model Tracker is one published rate that counts as an Edge under the rules above, all time. Every edge is listed; myths, rows explained by quality, context-only rows and rows below the sample floor are not models. The big number is the all-time rate in the spot (an average for scoring measures), shown next to the baseline (“typically”) and the real count, for example 73 of 181 QB games. The This season and Last season lines are the same measure recomputed from only those seasons’ games (the season windows below), never scaled from all time: “holding up” means that window is itself an edge in the same direction, “same direction, not clear yet” means it leans the same way without passing the tests, “not showing up” means it doesn’t lean that way, and “too few games to judge” means it is below the sample floor in that window. Situations that apply to this week’s games are listed first, then by the size of the gap (the quality-adjusted gap where there is one).

This week’s edges lists the upcoming games on the loaded slates and the models that apply to each, tagged from each team’s stored schedule before the game (rest, byes, last game, last meeting, division, back-to-backs). Weather, roof status and national TV aren’t forecast, so those models never appear there. Only edges are listed; the other spots checked for a game stay on the league guide.

Season windows

On Findings and on every finding page you can switch between all time (the default), this season, last season, and the last 5, 10 or 20 seasons. This season is the season underway: every stored game so far, capped at the same evidence date as all time (NFL through 2026-11-09, NBA and NHL through 2026-11-11). Early in a season most measures are below the sample floor and show as Too small. Last season is the most recently completed full season, never the one in progress. The last 5, 10 and 20 seasons include the current season if it’s underway. Each window is recomputed from the games in those seasons with the same rules: the same sample floors, intervals, false-discovery check (run separately inside each window) and quality adjustment. Nothing is scaled from the all-time rate, so a spot can be an edge all time and a myth or too small in a single season. When fewer seasons are stored than a window asks for, the window says so and equals all time.

All-time figures are the published findings computed 2026-11-12 14:16 ET. This season, last season and the other shorter windows were computed 2026-11-12 19:03 ET from the database at that time, capped at the same evidence dates. If games were added or corrected in between, a shorter window can differ slightly from the matching seasons inside all time. The late-and-close play-level rows only have play-by-play from 2024 on.

The games behind each finding

Every finding page lists the actual games behind each measure: date, matchup, final score, the number that was measured and, for rates, whether it happened. A green row marked Hit means it happened in that game (for example the QB reached 200+ passing yards, or the team won); a red row marked Miss means it didn’t. “X+” lines can’t tie, so there are no pushes on these measures. Averages (points scored, margin, combined points) show each game’s number without color, because an average has no hit or miss and is not an over/under record. Click a game for the full breakdown: venue, kickoff time (ET), the situation (rest, previous game, streak, road-trip game, playoff series) and, on NFL games, that day’s weather from the validated game-weather record: reported kickoff temperature and wind where the game record has them, otherwise Open-Meteo historical hourly reanalysis at the stadium for the four hours from kickoff (rain, snow and gusts are modelled). Each list follows the season window and adds up exactly to the published rate and sample; a list that doesn’t is hidden. No betting lines are shown: there is no verified start-of-game line history.

The earlier game behind a spot

On spots where one earlier game is the reason a game is listed (after a blowout win or loss, after overtime, on a 3+ game streak, revenge after the last meeting, first game back, bounce-backs), clicking a game shows that earlier game in its own box: the date, the final score, the margin, and for players his box-score line from that game. If he didn’t play in it, the box says so; if the game or his line can’t be matched, it says unknown. Nothing is filled in.

Two ways to read a game: the published mark, or the season average entering it

Player lists (and team scoring lists) have a switch, Hit or miss means. Published mark (the default) colors each game by the measure the finding is built on, for example 200+ passing yards, so the list adds up to the rate on the page. Season average entering the game colors each game by whether the player (or team) went over their own average for that stat from earlier games that same season: same player, same role group (for example QB games with 10+ attempts), same season, games dated before this one only. The game itself and anything later never count, and last season or career games are never used. At least 3 earlier games are needed; with fewer, the game is shown in grey as “No avg” and left out of the count. Over is green, under is red, an exact tie is a grey Push. The summary count follows whichever method is selected, and both numbers are shown on every row. The season average is a stand-in for a line on historical games, not a Vegas line: verified start-of-game prop and game lines aren’t stored, so none are shown. Season averages are not computed for whole-game combined totals (no single team’s average applies) or for wins and overtime.

Evidence window

NFL seasons 2020, 2021, 2022, 2023, 2024, 2025, 2026, through 2026-11-09. NBA and NHL from 2023-24, through 2026-11-11. The findings were last computed 2026-11-12 14:16 ET from free public box scores (ESPN, NHL), nflverse play-by-play and Open-Meteo weather.

