TennisEdge's AI model finds no betting value in Y. Starodubtseva vs M. Joint (Toronto on 2026-08-03) — the market is efficient, with Y. Starodubtseva favoured at 1.80.
Independent estimates of who wins, each from a different method. The blend is what the bot prices against the market.
| Method | Y. Starodubtseva | M. Joint |
|---|---|---|
| Elo ratings-based, surface-weighted |
41.3% | 58.7% |
| Machine learning 29-feature gradient-boosted model |
40.1% | 59.9% |
| Serve / return point-by-point simulation |
59.5% | 40.5% |
| Blend what the bot actually prices off |
48.9% | 51.1% |
| Market implied de-vigged from the opening odds |
52.6% | 47.4% |
We publish how well these score against the closing market, including where they lose: how accurate are tennis predictions?
| Player | Surface | Elo | Record | Matches |
|---|---|---|---|---|
| Y. Starodubtseva | hard | 1636 | 69–53 | 122 |
| M. Joint | hard | 1725 | 84–53 | 137 |
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. 3, 2026.
Level at 1–1 over 2 meetings. Full record.
| Date | Event | Surface | Winner | Score |
|---|---|---|---|---|
| Aug. 3, 2026 | Toronto | hard | M. Joint | 2 - 1 |
| June 15, 2026 | Nottingham | grass | Y. Starodubtseva | 2 - 1 |
| L | M. Joint | 1 - 2 | Aug. 3, 2026 |
| L | S. Stephens | 1 - 2 | July 28, 2026 |
| L | A. Blinkova | 1 - 2 | June 30, 2026 |
| L | A. Zakharova | 1 - 2 | June 23, 2026 |
| L | E. Navarro | 1 - 2 | June 17, 2026 |
| W | Y. Starodubtseva | 1 - 2 | Aug. 3, 2026 |
| W | A. Grubor | 2 - 1 | Aug. 1, 2026 |
| L | K. Liutova | 2 - 1 | July 30, 2026 |
| W | W. Osuigwe | 2 - 0 | July 27, 2026 |
| L | T. Valentova | 1 - 2 | July 22, 2026 |
All Aug. 3, 2026 results & picks →
Y. Starodubtseva (A) vs M. Joint (B) — WTA Toronto R1 (1/64), hard.
MODELS vs MARKET
ELO A 0.4131 | ML A 0.4013 | Sim A 0.4886 -> models favor B (~0.51-0.60)
Betfair A 1.73 / B 2.30 | Pinnacle no-vig A 0.5661 -> SHARP market favors A
Large model-vs-market split, but it resolves against the model (below).
RECENT FORM (most recent first)
A (Starodubtseva, 26, rank 57, ELO ~1561):
07-28 L Memphis-Q vs Stephens (1500) 4-6 7-5 2-6 -- lost qualy to a weak-avg player
06-30 L Wimbledon vs Blinkova (1623) 7-6 4-6 1-6
06-23 L Eastbourne vs Zakharova (1675)
06-17 L Nottingham vs Navarro (1769)
06-15 W Nottingham vs Joint (1718) 7-6 5-7 6-4 (H2H, on GRASS)
Shape: FADING on hard. Rank-57 is clay-inflated (Paris SF, beat Rybakina RG); ELO ~1561 (Avg)
is the honest hard-court level -- lost Memphis qualy to a 1500.
B (Joint, 20, rank 34, ELO ~1718):
08-01 W Toronto FINAL vs Grubor (1323) 3-6 7-5 7-6 -- won a 3-SET final YESTERDAY, grinding vs 1323
07-30 L Memphis-Q vs Liutova (1717) 3-6 6-3 6-3
07-27 W Memphis-Q vs Osuigwe (1500) 6-4 6-1
07-22 L Prague vs Valentova (1777)
07-21 W Prague vs Samson (1704)
Shape: match-sharp on hard but UNDER-performing (needed 3 sets vs a 1323 in the final).
PATH / SHARPNESS
B is deeply match-sharp (5 matches/14d on hard) but carries the extreme fatigue spot.
A is rested (last 07-28, 5d) but in poor hard-court form.
H2H: A beat B 6 weeks ago on grass (3 sets) -- surface differs, mild only.
FATIGUE (Step 2 calibrated)
B: won a 3-SET match YESTERDAY (08-01) -> 3-set-hangover extreme spot, -4.5pp base.
Age 20 (<=22, recovers fast) -> softened to ~-2.5pp. 518 min/14d + back-to-back = context.
A: 1 match/14d, well within limits. No penalty.
The sharp market (Pinnacle) has already priced B's fatigue -- this is NOT a soft casual
narrative but a measured effect confirmed by a sharp book, so no fatigue-fade value here.
NARRATIVE
B: won a small Toronto event Final 08-01 -> LETDOWN spot (title 1d ago), reinforces the fade.
A: none.
EDGE MATH (executable Betfair, 5% commission)
Back B @ 2.30: eff = 1+(1.30*0.95)=2.235, break-even 0.4474.
Pre-fatigue anchor = sim 0.5114 -> edge 6.4% (clears G1 numerically).
Apply 3-set-hangover haircut (softened, ~-2.5pp) -> B ~0.486 -> edge ~3.9% -> BELOW 5% bar.
ELO/ML's 0.59-0.60 for B is a stale-model artifact (ELO blind to yesterday's title final).
Back A @ 1.73: eff 1.6935, break-even 0.5905. Best A estimate ~0.51-0.52 -> NO edge (market too rich).
ALT MARKETS
Handicap A -1.5 @1.9: sim player_a_covers 0.5642 > sim win 0.4886 -- internally INCONSISTENT
(cannot cover -1.5 more often than you win) -> sim margin output untrustworthy, no articulable
margin edge. Fatigue is on B so cannot support any cover thesis anyway.
Totals 21.5: sim under 0.5126 vs under @1.93 (break-even 0.531) -> negative; over 0.4874 vs @2.0
(break-even 0.5128) -> negative. No totals edge.
AH: B +1.5 not quoted as an edge; B ML dog < 3.0 so no shadow-record requirement.
GATES
G1 edge-origin: PASS (numeric sim edge on B 6.4% pre-fatigue) -- but fatigue+sharp-market KILL it below the 5% bar post-adjustment.
G2 consensus: stake-tier only (lone-ish model edge, contaminated by stale ELO) -- moot, no bet.
G3 fatal-risk: n/a (ML side; not a margin bet).
VERDICT: PASS. The headline model-vs-market split is fully explained by B's measured
3-set-hangover + title letdown (won a final YESTERDAY), which sharp Pinnacle has already
priced. The sim already has B near a coin-flip; after the hangover haircut B's edge at 2.30
falls below the 5% bar, and A at 1.73 is too rich. No clean edge either side, no alt-market edge.