AWR — Research

The Literature Behind This Report

What It Says to Do, and What Each Point Means Here

A February review on using tracking data to quantify load. Six takeaways are actionable, and the hockey physiology literature supports the same direction (Vigh-Larsen & Mohr 2024).

The review labels its own exposure metric exploratory and its thresholds pragmatic rather than calibrated. A direction of travel, not a settled dose.

What We Can Estimate Today

Edge publishes season totals, not per-game detail. Most of the below is recovered by comparing a player against himself.

Every player is compared against his own baseline. Every column is labelled measured or modelled.

What the Public Feed Cannot Give Us

Properties of the feed, not gaps in the model. Published alongside the numbers.

Where the Load Concentrates, and Where the Rest Sits

84 games, 2 October to 10 April. Days rest is the only schedule effect that survives testing in our own work, so it is what the calendar is coloured by.

The two peaks are in different months. March is heaviest by games — 16, with three B2B and no break longer than two days. December and January are heaviest by travel, at 12,358 and 13,596 km with ten time-zone shifts between them. February is the recovery month: 10 games and a nine-day break from the 13th, the largest window in the season. The longest single leg is 3,695 km to Los Angeles on 6 January.

On the travel figures. Distance is great-circle between consecutive game venues, so it assumes the team flies directly rather than returning home mid-trip. Air time is block time estimated as 36 minutes plus distance at 780 km/h; legs under 50 km are treated as ground travel. Time-zone shifts count games where the venue offset moves by an hour or more, taken from the league feed rather than assumed.

Travel did not survive as a load driver. Distance flown in the previous seven days explained 0.006% of the variation in skating output, weaker than a simple home-or-away control, and home B2B showed a larger difference than road B2B — the opposite of what a travel mechanism would predict. It is included here because it shapes scheduling and staffing decisions, not because the data supports treating it as load.

October 2026 12 games 3 b2b

SMTWTFS

1

2@—

3@0

4

5

6

7vs3

8

9vs1

10

11vs1

12

13@1

14vs0

15

16

17vs2

18

19

20vs2

21

22

23

24vs3

25

26

27

28vs3

29@0

30

31

November 2026 12 games 1 b2b

SMTWTFS

1vs2

2

3vs1

4

5

6

7vs3

8

9

10@2

11

12@1

13

14vs1

15

16

17@2

18

19vs1

20

21@1

22

23

24

25vs3

26

27vs1

28@0

29

30

December 2026 14 games 1 b2b

SMTWTFS

1vs2

2

3@1

4

5@1

6

7@1

8

9

10@2

11

12@1

13

14

15@2

16

17vs1

18

19vs1

20

21@1

22

23

24

25

26vs4

27

28vs1

29

30@1

31@0

January 2027 14 games 2 b2b

SMTWTFS

1

2vs1

3

4vs1

5

6@1

7

8@1

9

10@1

11

12vs1

13

14vs1

15

16vs1

17

18@1

19vs0

20

21

22vs2

23

24@1

25

26

27

28

29@4

30@0

31

February 2027 10 games 3 b2b

SMTWTFS

1

2vs2

3@0

4

5

6

7

8

9

10

11

12

13@9

14vs0

15

16vs1

17

18

19vs2

20

21vs1

22

23vs1

24

25@1

26@0

27

28

March 2027 16 games 3 b2b

SMTWTFS

1@2

2

3vs1

4

5@1

6

7@1

8

9vs1

10@0

11

12

13@2

14

15vs1

16

17vs1

18@0

19

20

21@2

22

23vs1

24@0

25

26vs1

27

28@1

29

30vs1

31

April 2027 6 games 1 b2b

SMTWTFS

1@1

2

3vs1

4vs0

5

6@1

7

8@1

9

10@1

11

12

13

14

15

16

17

18

19

20

21

22

23

24

25

26

27

28

29

30

What the Tiers Mean

The tiers used in the next two sections are each club's own depth chart, ranked by ice time. D1 is a club's most-used defenceman, D6 its least; F1 to F12 the same for forwards. They are not labels we assign — ice time separates the slots cleanly enough to define them, accounting for 90% of the variation in a forward's ice time and 80% of a defenceman's.

How to read these two charts. Each shows average ice time per game, in minutes, for every slot in the league. The gold line joins the median player in each slot last season. The two shaded bands around it show how much players in that slot vary: the darker band covers the middle half, the lighter band the middle 80%. The blue lines behind are the four previous seasons, so a slot whose lines sit on top of one another is one that behaves the same way every year. The small figure under each slot label is how many players held it. Hover or tap any slot for its full spread.

What they show. Both lines fall steadily, and the bands stay narrow — a club's third defenceman plays much the same minutes as any other club's third defenceman. The blue seasons sit almost on top of the gold, so this is stable year to year rather than a feature of last season.

And the difference between the two. An F12 plays 0.48× what an F1 plays; a D6 plays 0.61× what a D1 plays — but across six slots rather than twelve.

Defencemen Cover Less Ground per Minute on the Second Night

Measured against a one-day turnaround rather than a rested one, since three quarters of the season runs on nought or one day. Cohort is 39,997 player-games across five seasons.

Output and deployment move in opposite directions. Ground covered per minute falls on the second night while ice time and shift length rise, in every group. The top four defencemen show the largest drop of any tier in the league, and their minutes hold.

A note on what this measures. Metres per minute is ground covered for each minute on the ice, not skating speed. A player can reach the same top speed and still cover less ground.

Last Season Measured, Next Season Projected

On a second night a defenceman loses about twice as much ground as a forward. Chychrun covered 711 m less than his one-day rate last season; Wilson, the most affected forward, 335 m less. Both dressed for all fourteen.

Per second night played that is 43 m for a defenceman against 18 m for a forward. Nine D account for 3,340 m in total, sixteen forwards for 3,003 m.

