MLB · Moneyline · Illustrative demo
SEA ML
Seattle Mariners vs Houston Astros. The board is showing +124 against a de-vigged fair estimate of +119 and a sharp consensus of +121. The demo model scores this 68 out of 100 and labels it WAIT.
Price history
de-vigged probability spacePlotted in implied-probability space rather than raw American odds. American odds jump across the -100/+100 boundary, so a chart drawn directly on them distorts distances; the axis is labelled back in American odds so it still reads like a betting chart.
| Minutes before snapshot | Board | Model fair | Sharp consensus |
|---|---|---|---|
| 60 | +118 | +120 | +120 |
| 50 | +120 | +120 | +120 |
| 40 | +124 | +120 | +121 |
| 30 | +126 | +119 | +121 |
| 25 | +123 | +119 | +121 |
| 20 | +120 | +119 | +120 |
| 15 | +122 | +119 | +121 |
| 10 | +125 | +119 | +121 |
| 5 | +123 | +119 | +121 |
| 0 | +124 | +119 | +121 |
Consensus
weighted sourcesNo single book is treated as the answer. The fair estimate is a weighted blend, and the weights are part of what makes the estimate a model output rather than a quote.
| Source | Weight | Line |
|---|---|---|
| Sharp consensus | 0.45 | +121 |
| Latent market | 0.30 | +120 |
| Model fair | 0.25 | +119 |
Book dispersion
5 sampledWhere the sampled books actually sit. A stale quote frequently looks like the best number on the board, which is exactly why it is marked rather than silently included.
| Book | Quote | Status |
|---|---|---|
| Book A | +123 | included |
| Book B | +124 | best sampled |
| Book C | +122 | included |
| Book D | +124 | included |
| Book E | +121 | included |
First movers
order carries information- 17:48Book D → +121
Moved, then moved back
- 17:52Book A → +123
Partial follow
Information events
what the market reacted to- 17:50Lineup
Two regulars out of the sample lineup; market reacted then partially retraced.
- 17:20Market
Two-way move and reversal inside 15 minutes of the sample window.
Playable threshold
where the label changesThe same market at a different price is a different decision. This is the price at which the demo model's label changes, which is a more useful thing to know than the label at one price.
| If available at | State would be |
|---|---|
| +132 | BET |
| +128 | WAIT |
| +124 | WAIT |
| +119 | PASS |
Reasoning
market → stateMarket
The board has moved through +118, +126, +120 and back to +124 inside the sample hour. Direction is not established.
Reversal
A reversal flag is set. Reversals usually mean the first move was information that turned out to be wrong, or a limit adjustment rather than a price read.
Persistence
Estimated half-life is 7 minutes. Anything that decays this fast is unlikely to survive the time it takes to act on it.
Uncertainty
The EV band crosses zero. When the lower bound is negative, the honest description is that the edge is not distinguishable from noise.
State
Score 68 is below the illustrative threshold of 75, so this row is labelled WAIT rather than BET.
This is the same sequence described on the methodology page, applied to one row.
Decay
How long this would last
Synthetic decay illustration based on the demo half-life estimate for this row.
- Half-life
- 7 min
- Edge now
- +1.02 pts
- After 7 min
- +0.51 pts
- After 14 min
- +0.25 pts
Demo estimate
Why half-life is scored, not just reported
A large edge that decays in three minutes is usually worth less than a small one that holds for twenty, because the time it takes to notice, check, and act is not zero. Persistence is a scoring component for that reason.
Snapshot 23 Aug 2026, 18:05 UTC · Sample snapshot timestamp — fixed value, not a feed reading.