M. Kostyuk vs K. Sebov — Prediction & Analysis

TennisEdge's AI model finds no betting value in M. Kostyuk vs K. Sebov (Toronto on 2026-08-04) — the market is efficient, with M. Kostyuk favoured at 1.03.

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 March 22, 2026.

Head to head

M. Kostyuk leads 1–0 over 1 meeting. Full record.

Recent form

All Aug. 4, 2026 results & picks →

Full analysis

M. Kostyuk (A, 24, #26) vs K. Sebov (B, 27, #225) — WTA Toronto, hard, R2 (1/32)

MODELS vs MARKET
  ELO      A 87.3%  (B 12.7%)
  ML       A 88.5%  (B 11.5%)
  Sim      A 79.0%  (B 21.0%)  <- see SIM VALIDATION, rejected
  Book     A 1.03 / B 12.0  (implied A 92.1%)
  Betfair  A 1.05 / B 17.50  -> B effective 16.68 after 5% comm = 6.0% required
  Pinnacle A 1.027 / B 17.31, no-vig A 94.4% (fetched 2026-08-03 22:55Z, ~fresh)
  Betfair and Pinnacle agree tightly (94-95% A). The book 12.0 on B is the stale
  quote here — the exchange prices B at 17.5, so the "value_pick B, edge 13%" flag
  is measured against a price nobody can take. Edge baseline is 17.5/16.68.

RECENT FORM (last 5 each)
  Kostyuk: 07-09 Wimbledon SF LOST Noskova (1931) 4-6 4-6 — competitive, no break of shape
           07-08 Wimbledon QF WON Paolini (1853) 6-3 6-2  DOM over Strong
           07-06 R16 WON Krueger (1874) 6-4 6-4            GRIND-ish over Strong
           07-04 R32 WON Navarro (1843) 6-2 4-6 6-1        3-set win over Strong
           07-02 R64 WON Blinkova (1644) 3 sets w/ TB
           Also 06-04 Roland Garros SF; 05-31 beat Swiatek (2046) 7-5 6-1.
           Two slam semi-finals in the window. This is elite, top-10-level tennis.
  Sebov:   08-02 Toronto R64 WON Ito (1649) 6-3 6-4        DOM over Avg — best win of window
           07-28 Memphis Q LOST Kalieva (1500) 1-6 6-7
           07-26 Memphis Q Final WON Teixido Garcia (1500) 6-1 7-5
           07-25 Memphis Q SF WON Soares (1500) 6-1 6-0
           07-09 Newport R16 LOST T. Maria (1803) 3-6 3-6
           Deeper window: LOST Zamarripa (1265), LOST Leonard (1540) 3-6 1-6,
           LOST Zaar (1570) 3-6 0-6, LOST Mert (1367), LOST Cengiz (1394).
           Ceiling is ITF/quali level; she loses routinely to sub-1550 opposition.

SHAPE / PATH (Step 1b)
  Kostyuk: SHAPE STABLE-to-SHARPENING (6-3 6-2 over Paolini, then a tight 2-set loss
           to a 1931 in a slam SF = neutral-positive). PATH: first match at Toronto
           (bye), last match anywhere 07-09 = 26 days ago, and grass -> hard
           TRANSITION. Genuine rust risk: depressed level and wider margin variance
           in match one. This is the only real case for the dog.
  Sebov:   SHAPE STABLE (6-3 6-4 over a 1649 is her best result of the window and
           above her recent bracket) — but preceded by a quali loss to a 1500.
           PATH: match-sharp, 2 hard-court matches this week at this venue,
           conditions-adjusted. Positive, and the market can see it.

FATIGUE (Step 2)
  Kostyuk: 0 matches/14d, fully rested (and then some — rust side of the ledger).
  Sebov:   4 matches/14d, 2/7d, 269 min, back_to_back 1, last played 08-02 (2 days
           ago, a straight-sets win). No extreme spot: not a 150min+ 3-setter
           yesterday, not a 3rd-in-3-days. 2-set win 2 days ago = ~neutral, no
           haircut applies. Age 27 = baseline modulation. Fatigue is NOT a factor
           in this match in either direction.

NARRATIVE (Step 3)
  Let-down: none for Kostyuk (SF loss, not a Final). Sebov lost an ITF-level Final
  06-25, outside the window that matters. Venue: no prior Toronto history for
  either in the window. TRANSITION on Kostyuk (grass/clay -> hard) — noted as a
  risk to any margin/cover thesis on her side, not as support for the dog ML.

