TennisEdge's AI model finds no betting value in A. Kovacevic vs N. Borges (Montreal on 2026-08-03) — the market is efficient, with N. Borges favoured at 1.36.
Independent estimates of who wins, each from a different method. The blend is what the bot prices against the market.
| Method | A. Kovacevic | N. Borges |
|---|---|---|
| Elo ratings-based, surface-weighted |
38.3% | 61.7% |
| Machine learning 29-feature gradient-boosted model |
43.7% | 56.3% |
| Serve / return point-by-point simulation |
38.5% | 61.5% |
| Blend what the bot actually prices off |
37.1% | 62.9% |
| Market implied de-vigged from the opening odds |
29.8% | 70.2% |
We publish how well these score against the closing market, including where they lose: how accurate are tennis predictions?
| Player | Surface | Elo | Record | Matches |
|---|---|---|---|---|
| A. Kovacevic | hard | 1669 | 153–121 | 274 |
| N. Borges | hard | 1773 | 128–74 | 202 |
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 | Montreal | hard | N. Borges | 2 - 1 |
| Oct. 15, 2023 | Shenzhen 2 Challenger Men | hard | A. Kovacevic | 2 - 0 |
| L | N. Borges | 1 - 2 | Aug. 3, 2026 |
| L | C. Norrie | 2 - 0 | July 28, 2026 |
| L | B. Van De Zandschulp | 1 - 3 | June 29, 2026 |
| L | D. Altmaier | 2 - 0 | June 23, 2026 |
| L | F. Cerundolo | 1 - 2 | June 16, 2026 |
| W | A. Kovacevic | 1 - 2 | Aug. 3, 2026 |
| L | R. A. Burruchaga | 1 - 2 | July 22, 2026 |
| W | O. Luz | 2 - 0 | July 21, 2026 |
| L | L. Darderi | 0 - 2 | July 17, 2026 |
| W | G. Dimitrov | 2 - 0 | July 15, 2026 |
All Aug. 3, 2026 results & picks →
A. Kovacevic (A) vs N. Borges (B) - Montreal, hard, 1/64-finals
MODELS vs MARKET (player A = Kovacevic, the dog)
ELO 38.3% | ML 43.7% | Sim 37.1% | Book implied A 31.2% | Betfair A 3.0 (comm-adj break-even 34.5%)
Edge on A ML: ML model +9.2%, but ELO only +3.8%, sim only +2.6%.
-> Lone-model split: only ML clears the 5% bar; ELO and sim both sit essentially WITH the market.
RECENT FORM (last 5 each)
Kovacevic: L Norrie 6-4 6-1 (hard, 1830 - dominated); L VDZ 3-6 7-6 4-6 0-6 (grass collapse);
L Altmaier (grass); L Cerundolo (grass); L Mpetshi Perricard 3-6 4-6 (London Final, grass).
Shape: FADING - best results were clay in May (Hamburg SF beating AAA/Ugo Carabelli), now
stale/off-surface. Live hard sample = one bad loss.
Borges: L Burruchaga 1-6 6-4 3-6 (clay, competitive); W Luz 6-3 6-3; L Darderi 6-7 4-6 (clay);
W Dimitrov 6-4 6-2 (DOM); W Kouame 6-4 6-2. Shape: STABLE.
PATH / SHAPE / TRANSITION
Kovacevic match-sharp on hard (last match Los Cabos, hard, 5d ago) but that match was a
6-4 6-1 loss - sharp but out of form.
Borges TRANSITION clay->hard, 11 days since last match (mild rust), but sounder hard-court player.
FATIGUE (Step 2)
Neither in an extreme spot. Kovacevic 1 match/14d. Borges 0/7d, 2/14d, last 7-22 (rest, not load).
Cumulative load = context only; no haircut applies to either side.
NARRATIVE
No first-Final let-down in window (last final was a LOSS). No title recency. Borges clay->hard
transition noted (depresses level slightly, widens margins - argues against a tight A cover, not for it).
ELO/RANK sanity: ELO (A 38%) roughly consistent with rank gap (95 vs 52). No divergence to discount.
ALT MARKETS
Handicap A +2.5 @2.12: sim covers 51.7% -> edge +4.55% (below 5% tight-AH bar; min 2.03). Pass.
B -2.5 @1.81: cover 48.3% -> edge -7%. Totals O/U 22.5 both ~ -1%. All sub-bar.
AH: A +2.5 @2.12 (shadow); passed on edge.
GATES
G1 edge-origin: FAIL - the only >5% edge comes from a lone outlier model (ML 44%); ELO and sim
both agree with the market (~37%). The ML outlier is plausibly inflated by clay-farmed stats,
while live hard-court form (6-4 6-1 loss to Norrie) refutes it. No clean numeric edge survives.
G2 consensus: stake-tier only - lone-model support -> would cap 0.5-1.0u even if betting.
G3 fatal-risk: n/a (ML) - but the fading-on-hard read + off-surface stat contamination is the
thesis-killer here.
VERDICT: PASS. A single-model edge on the underdog, contradicted rather than confirmed by his poor
live hard-court form and clay-inflated ML numbers. Two of three models and the exchange agree.
No qualitative validation -> no bet.