TennisEdge's AI model finds no betting value in K. Nishikori vs J. Shang (Washington on 2026-07-27) — the market is efficient, with J. Shang favoured at 1.50.
═══ K. Nishikori (A) vs J. Shang (B) — Washington, hard, 1/16-finals ═══
MODELS vs MARKET
ELO: Nishikori 51.3% / Shang 48.7% (coin flip)
ML: Nishikori 76.8% / Shang 23.2% (EXTREME OUTLIER)
Blended model: Nishikori 51.5% → drives the headline "value_pick A, edge 14.6%"
Point sim: Nishikori 44.3% / Shang 55.7% (best margin tool)
Market (Betfair): Nishikori 2.56 = 39.1% / Shang 1.61 = 62.1%
Market (book): Nishikori 2.57 = 38.9% / Shang 1.50 = 66.7%
The "14.6% value on A" is a MIRAGE. The blended model_prob_a (0.515) is dragged up
entirely by the ML's 76.8% — a wild outlier vs ELO (51%), the point-by-point sim
(44%) and the market (39%). The ML almost certainly keys on Nishikori's historical
peak / name, not his current post-injury level at age 36 (ranked 238). Strip the ML
and every credible signal (ELO ~coin-flip, sim, exchange) clusters at-or-against a
Nishikori edge. Shang (21) is the genuine favorite per the sim AND the market.
RESIMULATE INPUTS (baseline, not corrected)
Nishikori: spw .614 / rpw .369 / hold .787, 10 matches → win 44.3%
Shang: spw .606 / rpw .388 / hold .803, 16 matches → win 55.7%
Expected total games 23.3, expected margin -0.3 (near pick-em on games).
RECENT FORM
Brief returned EMPTY match lists for BOTH players (matches_a / matches_b = []),
and form_analysis_fallback = {}. No scorelines, no opponents, no DOM/GRIND, no
H2H, no path/shape, no surface-transition read available. The sim has underlying
stats (10/16 matches) but I cannot SEE opponent_elo to validate whether those
serve/return numbers were farmed on weak Challenger fields — so per framework I
trust the baseline and make NO numeric correction.
FATIGUE
All-zero for both (no data, not truly "well-rested"). Not a factor. Nishikori age
36 (slow recovery) / Shang age 21 (fast) noted but no schedule data to apply.
NARRATIVE
Veteran comeback (Nishikori, 36, rank 238, ELO likely stale/inflated) vs rising
young player (Shang, 21). Market sharply favors the younger man — a well-understood
spot the market prices efficiently. No let-down / venue / title / transition data.
MARKET EDGES (baseline sim, commission-adjusted 0.95)
ML A @2.56: +4.0% (below 5% bar; min 2.37) — no bet
ML B @1.61: -7.6% — no bet
AH A +2.5 @2.02: +8.0% (min 1.79) — ONLY numeric edge, but see gates
AH B -2.5 @1.91: -12.4% — no bet
Total O 22.0 @1.91: +1.1% — no bet
Total U 22.0 @2.04: -5.0% — no bet
GATES
G1 edge-origin: PASS — AH A +2.5 shows a genuine +8% off the baseline sim,
pre-narrative; the headline ML-driven "A ML value" is rejected as a model artifact.
G2 consensus: FAIL — the sim (A covers 58.8%) is contradicted by the sharper live
exchange, which prices A +2.5 at only ~49.5% (2.02). Exchange is the truth; I have
ZERO independent evidence (no H2H, no live-event results, no recent form — lists
empty) to adjudicate sim-over-exchange. A lone sim split is not enough.
G3 fatal-risk: FAIL — cover thesis = a rusty 36yo comeback vet stays within 2.5 of a
rising 21yo. Blowout risk (6-2 6-3 = 5+ games) is live and I cannot rule it out with
zero scoreline data; the market's wide handicap pricing directly signals it. Units
never buy back a negating risk.
Margin-evidence check for the cover also fails: need ≥2 of {backed-side scoreline
distribution, H2H margins, surface margin effect, style matchup}. I have only surface
(hard, moderate margins) — the other three are unknowable (empty lists). Insufficient.
VERDICT — PASS. No market clears its gates. The one numeric edge (A +2.5) rests solely
on a sim I cannot validate, is contradicted by the sharper exchange, and carries an
unmitigable blowout risk. Predicted winner: Shang (sim 55.7%, market ~62%).