Brazil vs. Japan - Post-Event Review

2026 World Cup, Round of 32 (Match #76) | Monday June 29, 2026 | NRG Stadium, Houston (roof closed)

KNOCKOUT | Predicted vs. Actual | Calibration: ~73% (top of C band, borderline B)

First match with the new WDM anchor (§6.6) + §7K knockout decision tree. Both innovations performed well: the exact modal scoreline hit and the advance path was correct. Assessment in §§7, 9, 12.

TL;DR

Result: BRA 2-1 JPN (HT 0-1; Sano 29', Casemiro 56' header, Martinelli 90+6'). Brazil trailed to a Casemiro error, equalised via a set-piece, then won it deep in stoppage through an Ancelotti sub-masterstroke — Martinelli from left-midfield — after Tanaka's turnover.

• Winner: ✅✅ Brazil (consensus 50% at 90' / 68% advance). Strict modal favorite in all 3 models; won in regulation (no ET), so the advance band resolved at 100%.
• Scoreline: ✅✅ EXACT modal — consensus modal was BRA 2-1 (14%); actual 2-1. One of the cleanest scoreline hits of the tournament.
• Who Scores: ❌ miss — both #1 picks blanked: Vinícius Jr (~33%, hit the post) and Ueda (~25%, 1 SoT saved). All top-3 on both sides missed; the winner came from Martinelli (rank 5, ~7% sub).
• Finding 11: ✅ — NRG roof closed → weather fully neutralized (predicted).
• Finding 22: ✅ — JPN injury crisis framed as −5pt (limited, not gutting); JPN competitive but capped (no Kubo cameo). Correct.
• Finding 21/bench-differential: ✅ — loaded BRA bench decided it (Martinelli). Mechanism fired (timing later than the 65–75' band).
• Finding 12 (Ancelotti-conservative-KO draw-tilt): ❌ — Ancelotti attacked throughout (4 subs, 2 attacking incl. the 65' winner-sub); the predicted "manage a lead quietly" tilt never applied because BRA were chasing.
• Watch (learning): In knockouts, a 5x-UCL manager's "conservative KO" reputation collapses the moment his team falls behind — conserving a lead is impossible when there is no lead. The conservation-draw-tilt is a conditional-on-leading mechanism, not a blanket.

Calibration Breakdown: Winner 25/25 · Scoreline 25/25 · Who Scores 0/20 · Findings 15/20 · Model-breaking 8/10 · deductions 0 = 73/100 → ~73% (C band, top edge / borderline B)

1. Winner Prediction — ✅✅ HIT

Consensus 90' modal = BRA 50% (range 48–53% across mm3/g4/g5, 5pt honest spread). Brazil won in regulation. The favorite not only won but did so inside 90', collapsing the ET/pens branch entirely. Advance probability (68% consensus) resolved at 100%. Direction was never in doubt; the only question was margin/timing, and the model was correctly bullish.

2. Exact Scoreline — ✅✅ EXACT MODAL HIT

Consensus modal scoreline BRA 2-1 (14%); g4 and g5 both carried 2-1 as modal, mm3 carried 2-1-in-90' as a co-modal path. Actual BRA 2-1. Exact hit. The two-goal BRA structure (one set-piece/header, one open play) and the single JPN counter-goal matched the modal shape almost perfectly. The only miss within the shape: WHO scored (see §3).

3. Who Scores — ❌ MISS (0/3 both sides)

The #1 pick on each side blanked; every top-3 pick missed on both teams. The match was decided by a rank-5 sub (Martinelli) — correctly identified as a bench threat but weighted at ~7%. No partial credit: Vini Jr did not assist (he hit the post), so even the assist-only band does not apply. This is the single biggest calibration drag on the score (0/20).

The scout's Finding-19-style note: the scorers came from the "secondary/bench" tier (Casemiro = rank-4 CB-header mechanism, Martinelli = rank-5 sub). The set-piece CB mechanism (Gabriel cross → Casemiro) was the kind of goal the Who Scores models under-weighted.

4. Weather — ✅ CONFIRMED (F11 neutralized)

• Forecast: roof closed, climate-controlled indoor, F11 heat penalty neutralized.
• Actual: NRG roof closed, ~21–22°C indoor (stated 35°C Houston outdoor is moot). Zero weather impact.
• Verdict: Finding 11 correctly excluded. No pressing degradation, no cooling-break inflation. Validated identically to POR-DRC (Jun 17, same venue).

5. Possession — ✅ DIRECTION CORRECT

The "BRA dominant possession" path was more accurate than the "JPN low block holds for 90'" path. 69% possession exceeded the upper estimate; BRA generated 35 touches in the opposition box. JPN's 0.23 xG → 1 goal was an overperformance (+0.77) driven entirely by Casemiro's defensive error on Sano's opener.

