Power v Sharks: how they play, why we lose, what to change
Performance analysis for the senior coach. Source: the full season's tracking and event data (252 games of 240 s,
including 20 Power v Sharks games and 116 Sharks games against other clubs). All numbers come from that data; methods,
code and a working notebook are in this folder (analysis/, notebook.md).
The short version. The Sharks don't beat us at the centre bounce. They beat us by keeping the ball in our
defensive half and waiting for our exit kick. They win the ball back inside our defensive 50 almost 5 times a game and
score from 62% of those, in a median 3.5 seconds. That is 13 of their 30 points against us. Our exits fail because our forwards
stand still in a flat line and the Sharks' defenders stand 3 m from them, our midfielders crowd the ball, and we kick at the first
opportunity (0.4 s) to whoever is nearest. The clubs that beat the Sharks exit the other way: midfielders higher, forwards
staggered and leading, and a 30-35 m kick to a player in space. Changing our exit shape and ball use is the priority, and it is
where the evidence is strongest.
1. Key findings
The game is lost at our defensive-50 exits. Against the Sharks, half of our possession chains that start in our
defensive 50 are lost again inside that 50 (47-52%); only 9-10% end in a score. The eight clubs that trouble the Sharks lose
only 15-16% inside the 50 and score from 22-29%. The Sharks then score off those turnovers at 62% (13.0 points a game; against
the strong clubs this source is worth only 1.9 points a game). High confidence
The Sharks press by covering our receivers, not by swarming the ball. Pressure on our kicker is ordinary (0.90 on the league's
0-2 scale, close to the 0.8-0.85 other clubs face), but our forwards have a Shark a median 2.9 m away and are an open target only
12% of the time (strong clubs' forwards against the Sharks: 5.3 m and 31%). High confidence
Our forwards don't move and our midfielders crowd the ball. With the ball in our defensive 50, our forwards are leading
(>5 m/s) 4% of the time, against 23-24% for the Sharks' and the strong clubs' forwards. They stand in a flat line (6.5 m of depth
between them, against about 12 m). We keep 3.9 players within 30 m of our own goal during exits (strong clubs 2.9). High confidence
We kick to contests, at the first opportunity. We release the ball at the 0.4 s minimum on 89-100% of disposals.
League-wide, a kick with 10+ m of space at release is kept 91-94% of the time, and one where the nearest defender is 5+ m closer to the landing spot than the
teammate only 17-30%. Only 8% of our kicks against the Sharks have more than 6 m of space (Sharks: 35%). High confidence
The clubs that beat the Sharks keep the ball. They keep 72% of their kicks against the Sharks (the other 22 clubs: 56%),
hold chains for 8.7 s, take marks and kick for goal from set shots (41% of shots). The Sharks get 13.9 possessions a game against them,
against 17.9 against us. Across all 136 Sharks games, the Sharks' margin is lower when the opponent holds the ball beyond the
minimum more often (r = -0.54) and kicks over the top from midfield more often (r = -0.57). Moderate confidence
We waste our forward-50 marks. After a mark or free inside 50 against the Sharks, we play on 87% of the time; the Sharks and
strong clubs kick for goal 90-96% of the time. None of our 71 shots against the Sharks were set shots, and 38 of them were
pressured shots from 30-50 m (goal rates 24% and 15%). High confidence (but a small number of occasions)
What already works: centre bounces. First touch is 50/50, and when we win it our centre chains average 1.37 points against the
Sharks, about the same as theirs (1.49). Our m1-to-f2 direct kick out of the centre is our best attacking pattern against them. Moderate confidence
Tested with the league's position rating: on 280 real Power defensive-50 possessions against the Sharks, giving our forwards 3 m more
separation and lifting our midfielders 12 m raises the expected value of the best open kick by +4.0 points of next-goal
probability per possession (95% CI 3.5-4.5). Picking better targets within the current shape is worth only about +1. Moderate confidence
(static test: the Sharks' reaction isn't modelled)
Figure 1. Results. The Sharks are 120-1-15 for the season (+25 a game) and 19-1 against us. The eight 'strong' clubs (Falcons, Magpies, Hawks, Dockers, Cats, Kangaroos, Bulldogs, Titans) account for 13 of the Sharks' 16 non-wins; the Sharks are 19-1-12 against them. We are 0-1-31 against those same eight clubs.
