Five days before the tournament, this site published a dark-horse checklist: five traits that supposedly mark a team about to outrun expectations, illustrated with numbers we explicitly labelled invented. The tournament is over, the 104-match dataset is final, and a checklist published before the fact deserves a grade after it. So: six teams beat their group-stage seed in the knockouts (Belgium, Morocco, Norway, Egypt, Canada and Paraguay). Of the checklist's five boxes, two cannot be tested with our data, one doesn't map to anything measurable, and the two that can be scored, the elite keeper and the kind draw, both failed. Only one of the six overperformers ranked in the field's top third on shot-stopping, and five of their nine knockout wins came against top-half seeds, not around them.

Sourcing. Every number here is computed from the bundled tournament dataset: all 104 completed matches with per-team box scores from ESPN's public scoreboard and summary feeds, served raw at /data/wc2026_results.json. Following the site convention, the first 72 matches are the group stage and the last 32 the knockouts; knockout draws are shootouts by construction, a shootout win counts as a knockout win, and the third-place game is excluded. The chart and every transcribed figure come from our chart script.

Two-panel chart. Left: scatter of all 32 knockout teams' group-stage seed against knockout wins, with a dashed staircase showing chalk expectation; Spain (seed 6, five wins), England, and the six dark horses (Morocco, Norway, Belgium, Egypt, Canada, Paraguay) sit above the stairs. Right: save-rate proxy against wins-minus-expectation; the six dark horses scatter at or below the field-median save rate while high-save-rate Colombia, Portugal and Switzerland sit at exactly zero overperformance.
Left: the 32 knockout teams, seeded by group record; above the dashed staircase means a deeper run than the seed implied. Right: the checklist's keeper claim. If elite shot-stopping marked overperformers, the green dots would crowd the right side. They don't; the best non-Spain save rates belong to teams that exited exactly on seed.

What the June piece actually claimed

Restated as predictions, the checklist said overperformers would tend to show: underlying performance (xG and xGA) ahead of results and reputation; a goalkeeper capable of stealing ties; a real set-piece edge; a kind side of the draw; and squad depth for the heat and schedule. It named no teams and flagged its one worked profile as invented in every detail, which is what makes an audit possible: the claims sit on the record, dated June 6.

The checklist rested on a definition: a dark horse is not a bad team that gets lucky but a team the consensus has under-rated, one whose true quality is higher than its public billing or even its power rating suggests. Each box came with a reason:

  • Underlying quality the rating hasn't caught. A side whose chance creation and prevention, by xG and xGA, ran better than its results or reputation, the kind of team results-only rating systems price too low.
  • A goalkeeper capable of stealing ties. Knockout football is low-scoring and variance-heavy, so a keeper a level above expectation can carry a team further than its outfield quality alone would, especially once matches reach shoot-outs.
  • A set-piece edge. Dead balls neutralise some of the gap to stronger opponents, and set-piece quality was argued to be more stable than open-play finishing.
  • A kind draw. The same quality goes much further from a soft bracket than a brutal one; since the draw is measurable variance, the path was treated as part of the profile, not an excuse.
  • Depth for the conditions. An overperforming run is by definition longer than expected, so more matches in heat and on short rest; depth lets a side survive a suspension or a third game in nine days.

Its worked example imagined a side rated in the fifteenth-to-twentieth range of the field whose underlying numbers looked more like a top-ten team, with a goalkeeper among the best at the tournament and a drilled set-piece routine, handed a group it should escape and a first knockout tie against another non-contender: every detail invented. And it set its own limit: even a team that ticks every box is more likely than not to go out early, so the checklist raises a side's probability of overperforming from small to less-small rather than making any single run likely.

The ledger: who actually beat their seed

"Dark horse" needs a definition that lives entirely inside the data. Ours: rank the 32 knockout qualifiers by group-stage record (points, then goal difference, then goals scored) and call that the seed. Chalk expectation is the staircase in the chart: seeds 1–2 reach the final, 3–4 the semis, 5–8 the quarters, 9–16 the round of 16, 17–32 go out immediately. An overperformer beat that expectation; a dark horse is an overperformer from outside the consensus-favourite set our June ratings piece put on record: Brazil, France, Argentina, Germany, Spain, England (how models rate the field). Eight teams finished above the stairs. Two were favourites: Spain, seed 6 by group record, won the whole thing (+3 rounds), and England reached the semis from seed 8 (+1). The other six are the ledger:

