TennisEdge's AI bot is backing Under 22.5 games at 1.92 in H. Rocha vs P. Martinez (Estoril on 2026-07-20). Result: LOST.
═══ H. Rocha vs P. Martinez — ATP Estoril 1/16-finals, clay ═══ MODELS: ELO 65.6% A | ML 46.7% A | Market 54.7% A (1.74/2.10; Betfair 1.83/2.16) — models SPLIT; ELO-vs-market edge on A (+10.9%) not validated by ML or live form RECENT FORM (last 5 shown): A Rocha: L Dzumhur (1646, Avg) 6-2 6-1 blowout 7d ago | L Mejia (1468, Weak) Wimbledon QF | L O'Connell (1475, Weak) GRIND | L Blanch (1695, Strong) GRIND | W Broom (1587, Avg) DOM — 1-4 in last 5, cold; Perugia clay title (5-0 run, June) is what props up his ELO B Martinez: W Skatov (1623, Avg) DOM — Estoril FINAL yesterday | W Jorda Sanchis (1640, Avg) DOM Estoril SF | L Hanfmann (1798, Strong) GRIND | L Kasnikowski (1667, Strong) GRIND | W Dev S D (1237, Weak) — red hot at this venue, 2 DOM wins FATIGUE: A — 1 match/14d, 3 sets, 81 min, last match 07-13 (fresh but possibly rusty) | B — 5 matches/14d, 15 sets, 617 min, 2 back-to-backs, played the FINAL YESTERDAY. Court-time diff >500 min → heavy disadvantage Martinez. NARRATIVE: B LETDOWN (won Estoril title 1d ago — peak spot); B venue momentum 2-0 at Estoril; A TITLE: Perugia 43d ago (clay). Live form outweighs ELO. ML VERDICT: ELO edge on Rocha killed by his cold form (blowout loss to a weaker opponent than Martinez 7d ago; ML model has him under 50%). Backing Martinez killed by extreme fatigue + letdown. No winner thesis either way → no ML bet. ALT MARKETS: AH A -1.5 @2.00 / B +1.5 @1.90 | Total 22.5: O 2.02 / U 1.92 | Model: U@21.5 = 58.8% → ~62% at 22.5 vs 52.1% implied. MARGIN THESIS: whichever player shows up flat — cold Rocha or exhausted post-title Martinez — the match projects lopsided. Over 22.5 needs a competitive grind, i.e. BOTH showing up, the least likely scenario here. Both players' recent losses have been blowouts. Model total independently agrees. VERDICT: BET Under 22.5 games @ 1.92 (Betfair), 1.5u. Est. 60% vs 52.1% implied (~8% edge), model-anchored + qualitative confirmation. RISKS: Martinez averaging 3.0 sets/match over 14d (grinder profile); model total derives from the same ELO-heavy sim I distrust on the ML side; clay lengthens sets.