B. Nakashima vs D. Altmaier — Prediction & Analysis

TennisEdge's AI model finds no betting value in B. Nakashima vs D. Altmaier (Montreal on 2026-08-04) — the market is efficient, with B. Nakashima 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. 1, 2026.

Head to head

B. Nakashima leads 1–0 over 1 meeting. Full record.

Recent form

All Aug. 4, 2026 results & picks →

Full analysis

B. Nakashima (A, 25, #33) vs D. Altmaier (B, 27, #64) — ATP Montreal, 1/32-finals, HARD

═══ MODELS vs MARKET ═══
  ELO A            0.7403
  ML A             0.5869   <- lone dissenter, favours B
  model_prob A     0.7403
  Sim A (baseline) 0.7952   spw_a 0.611 / rpw 0.421 vs spw_b 0.572 / rpw 0.399
  Sim A (calibrated spw_a 0.601, clamped=false) 0.7545
  Betfair ML       A 1.30 / B 3.90  -> exchange no-vig A 0.7500
  Pinnacle         A 1.301 / B 3.77 -> no-vig A 0.7434 (fetched 10:55Z, ~fresh)
  Pinnacle move    drift_a +0.25pp — sharp money marginally TOWARD Nakashima
  Book quote       A 1.32 / B 3.35 — note book B (3.35) is much SHORTER than
                   Betfair B (3.90). The book's "value_pick A, +2.3%" is an
                   artefact of the stale book line; against the exchange it is gone.

Consensus of the three sources I trust — ELO 74.0%, Pinnacle no-vig 74.3%,
Betfair no-vig 75.0% — is remarkably tight at ~74-75% for Nakashima. The
calibrated sim lands on 75.45% with a −0.010 SPW shade (well inside clamp), so
four independent estimates converge. ML at 58.7% is the outlier and I discount
it: this is the documented break-even-ML failure mode (mid-tier ATP matchup
hedged toward a coin flip). Every qualitative read below points the OPPOSITE way
to ML's dissent, so the disagreement gets no validation.

ELO/ranking sanity: A #33 with ELO-implied 74% vs B #64 — ELO and ranking agree
on both sides, no farmed-rating flag. ELO is credible here.

═══ RECENT FORM (5 most recent each) ═══
NAKASHIMA (A) — hard, Washington ATP 500 run:
  08-01 Washington SF   L  T. Fritz (1952)        3-6 6-3 3-6   competitive loss to Elite
  07-31 Washington QF   W  A. De Minaur (1991)    7-6 6-4       GRIND over ELITE
  07-30 Washington R16  W  J. Mensik (1917)       7-6 3-6 6-4   GRIND over Elite
  07-28 Washington R32  W  T. M. Etcheverry (1780) 6-3 6-4      DOM over Strong
  07-18 UTS Championship (exhibition format — excluded from reads)
SHAPE: SHARPENING. Beat two Elite (1991, 1917) opponents back-to-back on this
surface and took a set off Fritz — over-performing his bracket three matches
running. DOM over a 1780 in the opener.
PATH: 4 hard-court matches in the last week, ATP 500 semifinalist, last match
3 days ago (08-01). Match-sharp, no rust; first Montreal match but he has been
playing on this surface continuously.

ALTMAIER (B):
  08-04 Montreal R64    W  A. Vukic (1749)        6-7(2) 7-6(7) 6-4  GRIND, ~150min
  07-22 Kitzbuhel R16   L  A. Molcan (1819)       6-3 3-6 7-5   CLAY
  07-21 Kitzbuhel R32   W  R. Collignon (1850)    6-3 6-2       DOM, CLAY
  07-15 Bastad R16      L  L. Darderi (1916)      4-6 4-6       CLAY
  07-14 Bastad R32      W  H. Gaston (1654)       6-2 7-5       CLAY
SHAPE: FADING/UNDER-performing on this surface. Needed three sets and two
tiebreaks to escape Vukic (1749 — Strong-low) in his hard-court opener. His
quality wins in the window (Collignon 1850 DOM, Medvedev, Shelton) are all on
CLAY or GRASS, not hard.
PATH: TRANSITION — clay (Kitzbuhel 07-22) straight to hard, with a 12-day gap
before Montreal. First hard-court match of the swing was the Vukic grind.

