TennisEdge's AI model finds no betting value in M. Zheng vs M. Landaluce (Los Cabos on 2026-07-27) — the market is efficient, with M. Landaluce favoured at 1.70.
═══ M. Zheng (A) vs M. Landaluce (B) — Los Cabos, hard, 1/16-finals ═══
MODELS / MARKET (severe source disagreement)
ELO A 50.1% | ML A 48.0% | blended model A 50.4%
Sim (resimulate) A 73.8% / B 26.2% <-- 15pt OUTLIER above the sharp exchange
Betfair exchange: A 1.70 (58.8%) / B 2.36 (42.4%)
Book: A 2.18 (45.9%) / B 1.70 (58.8%) <-- FLIPPED vs exchange
The book and exchange disagree on who the favourite even IS. betfair_ml odds_a(1.70)
== book odds_b(1.70) and betfair odds_b(2.36) ~ book odds_a(2.18): the two price
sources are essentially swapped. One is stale/mislabelled; cannot trust either as clean.
DATA-INTEGRITY FLAGS (material)
- Brief alt_markets.model handicap said A covers -2.5 at 52.5% while win prob only 50.4%
-> a cover cannot exceed the win prob; internally impossible. Fresh resimulate instead
gives A covers -2.5 at 64.5% -> the brief's alt-model was corrupt.
- Match histories chronologically scrambled: Zheng "Won Wimbledon Final" 06-25 THEN played
Wimbledon 1/64-finals 06-29; multiple "Won" results carry losing scorelines. Round/date/
result fields are unreliable, so DOM/GRIND and narrative reads are low-confidence.
PATH / SHAPE (the reliable read)
A (Zheng, 22, rank 182): MATCH-SHARP on hard. 4 hard-court matches this week at Bloomfield
Hills, reached SF (beat Walton ~1767 Strong, Shelbayh 1605, Magadan), lost SF to Fearnley
1677 in straights. Same surface as Los Cabos. 2 days rest.
B (Landaluce, 20, rank 94): RUST RISK — fatigue shows 0 matches/14d, last_match_date null;
last listed match 06-29 (~28 days ago, >21d). Higher-ranked but pre-layoff form poor
(losses to Kwon 1566, Struff, Atmane 1441). ELO ~equal to Zheng despite rank 94 vs 182
-> ranking overstates current Landaluce; ELO likely fair-to-generous given the layoff.
Read: match-sharp Zheng vs rusty Landaluce SUPPORTS the exchange (A favoured), not stale ELO.
FATIGUE (Step 2)
Zheng 4 matches/7d, 409 court-min, back-to-back 1, age 22 (recovers fast) -> context only,
not an edge. Not an extreme spot. Landaluce fresh but rusty (see path). Net: neutral-to-A.
CONVERGENCE VERDICT ON THE SIM
Sim A 73.8% diverges from ELO (50%), ML (48%) AND the sharp exchange (58.8%) at once ->
inputs suspect (Zheng SPW 0.622 across a mixed 50-match grass/clay/hard window likely
inflated). No single input correction is justifiable from the corrupt history, so I do NOT
re-run to a free-handed number and do NOT lean on the sim's magnitude. Best estimate for A
defers to the sharp exchange: ~58%.
EDGE TESTS at executable (Betfair, commission-adj) prices, est A 58% / B 42%
ML A @1.70: implied 58.8%, min_odds 1.05/0.58 = 1.81 > 1.70 -> below min_odds, edge -2.1% -> PASS
ML B @2.36: edge -1.6% under my est (the +6% "edge" only exists on stale ELO's 50% for B,
which the exchange + rust read both contradict) -> PASS
AH A -2.5 @1.89: sim cover 64.5% is inflated by the same distrusted input; high-variance lay
line, no reliable margin thesis (histories corrupt) -> PASS
AH B +2.5 @1.81: sim rates it a LOSER (35.5%); backing B to stay close contradicts the
rusty-B / sharp-A read; would require a ~20pt free-hand override -> PASS
Under 22.5 @1.79: sim under 57.1%, edge ~0.0% (price already reflects it) -> PASS
GATES
G1 edge-origin: FAIL — no numerical model-vs-EXECUTABLE-price edge survives; the only apparent
edge (B ML) rests on stale ELO, and the sim edge (A) is a distrusted outlier.
G2 consensus: FAIL — for backing A the book (A dog 2.18) and ML (A 48%) both sit against it;
only exchange+sim agree, and sim is the outlier. Contradictory sources, no
hard independent evidence to override.
G3 fatal-risk: FAIL — pervasive data corruption (swapped ML odds, impossible cover prob,
scrambled histories) negates confidence in any staked probability.
VERDICT: PASS. Best qualitative read (sharp Zheng > rusty Landaluce) merely agrees with the
exchange, which offers no value at 1.70; every other apparent edge is manufactured by trusting
the weakest/most-corrupt source over the sharpest. No stake.