Companion to defend.html (proximity graph + attack Delaunay + Hungarian marking).
What am I looking at?
Not a match clock. It is one defensive phase that ends in a shot, sampled about twice per second. Each point is a freeze frame: how connected the back line is (λ₂), how much attack danger is uncovered, how many attackers look “free”, and structural pressure. You are watching whether the graphs move before the shot, not the whole game.
Surely pressing leads to regains and kills shots?
Often yes on regains — pressing and counter-press show up in regains. It is weaker as a pure shot predictor because coaches press when the ball is already in a dangerous zone. High pressure correlates with shots partly because the attack is already live, not only because pressing failed. Use pressure with shape + marking, not alone.
Does that mean once the shape breaks, the shot is inevitable?
No. The test was: does defensive λ₂ fall more in the 2 s before a shot than before phases with no shot? In this sample, it did not (if anything the opposite). So we did not find proof that late collapse is irreversible — only that a late graph break is not a reliable universal warning light.
We match each defender to the nearest attacker (Hungarian algorithm). Attackers free if the match is >8 m or the defender is not goal-side. New chart: visual EDA §16 — free roles at goals and shots (shots proxy “big chances” where xG labels are missing).
Next build: same logic filtered to organised blocks only; overlay body-pose facing on the two matches with pose data.
Each idea is now its own page — start at the cup ideas hub.