Eredivisie Matchweek 3: What We Learned — and What Changes Next

Matchweek 3 offered a useful warning: a correct result and a correct analysis are not the same thing. Sometimes the expected winner wins for a different reason. Sometimes the right question is asked but tested with the wrong evidence. And sometimes a match changes so dramatically that the original analysis is barely tested at all.

PSV won 5–1. Feyenoord scored five away from home. PEC Zwolle won 2–0 at Heerenveen. Go Ahead Eagles beat ADO Den Haag 3–1. FC Utrecht led three times at Sparta and still drew 3–3.

After reviewing all six Matchweek 3 fixtures FairEleven analysed, some early ideas have become materially stronger. Others need to be downgraded. Several of the most useful lessons came from matches where the result looked much simpler than the football underneath it.

PSV's most important number was 19, not five

PSV's 5–1 win over FC Groningen was emphatic, but the number worth carrying into the next analysis is 19: their shots from inside the penalty area.

The first two league matches produced 16 and 16.

Three matches remain a small sample, but the repetition against three different opponents matters. PSV are consistently generating attempts from dangerous areas, and that is now better supported than anything the 5–1 scoreline tells us about their finishing.

The five goals came from 3.87 xG, with an own goal and a penalty included. Groningen, despite losing by four, created four big chances and missed three. Neither "PSV are now a five-goal attack" nor "Groningen are defensively poor" follows from that match.

The repeatable signal is PSV's ability to reach the box and generate attempts from there.

The next question has also changed. Joey Veerman has left for Borussia Dortmund, while Lutsharel Geertruida has joined PSV on loan from RB Leipzig. PSV still created heavily against Groningen with a new midfield combination, but one performance cannot establish that Veerman's creative role has already been replaced. Geertruida's first competitive PSV appearance was only a short late cameo with the game already won.

The evidence for PSV's attacking volume is getting stronger. How that volume is created without Veerman remains open.

Related analysis: PSV v FC Groningen

Right result, wrong reason

PSV–Groningen also produced a useful correction.

FairEleven expected PSV to win and expected an open match with goals at both ends. Both calls landed. But the analysis had prioritised PSV's central three-against-two as the route most likely to create their clearest chances.

PSV did control central territory. The goals, however, came largely through other routes: wide delivery, a penalty, an own goal and attacks that bypassed the central overload entirely.

Ricardo Pepi, flagged beforehand as a likely beneficiary, scored.

From the penalty spot.

The name was right. The reason was not.

Go Ahead Eagles–ADO Den Haag gave FairEleven an even cleaner example.

FairEleven slightly favoured Go Ahead before kick-off, and they won 3–1. Yet the top-ranked reason behind that view was directly refuted by a measure chosen before the match.

ADO had conceded an extreme share of their opponents' shots from inside their own box in the first two matches. Goalkeeper Kilian Nikièma's workload was therefore nominated as a visible test of whether the same problem returned against Go Ahead.

He made two saves – nowhere near the 10 that had produced the earlier signal.

The share of shots ADO conceded from inside their own box fell from 83 and 82 per cent in their first two matches to 55 per cent. Go Ahead themselves produced only 11 shots after a previous 20-shot home performance.

And they still won.

Go Ahead created three big chances to ADO's one. They also recorded 1.71 xGoT – expected goals on target, which measures the quality of shots that actually hit the target – from 0.99 xG.

That helps describe how the match was won, but it should not become a new finishing claim: one game's xGoT can be noisy too.

The important point is simpler. The quantity-based explanation in the preview did not survive the match.

The result was right. The reason was not.

Related analysis: Go Ahead Eagles v ADO Den Haag

Cambuur–Feyenoord showed what a strong pre-match read looks like

Not every lesson was a correction.

SC Cambuur–Feyenoord contained several of Matchweek 3's strongest pre-match calls. FairEleven expected Feyenoord's control to be front-loaded; they led 4–0 at half-time.

Rafik El Arguioui had been identified as Cambuur's most proven attacking threat. He scored both of their goals. The analysis had also highlighted a later phase of the match as the period in which Cambuur were most likely to produce, and those goals arrived in the 70th and 80th minutes.

The scenarios worked well too. Once Feyenoord moved two goals clear, those that needed an open game were dropped, as agreed before kick-off, rather than being rescued afterwards.

But even this strong analysis contained a useful mistake.

FairEleven had assumed that if Feyenoord scored early and kept control, Cambuur's late attacking phase would never come into play.

Feyenoord did exactly that. Cambuur still scored twice.

The error was assuming Cambuur had to still be in the game.

Cambuur did not need to be competitive. They needed to be chasing.

That distinction is small in language and significant in analysis.

Related analysis: SC Cambuur v Feyenoord

Heerenveen–PEC: when the test measures the wrong thing

FairEleven slightly favoured sc Heerenveen against PEC Zwolle before kick-off.

PEC won 2–0.

The wrong call matters, but the more valuable lesson lies in why one of the supporting ideas survived all of its own checks and still failed.

Heerenveen were expected to benefit if they finally established territorial control against a PEC side that had conceded 26 and 27 shots in its first two matches.

They established the territory: 64 per cent possession, 616 passes, 114 attacks and 18 shots.

They still produced only 1.15 xG, compared with PEC's 1.79. Ten of Heerenveen's 18 attempts came from outside the penalty area.

