World Cup 2026 Knockout Stage Results, by the Numbers: All 32 Games

This page tracked the 2026 knockout rounds as they were played; the tournament is now complete, and what follows is the final ledger. Across the 104 completed matches in the results file, that means all 32 knockout games: 16 in the round of 32, 8 in the round of 16, 4 quarter-finals, 2 semi-finals, the third-place game and the final. Here is every knockout result in one table, and the numbers underneath. The short version: scoring dropped from 2.99 goals a game in the groups to 2.91 in the knockouts — and even that gap flatters the bracket, because a ten-goal third-place game did a lot of the catching up — more than half the knockout games were decided by exactly one goal, four ties went to penalties, and the last result in the file is the one that mattered: Spain 1–0 Argentina.

Sourcing, up front. Every 2026 figure comes from the completed matches in ESPN's public scoreboard feed, final at 104 matches (refreshed July 19, 2026) and served by this site at /data/wc2026_results.json so you can check any score against the raw file. Nothing is predicted and nothing is invented — and now, nothing is provisional either: the dataset is closed. One honesty note before the table: the feed carries no round labels. The stage split is derived, not guessed — the 48-team format plays exactly 72 group games, all before any knockout tie, so sorting by kickoff and cutting at match 72 separates the stages exactly; the rounds after that follow in strict sequence (16 games, then 8, then 4, then 2, then the third-place game, then the final). The shootout caveat gets its own note below.

Every knockout result

Here are all 32 knockout games, straight from the file, in kickoff order. Dates are the feed's UTC kickoff dates.

*The feed records the post-extra-time scoreline only — no shootout scores. A knockout game that ends level must be decided on penalties; that's the format. Each advancer is named only because it appears in a later round of this same file: Paraguay and Morocco played round-of-16 ties, Egypt showed up against Argentina in the round of 16, and Switzerland — after the tournament's only goalless knockout game — appeared in the quarter-finals. All four shootout winners are therefore identifiable without guessing a single kick.

The numbers: tighter in every direction (with one loud exception)

Now the aggregate picture, computed the only honest way — add every home_score and away_score, divide by games.

Computed from /data/wc2026_results.json, all 104 matches, refreshed 2026-07-19. “Finished level” means the recorded scoreline was a draw — in the knockouts, that means penalties.

The headline number needs its asterisk stated plainly: the knockout average only got within a tenth of the groups because of one absurd evening. Strip the ten-goal third-place game — France 4–6 England, the highest-scoring match of the entire tournament — and the other 31 knockout games averaged 2.68, a third of a goal below the group stage. The competitive knockout rounds really were tighter; the exhibition round threw a party.

The margin profile is where the tightening is unmistakable. One-goal games were a quarter of the group stage (18 of 72); in the knockouts they were 17 of 32 — 53%. Wins by three or more fell from 25% of group games to 4 of 32 (13%) — three 3–0s (France over Sweden, Spain over Austria, Morocco over Canada) and Belgium's 4–1 over the United States — and nobody won a knockout game by more than three, against group-stage extremes like Germany 7–1 Curaçao and Canada 6–0 Qatar. Two details stop this reading as “the football died”: both teams scored in 18 of the 32 knockout games (56%), slightly above the group-stage rate (54%), and only one knockout game finished goalless — Switzerland–Colombia in the round of 16 — while the groups produced seven 0–0s. The knockouts weren't cagey; they were close.

2022 ran the other way

Here's why we wouldn't file “knockouts get cagey” as a law of nature. In 2022 — from the bundled StatsBomb file of all 64 matches — the knockout rounds out-scored the group stage, 3.25 goals a game to 2.50, and 5 of 16 knockout games (31%) went to penalties against 4 of 32 (13%) here — and none of 2026's four came later than the round of 16, where 2022 put shootouts in the quarter-finals and the final itself. We unpacked in the 2022 deep dive why that 3.25 is partly an extra-time artifact. The same caution cuts the other way in 2026: at least four of these 32 games ran 120 minutes, which inflates goals-per-game, and the competitive knockout average still sat a third of a goal below the groups once the bronze final is set aside. Two tournaments, two opposite patterns — which is the real lesson: one edition of a knockout bracket is a small sample, whichever direction it points.

What the file says about the stories

Spain won it the way they played it. Their knockout run reads 3–0, 1–0, 2–1, 2–0, 1–0 — nine scored, two conceded, and the only goal they allowed in the whole knockout phase before the final week was Belgium's in the quarter. The full ledger — W7 D1, 14–1, seven clean sheets — is closed out in the final wrap.