═══ FATIGUE ═══
A Nakashima (25): matches_7d 4, matches_14d 4, 12 sets, 473min, back_to_back 2,
  last match 08-01. High cumulative load, BUT no extreme spot: no match
  yesterday, not 3-in-3 into this one, 3 clear days of rest. Per the 2026-07-27
  calibration cumulative minutes alone is context, not an edge. Age 25 =
  baseline recovery.
B Altmaier (27): matches_7d 1, matches_14d 3, 435min, last match 08-04 — a
  3-SET WIN (~150min, two tiebreaks) immediately before this match. This IS an
  extreme spot and it triggers the calibrated 3-set-hangover: −4.5pp,
  surface-agnostic, age 27 = baseline, applied ONCE and not stacked.
NET: fatigue favours NAKASHIMA. Note the market-overprices-fatigue check: if
anything the crowd should be shading Nakashima for the "tired after a 500 semi"
story — but his price (1.30) is if anything SHORTER than ELO/Pinnacle imply, so
there is no fatigue discount on him to exploit. That kills the contrarian angle
rather than creating one.

═══ NARRATIVE ═══
Let-down: NONE for either — Nakashima's deepest recent round is a SEMI-final
  (lost), not a Final; no Final in either window. No title (no Won+Final).
Venue momentum: Nakashima no Montreal matches in window. Altmaier 1-0 (the
  Vukic grind).
Surface transition: ALTMAIER transitions clay -> hard (last clay 07-22).
  Depresses level and widens his margin variance. Nakashima: none, all hard.
Player notes (a): two LWOT Washington previews confirming his run (beat
  Etcheverry 6-3 6-4, then Mensik, then de Minaur QF) — corroborates the form
  read only, no injury/fitness doubt. No notes for Altmaier.

═══ MARKET CHOICE ═══
One coherent distribution: sim calibrated to spw_a 0.601 -> winA 0.7545
(matches the ELO/Pinnacle/Betfair consensus). All three markets read off THAT run.

ML A: Betfair 1.30 -> comm-adj eff 1.285 -> implied 77.82%. My honest prob 75.5%.
  Edge NEGATIVE (−2.3%). min_odds = 1.05/0.755 = 1.391; Betfair 1.30 is BELOW
  the minimum -> automatic PASS under Pricing Discipline #3. Adding Altmaier's
  −4.5pp hangover would lift A to ~78-79% for a ~1% edge — but that edge is
  narrative-created and G1 forbids it. PASS.

ML B: Betfair 3.90 -> comm-adj eff 3.755 -> implied 26.63%. ELO gives B 26.0%,
  calibrated sim 24.6%, Pinnacle 25.7% — three sources at or BELOW the price.
  Only ML (41.3%) shows an edge, and it is the model I have judged broken for
  this matchup: it is contradicted by the surface transition, the Vukic grind,
  the 3-set hangover and Nakashima's two Elite wins. My honest probability for
  Altmaier is ~23-25%, i.e. BELOW the executable 26.63%. Edge is negative at my
  own number, so there is nothing to stake — I will not price a dog off a model
  whose dissent every other piece of evidence refutes. PASS.
  AH shadow: B +3.5 @ 2.10 (pass)

