TennisEdge's AI model finds no betting value in S. Shimabukuro vs M. Cilic (Montreal on 2026-08-02) — the market is efficient, with M. Cilic favoured at 1.50.
Independent estimates of who wins, each from a different method. The blend is what the bot prices against the market.
| Method | S. Shimabukuro | M. Cilic |
|---|---|---|
| Elo ratings-based, surface-weighted |
31.2% | 68.8% |
| Machine learning 29-feature gradient-boosted model |
48.5% | 51.5% |
| Serve / return point-by-point simulation |
66.7% | 33.3% |
| Blend what the bot actually prices off |
31.2% | 68.8% |
| Market implied de-vigged from the opening odds |
36.6% | 63.4% |
We publish how well these score against the closing market, including where they lose: how accurate are tennis predictions?
| Player | Surface | Elo | Record | Matches |
|---|---|---|---|---|
| S. Shimabukuro | hard | 1657 | 165–116 | 281 |
| M. Cilic | hard | 1813 | 64–52 | 116 |
Our own ratings, computed from match results since 2018. Players with fewer than 5 rated matches are omitted rather than shown at a provisional starting value. Last updated Aug. 4, 2026.
M. Cilic leads 1–0 over 1 meeting. Full record.
| Date | Event | Surface | Winner | Score |
|---|---|---|---|---|
| Aug. 4, 2026 | Montreal | hard | M. Cilic | 2 - 0 |
| L | M. Cilic | 0 - 2 | Aug. 4, 2026 |
| W | N. Budkov Kjaer | 2 - 0 | Aug. 1, 2026 |
| L | R. Pacheco Mendez | 1 - 2 | July 29, 2026 |
| L | J. Faria | 1 - 3 | June 30, 2026 |
| L | F. Tiafoe | 2 - 0 | June 17, 2026 |
| W | S. Shimabukuro | 0 - 2 | Aug. 4, 2026 |
| L | D. Medvedev | 0 - 3 | June 29, 2026 |
| L | U. Humbert | 2 - 0 | June 16, 2026 |
| L | D. Medvedev | 2 - 1 | June 13, 2026 |
| W | N. Borges | 1 - 2 | June 12, 2026 |
All Aug. 2, 2026 results & picks →
═══ S. Shimabukuro (A) vs M. Cilic (B) — Montreal, hard, R1 (1/64) ═══
MODELS vs MARKET (Shimabukuro / A perspective)
ELO 0.3125 (Cilic heavy fav)
ML 0.4846 (near coin-flip, break-even zone)
Sim 0.6674 (A heavy fav — OUTLIER, see below)
Betfair ~0.451 effective (odds_a 2.28) | Pinnacle no-vig A 0.4284
Book 0.3846 (odds_a 2.60)
Models are badly SPLIT; market sits ~0.44 for A, between ELO and ML.
SIM VALIDATION (Step 4b) — inputs contaminated, sim rejected as anchor
Baseline resim: A spw 0.603 / rpw 0.421 over 50 matches → win prob 0.667.
Convergence FAILS: 0.667 vs ELO 0.31 vs ML 0.48 vs market 0.44 — sim is a
far-high outlier. Shimabukuro's serve/return window is grass-season inflated
(Stuttgart grass TITLE + heavy grass slate vs modest fields, opponent-
unadjusted). On hard vs a ranked opponent those numbers overstate him.
No justified single-input correction rescues a 5% edge; anchor on ELO/ML/
market blend → true A ~0.46-0.49. Sim NOT trusted.
RECENT FORM (last 5 shown)
Shimabukuro (A): 08-01 Won Montreal "Final" vs Budkov Kjaer(1715) 6-3 6-4 (near-DOM);
07-29 LOST Los Cabos vs Pacheco Mendez(1327) 7-6 5-7 2-6 (UNDER-perf, lost to weak);
06-30 Lost Wimbledon (Faria) grass; 06-17 Lost Halle (Tiafoe) grass;
06-13 Lost Stuttgart QF (Shelton) grass. Won Stuttgart title on grass.
Path: match-sharp on hard (just won a hard title 08-01). Shape STABLE/mixed
— sharp title run but a bad loss to a 1327 four days ago.
Cilic (B): 06-29 Lost Wimbledon (Medvedev) 1-6 2-6 6... blowout; 06-16 Lost London
(Humbert); 06-13 Lost Hertogenbosch (Medvedev); 06-12 Won (Borges); 06-08 Won
(Shapovalov). Also lost to Kouame(1418) at RG. Shape FADING; last match 06-29 =
34+ DAYS AGO → heavy RUST RISK. Age 37, slow recovery.
FATIGUE / NARRATIVE
A: 2 matches/14d, 243min, 2-set win yesterday (NOT a 3-set hangover → no -4.5pp
haircut; ~neutral). Age 29 baseline. LETDOWN spot: won a title 08-01 (1d ago),
quick turnaround — a negative for backing A, not support.
B: rusty (34d idle), well_rested but rust >> rest here.
No surface transition either side (both last played hard / A on hard).
ELO vs RANKING: Cilic ranked 47, ELO ~0.69 fav — rank and ELO agree, but 34-day
layoff + string of losses to weak players means ELO likely OVERSTATES current
level. Market already discounts Cilic below ELO (56-57% vs 69%). Not a lean-on.
GATES
G1 edge-origin: FAIL — no model-vs-market gap survives once the outlier sim is
discounted. ML (0.485) vs market effective (~0.451) = ~3.4% on A, below 5%;
ELO opposes. No side clears the 5% executable-price bar.
G2 consensus: stake-tier only (non-blocking).
G3 fatal-risk: n/a (no margin bet staked).
MARKET CHOICE
ML A @2.28: need win_prob >= 0.501 for 5% edge; my est ~0.46-0.49 → PASS.
ML B @1.76: need Cilic >= ~0.631; market already ~0.58, rust argues against → PASS.
Totals: sim exp 21.9 games, under@22.5 0.551 / @23.5 0.608 → ~0.58 under at line 23;
under @1.92 effective implied 0.534 → ~4.6% edge, below bar AND built on the same
contaminated serve inputs → PASS.
Handicap: sim model handicap null; A +1.5 @2.0 — no margin-specific evidence, and a
cover thesis can't rest on the rust/letdown story → PASS.
AH: B -1.5 @1.91 (fav side) / A +1.5 @2.00 — not staked.
alt_markets present (betfair_ml, totals, handicap).
VERDICT: PASS. Real qualitative story (Cilic rust/fade vs Shimabukuro hard-court
sharpness) but it is already largely priced (market moved A well above ELO), the
sim's bullish A number is grass/weak-field inflated and rejected, and no market
clears the 5% edge bar at executable Betfair prices. Add A's letdown/quick-
turnaround as a counterweight. No robust numeric edge → no bet.