Three defencemen dressed for all 14. Chychrun, Roy and Fehervary played every second night last season. Carlson dressed for nine of the fourteen. The table is last season as measured: Carlson, van Riemsdyk and Chisholm have since left the roster.

Projected onto Next Season

Fourteen second nights again, four of them at home. Each returning defenceman's own minutes per second night applied to next season's count.

Three of last season's D have left, and three current D have no baseline. John Carlson, Trevor van Riemsdyk and Declan Chisholm are off the roster, so they are excluded from the projection. Dylan McIlrath, Vincent Desharnais and Jacob MacDonald are on it but have no Washington second-night history to project from, so they do not appear above.

Which makes these figures a floor, not a forecast. Carlson alone accounted for 24.2 minutes per second night, the highest on the team. Those minutes do not disappear — they are absorbed, most likely by the returning top four, who already carry the highest cost of any tier in the league. The realistic exposure for Chychrun, Roy, Fehervary and Sandin is above what is shown here.

How to read the ground-covered figure. Exposure is measured per player — games dressed on a second night and the minutes played in them. The per-minute difference applied to those minutes is a league estimate for that player's position tier, not a per-player measurement. Individual load tolerance does not replicate season to season (r = 0.02), so a per-player fade rate would be noise. Thirteen games inside the 28-day window after the Olympic break are excluded.

And the size of it. The effect runs at roughly 1% of a player's normal output, concentrated on 14 identifiable dates. The schedule is fixed; the roster is not.

What More Data Would Add

Three tiers, and they are different requests. Counts say how often a player went hard. A continuous record says what he was doing.

The third tier already exists. The league has captured player position at up to 15 times a second to roughly 2 cm, every player, every game, since 2021-22. The question is access.

One line to add to the request already with the league. Does the club feed include the underlying tracking data, meaning position through a game, or only summary metrics? At what capture rate, and can it be retrieved per shift? The rest of that request asks for more summaries; this asks for a different kind of data.

What Practice Data Adds, and How to Use It

Catapult in practice and league tracking in games share no unit and no session that both observe. That rules out adding their raw outputs together. It does not rule out a combined weekly load, which is the point of collecting practice data in the first place.

The index already does exactly this job. Ice time, metres per minute, sprint rate and contact have no common unit either. They are combined by standardising each against the player's own distribution and weighting the result. A practice series joins on the same basis, as a fifth input rather than a separate report.

• Validate the modelled change-of-direction column. It is derived from shift structure with nothing to check it against. Practice movement data is that check. Either it holds or we change it.
• Test an assumed constant. The position rate we ship is disclosed as unmeasured. Practice can show whether splitting by position is the right shape, or whether the variation is individual. Deployment tier explaining about 3.5% of the mechanical side suggests the latter.
• Size the gap per player. Where a speed-based and a direction-aware measure describe the same session, the difference says how much the speed view understates that player. A stable trait, and it differs by skating style.
• Give each player a movement profile. How much work comes from turning versus striding, and left-right evenness.

The weighting is a choice, not a finding. How much a practice minute counts against a game minute cannot be settled by the data. It is a structural decision, the same kind as the penalty-kill volume weight already carried in the index. Stated openly it is defensible; buried in a formula it is not.

One boundary that does hold. A combined load works for our own players, week to week, against their own baselines. It cannot be percentiled against the league, because no other club's practice data exists in the feed. So the combined number is an internal management tool and the league-comparable index stays game-only.

One assumption. Practice movement is more repetitive and mostly less intense than game movement, so a relationship fitted in practice may not hold in games. Scrimmage is the exception, running above official match intensity at 27.3% higher distance per minute and 22.3% higher decelerations (Front Sports Act Living 2022).

Notes on the Practice System

The hockey suite covers most of what the research asks for: around 45 metrics from inertial sensors, including skating load, strides, work bouts, work-to-rest ratio and skating symmetry. A bout is movement held above an intensity threshold for a set duration, which is the review's exposure idea under another name.

• The inertial side resolves direction. Accelerometer and gyroscope combined with gravity removed, reporting magnitude and direction regardless of unit orientation. Closer to what the review asks for than anything speed-based.
• Skating symmetry is the metric to ask about first. Instrumented-skate work puts the outside skate at 50 to 70% of body weight through a change of direction against 12 to 24% inside (Fortier et al. 2014).
• Trust the inertial path over the positioning path. An independent validation of their indoor positioning found position accurate to 0.21 m and distance error under 2%, but instantaneous speed off by 35% or more even in an ideal setup, concluding speed from raw positioning data is not valid (Luteberget et al. 2018). A different system tested in hockey reached the opposite conclusion on speed (Gamble et al. 2023), so each system needs validating in its own setting.
• Anchor placement is decided once. In the same study, an ideal versus poor layout moved position error from 0.21 m to 1.79 m, and distance error from under 2% to under 30%.
• A ready-made hockey-stop detector is not available. Studies classifying skating movements above 90% accuracy use skate-mounted sensors with per-player training.

Questions worth answering early. Does the setup give positional data or inertial only, since two of the four uses above need a speed-based and a direction-aware measure on the same session. Then: raw access or derived metrics only, how skating symmetry is calculated and whether it distinguishes turn direction, and what anchor layout the practice rink would achieve.

What to Do With This

What This Brief Can and Cannot Carry

• Turn loading comes from a small laboratory study. Eight players, instrumented skates, a controlled 90-degree manoeuvre. The direction is clear; the exact figures are not a population estimate.
• Nothing on the site is PK-specific yet. The derivation works on distance; it is not yet a published column.
• Nothing here requires testing the athletes. Where the framework needs a capacity measure we do not have, this says so rather than substituting a weaker proxy.