2. The game plan in one table
#
Change
Why the data says it works
Confidence
1
Exit shape. With the ball in our back 50: midfielders sit 35-40 m from our goal (now ~23-24 m), so one extra
player is in the 30-60 m band; forwards at two depths (one leading up to 45-55 m, one holding 60-70 m, one wide at |y| 15+), not a flat line of three.
Strong clubs set up this way and exit cleanly (Fig. 2, 6, 7, 11). Rating-tool test: +2.1 pts for the midfield move, +1.8 for forward
separation, +4.0 combined, per possession (Fig. 13). Open targets per possession rise from 1.0 to 1.6.
High confidence on the problem; Moderate confidence on the size of the gain
2
Forwards lead, on cue. When we win it back in defence, one forward leads at the ball carrier and one leads into space wide. Target: leading
about 20% of the time (now 4%).
Separation is what makes a kick safe (Fig. 8, 9). Our forwards are 2.9 m from a Shark; strong clubs' forwards 5.3 m, and our own forwards
against other clubs 6.8 m. A static forward is easy to stand next to.
High confidence
3
Ball carrier: take a second if free, then kick 30-35 m into space. No more 0.4 s releases to the nearest man. Avoid kicks under 20 m out of defence
unless the target is unmarked. Aim at the lead, and wide rather than central.
Under 20 m our exits are kept 49% of the time; 20-35 m kicks by strong clubs, 81-83%. Holding 0.8-1.2 s after the gather is tackled 2-3% of the time;
holding-the-ball frees are rare (29 all season). Beyond 1.2 s the risk rises (8-13%).
High confidence
4
Kick-ins: wait about 2 s, then 35-40 m to a wide lead.
Our kick-ins against the Sharks: taken after 0.4 s, 29 m, central, 58% kept. Strong clubs: 2.0 s, 37 m, wide, 73% kept. The Sharks' own: 39-43 m, 84-92% kept (Fig. 12).
Moderate confidence
5
From midfield, kick behind the Sharks' last defender when a forward is goal-side of them.
Kicks over the last Shark from 47-70 m: chain scores 30% v 19%; from 70-93 m: 45% v 17%, with no loss of retention. We do it 19% of the time from midfield;
strong clubs 37%. (From our own 50 it doesn't help, so don't bomb out of defence.)
Moderate confidence
6
Inside 50: take the set shot. Stop playing on from marks and frees.
Same distance (35-45 m): kicking for goal from the mark averages 2.93 points, playing on 1.73. From 45-60 m: 2.20 v 0.99. Only about 0.75 such
marks a game against the Sharks, so it's worth roughly a point a game.
High confidence on direction, small impact
7
Keep the centre-bounce pattern (m1 wins it and kicks direct to f2).
Our centre chains score 1.37 points per possession against the Sharks (against other clubs only 0.82); it's the phase where we match them.
Moderate confidence
Size of the prize (an upper bound, not a forecast). If our defensive-50 chains performed like the strong clubs' do
against the Sharks (0.91 points per chain instead of 0.34; about 10 such chains a game), we would add about 6 points a game. If the
Sharks' turnovers inside our 50 fell to the strong-club level (1.0 a game instead of 4.85), they would lose up to about 11 points
a game. The two overlap (our lost exits are their forward-50 turnovers), so they don't simply add. Halving the gap to the strong
clubs is worth something like 8-9 points a game: enough to turn a 21-point average loss into a contest, but not to make us favourites.
3. How the Sharks play
Attributes are drawn fresh each game and mirrored between the matching players of both clubs, so no club has better players.