What isn’t here

• No betting records. There are no verified closing lines, so there are no against-the-spread or over/under hit rates. Game totals are scoring averages, never “under” records.
• Birthdays are only known for players on current rosters, so older seasons are thinner.
• Weather is known for about half of NFL games (outdoor games with a validated record).
• Not measurable from local data: ATS / over-under hit rates (e.g. prime-time unders), Afternoon national TV (NFL), Contract year, First game after being eliminated, Friday 13th / jersey numbers, Homecoming (NHL), Homecoming by college city, Look-ahead / trap game, MLB (birthday HR, day after night, extra innings), Milestone chase, Must-win for seeding, NBA play-in, Return from injury (reason known), Rookie first playoff game (NHL), Sibling / parent matchups.
• No MLB findings in this version.

Coverage

The database pages

Team and player pages show box-score totals and averages exactly as stored. Records come from stored final scores; they aren’t official standings, and games that weren’t loaded aren’t counted. Player pages exist for everyone with a game in the latest season, or 5+ games the season before. Game logs cover those two seasons; season summaries cover every stored season.

Player and team pages also have recent-form windows: the last 5, 10 or 20 games, this season, last season, or both. A player’s games are the games he appeared in, not every team game; if fewer games are stored than the window asks for, the page says so and shows them all, and the last 10 or 20 reach back into last season when this season is short. These are descriptive numbers, not findings: no sample floors or edges apply.

NFL return-only lines: the database holds defensive and special-teams box lines for players with no offensive line that game, with offensive stats stored as zero placeholders. For offensive stats these lines are left out (the same rule the database’s own hit-rate tools use), so a receiver who only returned kicks doesn’t get a 0-yard game. Return, kicking and defensive stats still use them.

Prop hit rates

The Hit rates pages count how often each active player went over or under a line in a window of his games: the last 5, 10 and 20 games he appeared in (across this season and last season, so the last 20 can reach back a season), this season, last season, and against his opponent on the upcoming slate (H2H, only when the player’s team is on the loaded slate). A cell like 8/10 means 8 hits in 10 games. Fewer games than the window shows the real count.

Current book lines on this week’s slate (DraftKings and FanDuel, as of Thu Nov 12, 10:58 PM ET) are a separate, dated snapshot shown on the league guides and offered as an optional line on Hit rates; they are not closers and are not used in any finding.

The line starts at this season’s average for that player and market (this and last season together if he has fewer than 3 games this season, so one odd game doesn’t set it), rounded to a half: the .5 just above its whole-number part, so an average of 24.3 or 24.8 starts at 24.5 and a whole-number stat can never push on the starting line (minutes and time on ice are stored with decimals, so they can). The line is a research threshold you set, not a line from any book: there are no odds, prices, implied probabilities or expected values on the site. If you set a whole-number line, a game exactly on it is a push: shown separately (“1P”) and counted in the games but never as a hit, so the percentage is hits ÷ games and over + under + pushes always add up to the games played. Over means the stat went above the line, under means below it.

Colors: 80%+ dark green, 65–79% green, 55–64% light green, 45–54% neutral, 35–44% light red, 20–34% red, under 20% dark red. Combined markets (such as points + rebounds + assists) are sums of the stored parts. NFL offensive markets skip return- or special-teams-only box lines (the database’s own row-source rule), while tackles, sacks and kicking use them. NHL goalie markets count starts only, and power-play goals use the official box score. Opponent rank is the opponent’s per-game total of that stat allowed this season to all opposing players combined, computed from stored box scores, with 1 = most allowed. These are descriptive counts of past games, not findings: no sample floors or edge labels apply.

Big Bettors: how picks are graded

Big Bettors ranks big betting accounts on X by records we graded ourselves, not the records they advertise. We read 27 accounts’ posts from Oct 14 – Nov 13, 2026 (ET). A pick counts only if the post is still public, the bet is written in the post text (not in an image, video, link or paywall) and the post went up before the game started. Each pick is checked against the final score or the player’s box score, using the line written in the post; closing lines are never used.

• Straights and parlays are kept separate everywhere. A parlay counts only when every leg is written out; any losing leg makes it a loss.
• A push is never a win. Win % is wins ÷ (wins + losses).
• Not counted: after-the-fact “cash” / ✅ recaps, live in-game bets, repeats of the same pick, and sports we have no verified results for.
• Ranking: only accounts with 20 or more graded straights are ranked, by straight win %. Smaller samples are listed but not ranked. Break-even for a straight at −110 is 52.4%.
• Units appear only for straights whose post stated the odds, at 1 unit each.
• Limits: accounts can delete losing picks, which we can’t see, so a public record can look better than the truth. 30 days is a snapshot, and these samples are too small to tell skill from luck. Follower counts are as X reported them on Nov 13, 2026, 12:07 AM EST.

Disclaimer

These are historical scoring, usage and win rates. They are not betting records, picks or advice, and a past rate doesn’t predict any single game.