EV, arbitrage and middle rows are computed from these prices, so they are suppressed until the next capture lands rather than shown at a fair value nobody is still quoting.
Every captured price, held against the no-vig fair value of the sharp references — positive EVwhen a venue sells a side cheaper than the consensus says it's worth, arbitrage when both sides together cost under $1, and middleswhen two venues' totals leave daylight between the lines. The pattern is OddsJam's; the difference is honesty about speed: these rows are as fresh as the last capture (08:25Z), so treat them as candidates and verify at the venue link — not an execution screen.
Fair = the mean no-vig probability across Pinnacle, DK/ESPN, Bovada and BetRivers (the offering venue excluded from its own consensus; source count shown). Each book is de-vigged conservatively: the multiplicative and power methods bracket the honest answer, and every side takes the lower of the two — multiplicative flatters longshots, which is exactly where phantom edges breed. EV% = fair ÷ cost − 1, per $1 staked. Exchange costs are the ask plus Kalshi's taker fee; book costs include the vig. Stakes are quarter-Kelly on the ticket desk's 1,000-unit paper bankroll. ≥2% is the conventional alert bar; anything under ~3c of edge is inside de-vig + staleness noise. Qualifying exchange offers are also played forward as paper tickets (the edge-screen-ev shadow strategy on the ticket desk) — the screen's out-of-sample test of itself.
Capture — six venues, hourly through the slate
Polymarket and Kalshi order books (bid/ask), plus Pinnacle, DK/ESPN, Bovada and BetRivers lines, matched to one game by the canonical game key. One-sided or gappy exchange books (spread > 15c) are discarded here — an abandoned quote is not an offer.
De-vig every reference book — twice
Each book's two-way pair converts to implied probabilities that sum past 100%; the excess is the vig. The multiplicative method scales both sides down; the power method solves an exponent instead, keeping more on the favorite. They bracket the honest answer, so each side uses the LOWER of the two — errors bias against seeing an edge, never toward one.
Consensus — the fair price
The mean of the surviving conservative fairs across Pinnacle, DK/ESPN, Bovada and BetRivers, with the venue making the offer excluded from its own jury. Pinnacle also rides along per row as the single-sharpest sanity check.
Cost — what you'd actually pay
Exchanges: the executable ask plus Kalshi's taker fee. Sportsbooks: the raw line's implied probability with the vig still in, because you pay it when you bet. Fees live inside the cost, never in a footnote.
EV and stake
EV% = fair ÷ cost − 1, profit per $1 staked. Rows under 0.5% are noise; rows over 15% are stale-quote ghosts and get suppressed (counted, never silent). Quarter-Kelly sizes the paper stake, capped at 25u.
The forward test
Qualifying Polymarket/Kalshi offers are re-verified live and issued as paper tickets under the edge-screen-ev shadow strategy — CLV and P&L on the ticket desk are this screen's out-of-sample verdict on itself.
suppressed — EV rows are computed from the captured prices above, and those are too old to rank against. They return with the next capture.
0
one-sided bookno vig to strip
—
reference spread > 15¢an abandoned quote
—
EV over the 15% capa stale-quote ghost
—
arb over the 5% captoo good to be real
A 40% EV row is a stale quote, not a gift. These caps exist because the failure mode of this screen is confidently printing an artifact, so the count of what was thrown away is part of the output — never silent. The pipeline reports one total rather than a per-reason split; the four reasons are listed so you know what the total is made of.
Both sides of one market, cheapest venue each, total under $1. The $100 split is exact; “indicative” means at least one leg is a sportsbook line we can't see the order book for. Same-rules caveat as everywhere: verify both contracts before staking.
suppressed — arbitrage pairs are computed from the captured prices above, and those are too old to rank against. They return with the next capture.
Over at a low total on one venue, under at a higher total on another — land in between and both pay. Sorted by worst-case loss (our convention; OddsJam's ranking is unpublished). A negative worst case is also an arb.
suppressed — middle windows are computed from the captured prices above, and those are too old to rank against. They return with the next capture.
