This is today's featured match — its full report is free. Every board pick gets a page like this; members see all of them.

Jesper De Jong vs Vilius Gaubas

Outside band — caution

ATP · ATP Bastad, Sweden Men Singles · Clay2026-07-13 12:45 UTC

Model-favored side: Jesper De Jong · board of 2026-07-13

Edge outside the validated 5–12% band. A gap above 12% usually means a stale line or bad data, not a stronger signal — shown as caution, never conviction.

The math, in the open

Our model P (Jesper De Jong)
79.9%
Pinnacle fair P (vig removed, two-sided)
67.3%
Pinnacle raw implied P (Jesper De Jong)
69.3%
Best price implied P (1 ÷ odds)
69.4%
Edge (model P − best implied P)
+10.5%
Books priced
3

Vig removal: Pinnacle's two prices imply more than 100% combined; normalizing them out gives the fair line. This row is outside the validated filter (5–12% disagreement, odds 2.0–3.0, ATP) — shown for transparency, not flagged.

What the model is weighing

The model's own per-feature contributions for this matchup (LightGBM TreeSHAP) — the largest inputs first. Each bar shows the factor's pull, toward Jesper De Jong or toward Vilius Gaubas. These explain the raw model score; the published probability then passes through a per-surface calibration step and a small schedule adjustment, shown beneath.

  • Market fair probability 67.3%Jesper De Jong

    Pinnacle's two prices with the vig removed — the market's own estimate.

  • Market fair probability vs 50/50 +17.3ppJesper De Jong
  • Surface Elo gap +145Jesper De Jong

    Surface-specific Elo blended with overall Elo, difference between the players.

  • Return points won gap +4.7ppJesper De Jong
  • Style data available (player 2) noJesper De Jong
  • First-serve points won gap +8.1ppJesper De Jong
Raw model P (Jesper De Jong)
78.6%
After per-surface calibration
79.9%
Schedule/context adjustment
+0.0%
Published P (Jesper De Jong)
79.9%

Contributions are in the model's internal units (log-odds); bar lengths are relative within this match. Model v6-stacked. Values marked “neutral” are features the live pipeline doesn't populate — the model sees them at their neutral value, and we show that rather than hide it.

Head-to-head & recent form

No prior meetings in our results database.

Jesper De Jong: our results table ends 2024-11-04 — too stale to present as current form, so we omit it rather than show it.

Vilius Gaubas: our results table ends 2024-11-25 — too stale to present as current form, so we omit it rather than show it.

Clay Elo: Jesper De Jong 1610 (n=23) vs Vilius Gaubas 1484 (n=11)

Context is shown for orientation. What the model actually used is the factor list above plus the schedule/context adjustment; a 2–1 head-to-head is an anecdote, not a signal.

The market picture — Jesper De Jong price by book

BookPriceImplied PMove since first snapshot
Pinnaclefair-line source1.4469.3%
Unibetbest price1.4469.4%
Bet3651.4071.4%

Pinnacle fair P (vig removed): 67.3%. Implied P here is raw 1 ÷ odds per book. Only one snapshot per book is stored so far — movement appears after the evening capture.

Analyst read

Written analysis for this match hasn't been published yet.

Sizing, as education: if you act on filter matches at all, professionals size them flat (a fixed ~1% of bankroll) or by fractional Kelly (we model 0.15× Kelly with an uncertainty discount). Never more because a number looks big — a bigger gap above 12% usually means a stale line, not a better bet.

We don't beat Pinnacle — nobody reliably does. We find the books that disagree with it, and we grade every pick against the close, publicly — the receipts, including what we got wrong. Not betting or financial advice. 21+ where legal. About Baseline.

← Back to the Edge Board