Everything below is tactical: how they position and how they use the ball.
Figure 2. Average shape by phase against each other (each club drawn attacking left to right). Players don't hold fixed sides (mean signed lateral position is about 0 for every slot), so the lateral marker is mirrored. Top right: the Sharks' press when we have the ball in our defensive 50 (the Sharks' forward 50). Their forwards and midfielders are within about 20-30 m of our goal; their three defenders sit about 62 m from our goal, on top of our forwards.
3.1 With the ball
Long, forward kicking to forwards in space. Non-shot kicks average 32.6 m (ours 25-26 m) and gain 27 m (ours 21 m). From midfield, 91% of their kicks
target a forward. At release the nearest opponent to their landing spot is a median 15 m away (ours: 9-11 m). 35% of their kicks have more than 6 m of space,
against our 8%. High confidence
Few handballs, but safe ones. 14-21% of disposals are handballs; those average 12 m and 96-97% are kept (ours: 5 m, 82-87% kept). High confidence
They take time when it's on. 21-32% of their disposals come after the 0.4 s minimum (ours: 0-11%). High confidence
Forward line: three forwards, spread: one wide (f1, mean |y| 14 m), one semi-wide (f2, 11 m), one central (f3, 5 m) who also works up the ground (f3 takes 48% of
their midfield marks against us). Goals are shared between all three (f1/f2/f3: 33/26/30% against us). High confidence
Shooting: 10.6 shots a game against us. General-play shots from a mean 23 m; set shots from about 35 m, and they kick for goal from 90-96% of forward-50 marks. High confidence
3.2 Without the ball
High press in their forward half, built on covering receivers. When we have it in our back 50, their forwards and midfielders sit within about 20-30 m of our goal,
and their defenders push up about 62 m from our goal (higher against us than against anyone: 66-68 m against other clubs). Pressure on our kicker is
ordinary (0.90); the squeeze is on our receivers. High confidence
Defenders loosely matched, by lane. d1/d2/d3 sit a median 2.7-3.2 m from our f1/f2/f3 (from other clubs' forwards: 3.6-4.6 m). When the opponent's forwards hold their
lanes (us), each defender follows "their" forward (lateral-movement correlation about 0.5). When the strong clubs' forwards rotate, the pairings break
(d2 ends up on f3, d3 on f2) and the correlation falls to 0.15-0.3. This reads as lane-based matching rather than strict man-on-man. Moderate confidence
They win the ball back with spoils and ground balls, not intercept marks. 50% of our lost exit kicks were spoiled first; 92% of their turnovers are
ground-ball gathers. Their defenders win 37% of our lost exits, their forwards 35%, their midfielders 27%. They make 7.4 spoils and 5.0 tackles a game against us
(we make 3.7 and 1.35). High confidence
Their forwards don't come back. Without the ball the Sharks' forwards stay 23-26 m ahead of the ball, so every turnover finds them already in position. High confidence
Figure 3. Where the Sharks win the ball back (their attacking direction left to right; black dots = that chain ended in a score). Against us, the hot zone is inside our defensive 50 (right-hand side), where they win 4.8 a game and score from 62%. Against the strong clubs they win only 1.0 a game there.
3.3 Stoppages
Centre bounces are even. m1 always takes the first touch (it's the rule set-up). The Sharks win first possession 52-55% of the time against us, which is
indistinguishable from 50%. Their pattern: m1 handballs to m2, who kicks (24-33% of their centre wins), first gain 15-17 m. Centre chains score 1.49 points per
possession against us. High confidence
Kick-ins: they wait 2-4 s for leads, kick 39-43 m, often wide, and keep 84-92%. Moderate confidence (n = 13 against us, 114 in all)
Throw-ins are too rare (95 in the whole season) to analyse.