Brazil went out early. Norway — at their first World Cup since 1998 — beat them 2–1 in the round of 16, holding 66.4% possession and completing more than twice Brazil's passes (681 to 331). Norway's run ended one round later, 1–2 to England in the quarter-final; their tournament file got its own piece.

Germany's strange exit stands as written in July: a 7–1, a 2–1, a 1–2 loss to Ecuador, then 1–1 with Paraguay and out on penalties in the round of 32 — a third straight World Cup ending early, after the 2018 and 2022 group-stage exits.

Morocco's unbeaten run ended in the last eight. Six games: 1–1 with Brazil, wins over Scotland and Haiti, a shootout past the Netherlands, a 3–0 over Canada — then 0–2 to France in the quarter-final, the first defeat on their file. Ten scored, six conceded across the six.

Cape Verde's odd run. Three group draws (0–0, 2–2, 0–0), a knockout place, then a 3–2 exit that had Argentina rattled. The feed has no group labels, so we can't say whether they went through second or as a best third — only that three draws were enough.

All three hosts died in the round of 16. Canada 0–3 Morocco, Mexico 2–3 England, United States 1–4 Belgium — the same round, three straight days. The dated note on it has the full autopsy.

What this file can't tell you

• Rounds are derived, not labelled. The cut at 72 group games is exact by format, and every knockout round follows in strict sequence (16–8–4–2–1–1), with each advancing team appearing in the next round's team sheets — so the labels above are arithmetic, not judgment calls.
• No shootout scores. Level knockout games are recorded as draws; who advanced is visible only because the winner plays again. With the bracket complete, that identifies all four shootout advancers — but no kick-by-kick detail exists in this file.
• No minutes played. We can't separate 90-minute wins from extra-time wins for decided games, so the stage comparison is per-game, not per-90.
• Small samples, permanently. 32 knockout games is the entire population for this format's debut — enough to describe 2026 exactly, not enough to declare laws about 48-team World Cups.
• One source. Scores come from a single public feed (ESPN); checkable, but not triangulated.

The takeaway

Across all 104 matches, the 2026 knockouts ran tighter than the groups: 2.91 goals a game against 2.99 (2.68 without the bronze-final fireworks), a majority of games decided by exactly one goal, no knockout win by more than three, and a lower shootout rate than 2022's knockouts — four in 32, none past the round of 16. The bracket removed Brazil, Germany, the Netherlands and all three hosts, and it ended the only way Spain's tournament ever pointed: 1–0. The numbers are final, and the file behind them is public.

Reproduce it

Everything here recomputes from one file. The refresh script re-pulled ESPN's public scoreboard before each build and wrote completed matches (only) to data_layer/wc2026_results.json, served at /data/wc2026_results.json — final at 104 matches. Sort by date, cut at 72 for the stage split; goals per game is home_score + away_score over matches. charts/chart_wc2026_knockouts.py draws the figure from the same file, so chart and prose can't drift apart.

Where these numbers come from

• 2026 results: /data/wc2026_results.json (served copy of the bundled data_layer/wc2026_results.json) — ESPN public scoreboard feed (soccer/fifa.world), all 104 matches, refreshed 2026-07-19. Charted by charts/chart_wc2026_knockouts.py.
• 2022 comparison: bundled data_layer/wc2022_matches.json — StatsBomb Open Data (free public dataset; attribution required), all 64 matches.
• The full-tournament picture: The 2026 World Cup, by the numbers — goals, draws and blowouts across all 104 games.
• How it ended: Spain 1–0 Argentina: the final, in data.
• Context: Inside the 2022 World Cup: what all 64 results say — including why 2022's knockout scoring number is partly an extra-time artifact.
• Format background: the best-third-place qualification math behind the 48-team, 12-group, round-of-32 design.

The knockout autopsy: what predicted the winners

This section was first published on 24 July 2026 as a separate article.

Now that the tournament is over and the dataset is final, we can ask the retrospective question properly: across the 28 decided knockout games of the 2026 World Cup, what actually predicted the winner? Twelve indicators, one league table, and a result worth the whole exercise: a team’s pre-match possession average called 82% of decided knockouts — nearly matching shots on target measured in the game itself — while entering goals per game, the number every casual prediction leans on, finished dead last at 64%. The tournament didn’t reward form. It rewarded identity.

