England vs. Croatia - Post-Event Review
2026 World Cup, Group L, Match 21 | June 17, 2026 | AT&T Stadium, Arlington/Dallas, TX
Predicted vs. Actual | Calibration: ~68%
TL;DR
Result: England 4-2 Croatia (HT 2-2). Winner β (England, 55% pick) but SCORELINE β β 4-2 was in NO model's top-3 (modal was ENG 2-1; g5 picked 1-1 as single most likely). This is a DOUBLE UNDERESTIMATE: the models were too cautious on BOTH the favorite's win% (55% should've been ~65-70%) AND the margin (2-1 β 4-2). The decisive, unmodeled factor was TEAM-LEVEL AGE-DECLINE: Croatia's "fully fit golden generation" got overrun β ModriΔ (40) was subbed at 58' for being ineffective ("looked his age," per Guardian), and Croatia managed just 0.71 xG. g4's "outlier" Kane #1 pick (the scorer it ran hottest on) was RIGHT β a brace β refining the g4 bias rule (wrong on age-declined stars like Ronaldo, OK on in-form elite strikers like Kane). Turpin's pen-rate-as-band worked (0.40 β 1 pen, retaken) but the result didn't hinge on it; his 0-0 yellows was an extreme leniency case beyond the 20-25% discount.
- Winner: β England β consensus ~55% (mm3 58%, g4 55%, g5 52%). The favorite won; the pick was directionally right.
- Scoreline: β 4-2 β modal was ENG 2-1 (~17%); 4-2 was in NO model's top-5. Closest: ENG 3-1 (mm3 8%, g4 6%, g5 6%). g5's "1-1 single most likely" was decisively wrong (HT was 2-2, then it opened up).
- Who Scores: β β Kane (g4 #1) β scored TWICE (12' pen retaken, 42' header). g4's "outlier" scorer #1 inflation was VALIDATED here. Bellingham (#2-3 across models) β scored 47'. Croatia #1 PeriΕ‘iΔ β (didn't score; Croatia's goals came from Baturina + Musa).
- Penalty: β Turpin awarded 1 (retaken after Gvardiol encroachment + LivakoviΔ off-line). His 0.40 pens/game career rate HELD. But it didn't decide the match (ENG won by 2). Pen-rate-as-band calibration confirmed: useful probability input, not a load-bearing variable.
- **
[...truncated 145 lines...]
the draw band for 38+ star scorers (Finding 16). EXTEND it to TEAMS built around 38+ cores: a side whose engine is ModriΔ (40) + PeriΕ‘iΔ (36) + aging CBs will be overrun in transition by a pressing tier-1 opponent after ~60'. Don't just discount the individual β discount the team's pressing output and transition defense. Croatia's 0.71 xG and ModriΔ's 58' sub are the signature. |
| 2. "Favorite double-underestimate" is a 4th Finding 2 failure mode. When a tier-1 favorite has (a) an AGGRESSIVE manager (Tuchel, not Southgate), (b) a MILESTONE-chasing elite striker (Kane β Lineker record), AND (c) an AGEING opponent β push the favorite win% HIGHER (55% β 65-70%) AND weight multi-goal margins (3-1, 4-1, 4-2) higher. The "tightest consensus of the tournament" framing was wrong here. | doc-1 Finding 2 (add a 4th failure mode) + checklist. Add: "tier-1 + aggressive mgr + milestone striker + ageing opp = push favorite % up 8-12pts, add 4+ goal margins at 15-20%." |
| 3. g4's scorer #1 inflation is context-dependent. Wrong on age-declined stars (Ronaldo 41, 0 goals). RIGHT on in-form elite strikers (Kane, brace). Refine: g4's bull-on-scorer-#1 is a bias worth discounting for 38+ age-declined stars, but RELIABLE for in-form elite finishers chasing milestones. | _core.md g4 variant notes (refine the scorer-discount guidance). |
| 4. Pen-rate-as-band is the correct calibration β validated in BOTH directions across 2 matches. Turpin 0.40 β 1 pen (ENG-CRO); Al-Jassim 0.40 β 0 pens (POR-DRC). Net: the rate is a real probability input (~30-40% for a 0.40 ref). The post-POR-DRC "demote to band" fix is confirmed correct β don't treat it as deterministic, but don't ignore it either. | doc-1 Finding 3 (confirm band calibration). No change needed. |
| 5. Extreme referee leniency is possible β 0-0 yellows from a 3.32 Y/game referee in a chaotic 6-goal match. Tournament leniency can exceed the 20-25% discount entirely. Add a "whistle-swallowed" scenario (~10%) for elite referees in high-stakes group games. | doc-1 Finding 3 (add extreme-leniency scenario). |
13. Updated Cumulative Stats (21 matches)
| Metric | Before (20) | After ENG-CRO (21) | Trend |
|---|---|---|---|
| Winner hit | ~10/20 (50%) | ~11/21 (52%) | β (England won) |
| Exact scoreline top-3 | ~8/20 (40%) | ~8/21 (38%) | β (4-2 missed) |
| Who Scores β₯1 hit | 12/20 (60%) | ~13/21 (62%) | β (Kane brace + Bellingham) |
| Who Scores #1 hit | ~7/20 (35%) | ~8/21 (38%) | β (g4's Kane #1) |
| Referee pen prediction (band) | held | β held (Turpin 1 pen) | β |
| Referee card count | ~12/20 (60%) | ~12/21 (57%) | β (0-0 miss) |
| Overall | ~70-72% | ~70% | β (winner + scorers up, scoreline + cards down) |
14. Group L after MD1 (final)
| Pos | Team | P | W | D | L | GF | GA | Pts |
|---|---|---|---|---|---|---|---|---|
| 1 | England | 1 | 1 | 0 | 0 | 4 | 2 | 3 (+2) |
| 2 | Ghana | 1 | 1 | 0 | 0 | 1 | 0 | 3 (+1) |
| 3 | Croatia | 1 | 0 | 0 | 1 | 2 | 4 | 0 (-2) |
| 4 | Panama | 1 | 0 | 0 | 1 | 0 | 1 | 0 (-1) |
England top on goal difference (+2 vs Ghana +1). Croatia's -2 GD is a real MD2 problem β they likely need a win vs Panama and help elsewhere. MD2: England vs Ghana (the headline group match), Croatia vs Panama (must-win).
15. The g4 Question β Revisited
After GHA-PAN, I noted g4's bias-correction "worked" (it wasn't an outlier). ENG-CRO adds nuance:
- g4's WIN% was the middle of the three (55%) β not an outlier on the favorite margin this time.
- g4's SCORER #1 (Kane) was RIGHT β and it was the pick g4 ran "hottest" on. The bias-correction notes (which tell g4 to discount its scorer #1) would have WRONGLY pulled g4 off Kane here.
- Lesson: the g4 bias-correction is too blunt. It correctly pulls g4 off age-declined #1s (Ronaldo) but would wrongly pull it off in-form elite strikers (Kane). The refinement: g4's scorer #1 discount should fire ONLY for 38+ age-declined stars, NOT for in-form elite finishers. This is a content fix to the g4 variant notes β propagates via the build.
Post-event review for TASK-49. Sources: sources/matches/eng-vs-cro/post-event/ (5 files: Opta/Analyst stats, Guardian report, FA official lineups, BBC/Sky MotM+quotes, HRT Croatian perspective). No betting odds used. FINAL stats only (Opta xG 2.8/0.71). Yellow-card 0-0 flagged as a possible data-quality item but corroborated across 4 sources.