How we find +EV bets - the de-vig and closing-line value math

One experience

There used to be a Simple/Advanced toggle here. There is now just one view for everyone - a plain sentence and the numbers that matter, visible by default, with a "Show the math" expander on every card, row, and panel that opens the full worksheet: fair probability, EV%, consensus, line-move history, and the audit trail behind the suggested stake. Nothing was removed, it just all lives in one place now instead of behind a switch.

The de-vig, in plain words

Every sportsbook bakes a small profit margin ("the vig") into its odds - that is why both sides of a coin-flip bet are usually priced worse than 50/50. We take each book's own two (or three) prices for a market and mathematically strip that margin back out, proportionally, leaving that book's own honest estimate of the true probability. We do this for every book that quotes the market, then average those honest estimates together across books - that average is our fair probability. It is a calculation on real, public odds, never a prediction of our own.

A play is worth surfacing when one book's price pays out better than that fair probability justifies - that gap is the edge (EV%).

The consensus outlier filter

Before we average books together, we throw out stale or broken lines. If one book's price implies a probability more than 12 percentage points away from the middle (median) of every other book quoting that same market, we drop that book entirely for this calculation - it is far more likely a stale snapshot or a feed error (for example, a moneyline sitting at -10000 next to a market of -150-ish prices) than a real, tradeable price. We require at least 3 sane books left after that filter; if fewer remain, we do not compute a consensus at all rather than fabricate one from thin or contaminated data.

Data quality grade

Every play carries a letter grade describing how much we trust the number, separate from the size of the edge:

• Grade A - 5+ books, price between -300 and +300, snapshot under 2 hours old.
• Grade B - 4+ books, price between -500 and +500.
• Grade C - everything else, including any longshot price over +1000 (thin markets on huge prices get capped at C no matter how good the rest of the inputs look).

Earned label thresholds (never invented)

A small confidence chip sits on some cards, layered on top of the real grade and EV% - never in place of them. Every chip maps to one fixed rule, listed here in full:

• STRONG EDGE - Grade A, and EV% is +5% or higher.
• SOLID - Grade A or B, and EV% is +2.5% or higher.
• Anything that does not clear one of those two bars gets no chip at all. An unbadged play is not being hidden - it simply has not earned a tier yet.

Grading provenance

Every leg is graded one of two ways: the scheduled auto-grade job decides it from a final score (moneyline/spread/total, on a sport we track), or a member grades it by hand (props, futures, untracked sports, or anything the job couldn't decide). Auto-graded is the norm and shows no badge. When a slip finishes because a human clicked the grade dropdown, it carries a small "self-graded" chip next to its seal - on the slip itself, the slips list, shared links, and capper feeds - so nobody mistakes a self-reported grade for the machine's. Report cards and the leaderboard also show "N% machine-graded" for any record with a known mix. Slips logged before this tracking existed, with no reliable signal to backfill from, show neither the chip nor the percentage - we do not guess at history.

Today's Tails - the shortlist rule

Today's Tails is a maximum of 5 plays, selected only when ALL of the following hold: Grade A or B, fair probability 35% or higher, EV% of +2.5% or higher, and the game is today or tomorrow. Every play on the shortlist still carries a real, math-backed edge - we never list a likely loser priced to look good. Among the plays that clear that bar, we rank by how often the market says they hit (fair probability, highest first), then by edge size, then by proximity to kickoff - so the shortlist leads with the plays most likely to actually land, not just the biggest number. If fewer than 5 plays clear that bar on a given day, we show fewer than 5 - we do not pad the list to look fuller than the market actually supports.

Line shopping and "middles"

When two or more books price the exact same selection, we show one card at the best available price and list the rest as quiet alternates underneath - that is line shopping, not a duplicate. Separately, because every book sets its own lines independently, it is entirely possible for BOTH sides of the same market to clear our edge bar at once (for example, one book's number makes the home team +2.5 a good bet while another book's number makes the away team -2.5 a good bet). This is not a contradiction - it is called a middle, and it means the books disagree enough that there is a real, math-backed argument for either side depending on which book you use. Today's Tails only ever shows the stronger of the two sides, to keep the shortlist simple; the full Best Bets board shows both with a note, since that disagreement is itself useful information.

Suggested stake (quarter-Kelly)

The stake shown ("Suggested: 0.75u") is a quarter of the full Kelly-criterion bet size for that edge and price, clamped between 0.25 and 2 units, halved again for Grade C data, and capped to whatever daily exposure room you have left. Quarter-Kelly is a standard, conservative haircut on the mathematically "optimal" bet size, because the fair probability itself is a market estimate, not a certainty. The full worked formula is available one tap deeper on every play card.

