L. Tararudee vs A. Anisimova — Prediction & Analysis

TennisEdge's AI model finds no betting value in L. Tararudee vs A. Anisimova (Toronto on 2026-08-05) — the market is efficient, with A. Anisimova favoured at 1.32.

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 Aug. 3, 2026.

Recent form

All Aug. 5, 2026 results & picks →

Full analysis

L. Tararudee (A, 22, #99) vs A. Anisimova (B, 24, #6) — WTA Toronto 1000, R2 (1/32), hard
RE-ANALYSIS of an earlier PASS at owner request; fresh brief, current prices.

═══ MODELS vs MARKET ═══
ELO A 29.9% | ML A 34.2% | blend A 35.9% | sim (baseline) A 51.6%
Book A 3.30 / B 1.32 — but the EXECUTABLE price is Betfair: A 3.20 / B 1.45.
Betfair A 3.20 -> commission-adj eff 3.09 -> requires 32.36%.
Pinnacle (14:55Z, ~12min before Betfair snap): A 3.09 / B 1.407, no-vig A 31.3%, drift_a 0.0pp (no sharp move).
The brief's value_edge 7.3% is an artifact of a STALE BOOK quote (implied 28.6%). Priced off the
exchange the ML edge collapses: ELO -2.5pp, ML +1.9pp, blend +3.5pp. Exchange (32.4%) and Pinnacle
(31.3%) already sit ABOVE ELO. B at 1.45 on the exchange vs 1.32 at the book IS the market pricing
the rust story below — we are not first to it.

═══ SIM CALIBRATION (Step 4b) — the decisive finding ═══
Baseline sim says A wins 51.6% (spw_a 0.577 / rpw 0.447, hold 61.6%) — a coin flip against a top-10
player, diverging ~20pp from ELO/ML/Pinnacle/Betfair. Cause is textbook input contamination: A's
serve/return averages were farmed on Istanbul-2 QUALIFYING (five wins vs 1500-elo fields) plus clay
at Prague. They are unweighted and opponent-unadjusted, so they overstate her badly against elite
return pressure. Per Step 4b(2) the ML/ELO probability is trustworthy and the sim INPUT is broken.
Calibrated: ?spw_a=0.537 (baseline 0.577, used 0.537, clamped=false, inside +/-0.05) -> A win prob
30.96%, landing between ELO 29.9% and Pinnacle 31.3%. That run is the coherent distribution for all
markets. Corroboration that the calibration is right, not fudged: its totals line (over 20.5 = 48.5%)
independently matches Pinnacle's no-vig over of 48.3%.
CONSEQUENCE: blend model_prob_a 35.9% is itself contaminated (it carries the 51.6% sim), so the only
3-5% "probe band" number on the board is not usable as a pre-narrative edge.

═══ RECENT FORM (5 most recent each; counts use full window) ═══
A — Tararudee (hard->hard, NO transition):
  08-03 Toronto R64 W vs E. Jones (1748) 7-6 6-2 orientation-var, has TB -> GRIND over Strong
  08-01 "Toronto - Final" W vs Q. Zheng (1906) 6-3 6-4 -> DOM (see data flag)
  07-24 Prague (clay) QF L vs Snigur (1722) 0-6 3-6 -> heavy blowout loss, 11d ago
  07-23 Prague W vs Bartunkova (1704) 6-7 6-3 6-7 -> GRIND (two TBs)
  07-21 Prague W vs Kovackova (1573) 3-6 7-5 2-6 -> GRIND vs Avg (under-performing bracket)
  Earlier: 5-match Istanbul-2 qualifying run, mostly vs 1500-elo.
  DATA FLAG: a round labelled "Final" at tournament "Toronto" on 08-01 three days before a 1/32,
  vs an opponent shown at 1906 elo, is almost certainly a QUALIFYING final (mis-mapped). Treated as
  a quali win; no thesis is built on "beat Zheng".
  SHAPE: STABLE. Wins are tiebreak-heavy against sub-1800 (7-6 vs Jones 1748; 6-7/6-7 vs Bartunkova)
  = winning ugly, and there is a 0-6 3-6 blowout loss 11 days back. Not SHARPENING.
  PATH: 3 matches at THIS venue this week on hard -> genuinely MATCH-SHARP, conditions grooved.
B — Anisimova (grass 07-04 -> hard: TRANSITION):
  07-04 Wimbledon R16 L vs Keys (1905) 6-3 2-6 3-6 -> competitive loss to Elite, neutral shape
  07-02 Wimbledon W vs Kenin (1609) 6-2 4-6 7-6(10) -> GRIND
  06-30 Wimbledon W vs Gjorcheska (1609) 6-3 6-2 -> DOM
  06-12 London QF L vs Jovic (1872) 3 sets -> neutral
  05-30 RG R16 L vs Parry (1677) 3 sets -> UNDER-performing vs bracket
  SHAPE: STALE/UNKNOWN — last match 31 days ago. Last graded window mixed, with a poor loss to
  Parry (1677). PATH: first match at the venue (top-8 seed debut, per LWOS 08-03) -> RUST RISK.

