TennisEdge's AI model finds no betting value in J. Lehecka vs C. Wong (Los Cabos on 2026-07-29) — the market is efficient, with J. Lehecka favoured at 1.22.
J. LEHECKA vs C. WONG — Los Cabos, hard, 1/8-finals
═══════════════════════════════════════════════════
MODELS vs MARKET
Betfair (executable): Lehecka 1.27 (78.7%) | Wong 4.10 (24.4%)
Book snapshot: 1.22 / 4.33 — context only
ELO: Lehecka 77.1% → Wong 22.9% (BELOW market)
ML: Lehecka 73.1% → Wong 26.9% (1.5pp above comm-adj market)
Sim: Lehecka 49.1% → Wong 50.9% (<-- 30pp outlier, see below)
Blend (model_prob_a) 69.5% → drives the app's "value_pick B, edge 8.5%"
SIM VALIDATION — the headline edge is an input artifact
Baseline resimulate inputs: Lehecka spw .618 / rpw .411 ; Wong spw .614 / rpw .416.
Near-identical serve/return for a #19 and a #150 — the inputs are NOT
opponent-adjusted, and here the two players' fields are about as far apart
as the tour allows:
- Lehecka's 50-match window: grass, vs Zverev 1891, Ruud 1963, Fils 2004,
Shelton 1809, Majchrzak 1842, Carreno-Busta 1798. Elite returners on a
surface that suppresses rally-won%. His numbers are DEFLATED.
- Wong's window: Challenger hard (Lincoln, Bloomfield Hills, Jiujiang),
opponents 1561-1659 almost without exception. His numbers are INFLATED.
Convergence test: ELO 77.1, ML 73.1 and market 78.7 cluster inside 6pp; the
sim sits 25-30pp away. Divergence => distrust the sim, not the other three.
Bounded re-run (logged, clamped at the -0.05 limit): spw_b .614 -> .564,
rpw_b .416 -> .366 flips Lehecka 49.1% -> 90.3%. A max-allowed correction
overshooting to 90% shows the distribution is hyper-sensitive to inputs I
cannot pin down — it is not a usable margin anchor in either direction.
Working estimate = the convergent cluster: Lehecka ~76%, Wong ~24%.
RECENT FORM (5 most recent each)
LEHECKA (24, #19) — all grass:
07-07 L A. Zverev (1891) 4-6 5-7 6-3 6-7 competitive 4-set loss to Elite
07-04 W J. Munar (1687) 6-4 6-4 4-6 6-4 DOM over Strong — real edge
07-02 W A. Molcan (1594) 6-3 6-2 6-4 DOM
06-30 W A. Popyrin (1534) 6-4 6-2 6-4 DOM
06-18 L R. Hijikata (1721) 4-6 7-5 7-6 GRIND loss
WONG (22, #150) — all hard:
07-28 W Dar. Blanch (1561) 6-3 6-4 DOM (Los Cabos R32)
07-23 L J. Wolf (1640) 3-6 5-7 straight-sets loss
07-21 W S. Johnson (1591) 6-2 6-4 DOM
07-18 L S. Johnson (1586) 3-6 6-7 close loss
07-16 W A. Shepp (1576) 7-5 7-6 GRIND
SHAPE / PATH / SHARPNESS
Lehecka: STABLE-to-SHARPENING vs bracket — DOM over Munar (Strong) and a
4-set loss to Zverev with every set close. Level is not in question.
RUST RISK: last match 07-07 = 21 days idle, plus TRANSITION grass -> hard,
plus first match at the venue (bye into R16). Three rust vectors stacked.
Wong: STABLE, ceiling capped. Every result in the window is against a
1561-1659 band and he splits against it — beats Blanch/Johnson/Shepp, loses
to Wolf 1640 and Johnson 1586. Nothing above 1721 beaten in 90 days.
PATH: match-sharp, 1 match at Los Cabos already won in straights,
conditions-adjusted, no surface change. Genuine positive.
FATIGUE (Step 2 calibration)
Wong: 3 matches/7d, 5/14d, 471 court-min, last match 07-28. Last outing was a
2-SET win (~90min) — that is the ~neutral case, NOT the measured -4.5pp
3-set-hangover. Age 22 => recovers fast, soften a tier. Net: non-factor.
Lehecka: 0 matches/14d. "Well rested" in the payload is really rust, not an
advantage. Cumulative-minutes gap alone is noise per the calibration; it is
context here, and it is already in the price (see below).
NARRATIVE
Let-down: none for Lehecka (no Final in window). Wong's 05-21 Final loss is
68d ago and in a low-level draw — no signal.
Venue momentum: Wong 1-0 at Los Cabos; Lehecka no history in window.
Surface transition: LEHECKA grass -> hard. Depresses level and widens margin
variance in match one. Wong: none (hard -> hard).
ELO/ranking sanity: Lehecka ELO-favourite AND ranked 19 — agreement, ELO
credible. Wong #150 matches his 1560-1660 opposition band. No divergence
flag on either side.
THE MARKET HAS ALREADY PAID FOR THE WONG STORY
A #19 in good form against a #150 on hard normally prices ~1.15-1.18.
Betfair has Lehecka at 1.27. That ~10pp of extra generosity IS the rust +
transition + bye + match-sharp-opponent story, already bought and paid for.
Backing Wong at 4.10 requires the market to be underrating a layoff that is
public knowledge and visibly in the price. No articulable information the
exchange is slow to hold.
PRICING
Wong: Betfair 4.10 -> comm-adj 1+(3.10x0.95) = 3.945 -> 25.35% implied.
Need >=30.35% to stake. My estimate ~24%. Edge NEGATIVE.
min_odds = 1.05/0.24 = 4.38 > 4.10 available -> automatic PASS.
Lehecka: Betfair 1.27 -> comm-adj 1.2565 -> 79.6% implied.
Need >=84.6%. ELO 77.1 / ML 73.1 / my ~76%. Edge NEGATIVE.
Alt markets: alt_markets.handicap and totals are BOTH null — no exchange
prices exist for this match, so no handicap or total is stakeable regardless
of what the sim distribution says. ML is the only live market and neither
side clears its bar.
GATES
G1 edge-origin: FAIL — the only numerical edge (value_pick B, 8.5%) comes
from a sim whose serve/return inputs are opponent-unadjusted across a
#19-grass-vs-Elite window and a #150-Challenger window; strip that and ELO
(22.9%) and market (24.4%) leave Wong with no edge at all.
G2 consensus: FAIL — market and ELO both sit against Wong; ML is +1.5pp,
inside noise. No hard independent evidence: no on-surface H2H, and Wong's
live this-event evidence is one win over a 1561 in a field he splits with.
G3 fatal-risk: FAIL — class gap is real and my own read surfaces it: Wong
has beaten nobody above 1721 in 90 days and loses to 1586-1640 routinely.
If Lehecka's level travels off the layoff at all, the thesis is dead.
VERDICT: PASS
Three-gate failure. The app's 8.5% "value" on Wong is a measurement artifact
of unadjusted sim inputs, not a market error — the exchange, ELO and ML all
agree within 6pp and the layoff story is already in a 1.27 favourite. Wong's
price is short of the 4.38 minimum his fair 24% demands, and Lehecka at 1.27
is 5pp of probability short of his own bar. No exchange handicap or total
exists to express a lower-variance view. Nothing to stake.