6. Match Stats (FINAL)

7. Key Learnings

1. REFINE Finding 12 (conservation) — conservation is conditional-on-leading, not blanket. Pre-match, g5/mm3 applied an "Ancelotti conservative KO lead-management → quiet draw-tilt" that assumed BRA would manage a 1-0/2-1 lead tightly. In reality BRA trailed (Sano 29') and Ancelotti went full-throttle (Endrick at HT, Martinelli at 65'). The draw-tilt mechanism requires a lead to protect; with no lead, a 5x-UCL manager attacks. Proposed refinement: weight the conservation-draw-tilt at 0% when the favorite is the trailing team at HT — it only fires when the favorite scores first.
2. VALIDATE Finding 22 (tournament depth absorbs injuries) — again. JPN were missing Kubo/Mitoma/Minamino/Endo and still took a lead and held until 90+6'. The −5pt framing (limited, not gutting) was correct: JPN's floor stayed high, but their ceiling was capped (no Kubo cameo, only 0.23 xG). Same shape as the JPN-injury pattern noted in NED-JPN. Finding 22 holds for Tier-1 vs Tier-2 knockout contexts.
3. PROPOSE new nuance to Who Scores: weight the set-piece CB header mechanism higher vs low blocks. Casemiro's equalizer (Gabriel Magalhães back-post cross) is the classic "CB-from-deep-on-set-piece" goal (Finding-family: CB set-piece mechanism). vs a 3-4-2-1 low block that packs the box, the secondary set-piece threat is a disproportionate scoring source. The Who Scores models had Casemiro/rank-4 implicitly; this match says: vs low blocks, explicitly promote the aerial CB to the top-5.
4. VALIDATE Finding 21/bench-differential — the loaded-bench-decides mechanism, with a timing caveat. BRA's bench (Endrick, Martinelli, Neymar, Éderson, Fabinho) was the projected differentiator and it decided the match (Martinelli 90+6'). Timing nuance: the predicted 65–75' explosion window was too early — vs a hot GK (Suzuki, 4 saves incl. the Vini Jr denial) the resolution pushed to 90+6'. Refinement: add +15-20' to the bench-explosion window when the underdog GK is in hot-GK form (×0.25 calibrated).
5. WDM anchor (§6.6) debut — performed well as a SHARED STARTING POINT, not a precision instrument. WDM gave BRA 43%; consensus lifted to 50%; actual = BRA win (effectively ~100% advance). The anchor under-estimated BRA by ~7pt on the 90' win, but all 3 writers independently adjusted up with the same rationale (JPN injury, KO need-to-win), and the 5pt spread is textbook TASK-99 honest uncertainty. Verdict: the anchor's value is reproducibility + shared baseline (it killed the old 12pt inter-writer "naive gap" drift), not point-accuracy. No pathology. Pedigree-exclusion (weight 0.0) continues to look right.
6. §7K knockout decision tree — passing grade on first live test. The tree's modal path (BRA 2-1 in 90') was the actual result; the ET/pens branches correctly did not fire (no ET). The main over-weight was the 25% draw-at-90' allocation (actual 0%). This is the same draw-epidemic over-tilt seen in the group stage; in knockouts with a clear Tier-1 favorite + non-veteran-GK block, the draw band should be ~15-18%, not 25%. Calibrate §7K draw weight downward for clear-favorite knockouts.

8. Bracket Context (R32 → R16)

• Brazil ADVANCE to the Round of 16. Opponent/fixture (TBC pending remaining R32 results — do not fabricate).
• Japan ELIMINATED — WC 2026 ends at the Round of 32. Moriyasu's depleted side (Kubo/Mitoma/Minamino/Endo absent) took a lead and held until the 96th minute; the injury crisis capped the upset ceiling exactly as Finding 22 predicted.
• Casemiro injury flag (groin, off 90+1') — minor knock to monitor for the R16. Not in the calibration but a forward-looking data point.

⚠️ R16 opponent is not yet confirmed in the brief. Marked TBC — no fabrication.

9. Knockout Decision Tree Validation (§7K)

The §7K tree's modal path (BRA 2-1 in 90') was the exact outcome. The ET/pens sub-tree was never stress-tested (BRA won in regulation), which is itself a correct outcome — the tree did not over-allocate to chaos. The only miscalibration: the 25% draw-at-90' was ~10pt too high for a clear-favorite knockout with a non-veteran-GK underdog block. Recommend §7K draw-band = 15-18% in this profile.