═══ FATIGUE (Step 2 calibration) ═══
A: matches_14d 5, 7d 2, 510 court-min, back_to_back 1, played YESTERDAY (08-03) — but that was a
TWO-set win, so the measured -4.5pp 3-set hangover does NOT apply (2-set win ~ neutral). Age 22 ->
recovers fast, soften further. Cumulative 510min is CONTEXT ONLY, never an edge (2026-07-27 research:
next-match win rate 47% vs 48% after long matches).
B: zero matches in 14d — that is RUST, not an advantage (Step 1b). Net: fatigue is NOT a factor either
way. The usual "market over-prices visible fatigue -> value on the tired side" check does not rescue
A here, because the tired side is also the side the market has ALREADY drifted toward (1.32 -> 1.45).

═══ NARRATIVE ═══
Let-down: A won a "Final" 3d ago -> POSSIBLE by the letter of the rule, but it is a quali final at
this same venue, i.e. a path positive, not a peak-and-crash. Immaterial.
Venue momentum: A 2-0 at Toronto this week; B 0-0 in window (2024 finalist per press, outside window).
Transition: B grass->hard. Directionally supports an A ML lean; per Step 3 it may NOT support a tight
cover of B, and it cannot create numeric edge.
ELO/ranking sanity: A #99 with ELO implying 29.9% vs a #6 — coherent, ELO credible. B ranked #6 yet
ELO gives her only ~70% vs #99 — modest for a top-6, consistent with LWOS 08-03 noting she "sits well
outside the Top 10 of the WTA Race" (rank buoyed by her 2025 US Open final). So ELO already discounts
her; there is no stale-ELO edge left to harvest.
PlayerNotes: only a multi-match roundup preview, zero extracted facts, style_profile null. No usable
durable traits for a margin thesis.

═══ MARKET CHOICE ═══
All markets priced off the ONE calibrated distribution (spw_a 0.537, A win 30.96%):
  ML A 3.20 -> need 32.36% | calibrated 30.96% -> edge -1.4% (best-case ML model 34.2% = +1.9%, below
    the 3% probe floor; blend's +3.5% is contaminated by the broken sim). min 3.52 raw needed. FAIL.
  ML B 1.45 -> need 70.05% | calibrated 69.04% -> edge -1.0%. FAIL.
  A +4.5 @ 1.85 -> need 55.33% | sim 60.92% -> edge +5.6%. HIGH-VARIANCE tier (|line| >= 4.5) so the
    bar is 10% -> FAIL on the bar. Sharp cross-check also contradicts: Pinnacle A +4.0 devigs to
    48.6%, while the sim implies ~53% at +4.0 (it jumps 46.5% -> 60.9% across 3.5/4.5). Sharp says the
    cover is LESS likely than the sim -> that KILLS it independently of the bar.
  B -4.5 @ 2.08 -> need 49.36% | sim 39.08% -> edge -10.3%. FAIL.
  Over 20.5 @ 2.04 -> need 50.30% | sim 48.50% -> edge -1.8%. FAIL. (Brief's model over 53.7% came off
    the UNCALIBRATED broken sim; Pinnacle devigs over to 48.3%, matching the calibrated 48.5%.)
  Under 20.5 @ 1.91 -> need 53.63% | sim 51.50% -> edge -2.1%. FAIL.
AH shadow record: A +4.5 @ 1.85 (pass — high-variance bar + sharp contradiction)
No market clears its bar. No refutation web check run: PASS decisions get no web access by design.

GATES
  G1 edge-origin: FAIL — pre-narrative exchange edge is -1.4% on the calibrated distribution and at
    best +1.9% on the most A-friendly model; the only 3%+ number (blend +3.5%) is contaminated by a
    sim whose A serve inputs were farmed on 1500-elo quali fields. Below the 3% probe floor -> PASS.
  G2 consensus:   [tier: capped] — moot. ELO is BELOW the market on A; only ML is above, and barely.
    Lone contaminated signal; would have been 0.5-1.0u lone_signal: had any edge survived G1.
  G3 fatal-risk:  [none] — nothing left to cap; the blowout risk was encoded numerically via the
    spw_a shade rather than asserted, and the resulting cover still misses its variance-tier bar.

═══ VERDICT: PASS ═══
The real story here is attractive and it is exactly the kind the market is slow to price — a
match-sharp qualifier three matches deep on the surface against a #6 who has not played in 31 days,
is stepping off grass, plays her first match at the venue, and whose ranking is flattered by a 2025
result. The problem is that the market is NOT slow to it this time: Betfair has already moved B from
the book's 1.32 out to 1.45 and Pinnacle independently sits at 31.3% for A with zero drift. The
apparent 7.3% value is a stale book quote plus a simulation that thinks a #99 with quali-inflated
serve numbers is a coin flip against a top-10 returner. Once the sim is calibrated to the price we
actually trust, every market on the board — both ML sides, both handicap sides, over and under — is
negative or fails its variance bar. Nothing to stake; the earlier PASS stands on better evidence.