ELO / RANKING SANITY
  Sebov: HIGH ELO relative to a #225 ranking and, more damningly, relative to her
  actual results — her rating is farmed on W50/quali fields (opponents 1265-1570)
  that she still loses inside. Discount her ELO hard.
  Kostyuk: ELO UNDERSTATES her — 87.3% does not reflect a player who has just made
  back-to-back slam semi-finals beating Swiatek (2046) and Paolini (1853). Rating
  lags a breakout run.
  So the entire model-vs-market gap is explained by known ELO failure modes on BOTH
  sides pointing the same way. That is not market inefficiency; it is model error.

SIM VALIDATION (Step 4b, resimulate baseline pulled)
  Baseline inputs: Kostyuk spw .567 / rpw .426; Sebov spw .571 / rpw .428 — i.e.
  the sim believes the world #225 serves and returns BETTER than a Wimbledon
  semi-finalist, and outputs B 53.6% to win, expected margin -0.2 games.
  Why: SPW/RPW are unweighted averages over whatever field each player faced.
  Kostyuk's 50 matches are against Swiatek/Paolini/Navarro/Krueger/Noskova
  (1843-2050); Sebov's are against 1265-1650 ITF/quali opposition. Textbook
  opponent-unadjusted contamination, and in the wrong direction to fix with a
  bounded +/-0.05 override — closing a ~40-point gap would need far more than the
  clamp allows. The sim is REJECTED for this match, and with it every derived
  line (its handicap_lines top out at +/-3.5 and its totals anchor on a coin-flip
  match). model_prob_a 0.790 inherits the same contamination.

MARKET CHOICE
  ML B (dog) @ 17.50 Betfair: required 6.0% after commission; min odds at an honest
    7% = 15.0, so the PRICE is not the blocker — the PROBABILITY is. Stripping the
    two contaminated inputs, my honest number for Sebov is ~6-8%: she gets a live
    puncher's chance from Kostyuk's 26-day layoff + surface transition + first
    match at the venue, offset by a level that loses to 1367s and 1540s. That
    straddles the 6.0% break-even; there is no 5% edge.
  AH: A -7.5 @ 2.18 / B +7.5 @ 1.76. B +7.5 @ 1.76 (implied 56.8%). Wide line
    (|7.5| >= 4.5) => high-variance tier, needs 10% edge. No usable model anchor
    (sim rejected, alt_markets.model.handicap null) and Kostyuk's own margin
    distribution is mid (6-3 6-2 vs Paolini, 6-4 6-4 vs Krueger — she wins by 4-7
    games, not 8+), so -7.5 is genuinely near a coin-flip and I cannot price it to
    10%. Laying -7.5 on a player with a rust + transition flag is exactly the
    self-negating cover the margin rule forbids.
  Totals: 17.0, over 1.87 / under 2.06. Sim total (21.0 expected games) is derived
    from the rejected coin-flip distribution — unusable. Freehand: a ~94% favourite
    projects 15-19 games, straddling 17.0. No priced edge either way.
  alt_markets present (fetched 2026-08-03 23:07Z); no market clears its bar.

GATES
  G1 edge-origin: FAIL — the numeric edge exists only from ELO/ML/sim numbers whose
     error is identified and one-directional (Sebov's rating farmed on ITF fields
     she loses in; Kostyuk's rating lagging two slam SFs; sim inputs
     opponent-unadjusted). Corrected honestly, B is ~6-8% vs 6.0% required — under
     the 5% bar on ML, and no alt market reaches its own bar either.
  G2 consensus:   stake tier only — would have been the capped 0.5-1.0u band
     anyway (contaminated signal: farmed-ELO opposition + rejected sim). Moot.
  G3 fatal-risk:  FAIL for the margin markets (hard veto): any Kostyuk cover thesis
     rests on a player with a 26-day layoff and a grass->hard transition, i.e. the
     thesis negates itself. N/A for ML.

VERDICT — PASS
  Betfair and Pinnacle agree at ~94.5% and I have no information they lack: the
  layoff, the bye and the surface switch are all public and already in a price
  that has Kostyuk at 1.05 rather than the 1.02 her results alone would suggest.
  The only thing arguing for the dog is a set of model numbers whose failure mode
  I can name on both sides. Backing a 17.5 shot off compressed longshot ratings,
  against a player whose real level loses to 1367-rated opposition, is exactly the
  manufactured thesis the calibration anchor warns about. No bet, no lean.
  Predicted winner: M. Kostyuk.