Skip to content

Receipts, not hype

HOW WE GRADE

Proofing season · free picks, not yet a paid product

The rules behind every number on our record and picks pages. Same rules for wins and losses, and nothing gets deleted.

1. Picks are posted before the game starts

Each pick is saved once, with a timestamp, before the game starts. The save is write-once: our storage refuses a second write to the same pick, and the system refuses any pick for a game that has already started. A posted pick cannot be edited, re-priced or deleted.

Each card shows when it was posted, the line and price we took, the sportsbook that offered it, and the market consensus at that moment.

2. Units, sizing and “playable to”

We keep two tracks. The headline is flat 1 unit on every pick, so every day compares apples to apples. From Oct 26 we also track recommended size: 1, 2 or 3 units, set before the game and posted with the pick. Both are shown on the record with their own ROI.

  • Size by confidence: low = 1 unit, medium = 2 units, high = 3 units.
  • Playable to: the worst number still worth a bet. Example: Our pick: Lions −2 · Playable to −3 · 2 units (1 unit at the limit). At the pick number or better, the full recommended size. Worse than the pick number but not past the limit, 1 unit. Past the limit, it's a pass.
  • The limit gives away 0.5 points (low), 1 point (medium) or 1.5 points (high), never more than our model's edge, and in football it never goes through the key numbers 3 or 7. The worst acceptable price at any playable number is −120.
  • We grade the sized track at the number and price we posted. When our stored odds show the market moved past the playable-to number before the start, the card says so (information only).

Picks posted on Oct 25 have no playable-to or sizing and stay flat 1 unit. We don't size up on winners after the fact, and we don't quietly bet small on losers.

Where picks are priced

Our signals use every book we can see, including offshore and sharp reference books, for the consensus line, no-vig fair price, steam and sharp-led moves. But the line and price we take on a pick come only from major, regulated US sportsbooks subscribers can actually bet at: DraftKings, FanDuel, BetMGM, Caesars, BetRivers, ESPN BET / theScore Bet, Fanatics, Hard Rock Bet, Bally Bet, betPARX and bet365 (US), when our odds feed carries them. Availability varies by state.

Our Oct 25 launch-day picks used a mix of books (3 of 5 at LowVig.ag, an offshore book). They stay on the record as posted. From Oct 26, every pick is priced at a regulated US book.

3. Win, loss or push

After the game ends we grade it against the official final score and the exact line we took. Final scores come from The Odds API scores feed, and anything it can't settle is graded by hand from the official league result and marked as manual.

  • Spread: our team's score plus our line, minus the opponent's score. Above zero is a win, below zero a loss, exactly zero a push.
  • Total: the combined final score against our number. Over wins if it lands above, Under wins if it lands below, and landing exactly on it is a push.
  • Overtime counts, the same way sportsbooks settle these markets.

4. Units and ROI

Units are computed at the price we took, not a generic −110.

  • Win at a negative price (e.g. −120): +100 / 120 = +0.83u
  • Win at a positive price (e.g. +110): +110 / 100 = +1.10u
  • Loss: −1u. Push: 0u.

ROI = total units won or lost ÷ units risked on graded picks. Pushes count as risked and returned (0u). Win rate excludes pushes.

5. Closing line value (CLV)

Winning and losing over a few weeks is mostly luck. A better test of skill is whether we got a better number than the market closed at. If the line keeps moving toward our side after we post, the market is agreeing with us, and over a large sample that tends to matter more than a short-run win rate.

  • Closing line = the consensus (median) line across sportsbooks in our last stored odds snapshot before the start, if that snapshot was taken within 60 minutes of the start.
  • Points CLV = how many points better our number was than the close. Spread: our line − closing line. Over: closing total − our total. Under: our total − closing total.
  • Price CLV (when we took the same number the market closed at): closing no-vig probability − implied probability of our price, in percentage points. No-vig probability removes the bookmaker margin: p = i₁ / (i₁ + i₂) from both sides' prices, median across books.
  • We take each pick at one sportsbook, and the close is the multi-book consensus. Where that book was in the closing snapshot, we show its own closing number too.
  • Beat the close = a better number, or the same number at a better no-vig price.

When CLV is unavailable. We store odds snapshots twice a day, not minute by minute (our current data plan is limited). If the last snapshot before a game was more than 60 minutes before the start, we don't treat it as a close. The card says “CLV unavailable” and shows the last snapshot's time and line for information only, and it is left out of the CLV averages. We don't estimate or fill in missing closes.

Closing records are write-once and kept separate from the pick and the grade. We show the sample size next to every CLV average. Small samples are noisy.

6. What counts, and what doesn't

  • Every posted pick counts. We don't delete picks, and losing days stay on the board.
  • Void: a game that is postponed, cancelled or not completed is graded void (0u) and left out of the record and ROI. It stays visible, marked void.
  • Picks waiting on a result are shown as pending and are not in the totals.

7. Data sources and their limits

  • Odds and final scores: The Odds API (US sportsbooks, spreads and totals; college basketball spreads only). We take snapshots twice a day on a limited plan, so we can miss fast moves and true closes.
  • Team stats, form and injuries: public feeds (ESPN public data including college basketball AP rankings and scores, nflverse, the NHL public API). These are not licensed for commercial use. We use them for Proofing season and plan to switch to a licensed feed before charging.
  • Betting percentages: we have no public bet or money-percentage data. What we call “reverse line movement” is a labeled proxy based on line moves, not real ticket or handle splits.
  • We don't make up stats or percentages. If an input is missing, the pick says so.

Data plan in use right now

Odds data plan: free (500 credits a month).

  • Pre-game closing-line capture (about 15 min before each picked game): off
  • Frequent game-day snapshots: off (morning and one afternoon snapshot only)
  • College basketball totals: off (spreads only)
  • College basketball first-half and second-half lines: off. A half-game pick needs a stored half line from within 90 minutes of tip. Halves are graded from ESPN half scores (overtime counts toward the second half). If half scores aren't available, the pick stays ungraded until it is graded by hand. We never guess a result.

College basketball (NCAAB) rules

With 360+ Division I teams, many college basketball games have thin betting markets and little reliable information. So we add extra limits:

  • Only well-covered games: at least 6 books must post the spread, and they must agree within 1.5 points.
  • Only notable teams: at least one team is in the AP Top 25, or both teams are in the ACC, Big East, Big Ten, Big 12 or SEC.
  • At most 2 NCAAB picks a day, so college hoops never crowds out the rest of the slate.
  • Conservative early-season sizing: NCAAB picks stay low confidence (1 unit) until both teams have played 8 games, and can't be high confidence until both have played 15.
  • Spreads only for now: to fit our odds data budget, we track college basketball spreads, not totals.
  • Grading: NCAAB final scores come from ESPN's public scoreboard, with The Odds API as a backup. The win, loss and push rules are the same as every other sport.

8. Proofing season

From Oct 26 to Nov 21, 2026 we're in Proofing season: real picks, posted and graded publicly, with no money wagered or charged. This isn't a paid product yet. The goal is a record anyone can check before we ask anyone to pay.

For entertainment and informational purposes only. Sourdough Sports sells analysis, not wagers — we do not accept bets. Past results do not guarantee future outcomes; no pick is a sure thing. You must be 21+ (or the legal age in your jurisdiction) to bet. If you or someone you know has a gambling problem, call 1-800-GAMBLER.

See the record · See today's picks