Omenizer Research · July 2026
The life and death of a betting edge
Betting folklore says positive-EV prices vanish in minutes. Our data says otherwise — and it also says the market's closing line systematically disbelieves the biggest apparent edges. Findings from 20,981 settled positive-EV signals across 11 sports and 111 leagues, observed over 120 days.
Finding 1 — edges live for hours, not minutes
For signals detected at least 12 hours before the event (n=14,686), the median availability was 22.3 hours. Only 4.2% of flagged prices were gone within an hour; 6.7% within six hours — and 59.5% were still quoted at kickoff. "Available" means the price was still on the board; whether a book accepts your stake at full limits is a separate question our feed cannot observe.
Median availability by sport, signals detected ≥12h pre-event. Golf (n=322): 54.5h.
Finding 2 — the closing line disbelieves big edges
If an apparent edge is real, the market should move toward it by kickoff (positive closing-line value). That happens for modest edges — and stops happening as the apparent edge grows. Signals with 1–2% modeled EV closed +2.8% better than detection on average (44.7% CLV-positive). Signals claiming 12–20% EV closed −5.0% worse (29.1% CLV-positive). The bigger the claimed edge, the less the market ends up agreeing.
Average CLV by modeled-EV band; n per band 1,342–2,657 settled signals with valid closing prices.
Finding 3 — realized returns don't track raw EV
If modeled EV were the whole story, returns would rise with it. They don't. In this window, flat-staking every signal returned −2.0% at 1–2% EV and peaked at +5.4% in the 5–8% band, fading to +0.3% for 12–20% apparent edges — the band the closing line disbelieved most. Apparent EV overstates at the extremes; the middle of the distribution is where model and market meet.
| Modeled EV | Avg CLV | CLV-positive | Flat-stake ROI | n (settled) |
|---|---|---|---|---|
| 1–2% | +2.81% | 44.7% | -2% | 3,478 |
| 2–3% | -0.47% | 34.1% | -4% | 4,592 |
| 3–5% | -0.49% | 34.2% | -2.5% | 3,944 |
| 5–8% | -1.88% | 32.3% | +5.4% | 4,228 |
| 8–12% | -2.41% | 32.6% | +2.8% | 2,578 |
| 12–20% | -5.03% | 29.1% | +0.3% | 2,191 |
Why this matters
Three practical consequences. First, speed matters less than folklore claims — selection matters more: most edges wait hours, but knowing which ones deserve the wait is the hard part. Second, raw EV is not a ranking: a screen sorted by biggest apparent edge sorts largely by model error. Third, honest measurement requires closing-line value, not win–loss records — which is why every Omenizer signal is tracked to its close and published, wins and losses alike, in our market records. These findings are also why the platform ranks signals by measured expected return per segment rather than by raw EV.
Method and data
Sample: signals with a BET recommendation, modeled EV between 1% and 20%, detected within the 120 days ending 2026-07-15, excluding records with data-quality flags. Availability = time from detection until the signal's live EV first went negative, censored at event start; the ≥12h-lead subsample removes censoring bias. CLV compares the flagged price with the recorded closing price (valid-close records only). ROI is flat 1-unit staking on settled signals. Fair prices per our methodology; the aggregate tables are downloadable as CSV — reuse with attribution ("Source: Omenizer", see press).
Historical, aggregate market data — not betting advice and not a guarantee. Markets adapt; edges fade; variance is real.