Performance analysis. Source: season tracking database (244 games, 10 Hz positions, all events), the league position rating
(tools/rate_position.py) and the league match engine with behavioural models of both clubs. Working notes, code and every number
are in notebook.md, analysis/ and sim/.
The short version. The Eagles play the same way against everyone. They kick the ball long and down the middle as soon as they get it,
and they leave two forwards in the goal square for the whole game. They never handball (zero all season). On our side, the Power already win this
match-up narrowly (11-9 in head-to-heads, +0.8 points a game). We win the ball where the Eagles' long kicks come down and we keep it with short kicks.
We lose when we give it back to their man-on-man defenders, mostly from kicks after a mark. The change most likely to swing the next game: when the
Eagles have the ball, our two midfielders should drop 8-12 m deeper, so they stand where the Eagles' kicks land, not near the kicker.
Keep everything else as it is: forwards where they are, d2/d3 on the Eagles' goal-square forwards, and keep spoiling. Then fix one leak: our kicks after a mark.
In the simulator the midfield change alone turns a -2.7-point game into +9.4 (Power win rate 42% → 76%).
It was the best option under every version of the Eagles model I tried.
1. Key findings
The Eagles are a fixed 3-2-3 "kick it long to the goal square" team. They have the ball for only 0.34 s on average before disposing, never run with it and
never handball. Their kicks average 37 m (league 29 m) and come down in the central corridor (8 m from the centre line, league 15 m). The f2 stands 10-22 m out (16 m on average), dead
centre, in every phase, even at their own kick-ins; the f3 and f1 stay close by. Against us, f1+f2+f3 score 16.6 of their 17.3 points a game, f2+f3 alone 71%. High confidence
Their ball movement is a lottery and they don't change it. They keep only 44% of their kicks (league 67%), and 29% against us. Their kick-ins are long and central:
we won 36 of their 37 kick-ins. Their style is the same against every opponent (kick length 37 ± 2 m, f2 depth 124 ± 4 m across 29 opponents),
including four straight losses to the Seagulls. High confidence
Pressing the Eagles' kicker does nothing. Outnumbering them where the ball lands does everything. Their kick retention is the same whether the kicker is
under pressure or not (29-34% from defence at every pressure level). It falls from 42% to 23% / 13% / 10% when the opposition has 1 / 2 / 3 more players than the Eagles within 15 m of the
landing spot. High confidence
Their defence is five players marking man-on-man; their forwards never defend. d1/d2/d3 each mark one opposition forward 2-3 m goal-side.
Their depth follows the opposition forwards (r = 0.99 across games), so our forwards decide where their defenders stand. With the ball inside
the Eagles' defensive 40, opponents face pressure of 0.38, against 1.4 when the Eagles attack. High confidence
They take bad shots. 28-30% of their shots come from 40 m or further and return 0.1-1.3 points per shot, against 3.9 inside 15 m. High confidence
How we beat them now: we keep the ball (69% of kicks, against 64% vs other clubs). We win their long kicks in midfield; our m1/m2 collect 41% of them. We score 4.9 points
a game from spoils of their kicks. How they beat us: their defenders intercept-mark our kicks (3.5 a game, worth 6.1 points). The worst
leak is kicks after we take a mark or free kick: 36% of those are marked by the Eagles, against 8% in general play and 8% vs other clubs. Our margin tracks our
kick retention (r = 0.72) and these set-play turnovers (r = -0.63). High confidence for the mechanics; Medium confidence for cause and effect (20 games).
Other clubs show the same weakness. Clubs that do better against the Eagles than against us play their midfield higher and narrower around the ball (r = -0.72 across 28 clubs).
The Seagulls beat the Eagles by 18 a game but lost to us by 8. They did it with a compact block (35 m long, 16 m wide) sitting where the Eagles' kicks land. Medium confidence
2. Recommendations for the Power
R1. When the Eagles have the ball, our two midfielders drop 8-12 m deeper (towards our goal), into the corridor where their 35-40 m kick will land.
Don't chase the kicker. He releases in a third of a second, and pressure on him doesn't change where the ball goes. Why the data says it works: Eagles retention falls from 42% to 23% when we have one extra player near the landing spot, and to 13% with two. The Seagulls' block shows it in real games.
Simulator: mids 4 / 8 / 12 / 16 m deeper without the ball gives margins of +4.0 / +8.8 / +9.4 / +10.2
against -2.7 now. The gain stops growing beyond about 8-12 m. Dropping deeper with the ball gives nothing (-3.2), and pushing the mids higher is a disaster (-10.4).
