Was it the penetration, or the pressing?
Korea Republic vs Jordan, Group Stage

00

Two claims

The work started from a proposition raised in a coaching session.

That sentence bundles a factual claim with a causal claim. “Penetration fell” is something you check. “And that is why the performance dropped” is something you test. I separate the two and set a competing explanation against the causal half.

Part I tests Claim 1. Part II tests Claim 2. For each, the observations that would have to follow are drawn from the claim itself and checked against the data one at a time.

Before either, one thing has to be fixed: what counts as penetration. With different definitions the same numbers produce different conclusions.

01

Common ground — what counts as penetration

Both claims use the word. The first job is to make sure they use it the same way. This analysis adopts the FIFA Training Centre definition.

Four things the definition settles

Point ④ becomes the basis of the strongest objection later. Flagging it now.

Where — the spatial definition

The definition alone is not enough. Without fixing which zone is being counted, the number moves. This analysis counts only runs breaking the opposition defensive line inside the final third.

Direction

From definition to metric

The definition is set. Now to Claim 1, one observation at a time.

02 · Point ①

Did penetration actually fall?  As expected

Before disputing the claim, confirm what it rests on. It fell — sharply.

Point ① holds. To this extent the proposition is right, and the data does not dispute it.

On to the next point. If penetration halved, attacking output should have fallen with it.

03 · Point ②

Did attacking output fall with it?  Contrary

This is where Claim 1 first parts company with the data.

Given that Korea had ten percentage points less of the ball, the gap widens once the two matches are put on the same footing.

Point ② comes out the other way. Penetration halved, yet entries rose and shots went up by a little over half. Claim 1 does not permit that combination.

One correction before moving on

Korea's two goals against Jordan were a Son penalty (9′) and an own goal (90+1′). Twenty-two shots produced no open-play goal. All three against Bahrain were open play.

So the point is not “more shots, therefore better”. The accurate statement is “entries and shots rose while the goals disappeared” — which is precisely why the cause has to be looked for somewhere else.

Two points down, but not decisive yet. The next test is not in the Jordan match at all — it is in the Bahrain match. If penetration won that game, the win should carry its fingerprints.

04 · Points ③④⑤

Does the Bahrain win carry the fingerprints of penetration?  Contrary

Three places to look.

The middle one is decisive. If penetration won the match, the nine successful runs should have produced something. They did not. Bahrain's three goals came from progression through midfield and from Lee Kang-in's individual finishing.

Son's success rate actually rose — and I will not use it

Final-third penetration success was 9/40 (22.5%) against Bahrain and 3/6 (50.0%) against Jordan. On the face of it, more than double. But the Jordan denominator is six.

The intervals overlap heavily. Fisher exact test p = 0.173. So this analysis does not say Son's penetration got better. The most it can say is that it did not get worse.

Deciding not to use a favourable number is part of the analysis. 50% is an attractive figure, but an argument built on a denominator of six falls at the first objection.

05

Claim 1 — where it stands

The five points, set beside what was measured.

So what did cause it? The fall in penetration is itself something that needs explaining — it is not the explanation. Part II tests the alternative.

06 · Observation ①

What did Jordan do differently?  Confirmed in the data

Start by clearing away the most common reading. “Jordan sat deeper” is not supported.

The blocks were in almost the same place

Blocks sit at different heights by type, so like has to be compared with like. Taken as a whole-match average, “low block 27%” makes Jordan look like a deep-sitting side. Split by type, it does not.

What changed was time, not height

These are whole-match averages. The report does not break phase shares down over time. If Jordan pressed higher early and dropped off after taking the lead, the average blends the two states. The A–B–A analysis in section 8 points strongly that way, but settling the phase data by passage requires video measurement. The averages alone support “not higher, but faster” — and no more than that.

This matches the observation. The difference between the two opponents is not the height of the block but the fact that Jordan closed on every turnover instead of setting up.

Did that pressing actually do something to us? The first place it lands is the back line.

07 · Observations ②③④

What did that pressing do to us?  Confirmed in the data

Where the pressure landed, what it pushed back, and where the chain broke.

Observation ② · The centre-backs could not circulate

One metric was revised here. The first version used “the holding midfielder absorbed the build-up load” as a standalone indicator. But a holding midfielder taking part in build-up is his job, and CB ↔ DM holding at −3% confirms nothing unusual happened there. The anomaly is not the midfielder's involvement — it is the lateral link between the centre-backs going missing.

Observation ③ · So the front line dropped in

Both attacking reference points came towards the ball. With nobody left ahead, 59 through-passes still went in.

Observation ④ · So the passes stopped connecting

Through passes per run rising while conversion falls is the signature of “played in with nobody to receive, and so it did not come off”.

Statistically, this pair alone settles nothing. Put 38.8% → 22.0% through a Fisher exact test and it returns p = 0.090, short of the usual 0.05. The direction is consistent, but a two-match sample cannot rule out chance. This metric does not carry the conclusion — it is one of several pointing the same way.