10. Pre-Match Calibration Table

11. Referee, GK & Lineup Validation

• Referee (Mariani): predicted lenient 2-4Y, 0R, 15-25% pen band. Actual 5Y (slightly above range), 0R ✓, 0 pens ✓ (within band). Slightly stricter than predicted — JPN's tournament discipline (2Y in 3 MD matches) fractured under pressure (3Y incl. late Junnosuke Suzuki foul). VAR (Di Bello) active, correctly waved off a 65' handball. No material error.
• GK: Both confirmed starters correct. Alisson (7.4) 1 save, no error (the goal was a Casemiro defensive mistake, not GK). Suzuki (7.2) 4 saves incl. the crucial Vini Jr denial at 58' — the hot-GK scenario went briefly live but was broken at 90+6'. No GK change; no model-breaking input error.
• Lineup: BRA XI predicted 11/11; JPN XI correct (3-4-2-1, Itakura out as predicted, Taniguchi started). Surprises: Kubo benched but UNUSED (cameo prediction failed); Neymar benched and UNUSED (super-sub prediction failed). Both managers kept their marquee bench options on the bench.

12. Finding Hit/Miss Summary

Honest flag (no −5 deduction applied): the exact modal scoreline hit was driven by a Casemiro defensive error (the JPN opener) and a sub winner — i.e. the right answer via a partly different mechanism than the modal-path narrative (which imagined Vini Jr + Cunha scoring, Ueda pulling one back). The reasoning "BRA dominate possession + win 2-1" held; the scorer-level reasoning did not. Flagged explicitly here, so no misleading deduction.

13. TOML Calibration Block

[match]
code = "bra-vs-jpn"
date = "2026-06-29"
stage = "R32"
match_no = 76
venue = "houston-nrg"
referee = "mariani"

[actual_result]
team_a = "BRA"
team_b = "JPN"
score_a = 2
score_b = 1
outcome = "win"           # from BRA (team_a) perspective
ht_score = "0-1"
scorer_a = ["Casemiro 56'", "Martinelli 90+6'"]
scorer_b = ["Sano 29'"]
extra_time = false
penalties = false

[win_probability]         # consensus pre-match, team_a = BRA perspective
favorite = "Brazil"
win = 50                  # BRA win in 90' (ACTUAL)
draw = 25                 # level @ 90' → ET/PK
loss = 25                 # JPN win in 90'
advance = 68              # BRA advance (actual 100%)

[scorelines]              # consensus pre-match, top band (BRA-JPN format)
"2-1" = 14                # BRA 2-1 ← ACTUAL (modal, EXACT HIT)
"1-1" = 13                # → ET/PK
"2-0" = 10
"3-1" = 9
"1-0" = 8

[who_scores]
team_a_1 = { player = "Vinicius Jr", probability = 33 }   # unanimous #1 — MISSED (hit post 58')
team_b_1 = { player = "Ueda", probability = 25 }          # unanimous #1 — MISSED (1 SoT saved 63')
winner_scorer = { player = "Martinelli", rank = 5, probability = 7 }  # sub winner — in-list but rank-5

[calibration]
winner_correct = true
scoreline_in_top3 = true
scoreline_exact_modal = true         # modal BRA 2-1 == actual
scorer_1_hit = false                 # both #1 picks blanked; all top-3 missed both sides
scorer_was_secondary = false         # winner = rank-5 sub, not in top-3
hot_gk_activated = true              # Suzuki 4 saves incl. Vini Jr denial
hot_gk_forced_et = false             # broken at 90+6'
finding_12_conservation = 0          # 0% both teams (energy)
finding_12_draw_tilt_missed = true   # Ancelotti-conservative-KO tilt INVALIDATED (trailing → attacked)
finding_4a_triggered = false
finding_11_neutralized = true        # NRG roof closed
finding_22_injury_limited = true     # JPN -5pt framing correct
finding_21_bench_decided = true      # Martinelli sub winner (timing 90+6', late)
wdm_anchor_first_use = true
wdm_anchor_bra_pct = 43              # consensus lifted to 50%
rubric_winner = 25
rubric_scoreline = 25
rubric_who_scores = 0
rubric_findings = 15
rubric_model_breaking = 8
rubric_deductions = 0
rubric_total = 73
rubric_band = "C"                     # 60-74 (top edge, borderline B)
calibration_pct = 73

[meta]
first_wdm_anchor_match = true
first_s7k_knockout_match = true
model_spread_90 = 5                  # mm3 49 / g4 53 / g5 48 — honest (TASK-99)

Pre-Match vs Post-Match Spread Accuracy

5pt pre-match spread (TASK-99 honest-uncertainty band) → actual BRA win in 90'. g4 was closest on the winner number (53%); all three nailed the modal scoreline (2-1). The ensemble's directional + scoreline convergence was the cleanest of the knockout rounds so far; the miss was purely at the scorer level.

Post-event for Round of 32 Match #76. Category 0 PASSED: BRA 2-1 JPN confirmed from ≥5 independent sources (BBC Sport, Guardian live blog + match report, FotMob/Opta, ESPN, Sky Sports, Law5-TheRef). FINAL stats only (Opta via FotMob + BBC). No ET, no penalties. BRA advance to R16 (opponent TBC); JPN eliminated. Sources: sources/matches/bra-vs-jpn/post-event/ (4 archived files). First live use of WDM anchor (§6.6) + §7K knockout decision tree — both assessed in §§7, 9, 12.