The league position rating also prefers a set-up with more of our players between the Eagles' kicker and his target (0.61 → 0.54-0.56 Eagles-score-next).
Medium-high confidence on direction; Medium confidence on size.
R2. Leave the forwards where they are (about 49 m out when the ball is in midfield). Don't push them higher, and don't also push them deeper once R1 is in place.
Keep "forwards 10-12 m deeper" in reserve in case the Eagles adapt to R1. Evidence: the Eagles' defenders follow our forwards (r = 0.93 within games, 0.99 across games). On its own, moving the forwards 12 / 20 m deeper helped in the simulator
(+3.0 / +4.0 against -2.7), and Eagles intercept marks of our kicks fell from 5.3 to 3.3 a game.
Combined with R1, though, it took away from R1's gain: +8.2 with forwards 12 m deeper and +1.1 with 20 m, against +9.4 for R1 alone.
Deep mids plus deep forwards leave too big a gap between the lines, and our kick retention fell to 53-64%. The same happened against the stronger Eagles model.
Forwards 12 m higher was worse (-3.7). The position rating sees no difference between these forward depths. Medium confidence
R3. Keep d2 and d3 on the Eagles' f2 and f3, goal-side, and keep spoiling rather than marking their kicks. Those two players produce 71% of the Eagles' scores
against us. Spoils already give us 4.9 points a game. The high-block alternative that leaves f2 alone was no better in the simulator than one that keeps d2 on him
(+3.2 vs +3.5). Standing in front of f2 instead of behind changed nothing (-1.2). Medium-high confidence
R4. After a mark or free kick, don't kick to a forward who has an Eagles defender standing goal-side of him. Use the 6 seconds for a lead, go short to a free
midfielder, or play on with a handball. Evidence: 36% of our set-play kicks were marked by the Eagles (27% if shots at goal are counted), against 8% in general play. When we played on by running we kept it 77% of the time
(n = 30), and by handball 62% (n = 40, no intercepts). Caveat: the samples are small, and the simulator could not test this fairly. It does not reproduce the leak (16% vs 36%),
and its blanket "always play on" version was worse (-6.6) because the kicker then kicks under pressure. Treat this as a targeted
fix to rehearse, not a blanket rule. Low-medium confidence
R5. Keep doing what already works: the short-kick possession game (69% retention), the set-up at Eagles kick-ins (36 of 37 won), and protecting the 25 m arc while conceding long shots
(their 40 m+ shots return about 1 point each). Don't shorten kicks further (-4.6), switch kick-ins wide (-3.9), chase tackles
(-1.8) or tighten our own shot selection (-3.7). None of these helped in the simulator.
3. How the Eagles play
3.1 Shape: a fixed 3-2-3
Average positions with the ball (circles) and without it (squares). Each club attacks left to right; y is drawn by side (d1/f1/m1 top, d3/f3/m2 bottom), using the average distance from the centre line.Where each line spends its time. Eagles forwards (top right) live in the goal square and on two lanes beside it. Their defenders sit wherever the opposition forwards are. Our forwards (bottom right) roam a 50 m band.
Forwards: f2 averages x = 124 (16 m out) and 2 m from the centre line in every phase. Even when the ball is in the Eagles' own defensive 40 he stands at x = 128-130. f3 sits about 119 and 12 m wide, f1 about 105 and 11 m wide.
When the Eagles lose the ball the forwards stay; they are 8-9 m from any opponent (our forwards are 5-6 m).
Midfield: m1 and m2 follow the ball and both go to the opposition m1 (3-5 m away). m1 takes every centre bounce. They win about 50% of first possessions (46% vs us).
Defence: d1/d2/d3 mark f1/f2/f3 one-on-one, 2-3 m goal-side. There's no zone and no spare man. Only about 3 Eagles are goal-side of the ball when defending (opponents keep 3.4-3.7).
3.2 With the ball
Where kicks land (not shots). The Eagles go long and central; we and other clubs spread the ground and go shorter.Kick length and how often the kicking side keeps it. The Eagles' long kicks are kept less than 45% of the time at every length.
per game, vs other clubs
Eagles
Power
other clubs
possessions
22.6
34.7
34.0
kicks
24.8
28.3
32.9
handballs
0.0
8.0
8.3
marks
4.7
5.0
9.0
tackles
5.9
1.8
5.7
shots
8.8
5.7
6.9
goals per shot
0.33
0.37
0.36
average kick (not shots)
37 m
26 m
29 m
kicks kept
44%
64%
67%
seconds holding the ball
0.34
0.43
0.58
36% of Eagles kicks are shots. Their chains are short and direct: 5.2 s and 1.3 kicks on average; a scoring chain lasts 6.2 s, against 11.5 s for their opponents.