TeamSeed (group record)Runvs seedSignature result
Belgium18 (5 pts, +4)QF, lost 2–1 to Spain+24–1 over #12 United States
Morocco9 (7 pts, +3)QF, lost to France+1shoot-out over #4 Netherlands
Norway14 (6 pts, +1)QF, lost to England+12–1 over #5 Brazil
Egypt19 (5 pts, +2)R16, lost 3–2 to Argentina+1shoot-out over #26 Australia
Canada20 (4 pts, +5)R16, lost 0–3 to Morocco+11–0 over #28 South Africa
Paraguay30 (4 pts, −2)R16, lost 0–1 to France+1shoot-out over #11 Germany

Two names the narrative wanted on this list are honestly not on it. Switzerland's quarter-final was not an overperformance: they were seed 7, and seeds 5–8 are supposed to make the quarters. And Colombia went home unbeaten (two group wins and a draw, 1–0 over Ghana, a 0–0 with Switzerland lost on kicks) yet never beat their seed: 10th, round-of-16 exit, exactly on the stairs. Paraguay's shoot-out win over Germany, a 19-place seed gap, was the largest upset of the tournament by this measure; Norway over Brazil (+9) was second. Both runs have their own files: Norway's numbers and the shootout retrospective.

Scoring the checklist, box by box

Box 1 (xG/xGA ahead of results): untestable. Our feed carries box scores, not expected goals, and the claim is about the pre-tournament cycle besides. We won't improvise a substitute and call it xG: this box needs data we don't bundle.

Box 2 (the elite keeper): testable, and it failed. Our proxy is tournament-long save rate: one minus goals allowed over shots on target faced, across all matches (shoot-out kicks excluded). The six dark horses rank 7th (Paraguay, 77.8%), 17th, 18th, 20th, 21st and joint-bottom (Canada, 50.0%) among the 32 knockout teams: one of six in the top third, four of six at or below the field median of 70.7%. Norway's +1 run happened on a 64.5% save rate. Meanwhile the best save rates in the field behind champion Spain (90.0%) belonged to Portugal (86.4%) and Colombia (85.7%, one goal allowed on seven on-target shots in five games), who exited precisely on seed. The proxy is crude (it blends keeper skill, defensive shot quality and opponent finishing, on small denominators), but the claim was directional, and the direction did not show up.

Box 3 (the set-piece edge): unmappable. The feed logs corners won, not set-piece goals, and corner volume is not dead-ball prowess: we showed during the group stage that out-cornering the opponent was a coin flip. For the record, the six dark horses span the entire corner spectrum, Paraguay's 2.0 per game (joint-lowest in the knockout field) to Canada's 10.0 (highest): no common signature, and no honest way to score the actual claim.

Box 4 (the kind draw): mostly failed. Directly measurable: the seeds of the teams each dark horse knocked out. Five of their nine knockout wins came against top-16 seeds: Morocco through #4 Netherlands, Norway through #5 Brazil and #13 Ivory Coast, Paraguay through #11 Germany, Belgium through #12 United States. Only Belgium's opener (#31 Senegal), Egypt (#26 Australia) and Canada (#28 South Africa) fit the soft-path story. The 2026 overperformers did not slip around contenders; most of them went through the front door.

Box 5 (depth for heat and schedule): untestable. No lineups, no rotation, no venue weather in this feed. Unscored.

The verdict, and what the checklist should say now

Final scorecard: zero boxes validated, two failed, one unmappable, two untestable. What survived is the June piece's definition (a dark horse as a mispricing rather than a lucky weak team) and its refusal to name one. Belgium is the definition wearing a shirt: a squad nobody would call weak, priced to seed 18 by a bad group, then two rounds better than the bracket's implied expectation. The constructive lesson: the honest anchor was the group table itself; nine-point teams went 3-for-3, four-point teams 2-for-11 (did the groups predict the bracket); and dark horses are the residuals around it, residuals that in 2026 shared no box-score profile at all. Paraguay overran their seed with 29.7% possession, the lowest in the knockout field; Morocco did it with 59.0%. Which pre-match numbers did sort knockout winners from losers (possession identity over goals form) is already settled in the knockout autopsy; this piece defers to it.

The caveats, at full volume

This is one tournament and six teams: an audit, not a refutation; a checklist can lose a single World Cup and still be right on average. The definition drives the result: seeding by group record measures in-tournament expectation, and a different anchor (FIFA ranking, a pre-tournament rating model) would redraw the list. It would never call sixth-seeded Spain an overperformer, and might not call 2022 semi-finalist Morocco "dark" at all. Our proxies are box-score shadows of the checklist's real claims. And a five-box checklist scored against six teams is exactly the small-sample regime we warn about. The June piece now carries a dated pointer here: a prediction made in public, and its grade.

Sources

Published by SoccerAnalytics from public data and the sources named in this piece. How this site is made.