FairEleven had set clear conditions that should have challenged the preferred scenario. None of them were met.

The scenario still failed.

The problem was in the test itself. The underlying idea depended on shot quality, while the checks watched mainly shot count and possession.

The lesson is straightforward:

The test has to measure the variable the underlying idea actually depends on.

The match also changed how PEC should be read. Their 26 and 27 shots conceded had come against Ajax and FC Twente. Against Heerenveen, the number fell to 18 and PEC kept a clean sheet.

Some of the earlier signal belonged to the opponents rather than to PEC's defence.

PEC also took the lead and held it, providing the first counter-example from a competitive match to the early observation, from their opening two games, that going ahead had not brought them control.

One counter-example does not establish the opposite pattern. It does weaken the old one.

Related analysis: sc Heerenveen v PEC Zwolle

Territory and chance quality kept diverging

Heerenveen–PEC was not an isolated case.

Sparta had 56 per cent possession against FC Utrecht, more dangerous attacks and more key passes. Utrecht nevertheless produced the higher xG: 2.13 to 1.41.

In fact, Utrecht's best attacking output of the season came in the match in which they had their lowest possession share: 44 per cent.

ADO offered another version. They had 53 per cent of the ball against Go Ahead and took 16 shots, but nine came from outside the penalty area. The result was only 0.84 xG and 0.45 xGoT.

Across several Matchweek 3 games, one distinction became increasingly important:

Territory tells us where a match is being played. It does not automatically tell us who is creating the better chances.

Possession, attacks and territorial pressure remain useful descriptions of a game. They should not become substitutes for chance quality.

Utrecht surrendered three leads — and that still is not a pattern

FC Utrecht led Sparta three times.

They were pegged back three times, the last equaliser arriving in the 96th minute.

It is tempting to turn that immediately into a story about Utrecht having a problem closing out games. The evidence does not support that yet.

Utrecht made five substitutions between the 62nd and 77th minutes, with two apparently forced by injury according to the match data. One game cannot separate those circumstances from a lasting problem seeing games out.

The accurate conclusion is narrower:

Utrecht surrendered three leads against Sparta.

That happened. Whether it describes the team remains open.

The same match also showed why restraint matters. Utrecht had conceded shortly after half-time in each of their first two league matches. FairEleven explicitly declined to turn those two observations into a pattern.

Against Sparta, Utrecht scored in the 46th minute.

Had two instances already been promoted into a structural claim, the next match would have contradicted it.

Sometimes one of the best analytical calls is the one you do not make.

Related analysis: Sparta Rotterdam v FC Utrecht

Fortuna–AZ: the biggest xG number, and one of the least useful

Fortuna Sittard–AZ belongs in a different category.

Ole Romeny was sent off after five minutes at 0–0, before either side had registered a shot. AZ subsequently finished with 72 per cent possession, 26 shots, 18 attempts from inside the box and 4.65 xG.

No other team in the six Matchweek 3 fixtures FairEleven analysed produced a larger xG figure.

And almost no number from the round is less portable into the next match.

AZ spent roughly 85 minutes playing against 10 men. The red card replaced the tactical contest the pre-match analysis had been written to examine. Fortuna's normal attacking and defensive structure was barely tested, while AZ spent almost the entire match trying to break down an unusually deep, numerically reduced opponent.

The numbers are real. The context makes them exceptional.

This was an outlier event, not evidence that the original analysis had failed. It does not vindicate Fortuna's pre-match profile either: the intended eleven-against-eleven test simply never happened.

One FairEleven error did survive the dismissal.

Fortuna's confirmed XI had been treated as confirmation that the team played with a back three. A later average-positions map showed the identical 11 players in a 4-3-3.

The correction is not that Fortuna therefore "really" play 4-3-3.

It is simpler:

A team sheet is not a formation.

Fortuna's structure remains unresolved, and that uncertainty has to travel into the next analysis rather than being silently filled in.

Related analysis: Fortuna Sittard v AZ

What changes for Matchweek 4

Three rounds in, a few signals have earned more weight. PSV's consistent volume of shots from inside the box is one.

Others should be downgraded. PEC's extreme shot-conceding numbers were more opponent-dependent than they first appeared. Utrecht's post-half-time sequence did not survive a third observation. ADO's extreme inside-box concession figures changed sharply against different opposition.

Several attractive questions should remain open: PSV's central creation after Veerman, Fortuna's formation, AZ's finishing profile and whether Feyenoord's late concessions amount to anything more than a watchpoint.

The transfer market is changing the evidence at the same time. PSV have lost Veerman and added Geertruida. Ajax have loaned Josip Šutalo to Lazio. Feyenoord have added Javi López and sold Anel Ahmedhodžić to Burnley. Those changes affect the personnel available to answer questions raised by the opening weeks.

The analysis has to change too.

Tests need to measure the ideas they are supposed to test. Open questions should stay open until the evidence points somewhere. Two observations should not become a pattern simply because the story is attractive. Territory should not be mistaken for chance quality.

And a correct result should never be allowed to conceal incorrect reasoning.

Three matchweeks have not given us a stable Eredivisie. They have given us something more useful: the first meaningful separation between signals worth carrying forward, assumptions that need downgrading, and questions for which the evidence still says wait.

That is what changes next.