This is the reference for every figure Supaliner displays. It is written to be checked. Where a method has a known bias or limit, we say so here rather than hide it. Independently reviewed by a PhD-quant pass and a professional-bettor pass.
A bet has two independent sources of value: the line (the handicap, +3 vs +3.5) and the price (the juice, −110 vs −105). We report two distinct, separately-labeled quantities.
The price improvement of the best available price over the field (the median book) on the same bet, in cents per $1 staked:
This is what line shopping actually delivers on every bet, and it is always ≥ 0. It is a factual statement about price dispersion — not a claim of profit. It is the headline number on each card.
Expected value per $1 staked, versus the vig-free fair price:
EV is usually ≤ 0 — most markets are not beatable even at the best available price. That is the truth, and we show it as such. A positive EV is the rare, genuine edge; it drives the "+EV" flag and the CLV proof metric, never the everyday board.
De-vig — power method. For a market's outcomes with vigged implied probs qi (summing above 1), we solve for the exponent k such that
One method everywhere: near pick'em it agrees with simple normalization to within 0.2%, and on lopsided prices it correctly shifts probability off the longshot and onto the favorite — the empirically right direction (favorite-longshot bias). This dominates MLB moneylines, where the board currently lives.
Stated honestly: simple multiplicative de-vig overstates the underdog and understates the favorite; it is exact only near 50/50. The power method reduces this; Shin's method is a further refinement we may adopt for extreme moneylines.
1. Sharp (Pinnacle). When Pinnacle quotes both sides, we de-vig its two prices. A low-margin sharp price-maker is a better truth than a consensus of correlated retail books. reference only — never a bettable tile
2. Consensus fallback. Otherwise we de-vig the median implied prob per side across bettable books, and label the number "vs consensus" rather than "vs fair value."
Using the median (not the mean) keeps the benchmark robust: the best, extreme quote never moves the benchmark it is measured against, so an edge is never manufactured by self-reference.
Realistic magnitudes: best-of-six line-shopping edge vs a sharp anchor is small — field gaps of a cent or two, true-EV board averages well under a cent, genuine +EV spots rare. Anything materially above this is a data artifact, and the guards quarantine it.
The "Same picks, opposite seasons" charts are models, clearly labeled — not measured results.
The bettor's "true 52%" is an illustrative assumption, stated on the page — we cannot know any user's true rate. The point is narrow and correct: at −110 a 52% bettor loses, and price alone flips them.
Closing Line Value (% of taps that beat the closing line) is our core, stake-free proof of value. To be defensible it is computed against a sharp, no-vig close (Pinnacle/Circa), with the vig removed, guarded against steam-timing and survivorship, and restricted to liquid markets. Beating the close is a leading indicator of value, not a guarantee of profit, and is never equated with winning money.
Any performance figure shown before real settled data exists (a dashboard "Beat close %" or "Sharpe") is simulated / Monte-Carlo modeled, not actual results, and is marked "SIM" with a pointer to this page. Tiles carry a light marker rather than a repeated disclaimer; the full model lives here. Once real taps settle, these switch to measured values computed per §6.