Sourcing. Every number below is computed from our bundled dataset — all 104 matches of the completed tournament, from ESPN’s public scoreboard and match-summary feeds, served raw at /data/wc2026_results.json. “Entering” indicators are each team’s tournament-to-date per-game averages before kickoff of that knockout game; “in-game” indicators are the box score of the game itself. Ties on an indicator are excluded per comparison, so denominators vary and are printed on every bar. Chart script: charts/chart_wc2026_predictors.py.

The league table of predictors

Read the top of that chart carefully, because it’s stranger than it looks. That shots on target led the table is no surprise — this series tracked that pattern all tournament (the familiar 24-of-28 ledger counted teams that matched or beat their opponent on target; today’s 22-of-26 is the strict higher-count version, with two ties set aside — same pattern, tighter definition). But shots on target is measured during the game. It predicts nothing; it describes. The remarkable line is the one just below it: entering possession average, 23 of 28 (82%) — information fully available before kickoff, outperforming every other pre-match indicator and essentially tying the best in-game one.

The possession finding, stress-tested

The five misses are instructive. Three belong to one team: Norway, whose counterpunching run beat higher-possession Ivory Coast and then higher-possession Brazil — the tournament’s designated exception, profiled in its own note. One was Mexico over Ecuador in the round of 32. And the fifth was the bronze final, where England’s 54.1% entering average beat France’s 57.1% in the tournament’s least consequential fixture. Which means: from the quarter-finals onward, the higher-possession-entering team won six of the seven games that produced a winner — and the seventh was the third-place game. The final itself was a hold: Spain entered at 63.8% to Argentina’s 60.7%, and Spain are champions.

Why would a style statistic outpredict every results statistic? The honest answer has two layers. The kind one: possession average is the most stable number in this feed — goals arrive in single digits and lie about small samples, but a team’s share of the ball barely moves game to game, so seven matches of it measures something real about how a team imposes itself. The skeptical one: possession is also a proxy for plain quality — good teams hold the ball and win, and nothing in a who-led-won count separates style from strength. Both layers are true at once. What the table actually establishes is narrower and still useful: among the numbers this feed carries, the stable identity measures aged well, and the volatile scoring measures aged terribly.

How badly goals form aged

Entering goals per game called 16 of 25 — 64%, last place, barely clear of the coin flip line. And the misses aren’t noise at the fringes; they run straight through the podium. Spain, the eventual champion, was the goals-form underdog in two of its own knockout wins — entering the Austria game scoring 1.67 a match to Austria’s 2.00, and entering the Belgium quarter-final at 1.80 to Belgium’s 2.60. Morocco beat higher-scoring Canada; Belgium beat higher-scoring Senegal and the United States. Nine times in twenty-five, the team that had been scoring less advanced anyway. A month of “team X is averaging three a game” television graphics predicted the bracket worse than a possession column nobody puts on screen.

The reason is the one this series kept meeting all tournament: goals are rare events. Seven games of goals is twelve to twenty observations wearing a per-game costume, and group-stage schedules inflate them unevenly — a soft group buys a gaudy scoring average that says little about a quarter-final. Possession comes in continuous percentages every match; goals come in lumps. When you rank indicators by sample stability, you have roughly ranked this chart.

The caveats, stated at full volume

Twenty-eight games is a small book to bet a theory on, and this piece tested twelve indicators against it — with that many comparisons, the spread between 64% and 82% contains real information and real luck, and one different result moves any bar by about four points. The entering averages blend group-stage opposition of wildly different quality. The who-led-won method ignores margins entirely (leading possession by 0.2 points counts the same as by 20). And nothing here is causal — nobody should hoard possession because of this table. What the table earns is a prior for 2030: when you fill in your next bracket, weight who teams are over what they just scored. That prior would have gone 23-for-28 this summer.

This closes the live chapter of the World Cup series and opens the retrospective one — the complete dataset supports slower questions now, and the next ones on the list are how the 48-team format actually redistributed goals, and whether the group stage told us who would survive the bracket. The hub keeps collecting all of it.

Sources

• Match results and team stats: ESPN public scoreboard + match-summary APIs, parsed to data_layer/wc2026_results.json (all 104 matches, tournament complete); served at /data/wc2026_results.json. All rates computed by charts/chart_wc2026_predictors.py; ties on an indicator excluded per comparison.
• The in-tournament tracking this retrospective formalizes: the final wrap (the 24-of-28 matched-or-beat ledger) and the early what-wins-games note from the group stage.
• The counterexample’s file: Norway’s run, quantified.
• The World Cup hub collects the complete series.