AH A −3.5 @ 1.84: implied 55.62%, calibrated cover 61.90% -> edge +6.28%.
  Class = HIGH-VARIANCE (laying the favourite wider than −2.5) -> bar is 10%,
  min 1.70. 6.28% < 10% -> FAILS the numeric bar.
  Sharp cross-check: Pinnacle −3.5 (1.826/2.04) -> no-vig A covers 52.77%. My
  61.90% is 9.1pp above the sharp line — a material contradiction, which KILLS
  the bet independently of the bar.
  Margin evidence also opposes it: Nakashima's own Washington scoreline
  distribution is +3 (De Minaur), +2 (Mensik), +5 (Etcheverry) — he wins TIGHT,
  clustered right at or below −3.5, not by blowout. Altmaier's clay->hard
  transition widens variance both ways, which is a risk on a cover, not support.
  PASS.
AH B +3.5 @ 2.10: implied 48.90% vs sim 38.10% -> edge −10.8%. Also barred by
  the self-contradiction guard: I haircut Altmaier −4.5pp for the win, so I
  cannot claim he stays close enough to cover. PASS.

Total UNDER 22.0 @ 2.00: implied 51.28%, calibrated under ~56.92% -> edge +5.64%,
  which clears the standard 5% bar. But it dies on two independent checks:
  (1) Sharp cross-check — Pinnacle T22.0 (1.885/1.971) -> no-vig under 48.88%.
      Betfair and Pinnacle independently set the SAME 22.0 line at near-even
      prices; my 56.92% is 8.0pp above the sharp number. Material contradiction
      -> kill. The sim's expected total is 21.1 games (raw 23.7 with a hand-tuned
      −3.0 bias_correction) — when the sim disagrees with two markets on total
      games, the suspect number is the sim's, not the market's.
  (2) Margin-specific evidence runs the other way: Altmaier is a long-match
      grinder (avg court time 145min; 7-6(2)/7-6(7)/6-4 vs Vukic, 7-6 vs
      Medvedev, tiebreaks vs Tiafoe, a 5-setter vs FAA), and Nakashima's whole
      Washington run was 7-6 6-4 / 7-6 3-6 6-4 — two players whose recent
      scoreline distributions both point LONG. A 21-game match is not the read.
  PASS. Total OVER 1.92: implied 53.36% vs 43.08% -> edge −10.3%. PASS.

Lowest-variance expression of the only edge anyone has (A to win) would be the
ML, and the ML is priced through fair on the exchange. Nothing left to bet.

GATES
  G1 edge-origin: FAIL — no market has a 5%+ pre-narrative edge at the executable price that survives its own bar: A ML is −2.3% (below min_odds 1.391), B ML is negative at my honest 23-25%, AH −3.5 has 6.3% against a 10% wide-AH bar, Under 22.0's 5.6% is contradicted 8pp by the sharp line. The only edges that "appear" come from adding Altmaier's fatigue haircut — exactly what G1 forbids.
  G2 consensus:   [tier: capped] — ELO+sim+both market sources agree on ~74-75% A; the one dissenting model (ML 58.7%) is a lone, contaminated signal (break-even hedge, refuted by surface transition and live this-event form). Would cap 0.5-1.0u lone_signal: if anything were stakeable.
  G3 fatal-risk:  [none] — no unencodable thesis-negating risk; the margin risks (Altmaier's transition-widened variance, Nakashima's tight winning margins) were encoded numerically via the spw_a 0.601 recalibration rather than asserted as a story.

═══ VERDICT: PASS ═══
This is a textbook efficient price. ELO 74.0%, Pinnacle no-vig 74.3%, Betfair
no-vig 75.0% and the calibrated sim 75.5% all agree, and the qualitative read —
Nakashima SHARPENING with wins over 1991 and 1917 on this surface and 3 days
rest, versus Altmaier transitioning off clay, grinding past a 1749, and carrying
a −4.5pp 3-set hangover — points the SAME way the market already does. Full
model+market consensus on the winner and a sharp-contradicted margin
distribution: there is no side to take. The one apparent value signal (book B at
3.35, and value_pick A +2.3% off the 1.32 book quote) is a stale book line the
exchange contradicts in both directions. Nakashima wins this more often than
not, but not more often than 1.30 requires. No bet.