L. Sonego vs T. Griekspoor — Prediction & Analysis

TennisEdge's AI model finds no betting value in L. Sonego vs T. Griekspoor (Montreal on 2026-08-03) — the market is efficient, with T. Griekspoor favoured at 1.80.

Model probabilities

Independent estimates of who wins, each from a different method. The blend is what the bot prices against the market.

We publish how well these score against the closing market, including where they lose: how accurate are tennis predictions?

Elo ratings

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 Jan. 22, 2026.

Head to head

L. Sonego leads 2–0 over 2 meetings. Full record.

Recent form

All Aug. 3, 2026 results & picks →

Full analysis

L. Sonego (A, 31, #66) vs T. Griekspoor (B, 30, #25) — ATP Montreal R64, HARD
RE-ANALYSIS on fresh prices (brief alt_markets 15:31Z, Pinnacle 15:25Z).

═══ MODELS vs MARKET ═══
ELO      A 42.9%  (B 57.1%)
ML       A 44.8%  (B 55.2%)
Blend    A 44.1%  (B 55.9%)
Sim      A 40.4%  (B 59.6%)  — spw/rpw A .587/.405, B .606/.404
Betfair ML (executable)  A 2.04 / B 1.93 → devig A 48.6% / B 51.4%
Pinnacle no-vig          A 49.5% / B 50.5%, drift_a_pp 0.0 (fresh, no sharp money moving)
Book quote               A 2.00 / B 1.80 (context only)

All models lean B by 4–9pp over the market; the SHARPEST estimate (Pinnacle, 6 min
older than the exchange snapshot) is a coin flip — i.e. the sharps sit BELOW the
market's own B price, not above it. That inverts the usual "model finds value" read.

Edge test on B at the executable price: eff = 1+(1.93−1)×0.95 = 1.8835 → 53.1%.
  blend 55.9% → +2.8pp   ML 55.2% → +2.1pp   ELO 57.2% → +4.1pp
vs Pinnacle fair (50.5%) → B is OVERPRICED at Betfair by ~2.6pp.
A side: needs 50.3% at eff 1.988; no model or sharp puts A above 49.5% → dead.

ELO/rank sanity: Griekspoor is #25 with a B-favouring ELO, but his live results
(below) are those of a much lower-rated player — this is a rating that has not yet
repriced a run of losses to sub-1800 opposition. Do not lean on it.

═══ RECENT FORM (5 most recent each; all fetched matches used for counts) ═══
SONEGO (A) — clay/grass only, ZERO hard-court matches in window
  07-15 clay Gstaad  L  R. Collignon 1830   7-6 7-6      GRIND-loss, narrow vs Strong
  07-13 clay Gstaad  W  Schwaerzler 1639    6-4 7-6(TB)  GRIND vs Avg
  07-04 grass Wimb   L  T. Fritz 1979       4-6 6-3 6-4 7-6  competitive vs Elite
  07-02 grass Wimb   W  G. Diallo 1730      5 sets, 3 TBs  GRIND
  06-30 grass Wimb   W  Etcheverry 1592     6-4 6-4 6-7 7-6  GRIND vs Avg
  Also: lost Halle FINAL 06-14 to Basilashvili 4-6 4-6 (51d ago — letdown expired).
  SHAPE: STABLE-to-FADING. Almost every win is a GRIND, incl. tiebreak sets against
  Avg opposition (Schwaerzler 1639, Etcheverry 1592) = under-performing the bracket;
  the one positive is a 4-set loss to Fritz. No DOM over Strong/Elite in 90 days.
  PATH: last match 07-15 → 20 DAYS IDLE, first match at the venue. Rust risk, and
  age 31 shakes rust slower. TRANSITION clay → hard (level depressed, margins wider).

GRIEKSPOOR (B) — on surface, but the results are ugly
  07-27 hard Washington  L  A. Tabilo 1772   lost 3 sets (7-6 4-6 6-4)
  07-18 hard UTS         L  J. Fonseca       EXHIBITION (UTS scoring) — discarded
  07-18 hard UTS         L  C. Norrie        EXHIBITION — discarded
  07-16 hard UTS         W  C. Moutet        EXHIBITION — discarded
  06-30 grass Wimb R1    L  J. Duckworth 1541  4-6 6-4 5-7 4-6
  Earlier: L Shimabukuro 1719 (Halle), L Z. Zhang 1622 (Hertogenbosch),
  L Arnaldi (RG R1), Bordeaux CHALLENGER SF (beat Bonzi 1585, Molcan 1812, lost
  Collignon 1831), L Blockx 1856 (Rome).
  SHAPE: FADING, clearly. In real (non-exhibition) tennis: one tour-level win since
  06-10 (Van de Zandschulp 1569, in 3). Losses to 1541 / 1622 / 1719 are UNDER-
  performing the bracket by any reading. A top-25 player dropping to a Challenger
  and losing its semi is the shape of a slump, not of a 60% hard-court favourite.
  PATH: last real match 8 days ago, on hard. No transition. Not match-sharp either
  (a 1st-round loss, then exhibitions).

