Open audit trail — Soccer & MLB
The full ledger.
Timestamped. CLV-graded.
Last updated: 2026-07-27 10:00 AM ET · Current era only · Deduplicated · Soccer + MLB only — WNBA props trail here
Every soccer and MLB pick the EdgeLeverage scanner surfaces is timestamped before kick-off, devigged off six marketmaker books, and graded against the closing line after the result. This page is the public ledger of that work. WNBA props use a different pricing methodology and are tracked separately.
On sample size: the data below represents 165 unique graded picks across the current tracking era (June 22, 2026 onward). By professional quant standards, meaningful CLV claims require 1,000+ picks. We do not claim otherwise. We publish anyway, because an open, growing audit trail is how a serious research operation operates. Sample expands as the European and South American regular seasons resume in August.
Process integrity — non-statistical metrics
What we commit to
These are the metrics that don't depend on sample size. They describe the discipline of the operation itself.
Marketmaker books
6
Pinnacle · Prophet · Bookmaker · Sporttrade · LowVig · BetOnline
Picks unlogged
0
100% of picks timestamped pre-kickoff
CLV tracking
100%
Every entry graded vs close
Model layers
5
xG · ELO · form · referee · weather
Sample so far — variance dominates at this size
Aggregate stats (165 unique picks)
Reported in good faith with the explicit caveat that this sample is too small for statistical confidence. Numbers will shift as the sample grows toward 1,000+.
Unique picks graded
165
83W – 82L – 6P
Hit rate
50.3%
ignoring pushes · sample too small
ROI (tiered Kelly)
+10.7%
25–50% Kelly staking · sample too small
Avg CLV
+2.06pp
closing line value · process metric
Avg Odds
+200
3.00 decimal · 33.3% implied
Target sample
1,000+
When CLV becomes statistically meaningful
CLV by EV band — where edge concentrates
Closing line value breakdown
If the model is working, higher-EV alerts should show stronger CLV. This is the key diagnostic.
| EV Band | Avg CLV | Sample |
|---|---|---|
| 3–4% | +1.92pp | 32 graded |
| 4–6% | +1.35pp | 56 graded |
| 6–9% | +1.97pp | 33 graded |
| 9%+ | +3.45pp | 38 graded |
Favorite–longshot bias — soccer (Buchdahl research applies here)
Soccer odds bucket breakdown
The favorite–longshot bias is a documented phenomenon in fixed-odds football markets. If our CLV and hit rate degrade at longer prices, the model may be overconfident on longshots. Soccer only — MLB operates under different market dynamics and is tracked separately below. Avg CLV in this table is live-captured only — reconstructed figures are excluded here, because at the per-bucket level they are too thin to be meaningful (see data provenance above).
| Odds Range | Picks | Hit Rate | Avg CLV (live) | Edge context |
|---|---|---|---|---|
| 1.57–2.00 | 3 | 33.3% | +7.28pp (n=3) | 6.2% avg EV |
| 2.00–2.50 | 42 | 45.9% | +1.48pp (n=38) | 5.7% avg EV |
| 2.50–4.00 | 11 | 54.5% | -0.15pp (n=7) | 15.6% avg EV |
| 4.00–6.00 | 3 | 0.0% | +0.17pp (n=3) | 22.8% avg EV |
| 6.00+ (longshot) | 2 | 0.0% | -1.81pp (n=2) | 11.4% avg EV |
MLB — odds bucket breakdown (independent of FLB)
MLB odds bucket breakdown
MLB spreads operate under moneyline market dynamics, not fixed-odds football pricing. Tracked separately so it doesn't contaminate the soccer FLB analysis above.
| Odds Range | Picks | Hit Rate | Avg CLV (live) | Edge context |
|---|---|---|---|---|
| < 1.57 (heavy fav) | 2 | 50.0% | +2.85pp (n=2) | 7.7% avg EV |
| 1.57–2.00 | 23 | 69.6% | +0.10pp (n=20) | 6.2% avg EV |
| 2.00–2.50 | 50 | 61.2% | +2.03pp (n=28) | 7.2% avg EV |
| 2.50–4.00 | 32 | 37.5% | +3.54pp (n=27) | 5.8% avg EV |
| 4.00–6.00 | 2 | 0.0% | +0.00pp (n=1) | 4.8% avg EV |
| 6.00+ (longshot) | 1 | 0.0% | — (n=0) | 9.7% avg EV |
Day-by-day log
Raw ledger
Each session shows the W/L record and average CLV for that day's graded picks. Deduplicated — one row per unique bet.
