How picks are made — and graded.
The full pipeline: how a pick gets made, how the record gets graded, why we don't sell win rate, what happened to our old CLV stat, and — written down in advance — what a losing streak does and doesn't mean. No marketing math.
The live, dated record — wins and losses — is on /proof. That's the only place we report results as they accrue.
How a pick gets made.
Every quoted price carries the book's margin, so the first step is always removing it. We run five de-vig methods — Shin is the default — so the vig never flatters a number before any comparison happens.
The fair line is a weighted multi-book consensus: no single book is treated as the truth, the market is. And when a book's price is graded, that book is excluded from the consensus it's compared against — leave-one-out — so a book can never confirm itself.
A pick is only issued when at least 3 books confirm the number. One outlier price with no market agreement behind it is not a pick; it's noise.
- De-vig
- 5 methods; Shin is the default
- Fair line
- Weighted multi-book consensus — no single book is the truth
- Grading guard
- Leave-one-out — the graded book is excluded from its own consensus
- Confirmation floor
- Minimum 3 confirming books, or no pick is issued
All comparisons happen at the no-vig level — both the bet odds and the reference odds are de-vigged first, so the metric isn't flattered by the book's margin.
Locked when issued, graded when the game ends.
Every pick is locked the moment it's issued and graded when the game ends. Nothing on the record is written — or rewritten — after the result is known.
The ledger is append-only: rows are added, never edited, never deleted. 99.4% of issued picks are graded (measured 2026-08-28, every sport and market) — the remaining fraction is a player-matching gap, not a quietly-dropped bucket where the losers go: those picks fail to match on identity before any outcome exists, so the gap cannot lean toward wins or losses. Wins and losses are both public.
Erratum, 3 September 2026. An audit of the earliest MLB rows found a grading error. Picks from 20 June to 25 July carry no stored start time, and a night game rolls onto the next UTC day — so for 54 matchups the grader matched the same teams' NEXT game and graded against the wrong box score. Re-grading those against the correct game changes 516 outcomes.
Two things about that number matter more than the number. First, it moves the affected rows about $474 our way at the $10 flat stakes this record is priced in, which is exactly why it is published rather than quietly repaired. Second, none of those rows are on the board or inside the window any published figure is drawn from, so no number we have ever shown you changes. The rows stay as they are, because the ledger is append-only and that pledge is worth more than a flattering correction. A further 58 matchups sit on split doubleheaders where the true game cannot be identified from what was stored; those are excluded from the count rather than guessed.
- Lock
- Every pick locked the moment it's issued
- Ledger
- Append-only — no edits, no deletions
- Coverage
- 99.4% of issued picks graded — the gap is measured, not silent
- Visibility
- Wins AND losses public, pick by pick, on /proof
Why we don't sell win rate.
Win rate is the easiest number in betting to dress up and the least connected to profit. Here's the proof from our own ledger: across 6,503 settled picks (as of the July 14 freeze), heavy favorites won 72.5% of the time at +3.1% ROI, while slight underdogs won just 49.0% of the time at +8.4% ROI. The “worse” win rate more than doubled the return per dollar staked.
A tout who wanted a shiny percentage would sell you the favorites. Win rate is the wrong metric — profit is the point — which is why ROI leads every record we publish and win rate never does.
- Sample
- 6,503 settled picks from our own graded ledger, as of the July 14 freeze
- Heavy favorites (−200 and shorter)
- Won 72.5% · +3.1% ROI
- Slight underdogs (+100 to +149)
- Won 49.0% · +8.4% ROI
- Takeaway
- Win rate is the wrong metric — profit is the point
Our old CLV stat is retired.
Closing-line value asks one question: did you get a better price than the market's final word? Measured honestly, it's the strongest single signal that an edge is real. Measured sloppily, it's just another vanity number — and ours was measured against the wrong anchor.
Our previous CLV stat was anchored to our own multi-book consensus. As a game approaches, the set of books quoting a market changes, and the consensus moves with it — so that stat partly reflected book-mix changes near game time, not market movement. That's not good enough to publish, so we retired it.
Its replacement is sharp-anchored: CLV measured against Pinnacle's closing line. That measure is accruing now and publishes once its pre-registered sample floor is met — not a day sooner, and regardless of whether the number turns out to be flattering.
- Old stat
- Consensus-anchored CLV — retired (reflected book-mix changes near game time, not market movement)
- New anchor
- Pinnacle's closing line
- Publication
- Publishes once its pre-registered sample floor is met — flattering or not
What a losing streak does and doesn't mean.
This section was published on 2026-07-14, before any losing streak — so when one arrives, this isn't an excuse written after the fact. It's the standard we committed to in advance.
The math: at the settled win rate the record showed that day (53.9%), any single pick starts a 5-loss run about 2% of the time — at a few picks a day, that's a streak roughly every couple of weeks. It will happen, on schedule, to any bettor with a real edge — and it means nothing by itself. Judging an edge by its worst week is like judging a coin by five flips.
What would make us say the edge is gone: the pre-registered forward test failing its written, pre-committed criteria. Not a cold stretch, not a gut feeling — the test. And we will publish that outcome either way, pass or fail, on the same record as everything else.
- Published
- 2026-07-14 — before any streak, not after one
- Expected variance
- At a 53.9% settled win rate, a 5-loss run lands roughly every couple of weeks at a few picks a day
- What a streak means
- Nothing by itself
- What would end it
- The pre-registered forward test failing its written criteria — and we publish that outcome either way
The Books, the regions, the cadence.
Whet scans player-prop and game-line markets across 25+ sportsbooks, refreshing about every 3 minutes. That price set includes offshore and sweepstakes books — they move early and their numbers make the consensus honest — so a surfaced pick can carry one of their prices. Bet links only ever point at US-licensed books. The list shifts as books enter or exit markets; the integration count published on the landing is the floor, not a fixed ceiling.
- Scan cadence
- About every 3 minutes
- Regions
- US states where the books are licensed (varies by book)
- Markets
- Player props (15 stat types), spreads, totals, moneylines, parlays
- Update window
- A surfaced bet carries the de-vigged price it was flagged at, timestamped
- Never routed to
- Offshore and sweepstakes books can price a pick, but we never send a US bettor to one
- Excluded entirely
- Peer-to-peer exchanges, books without public price feeds
Want to see the math live?
Every bet on Whet ships with its edge, its closing-line target, and the Kelly-sized stake. The methodology applies to the live feed, not just the snapshot.