The Tail Rating

Every +EV play also carries a 1-to-5 Tail Rating - a quick visual read of the same grade and EV% shown elsewhere on the card, compressed into one number so you can scan a list fast. It is not a new calculation and never a prediction; it is a fixed mapping over numbers you can already see:

• 5 tails - Grade A, and EV% is +5% or higher (the STRONG EDGE bar).
• 4 tails - Grade A or B, and EV% is +3.5% or higher.
• 3 tails - Grade A or B, and EV% is +2.5% or higher (the SOLID bar).
• 2 tails - everything else still on the board (EV% at or above the board floor), or any Grade C row at all.
• 1 tail - informational rows below the board's EV floor, on the surfaces that show them.

Grade C always caps at 2 tails no matter how large the EV% looks - thin market data (few books, or a longshot price) does not earn conviction, the same rule the letter grade already applies. Tap the rating on any card ("Why this rating") for the exact grade, books, and edge behind that number.

The honest bottom line: the Tail Rating compresses data quality and edge size into one glance - it does not predict outcomes, and it is not a substitute for reading the grade and EV% themselves.

The tails record

Every play that actually renders as a Today's Tail is quietly logged the moment it appears - not after the fact, not cherry-picked. Once that game's line closes, we check the closing price against the price we surfaced and record two honest numbers: whether we beat the close, and by how much (the same CLV - closing-line value - concept explained above). The stat line you may see above Today's Tails ("Our tails: N graded - X% beat the close - avg CLV +Y%") is a rolling 30-day read of that log.

We only show that line once at least 10 tails have been graded in the window - fewer than that is too small a sample to mean anything, so we say nothing rather than show a shaky percentage. Plays that can't be graded against a closing line yet (the game hasn't started) or that have no featured closing line to check against (player props) are excluded from the percentage but never hidden from the underlying count - "graded" always means graded, nothing is padded to look better than the sample actually supports.

Parlay math

The Parlay lab combines legs the standard sportsbook way: the combined price is every leg's decimal odds multiplied together, and the combined fair probability is every leg's fair probability multiplied together. That second multiplication assumes each leg's outcome is statistically independent of the others - a real, stated assumption, not a fact about the games themselves.

That assumption is false for same-game legs (two bets on the same game are not independent events - a team covering a spread and that same team's moneyline hitting are correlated). When two or more legs in a parlay share the same game, the lab flags it with a same-game warning and never awards a STRONG EDGE or SOLID label to that parlay, because the fair probability it is showing you is optimistic - the real chance of hitting is lower than the math on the page implies.

Manual legs (anything not pulled from the board) use a de-vigged fair probability if you give both sides' odds, or a plain implied probability - clearly caveated "no consensus" - if you only give one side. Either way, a manual leg with no consensus behind it is treated as grade C data, the same as anywhere else in the app.

The honest bottom line: multiplying good individual edges together does not create a bigger edge for free - it multiplies the number of ways to lose. A parlay of three +EV singles is very often -EV once the multiplication runs, and the lab will tell you plainly when that's the case rather than dress it up.

TrueTail Sim (the Sim Lab) - experimental, plainly labeled

The Sim Lab is a separate, experimental engine from the de-vig math above - it does not read sportsbook odds at all. It runs a Monte Carlo-style simulation of each game/player prop from real inputs (recent form, matchups, weather, park factors, injuries) and reports the resulting probabilities plainly, next to the market's own number when one is available for comparison. Nothing on that page is ever called a "pick" - it is a model output, shown next to a live, honest scoreboard of how that model has actually performed (see below).

The record, never hidden: every sim carries a running backtest against real closed games - its Brier score (lower is better; a perfect model scores 0) compared against two honest baselines: always-picking-the-home-team, and picking-by-season-record. The Sim Lab states plainly which baselines it currently beats and which it does not, plus the sample size (n) behind that claim - a model is not called "better" on a handful of games.

Live win probability, in games already underway: once a game goes live, the Sim Lab's Live Game Center (linked from any live game on the Games page) shows a live win% computed as follows, exactly:

1. Take each team's pregame simulated run-scoring mean (mu) over a full 9 innings.
2. Scale each team's mean down proportionally by the fraction of ITS OWN innings still left to bat (the away team bats first each inning, so the two teams aren't always at the same point).
3. Treat each team's remaining runs as Poisson-ish (variance equal to the mean - the standard property of a Poisson count), and model the final run differential (home minus away) as approximately Normal, using the current score plus the two teams' remaining-run means, with variance equal to the sum of the two remaining-run variances.
4. Win probability is the normal-approximation area where that differential favors the leading side.

That is a documented, honest approximation - not a certainty, and not the same de-vig math the rest of this page describes. It is always labeled "TrueTail Sim - experimental" wherever it appears.