TennisEdge's AI model finds no betting value in M. Joint vs A. Grubor (Toronto on 2026-08-01) — the market is efficient, with M. Joint favoured at 1.02.
Independent estimates of who wins, each from a different method. The blend is what the bot prices against the market.
| Method | M. Joint | A. Grubor |
|---|---|---|
| Elo ratings-based, surface-weighted |
90.2% | 9.8% |
| Machine learning 29-feature gradient-boosted model |
90.4% | 9.6% |
| Serve / return point-by-point simulation |
55.3% | 44.7% |
| Blend what the bot actually prices off |
83.1% | 16.9% |
| Market implied de-vigged from the opening odds |
94.9% | 5.1% |
We publish how well these score against the closing market, including where they lose: how accurate are tennis predictions?
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.
M. Joint leads 1–0 over 1 meeting. Full record.
| Date | Event | Surface | Winner | Score |
|---|---|---|---|---|
| Aug. 1, 2026 | Toronto | hard | M. Joint | 2 - 1 |
| 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. 1, 2026 results & picks →
═══ M. Joint (A) vs A. Grubor (B) — Toronto (hard), Final ═══
MODELS vs MARKET
Market: A 94.9% (odds 1.02) / B 5.26% (odds 19.0), book quotes only (no exchange).
ELO: A 90.2% -> B 9.8%
ML: A 90.4% -> B 9.6%
Sim: A 83.1% -> B 16.9% <-- outlier, source of the flagged 11.8% "value" on B.
All three models rate A BELOW the market, but the only material dog edge comes from the sim.
SIM DISTRUST (opponent-unadjusted inputs)
Grubor's serve/return numbers were built almost entirely against ITF fields with
opponent ELO 1086-1658 (Bhopal 1086, Smith 1222, Ahn 1401, Schoppe 1297...).
The sim does NOT quality-adjust those inputs, so it materially OVERSTATES her
competitiveness against a tour-level opponent. The 16.9% dog prob is inflated;
ELO/ML (~9.7%) are the credible anchors here.
RECENT FORM
A (Joint, rank 34, age 20): mixed but clearly tour-level. Beat Osuigwe (1500) 6-4 6-1,
beat S.Williams (1387) in a 3-set grind, lost to Liutova (1717), Valentova (1777),
Eala (1846). Superior player by a wide margin.
B (Grubor, rank 733, age 22): FADING / weak ITF form. Lost to Schoppe (1297), Ahn (1401),
Cross (1596), Hosogi (1500). Beating only sub-1300 opposition. No qualitative case for
the dog; the shape read argues AGAINST backing B.
FATIGUE
A last played 07-30 (3-set LOSS, 2 days rest) — no extra-set hangover (lost, not won).
B last played 07-25 (7 days). No extreme spot either side. Neutral.
DATA INTEGRITY
Rank-34 Joint shown losing in Memphis *qualification* 07-30, then a WTA Toronto final
08-01 (implausible timeline). Rank-733 Grubor (ITF-level, losing to sub-1400 ELO) in a
WTA-1000 final is implausible. Inputs are suspect — a further reason to demand a clean,
large exchange edge before staking, which does not exist.
GATES
G1 edge-origin: FAIL — the only >5% edge on B comes from the opponent-unadjusted sim
(16.9% vs 5.26%). Stripping that biased input, ELO/ML give B ~9.7% vs book 5.26% =
~4.4% edge, below the 8% book-priced bar. No numeric edge survives on credible models.
G2 consensus: stake-tier only (never blocks). Would be lone_signal (sim-only) anyway.
G3 fatal-risk: n/a (ML market).
MARKET CHOICE
alt_markets: none at analysis time
AH: not quoted
No exchange ML price -> book-priced dog ML requires edge >=8%; credible models give ~4.4%.
VERDICT: PASS. Backing favorite @1.02 has no edge (min-odds discipline). The dog's apparent
11.8% value rests entirely on a sim whose inputs were farmed against weak ITF fields; ELO/ML
give a sub-bar book-priced edge, there is no exchange liquidity (the losing segment), the
qualitative read is anti-dog, and the match data is internally inconsistent. No stake.