Their biggest scoring source is ball won in their forward half. Against other clubs, chains that start in their forward 40 produce 7.4 points a game (2.1 points per chain). The forwards who never leave the goal square
pick up the spills from their own long kicks.
Shots. Left: Eagles shots against other clubs. Middle: our shots against the Eagles. Right: points per shot by distance.
3.3 Without the ball
Defending, by where the ball is. Inside their own defensive 40 the Eagles put almost no pressure on the ball carrier (0.38) and have about one player goal-side.
The Eagles defend with five players: two midfielders around the ball and three man-markers. When an opponent has the ball deep in the Eagles' half there is no
spare defender and almost no pressure on the ball. The defenders are good at intercepting, though. Against us they mark 3.5 of our kicks a game (we lose 1.5 a game against other clubs).
Of the 70 Power kicks they marked, 59 went to d1/d2/d3.
3.4 Stoppages and transition
Kick-ins. The Eagles always kick long down the middle; we won 36 of their 37 kick-ins. Other clubs go wide and keep about two-thirds.
Centre ball-ups are set by the rules (everyone starts in the same spots), and first possession is about 50/50 for every club. After winning a centre clearance against us, the Eagles
score the next goal 59% of the time (we manage 56%).
Kick-ins: long (45 m) and central, with f2 left in the opposite goal square. They kept 31% against other clubs and 3% against us.
Transition: the Eagles score fastest off turnovers. Their scoring chains start on average 55 m from goal, ours 73 m.
Points a game by how and where possession was won before the score.
3.5 Where they are vulnerable
The landing zone. The ball goes long, early and central, to space rather than to a teammate (the nearest Eagle is 16 m from the landing spot when the ball is kicked).
A side that has more bodies there wins the ball back (finding 3).
Five defenders against our eight. Their forwards don't defend, so once we get the ball past their two midfielders we attack 5-on-5 or better, against defenders who follow our forwards wherever they go.
Shot selection and kick-ins, as above.
No plan B. Nothing in 132 games suggests they change their approach to suit the opponent.
The Seagulls' answer (they won all 4 games against the Eagles): a compact block of all eight players where the Eagles' kicks come down, leaving f2/f3 alone.Clubs that play their midfield and d2 higher against the Eagles do better against them than against us (each dot is one club).
4. How the Power play now, and why the head-to-heads go the way they do
Against other clubs we play a compact block (58 m long against the Eagles' 89). Our three forwards stand in a line about 20 m ahead of the ball and our defenders mark man-on-man.
We go forward with short kicks (26 m) and very short handballs (4.6 m), shoot from 27 m on average, and barely tackle (1.8 a game; our defenders and forwards made no tackles all season).
Against the Eagles our d2/d3 follow their f2/f3 into the goal square (x about 20), and we have more of the ball than they do (42 possessions to 19).
Our kicks marked by the opposition. Against the Eagles, kicks of 30-40 m get picked off 19% of the time.The head-to-heads, game by game. Left: we win when we keep the ball. Middle: we lose when our set-play kicks are marked. Right: the set-play leak only appears against the Eagles.Typical turnover (game 225, 0:23). A 37 m kick to a forward standing in front of his man; the Eagles' d1 is goal-side and marks it. (The renderer gives the home club the red kit, so here the Power are red.)Typical Eagles exit (game 224, 1:07). d3 kicks 43 m into the middle, where our m1/m2 are waiting. f2 stands in the goal square at the far end with our d2 beside him. (Eagles in red.)
When we win: we keep 70%+ of our kicks, our mids pick up their long kicks, and we score from spoils and centre clearances. In our wins the Eagles had 13-22 possessions and shot from long range.
When we lose: our retention falls to the high 50s, and their defenders mark our kicks, especially from set plays, and turn them into f2/f3 shots (6.1 points a game from intercept marks).
The two heaviest losses (0-33, 0-38) had 7 and 10 Eagles intercept marks. We had no real home advantage (5-5 at home, 6-4 away), and nor did anyone else this season.