| Date | Record | Avg CLV | Leagues |
|---|---|---|---|
| Jul 26 | 0W–1L | +0.00pp | MLB |
| Jul 25 | 0W–1L | +7.73pp | MLB |
| Jul 24 | 2W–1L | +4.63pp | MLB |
| Jul 23 | 1W–3L | -0.36pp | MLS · MLB |
| Jul 22 | 1W–1L | +3.72pp | MLB |
| Jul 19 | 55W–44L | +2.37pp | MLB · WC |
| Jul 15 | 1W–1L | +0.86pp | WC |
| Jul 14 | 1W–0L | -5.74pp | WC |
| Jul 12 | 2W–1L | +1.68pp | WC |
| Jul 10 | 0W–0L | +12.67pp | WC |
| Jul 06 | 1W–1L | +1.92pp | WC |
| Jul 04 | 2W–3L | +1.30pp | WC |
| Jul 03 | 3W–6L | +2.37pp | WC |
| Jul 01 | 1W–1L | +4.74pp | WC |
| Jun 30 | 0W–2L | +1.69pp | WC |
| Jun 28 | 5W–4L | +0.84pp | WC |
| Jun 27 | 1W–3L | +3.55pp | WC |
| Jun 26 | 2W–3L | -0.51pp | WC |
| Jun 25 | 2W–2L | -1.07pp | WC |
| Jun 24 | 2W–3L | +1.79pp | WC |
| Jun 23 | 1W–1L | +0.40pp | WC |
How we measure CLV — and the benchmark we hold ourselves to
The line we grade against
A closing-line-value figure means nothing until you state what it's measured against. Here is ours, precisely — and it is a deliberately harder standard than the one most services quote.
The calculation
For every pick, CLV is the difference in implied probability between the price we logged at alert time and the closing price, expressed in percentage points (pp). A positive figure means we secured a longer price than the market's final, sharpest assessment of the game — the single most reliable long-run predictor of a genuine edge. Win/loss outcomes are noise over short windows; CLV is signal. We compute it on every graded pick, with no cherry-picking.
The benchmark: the best price at close — which means we beat Pinnacle
We do not grade against a soft recreational book, and we do not grade against a single convenient reference. We grade against the best closing price available across every book we track — the longest odds anyone, anywhere, could have secured at kick-off. Pinnacle — the sharpest and most respected closing line in the industry, and the book nearly every serious operation quotes its CLV against — is one of those books.
That choice carries a precise, provable consequence. The best available closing price is, by definition, always at least as sharp as Pinnacle's. So every time we post a positive CLV figure, we have not merely beaten the soft books — we have beaten Pinnacle's own closing line. Clearing the best-of-all benchmark means Pinnacle was cleared on the way there. Our reported CLV is therefore a floor: measured against Pinnacle alone, the number could only be equal or higher.
Most services quote "CLV vs Pinnacle close" because it is the easier bar to clear. We hold ourselves to the harder one — beating the best price on the entire board — and we disclose it in the open, because a CLV number is worth exactly what its benchmark demands of it, and no more.
Data provenance — how each CLV figure was captured
Measured vs reconstructed
A closing-line-value number is only as trustworthy as the record it came from. We disclose exactly how every CLV figure on this page was captured — none of it is a black box, and we separate what was measured live from what was recovered after the fact.
Live-captured CLV
133
84% of graded CLV · avg +1.83pp · closing line pulled in the hour before kick-off, in real time
Reconstructed CLV
26
16% of graded CLV · avg +3.43pp · closing line recovered from the historical odds archive
What “reconstructed” means — and why it still counts
Our default is live capture. An automated job fetches the market's closing price across our six marketmaker books within the hour before kick-off and records it in the moment. That is how 84% of the CLV on this page was measured — as it happened, no hindsight involved.
The remaining 16% is reconstructed. When a capture-system outage let some games close without a live snapshot, we faced a choice most services resolve silently: drop the bets, or recover them honestly. We recovered them — querying The Odds API's historical odds archive for the exact snapshot at each game's kick-off, then computing CLV with the identical formula used for live captures. This is genuine closing-line data, from the same marketmaker books, priced at the same moment the game began. The only difference is that it was retrieved after the fact rather than in real time.
We publish the split instead of blending it because a reconstructed close is a faithful proxy, not a byte-for-byte identical measurement. It treats the kick-off-time snapshot as the close — accurate to the minute, but not the same as a line captured live and continuously through the final market moves. Every reconstructed entry is flagged as such in our database, so the line between measured and recovered is explicit and independently auditable. If you want to weight them differently — or judge our live-captured record on its own — you have the numbers to do exactly that.
How to read this
Methodology notes
Timestamp before kick-off
Every pick is written to the audit ledger with a server-side timestamp before the game's scheduled kick-off. There is no publishing wins after the fact. If a pick isn't in the ledger before kick-off, it doesn't count.
Devigging from marketmaker books
The fair price for every bet is computed by removing the book's hold from six marketmaker books (Pinnacle, Prophet, BookmakerEU, Sporttrade, LowVig, BetOnline). Edge is measured against this fair price, not against the recreational book's quoted price.
CLV graded at close
After each game, every pick is graded against the best closing price available across all books we track — Pinnacle included. Because that benchmark is always at least as sharp as Pinnacle's own close, a positive CLV figure means we beat Pinnacle's closing line, not just the soft books. See "How we measure CLV" above for the full method. Positive CLV is the only reliable long-run predictor of profitability; win/loss outcomes are noise over short windows.
What this isn't
This isn't a guaranteed-pick service. It isn't a lock-of-the-day channel. It isn't a backtest fitted to past results. It's the live, running output of a rules-based scanner operating under a single decision rule: positive expected value off the sharp fair price, confirmed by a Poisson goal model.