Methodology

For any American price a we use the standard conversion:

d = 1 + (a > 0 ? a / 100 : 100 / |a|)
q = 1 / d        // raw implied probability (still vigged)

The raw probabilities q_i for every selection in a market sum to more than 1 — the excess is the book's overround (vig). De-vig removes it.

The simplest method — divide each q_i by the overround.

p_i = q_i / Σ q_j

Solve for an exponent r ≥ 1 such that Σ (q_i)^r = 1. Bisect on r in [1, 8] — the sum is strictly decreasing in r, so it converges. Power is our default for 2-way and 3-way side markets because it preserves the relative price between favorites and dogs better than mult on high-vig books.

Σ (q_i)^r = 1   ⇒   p_i = (q_i)^r

For every selection we average de-vigged probabilities from sharp / low-margin books (Pinnacle, Circa, Novig, Betfair Exchange) — these books are price-discovering, not promotional. We then blend that sharp consensus with the market-wide power-devig according to anchor weight.

p_fair = w · p_anchor + (1 - w) · p_market

where w ∈ [0, 0.6] grows with:
  • # of anchor books available for the selection
  • freshness of the anchor quote
  • inverse spread (tighter anchor → higher weight)

For every selection we compute the fair probability under multiplicative, power, and (for 2-way) Shin, then take the minimum. This is the de-vig method that gives the smallest edge — the one we report when we want zero benefit-of-the-doubt.

p_worst = min(p_mult, p_power, p_shin)

A pick that still shows +EV under conservative is the most defensible. The drawer's de-vig toggle lets you flip between methods live; the delta between Power and Conservative is your edge-robustness check.

EV% = (p_fair · d_offer - 1) · 100
kelly = (p · d - 1) / (d - 1)         // fraction of bankroll
We display 0.25 × kelly by default — a quarter-Kelly stake.

Conviction is a single 0-100 score per pick. It combines:

• EV% magnitude (most of the weight)
• Market width in cents (tighter → more)
• Books contributing (more independent prices → more)
• Anchor inclusion (Pinnacle/Circa in the set → more)
• Movement-toward score — line drifting toward our side

Tiers: S ≥ 80, A ≥ 65, B ≥ 50, C < 50. Thin markets (< 3 books and no sharp anchor) are forced to Lean and capped at conviction 40.

The 10-game hit-rate strip and the drawer's "Recent form" bars are sourced from real settled box scores — not modeled, not padded. For each player we read the last N official games for the matching stat type, compute over/under the offered line, and require N ≥ 3 to render. Fewer than 3 → we say so explicitly ("Not enough settled games (n=X)") instead of inventing a number.

Every fire is captured at fire time with the exact price, line, book, fair odds, and timestamp. At kickoff we re-read the market and record a median closing fair price (requires ≥ 2 books at close). CLV is the move from fire to close in percentage points of implied probability. We never re-price after the fact.

CLV_pp = (p_close - p_fire) · 100         // in probability points

On the track record we lead with average CLV and sample size, not a binary win-rate. We mark samples as "accruing" until N ≥ 200 — that's where confidence intervals on CLV tighten meaningfully.

The no-vig calculator and the no-vig fair-odds tool let you paste any two-way or three-way set and reproduce our power / multiplicative numbers exactly. If a Parlae fair price is ever off by more than 0.5pp from what those tools say for the same input, that's a bug — file it and we'll fix it publicly.