How we record and grade NBA picks

NBA on Hunchmeter: how a daily league works

The same core rules apply to the NBA as to the NFL: we record public picks, note when each one was seen, grade every pick against the official score, and link every record back to the page, video or post it came from. Because basketball is played almost every day and published across different media formats, a few tracking and grading rules work differently from the NFL.

Why the primary leaderboard tracks sources (outlets and channels)

In the NFL, almost every tracked pick comes from a weekly media table where each column belongs to a named writer. In the NBA, most daily picks are published as an outlet column, a YouTube segment or a Telegram post without an individual writer's byline. To reflect what is actually published without inventing bylines or discarding unbylined picks, NBA leaderboards use two views: sources and channels (every tracked website, editorial desk, YouTube channel and Telegram channel) and named analysts (individual writers and speakers whose names are explicitly attached to their picks; one person gets one profile; a writer who publishes under the same full name at more than one outlet is one record, listed under the outlet with the larger graded sample, and a pick posted at both outlets is counted once). Each source is labeled as a single-voice source (one editorial call per market per game) or a multi-writer panel (several named writers at the same outlet who may pick against each other); a panel's page shows all writers combined and the games where every participating writer agreed.

When and how NBA picks enter the record

Every pick carries one of three labels. Recorded before tip-off: we captured the pick from the source's public page before the scheduled start of the game. Timestamped before tip-off: we read the page after the start, but the publishing platform's own post time (a public Telegram post or a YouTube publish time) is before the start. Read after the game: read from the source's page after the game started, without a pre-game timestamp; shown in recording history only, never in records, points or leaderboards. Picks the archive marks as duplicates, parlays, live calls or unreadable as a single pick are listed as Not graded with the reason. Game dates use US Eastern Time: late West Coast games often start at 10:00 or 10:30 p.m. ET, which is already the next day in UTC, so daily slates and game addresses use the Eastern game date with the daily boundary at 4:00 a.m. ET, while machine-readable timestamps store exact UTC times.

Picks we could not match to a game

Before a pick reaches a game page it passes a last check against the schedule: the participant it backs must be a team of that game, the quote must not be about another sport or league, a team-statistic line (rebounds, assists, three-pointers) is not graded as a point spread or total, and the recorded side must match the team the quote backs. Picks that fail this check are not shown and not counted anywhere on the site: 141 in the NBA section so far. They stay in the archive and are reviewed as binding errors.

What we publish: the publication gates

Before a pick reaches a page it must pass the same set of gates. Markets: only moneyline, point spread, over/under and team totals are published at launch; player props, quarter and half markets and other markets stay in the archive for a later phase. Lines: for point spread and totals the line on the card must appear in the quote itself. Teams: the quote must name at least one team of the game (table rows must name both). Quotes that carry a result mark (✅/❌, WIN/LOSS, units won) were written after the game and are not published; headlines and previews without a pick are not published; one quote cannot stand for both sides of a market. Picks read after the game started are never graded. Held back by these gates so far: 1,357 in the NBA section (523 markets held for a later phase (props, quarters and halves, other), 320 picks from sources outside the launch list, 165 picks by probability models (held for a Models view), 73 quotes naming no team of the game, 72 quotes without a line or a price, 67 table rows without both teams, 45 table cells without an expert or a pick marker, 43 result marks in the quote, 9 lines not in the quote, 9 previews without a pick, 7 alternate lines and other markets, 6 team totals without a team, 4 bookmakers' own channels, 3 either-or bets, 3 site widgets counting other people's picks, 3 quotes whose written date points to another game day, 2 combined bets, 2 quotes recorded as both sides, 1 European 3-way handicap). At launch we publish picks only from a reviewed list of sources: English-language outlets and named experts, plus sources that passed our manual checks without an error; the rest stay in the archive until they pass. The archive keeps every one of them.

NBA grading rules

We grade all full-game NBA markets against the official final score, including overtime. Moneyline: the picked team wins the game, the pick wins. Against the spread: we add the spread printed with that pick (or recorded when the pick was captured) to the picked team's final score, including overtime; ahead is a win, behind is a loss, exactly level is a push. Two sources picking the same game at different times may be graded against different numbers if their published lines differed. Over/under and team totals include all overtime periods. Quarter and half picks are graded strictly on points scored in that quarter or half; overtime points do not count toward fourth-quarter or second-half picks.

Player props

Player prop picks (points, rebounds, assists, 3-pointers made) are graded against the official final box score where the archive holds it. Overtime statistics count. A prop on a player who did not play is not graded, and the reason is shown.

Priced picks vs. result-only picks

In the NFL, weekly media panels publish uniform tables where spread and total picks sit near standard −110 odds. In the NBA, sources publish picks across markets with widely different prices, and some outlets publish a team call without quoting odds or a timestamped price. Our archive applies one strict rule: a pick recorded without a market price counts for the win-loss result only and never enters market-adjusted points or CLV. All graded picks (Graded W-L-P and Win %) include every settled pick, priced or not; next to every win percentage we show the 95% Wilson score interval, the likely range. Priced picks (market-adjusted points, points per pick and CLV) include only settled picks with a valid recorded market price. We never invent a price for an unpriced pick and never pad unpriced picks as zeros into a per-pick average.

Market-adjusted points

As in our NFL records, a pick earns more when the market thought it unlikely and less when it was the obvious call. A right pick earns (1 − p) / p points, where p is the no-margin market probability of the picked outcome at the time of the pick; a wrong pick costs 1 point; a push counts zero. Zero is the expected score if the no-margin market odds are right, so a source that picks heavy favorites and a source that picks underdogs are measured on the same scale. These points measure pick accuracy relative to market difficulty; they do not represent betting returns.

Closing line value (CLV)

For priced picks we also compare the recorded price with the no-margin closing price of the same outcome right before tip-off. CLV is the change from the recorded price to the closing price as a share of the closing price; positive means the pick beat the closing line. Like market-adjusted points, CLV is a reference metric showing whether a source's picks anticipated pre-game line movement; it is not a calculation of betting payouts.

Leaderboard thresholds and early-season tiers

In the NFL, numbered leaderboard rankings require at least 30 scored picks on a market. Because NBA tracking launched more recently and builds game by game, a 30-pick gate on day one would leave most tables nearly empty, while no threshold would let a 3–0 start top the page. Early-season thresholds: the main table requires at least 15 scored picks on that market (and at least 10 priced picks to be ranked by market-adjusted points); the early-sample tier (8 to 14 scored picks) is shown separately with its likely ranges; fewer than 8 scored picks are shown on the source's own page but not ranked. Full-season threshold: as soon as at least 25 tracked sources in the league reach 30 scored picks on a market, the main threshold for that market steps up to 30 scored picks (the NFL standard) and the 15 to 29 group moves into the secondary tier. Week and day leaders start at 8 graded picks in one market. Game pages are open to search from 3 picks, source pages from 5 graded picks. Milestone badges stay off in the NBA until a deep competitive field exists.

Corrections

If a pick, a score or a time is wrong, write to editors@hunchmeter.com. A confirmed error is fixed everywhere it appears and listed on the corrections page with the date.