3.4 Transition
Lethal off turnovers in their forward half: median 3.5 s and 1.4 kicks from winning it to scoring; 62% of those chains score against us. High confidence
From their own back half they are patient: scoring chains from defensive-50 turnovers take 14-18 s and about 3.5 kicks, usually via a mark. High confidence
Figure 4. Points per game by where the scoring chain started. Against us the Sharks' biggest source (red) is turnovers inside our defensive 50: 13.0 of their 30.1 points. Against the strong clubs that source is worth 1.9.Figure 5. Shot maps (attacking right). Stars = set shots. The Sharks get close; the strong clubs shoot from further out but from set shots (39%); we have no set shots and many pressured shots from 30-50 m.
4. Where the Sharks are vulnerable
The best evidence is the eight clubs that took 13 of the Sharks' 16 non-wins. Against them the Sharks average 18.8 to 16.1, a near-even contest.
Opponents who keep the ball shrink the Sharks' game. Strong clubs keep 72% of their kicks and 96% of their handballs against the Sharks. The
Sharks' possessions fall from 17.9 a game (against us) to 13.9, and their score rate per possession from 46% to 33%. Moderate confidence
(the strong clubs are also good in general; the style and the quality can't be fully separated)
The lane-matched defence can be pulled apart. Kicks into the Sharks' defensive 50 that land 15+ m from the nearest Shark are kept 75% of the time, and 53%
of those chains score; under 5 m, 40% kept and 18% scored. The best landing zone is 25-35 m out (44-47% of entries there lead to a score). Strong clubs'
entries from a set position (after a mark): 90% kept, 75% scored. Moderate confidence
Space behind the high line, from midfield. With their defenders pushed up, kicks that land behind the last Shark lift the chain's scoring rate from 19% to 30%
(kicked from 47-70 m out of our goal) and from 17% to 45% (70-93 m), at the same retention. It does not help from inside our own 50. Moderate confidence
No evidence that they fade. Speeds drop after the first minute for every club alike, and the Sharks score evenly across all four minutes. I rejected "run them
off their legs" as a plan. Moderate confidence
Where there is no edge to find: centre bounces (first touch is even, and both sides' centre chains are worth about 1.4-1.5 points). Their ball use against us (69% of kicks kept) is no worse than against the strong clubs (69%), so the gain is in our ball use, not in pressuring theirs.
5. How the Power play now, and why the Sharks beat us
5.1 Our current style (all 136 games, and against the Sharks)
Short ball movement: kicks 25-26 m, handballs 5 m, 89-100% released at the 0.4 s minimum.
Midfielders glued to the ball (lateral position correlates about 0.65 with the ball's). With the ball in our back 50 they sit about 23 m from our goal, level with the ball (0.6 m behind it on average) rather than ahead of it; the strong clubs' midfielders sit 8 m ahead of the ball, giving the kicker a target.
Forwards in a fixed flat line across the ground: f2 central (|y| about 2 m, spread 5 m), f1 and f3 wide at |y| 14 m, all at the same depth, about 60 m from our goal, and nearly static.
Without the ball our defenders push up close to the ball (10-11 m behind it; the Sharks' defenders sit 16-22 m behind) and our forwards stay very high (37-38 m ahead of it).
Forward 50: we play on from marks and frees (87% against the Sharks) instead of taking set shots.
5.2 The losing cycle
The Sharks enter their forward 50 about 9.4 times a game. Our defence copes reasonably: they keep 64% of entries against us, the same as against the strong clubs (66%).
We win it back deep: 6.1 turnovers a game in our back 50, plus 3.85 kick-ins. About 10 exits a game, and almost 20 disposals a game inside our own 50 (strong clubs: under 10).
Our exit kicks are 24 m and central, aimed at spots a Shark is already 9 m from. Only 27% land with no Shark within 5 m; 8% are marked by our player.
The ball dies in our 50: on average our exit chains never get past 38 m from our own goal. The Sharks gather the spill and score within seconds.