═══ FATIGUE (brief, precomputed) ═══
A: 0 matches/14d, 0 min — fully rested; the issue is rust, not load.
B: 3 matches/14d, 1046 min BUT that total is inflated by three UTS exhibition
   "matches"; last real match 07-27. back_to_back 0, no 3-set win yesterday.
No extreme spot on either side → fatigue is CONTEXT ONLY, worth nothing here
(2026-07-27 calibration). No 3-set-hangover haircut applies to either player.

═══ NARRATIVE ═══
Let-down: none live (Sonego's lost Halle final is 51d old).
Venue momentum: neither player has a Montreal match in the window.
Title recency: none.
Surface transition: SONEGO clay → hard (risk on any tight-cover thesis for him).
player_notes: null — no press-derived facts, no style_profile for either man.

═══ GATES ═══
GATES
  G1 edge-origin: FAIL — pre-narrative ML edge on B vs executable Betfair 1.93 is
    +2.8pp (blend) / +2.1pp (ML): below even the 3% probe bar. Only the stale-looking
    ELO reaches 4.1pp, and Pinnacle prices B 2.6pp WORSE than Betfair, so the sharp
    reference removes the gap rather than corroborating it. No numeric edge to stake.
  G2 consensus:   [tier: capped] — models agree on B but the agreement is contaminated
    (ELO not yet repriced for the slump; sim serve inputs off-surface / exhibition-fed).
    Moot given G1.
  G3 fatal-risk:  [none] — no unencodable thesis-negating risk; nothing to cap.

═══ MARKET CHOICE ═══
ML: B +2.8pp at eff 1.8835 (min 1.92 raw needed at p=.559, and p is not defensible
    above the sharp's 50.5%) → PASS. A side needs 50.3%, no estimate clears → PASS.
HANDICAP (Betfair A +0.5 @1.93 / B −0.5 @2.00):
    A +0.5 @ 1.93 — sim A covers 42.7% vs 53.1% needed → dead.
    B −0.5 @ 2.00 — sim B covers 57.3% vs 51.3% needed = +6.0pp, TIGHT tier (bar 5%),
    so it clears on paper. KILLED by the sharp cross-check: Pinnacle's own +0.5 games
    line (1.885/1.98) devigs to A covers 51.2% / B covers 48.8% — the sharps make B's
    −0.5 cover an 8.5pp LONGER shot than our sim does. A Betfair AH price the sharps
    contradict is a stale quote, not value.
    Sim calibration confirms the sim is the broken side: shading spw_a from .587 to
    .603 (+0.016, inside the ±0.05 clamp, clamped=false) lands A's win prob at 48.7%
    ≈ market 48.6% ≈ Pinnacle 49.5%, and at that calibration B −0.5 covers only 52.8%
    → +1.5pp, no bet. The shade is defensible: Sonego's 50-match input window is
    entirely clay+grass (his serve plays bigger on hard), while Griekspoor's .606 SPW
    is fed partly by UTS exhibition tennis. The whole sim edge closes inside a
    justifiable input shade → not stakeable.
TOTALS (Betfair 23.0, over 1.92 / under 2.04): sim under-23 = 52.4% (calibrated 51.3%)
    vs 50.3% needed → +2.1pp / +1.0pp, below bar. Over needs 53.4% vs sim 42.4% → dead.
    Pinnacle 23.0 devigs to over 50.9% / under 49.1% — contradicts the under lean too.
AH: A +0.5 @ 1.93  (shadow quote; handicap passed)
refutation_check: not run — PASS decisions get no web access (Step 4c).

═══ VERDICT: PASS ═══
This is the shape of a genuine no-bet. The one direction the models point (Griekspoor)
is (a) only worth 2–3pp at the price we can actually take, (b) priced WORSE on Betfair
than the sharpest book in the world prices it, and (c) qualitatively unsupported — the
live evidence on Griekspoor is a top-25 player losing to Duckworth (1541), Zhang (1622)
and Shimabukuro (1719), which is the opposite of the confirmation a model edge needs.
Meanwhile the side the qualitative slump read would push you toward (Sonego) is 20 days
idle, transitioning off clay, with no hard-court evidence at all and a grind-heavy shape
— so there is no honest case for him above 49% either. Two players in poor shape, an
exchange price sitting on top of the sharp price, and a sim edge that evaporates under a
0.016 input shade. Nothing to bet in any of the three markets.