5. How the ideas were tested
Simulator. The league engine ships with this workspace without any club's coaching code. I wrote behavioural models of both clubs from the measurements above. Eagles: the fixed 3-2-3, man-marking,
long kicks to space, shots from up to 55 m, no handballs. Power: forwards sliding with the ball (fitted: x = 0.45 × ball x + 56), short kicks to a teammate who can win the race to the ball, spoiling.
Calibrated against the head-to-heads, the models reproduce the Eagles' possessions (18.6 vs 18.8 real), Eagles shots (8.0 vs 8.5) and our kick retention (68% vs 69%). The simulated
margin (-2.7 ± 0.8) sits within the real head-to-head range (+0.8 ± 4.6). Every variant was played over the same 160-280 seeds, with ends and player ratings balanced.
Power set-up tested
sims
Power win
Power margin
score P – E
Eagles kick retention
Power kicks marked by Eagles /g
note
Baseline: the Power as they play the Eagles now
280
42%
-2.7 ± 0.8
13.4 – 16.1
50%
5.3
Midfielders 4 m deeper when the Eagles have the ball
160
62%
+4.0 ± 1.0
16.4 – 12.4
40%
4.6
Midfielders 8 m deeper when the Eagles have the ball
160
74%
+8.8 ± 1.2
19.5 – 10.7
36%
4.3
R1: midfielders 12 m deeper when the Eagles have the ball
160
76%
+9.4 ± 1.2
20.7 – 11.3
35%
4.2
Midfielders 16 m deeper when the Eagles have the ball
160
75%
+10.2 ± 1.1
20.2 – 10.0
37%
4.0
Midfielders 8 m deeper when the Power have the ball
160
41%
-3.2 ± 1.1
14.4 – 17.6
51%
5.5
no gain
Midfielders 8 m higher (all phases)
160
20%
-10.4 ± 1.1
10.3 – 20.7
63%
6.2
reverse test
Forwards 12 m deeper (closer to goal)
160
52%
+3.0 ± 1.2
19.5 – 16.6
52%
4.5
Forwards 20 m deeper
160
59%
+4.0 ± 1.0
20.7 – 16.7
52%
3.3
Forwards 12 m higher
160
39%
-3.7 ± 1.0
12.2 – 15.9
51%
4.4
Mids 10 m deeper w/o ball + forwards 12 m deeper
160
74%
+8.2 ± 1.1
22.2 – 14.1
37%
3.9
less than R1 alone
Mids 10 m deeper w/o ball + forwards 20 m deeper
160
53%
+1.1 ± 1.2
16.4 – 15.3
40%
3.0
lines too far apart
As above + shoot from up to 50 m
160
73%
+8.2 ± 1.2
22.6 – 14.4
40%
3.8
Seagulls-style high compact block, f2 left free
160
54%
+3.2 ± 1.3
23.7 – 20.5
40%
3.5
High compact block, d2 stays on f2
160
61%
+3.5 ± 1.4
23.0 – 19.5
41%
3.9
d2 plays in front of f2 instead of behind
160
45%
-1.2 ± 1.1
14.4 – 15.6
48%
5.1
Kick no further than 25 m
160
41%
-4.6 ± 1.1
13.9 – 18.5
49%
3.8
fewer intercepts, less ground
Avoid 30-40 m kicks
160
43%
-2.4 ± 1.1
13.8 – 16.2
50%
5.0
Play on immediately after every mark
160
30%
-6.6 ± 1.0
10.9 – 17.4
51%
4.9
see caveat
Midfielders tackle the Eagles ball carrier
160
43%
-1.8 ± 1.0
13.9 – 15.6
50%
5.3
Mark the Eagles' kicks instead of spoiling
160
48%
-1.0 ± 1.1
15.1 – 16.2
49%
5.2
Power kick-ins wide
160
37%
-3.9 ± 1.1
12.7 – 16.6
51%
4.3
Power shoot only inside 32 m
160
38%
-3.7 ± 1.1
13.4 – 17.1
51%
5.2
Power shoot from up to 50 m
160
49%
-1.3 ± 1.1
15.5 – 16.8
52%
5.1
Robustness. The model Eagles keep the ball more than the real ones (50% vs 29%), so I reran the key plans against two other versions of the Eagles model (120 games each; margin, with Power win % in brackets):
Eagles model
baseline
R1: mids 12 m deeper w/o ball
R1 + forwards 12 m deeper
high block, d2 on f2
forwards 12 m deeper
Eagles clone as calibrated
-2.7 (42%)
+9.4 (76%)
+8.2 (74%)
+3.5 (61%)
+3.0 (52%)
Eagles clone chases its own kicks less (keeps the ball less: closer to the real Eagles)
+2.7 (62%)
+12.0 (81%)
+11.8 (82%)
+6.1 (69%)
+7.1 (66%)
Eagles clone chases more and defends harder (stronger Eagles)
-8.0 (25%)
+7.0 (68%)
+3.1 (58%)
+1.5 (51%)
-4.4 (40%)
Position rating. I took real average positions in three head-to-head situations and edited our players' positions:
Situation
Power set-up
rating
Eagles defender has the ball 33 m from his goal
as now
0.610 Eagles score next
compact block between kicker and target, d2 stays on f2
0.542
compact block, f2/f3 left free