Figure 6. How possession chains that start in a club's own defensive 50 end, against the Sharks. Ours (top): half are lost again inside our 50. The strong clubs lose 15-16% there and score from about a quarter.Figure 7. Kicks out of defensive 50 against the Sharks, by landing point (attacking right). Ours are short and central, landing in the Sharks' press; the strong clubs' are longer and wider, and 32% are marked by their own player.Figure 8. Forwards while their own team has the ball in its defensive 50. Against the Sharks our forwards are 2.9 m from the nearest opponent, open 12% of the time, and leading 4% of the time. Against other clubs they have more room (6.8 m) but still don't lead, so the 'stand still' habit is ours; the Sharks are simply the side that punishes it.Figure 9. Left: in every game this season, the share of kicks kept rises steeply with space at release (for kicks of 15 m or more; the same pattern holds within each length band). Right: how much space each side's kicks have. Ours against the Sharks bunch at 0-3 m; the Sharks' and the strong clubs' spread to 10 m+.
Picture of the problem: game 245, 2:12. Five Power players are within 13 m of our goal. Our three forwards stand in a line about 30 m out,
each with a Shark beside them. d3 kicks 17 m into the gap, Sharks d1 runs onto it, and they kick a goal 6 s later. That sequence (our defensive-50 kick lost, then a
Sharks goal within seconds) happened 27 times in the 20 games. In the broadcast stills below the home side wears red.
Figure 10. Game 245 (Power home, red). Our exit: no lead, no depth, nobody between 13 m and 30 m from our goal.Figure 11. Game 72 (Falcons, away in blue, won 32-2). The contrast: Falcons midfielders 35 m up the ground, forwards at two depths, f2 leading into the space in front of Sharks d3; a 35 m kick, uncontested mark, set play.
5.3 Kick-ins
Figure 12. Kick-ins (attacking right). Ours are taken immediately, short and central. The strong clubs and the Sharks wait about 2 s and go 37-39 m, wider.
5.4 Why we won game 235, and the close ones
Our one win (18-15) and the three games within 7 points share the same signature: more kicks (31-36 against 18-28 in our seven heaviest losses), more marks (4-8),
longer chains (6.4-8.2 s). Across the 20 games our margin correlates with our kick count (r = 0.85), chain length (0.59) and marks (0.55), and against our handball
share (-0.42). The Sharks' own kicking accuracy also matters (their kick retention against our margin: r = -0.60), so some of game 235 was their off day.
Low confidence (20 games; indicative only)
6. Testing the changes
Method. The league's position rating (tools/rate_position.py) values a position as the probability that each side scores the next goal. I checked it
against the database first: on held-ball positions rebuilt from tracking it reproduces the stored rating closely (r = 0.89, mean error 0.08). On loose and in-flight
balls it does poorly (r about 0.45, because the flight details can't be fully rebuilt), so I used held-ball positions only.
Valuing each kick option on 280 real moments when we held the ball in our defensive 50 against the Sharks. For every teammate in kicking range (from each
kicker's kick power): value = P(kept) x rating if we hold it there + P(lost) x rating if the nearest Shark has it there. P(kept) comes from a model fitted on 16,000 league
kicks (it uses space at release, length and pressure: 59% / 77% / 89% kept at 0 / 5 / 10 m of space for a 30 m kick).
Result in the current shape: nearest teammate 0.466, most-open teammate 0.477, most-open forward 0.448 (worse, because the forwards are covered), best available 0.486.
Better decisions alone are worth about +1 point.
Result with the shape changed (only our players moved; the Sharks left where they really were): forwards 3 m more separation +1.8; midfielders 12 m higher +2.1;
both +4.0 points per possession (standard error 0.3). Open targets go from 1.0 to 1.6 per possession.
A cruder test, rating the static position itself, agrees in direction: midfielders up +0.9, forward separation +0.6. Setting the forwards running at the ball
lowers that static rating (-1.3). The rating can't see the next kick, so I don't read this as evidence against leading. The evidence for leading is the
retention data in Fig. 8-9.
Figure 13. Rating-tool test: change in the expected value of the best open kick when the Power's exit shape is changed, averaged over 280 real defensive-50 possessions against the Sharks.