Everything so far compares two matches, which invites the objection that the opponents were different. The last observation sidesteps it — inside a single match.

08 · Observation ⑤

What changed in the passage where the pressing stopped?  Confirmed in the data

Up to here it was two matches. Now the condition enters and leaves inside one match — same opponent, same players, so the pressing can be isolated.

Al-Naimat's goal in first-half stoppage time put Jordan 2-1 up, and in the second half they shifted from pressing to holding shape. Laid side by side — Bahrain (A), Jordan first half (B), Jordan second half (A′):

Korea's penetration did not rise in the second half. What changed was the freedom of the back line. Once Jordan stopped pressing, the centre-backs could circulate again and shots went from 6 to 16.

An honest qualification

Substitutions are mixed into the second-half improvement. Hong Hyun-seok's pass reception rate of 60.5% was the highest in the team, and Hwang In-beom → Hong was the team's second-most-used link. Saying both the disappearance of the press and the substitutions were at work is more honest, and in the end more persuasive.

One question remains. Why did Jordan stop pressing? Without an answer, “the pressing disappeared in the second half” reads as coincidence.

09

Why Jordan stopped pressing

Two reasons arrived at the same moment — the scoreline and the physical load.

The scoreline — something worth protecting

Al-Naimat's goal at 45+6 put Jordan 2-1 up. Until then they had not led, and so had a reason to win the ball back. Once ahead, that reason disappears.

The load — not sustainable

The two possession figures do not add to 100. The report splits it three ways — Jordan 31.3% · contested 14.7% · Korea 54.0%. Writing “Jordan had 46%” would be wrong.

10

Putting it together

What six sections established, in causal order.

One thing stated plainly. The five rows above are what was observed across two matches and three passages within one of them. No individual metric reaches conventional statistical significance, and what can be claimed is that all five point the same way — no more. Magnitudes and thresholds need more matches before they mean anything.

Accumulating those matches requires metrics that can be measured repeatedly.

11

KPI design — what to keep measuring

Thirteen candidates were screened with 40,000 Monte Carlo resamples.

The test

Event counts were modelled as Poisson, “X out of N” proportions as binomial. Using the observed values as parameters, 40,000 resamples were drawn and I measured how often the direction of change between the two matches survived — sign stability. The question it answers is a single one: is this difference real, or is it one match of luck?

Sign stability is not a standard statistic. It is a bootstrap heuristic I built for this screening, and the A/B/C bands below are my own cut-offs, not an established convention.

The metric that passed and was dropped anyway

Progression-load concentration reached 98.1% sign stability — grade A. I dropped it. It was meant to measure “did the holding midfielder absorb the build-up load”, but a holding midfielder joining the build-up is his normal job. The data agrees: CB↔DM went 33 → 32, and concentration across the top five pass links held at 13.9% → 13.7%. What actually broke was CB↔CB, at −40%.

Adopted set

How to use them

1. Read them as a set. Cause ↑, back ↓, front ↑, link ↓ together is the full chain.
2. Front and link always as a pair. Through-per-run up with conversion down is the signature of forcing it.
3. Match-status responsiveness must be computed for both teams. One side alone invites the charge of bias.
4. Thresholds are provisional. They come from two matches and need at least ten to be recalibrated.

That is as far as the data goes. One objection remains that the available material cannot answer.

12

The strongest objection

It goes last because it is the most interesting unresolved point in the analysis.

I accept it directly. The data I have cannot refute this. A decoy effect is a contribution made without touching the ball, so it leaves no trace in event data. How many metres a run pushed the line back exists only in tracking data.

The circumstantial evidence cuts both ways. Jordan's low block sat 18 m from goal and 26 m deep, shifting to a back five — a deep backline in FIFA's terms, with little room to be pushed back. But Bahrain's low block sat 20 m from goal and 21 m deep, and they held it for the same 27% of the match. Identical conditions with different outcomes weighs against the objection. Taken only within that 27%, however, the effect of Son not pushing the line may well have been real.

For what it is worth, In Between movements also fell, 120 → 63. Whether the space vanished because nobody pushed the line, or was never there because the block was already compressed, cannot be separated with event data alone.

I cannot settle it now. What is needed is tracking data — the distance between runner and defensive line, and how far the line retreats immediately after a run. Until that is measured, this analysis argues Claim 2 as the principal route and does not claim the decoy effect was zero.

13

What is settled, and what more data would settle

This was done with two public match reports. What that range allowed, and what is still unsaid for want of the right tools, are set out side by side. The second column is a limit of the material, not of the method — each line is somewhere the work continues the moment the data exists.

What was left out

• Offers Made by Third — the two reports define it differently, so it is not used.
• Bahrain physical data — that section is absent, so the two matches are not compared on running.
• Goalkeeper metrics — Kim Seung-gyu and Jo Hyeon-woo are different players; excluded from between-match comparison.
• Whether In Behind runs were received — the report gives offers only. I tagged Son's video myself to fill the gap, and left that sample (3/6) out of the argument because the denominator is too small.