0.551
Power have the ball at midfield
forwards as now (about 49 m out)
0.635 Power score next
forwards 10 m / 20 m deeper (their defenders follow)
0.641 / 0.639
forwards 10 m higher
0.609
Eagles kick in the air towards f2
d2 behind f2 / in front of f2
0.611 / 0.612
The rating tool is useful but limited. It doesn't know about player ratings, fatigue or club styles, and it is under-confident: when it says 25% the real rate was 14%, and when it says 74% the real rate was 87%. It also under-rates the Eagles
against other clubs (0.47 predicted, 0.56 actual). I used it only to compare set-ups, never as an absolute probability.
6. What I could not establish, and how sure I am
Coaching instructions and fatigue aren't in the data. Everything about "intent" is inferred from behaviour. Running speeds fall the same way for every club (midfield top speed 7.6 → 5.4 m/s within 90 s),
and the Eagles don't fade late (scoring by minute is flat). So I found no fatigue edge, but I couldn't measure fatigue directly.
Twenty head-to-heads is a small sample (real margin +0.8 ± 4.6). The per-game correlations are suggestive, and the set-play one leans partly on one game (-38).
The rank correlation is -0.50 (p = 0.03), and -0.46 without that game.
The simulator is a model of both clubs, not the clubs. It flatters the Eagles' retention (50% vs 29%) and doesn't reproduce our set-play leak. That's why R4 rests on the real data only.
The size of the R1 gain (+9 to +15 points depending on the Eagles model) should be read as "large and positive", not as a forecast.
The Eagles might adapt. They never did in 132 games, including four straight losses to a block, but their coaches could change things before this game.
If they shorten their kicks against our deeper mids, R1 loses force, and we should go back to our current shape.
The other-club comparison (finding 7) uses 28 clubs and many candidate features, and opponent style is measured in games where they were reacting. It's a pointer, not proof.
The midfield-height link survives dropping the Seagulls (r = -0.69; rank correlation -0.66, p < 0.001), but the d2-height link mostly does not (-0.56 → -0.29).
The midfield measure is taken while the opponent has the ball, so it is partly a result of dominating rather than a cause.
Player ratings don't explain results. Matching players have identical ratings in every game. Across 80 role-by-rating checks I found nothing beyond chance (largest |r| = 0.28 over 112 games).
Not established: whether the Eagles' f2 position depends on who marks him, how they defend kick-ins in detail beyond the summary above, or anything about the 28 other clubs' games against each other (not in the data).
Conclusion
Confidence
Eagles' fixed structure, no handballs, long central kicks, f2/f3 dependence
High confidence
Eagles' retention is decided by numbers at the landing spot, not pressure on the kicker
High confidence
Eagles defend man-on-man with five; their defenders follow our forwards
High confidence
Our losses come from Eagles intercept marks, especially after our set plays
Medium confidence
R1: drop the mids into the landing zone
Medium-high confidence
R2: forwards stay where they are (deeper only as a fallback)
Medium confidence
R3: keep d2/d3 on f2/f3 and keep spoiling
Medium-high confidence
R4: change our set-play kicking
Low-medium confidence
Size of the simulated gains
Low confidence
Method notes: all positions are flipped so the club being described attacks x = 140. "Kept" means the kicking side's player made the next mark or possession.
Pressure uses the league formula (sum of exp(-(d/5)²) over opponents, capped at 2). Possession chains run from gaining the ball to losing it or scoring.
Code: analysis/*.py (data), sim/bots.py, sim/run.py, sim/variants.py (simulator), analysis/rp_tests.py (position rating).