Limits of this test: the Sharks don't react in it; a moved Power player still has the Sharks' real positions around them; and the option model values a kick to
where a player is, not to where they are leading. It shows which direction is better and the rough size; it isn't a match simulation.
7. Confidence, and what I could not establish
Conclusion
Confidence
Basis
Our losses come mainly from turnovers inside our own 50 and the Sharks' instant scoring off them
High confidence
Chains reconcile exactly with the scoreboard (504/504 club-games); 20 games; large, consistent gap to the strong clubs
Space at release drives kick retention
High confidence
18,790 disposals, every game, monotonic at every kick length
Our forwards are static and covered; the Sharks cover receivers rather than pressing the ball
High confidence
Tracking at 0.5 s, all 20 games
Midfielders higher plus forwards leading will improve our exits
Moderate confidence
Strong clubs' example, the retention curve and the rating test. Not observed for the Power itself; the Sharks may adjust
Size of the improvement (+4 points of next-goal probability per possession; 8-9 points a game if we halve the gap)
Low confidence
Static counterfactuals; no reaction modelled
Sharks' defenders match by lane rather than strict man-on-man
Moderate confidence
Movement correlations; an alternative reading (loose man-on-man that gets lost) fits some of the data
Set shots over playing on inside 50
High confidence on direction
286 Power forward-50 set plays all season; only 15 against the Sharks
Kick-in change
Moderate confidence
59 of ours v 48 of the strong clubs' against the Sharks
What I could not establish:
The Sharks' instructions. Coaching orders are not in the data, so everything above is inferred from what they do. I can't tell whether their defenders are told
to match by lane or to mark a player. I also can't tell whether they will change their set-up if we change ours. Strong clubs exit well against them, and they keep
the same press in those games (defenders about 65 m from the opponent's goal), which suggests they don't adapt much within a game. That is a weak inference.
Fatigue. Energy is not recorded. I looked at running speeds and found no Sharks-specific fade; I can't measure whether our proposed extra leading would tire our forwards.
GAME_MECHANICS says a player sprinting continuously empties in about 7 s. Leads should be short and timed, not constant.
Cause v correlation for the strong clubs. They play a ball-keeping style and are good clubs. Within-game evidence (the retention curve, the over-the-top
analysis controlled for field position) supports the mechanism, but the game-level correlations (r = -0.4 to -0.6) mix style with quality.
Player selection. Attributes are mirrored between both sides in every game, so there is no personnel edge to find, and the role names say nothing about any individual.
Match simulation. I didn't simulate full games. The engine in _engine/ holds physics only, and I have no model of either club's decision-making to put in it.
The rating tool ignores set plays and energy, and is unreliable for balls in flight, so I used it only on held-ball positions.
Appendix: definitions and method
Coordinates. Every club is analysed in its own attacking direction (x = 140 is the goal it attacks; away games rotated 180 degrees). "Defensive 50" = x < 47, "forward 50" = x > 93.
Possession chain: from a club gaining control (possession, mark, set play, kick-in, free kick) until the other club gains control, a score, or full time. Chain points match the scoreboard in
all 504 club-games; chain turnovers match the provider's 5,958 turnover events exactly.
Kept: the next control event after a disposal is by the same club (or it scores). Space at release: nearest opponent's distance to the landing spot minus nearest teammate's,
measured at the moment of the kick. Open target: a teammate 15-50 m from the ball with no opponent within 6 m.
Strong clubs: Falcons, Magpies, Hawks, Dockers, Cats, Kangaroos, Bulldogs, Titans (each played the Sharks and the Power 4 times).
Code:analysis/build.py (extraction), common.py (chains), disposals.py and context.py (disposal outcomes and tracking context), shape.py,
separation.py, matchups.py, follow.py, fronting.py (structure), options.py, overtop.py, test_exit.py, test_options.py,
test_options2.py (rating-tool tests), figs.py, build_report.py. Working notes: notebook.md.