TennisEdge's AI model finds no betting value in U. Humbert vs A. Martin (Washington on 2026-07-28) — the market is efficient, with U. Humbert favoured at 1.10.
═══ U. Humbert (A) vs A. Martin (B) — Washington, hard ═══
MODELS (prob A): ELO 82.6% | ML 60.2% | blended 74.2% | sim 72.9%
MARKET: Betfair ML A 1.26 (comm-adj 80.2%) / B 4.30 (comm-adj 24.2%)
Book A 1.10 / B 6.50 (soft — exchange is 12pts tighter, ignore book)
RANKINGS: Humbert #36 | Martin #269
*** DATA-QUALITY RED FLAG (decisive) ***
Player B "A. Martin" history is CORRUPTED — a merge of 3-4 different
players sharing the name, with physically impossible simultaneous
schedules:
- 07-26: Washington (USA, hard) Final AND Liberec (Czech, clay) SF
- 07-25: Washington SF (USA) AND M25 Ollersbach (Austria) SF
- 07-24: Bloomfield Hills (USA) AND M25 Ollersbach (Austria) QF
One person cannot be in the USA and Czech/Austria on the same day.
Consequence: Martin's form/shape/path CANNOT be read, and the two
model outputs that ingest recent form are poisoned:
- ML (Martin 39.8%) is inflated by a fat merged win-log.
- Sim SPW 0.577 / RPW 0.403 (50 "matches") aggregate multiple
players across mixed clay-ITF and hard — not one player, not
opponent-adjusted. Sim Martin 27.1% is not trustworthy.
The precomputed fatigue block (B: 1781 court-min, 13 matches/14d,
back_to_back 3) is likewise the SUM of several players — meaningless.
RECENT FORM
Humbert (A): all recent play is GRASS (Eastbourne Final, Wimbledon
R64 L to Bergs in 5) and earlier CLAY. Last match 06-30 → ~27 days
idle. RUST RISK (>21d) + SURFACE TRANSITION grass→hard. Legit #36,
ELO-fav, but coming in cold onto a new surface.
Martin (B): unreadable (corrupt merge, see above).
FATIGUE: Martin block corrupt → discard. Humbert well-rested but rusty
(27d off). No usable extreme-spot signal on either side.
NARRATIVE: Humbert grass→hard transition = depressed level + wider
margin variance in match one. This is a REASON NOT to lay him at
-3.5, not a reason to back an unreadable dog.
ELO vs RANK: Humbert ELO≈rank (credible). Martin ELO can't be trusted
as a single entity given the merge — treat the whole B side as noise.
EDGE SCAN (all vs commission-adjusted executable Betfair prices)
Martin ML @4.30: breakeven 24.2%; best model 27.1% → edge ~1-3%. FAIL.
Humbert ML @1.26: breakeven 80.2%; sim/blend say only 72.9/74.2%
→ NEGATIVE edge; min_odds 1.05/0.826=1.27 > 1.26 → auto-PASS.
Martin +3.5 @2.24: breakeven 45.9%; model cover 47.4% → edge ~1.5%. FAIL.
Humbert -3.5 @1.73: breakeven 59.1%; model cover 52.6% → NEGATIVE. FAIL.
Under 22.0 @1.91: breakeven 53.6%; sim ~54.8% → edge ~1%. FAIL.
Over 22.0 @2.02: NEGATIVE. FAIL.
Every market is thin AND every model input on B is corrupt.
GATES
G1 edge-origin: FAIL — the only real "gap" (ML/sim vs market on B)
comes from poisoned merged form data, not a genuine model edge.
G2 consensus: FAIL — ELO + exchange agree (Humbert ~80-82%); the
dissenting ML/sim are the corrupted signals, not hard evidence.
G3 fatal-risk: FAIL — cannot verify Martin is a hard-courter or even
one player; backing an unreadable dog on the favourite's rust is a
manufactured edge.
VERDICT: PASS. No executable edge on any market, and the entire B-side
model stack is built on merged multi-player data. Nothing to price.