A Blog

Estimating Lateral Puck Movement Before Goals

2026-06-07T14:01:24+00:00

Introduction

Moving the puck from one side of the other gets the goalie moving and can make it easier to score. So I wanted to see how much lateral puck movement there is right before a goal. To do so, I used both goal tracking data and shot location data from NHL.com’s API. Combining the two, I was able to isolate the parts of the tracking data right before the shot.

The code for the analysis can be found here.

Data

One part of the data I used was the goal tracking data. I used this data previously when clustering power play goals and when coming up with criteria for rush goals. The tracking data comes from NHL.com’s goal visualizations.

The other data I used was the shot location data provided by NHL.com’s API. Specifically, I used the play-by-play endpoint:

https://api-web.nhle.com/v1/gamecenter//play-by-play

Since this shot data uses a different coordinate system, I had to convert the shot locations to use the same coordinate system the tracking data uses.

Pre-Processing the Data

While the tracking data from the goal visualizations uses a coordinate system where the rink is about 2400x1015 units, the shot locations from the API are given in feet. Also, the y-axis is flipped between the two. So I converted the shot location coordinates by flipping the y-axis and converting to the tracking data’s coordinate system.

Another part of pre-processing the data is finding the original shot location for tips and deflections. In the play-by-play API data, tip and deflection goal locations are recorded where the tip or deflection happened. I wanted to use the location of the original shot when calculating lateral puck movement to avoid counting the shot’s lateral movement. That meant I had to find the original shot using the tracking data. For each tip and deflection goal, I went backward through the tracking data, starting at the end, to see if the puck either changed direction above a threshold or changed speed above a separate threshold. The rationale being a puck shouldn’t change direction or speed too much directly after being shot.

Here is one example of a deflection goal:

The green dot is the location of the tip, and the blue dot is the estimated location of the original shot. The green dot doesn’t line up with the puck tracking data, shown as the black lines, because the play-by-play data uses a coarser coordinate grid.

I checked the original shot location logic by looking at a dozen or so deflection goals and adjusted the direction and speed thresholds as needed.

Since the coordinate system used by the play-by-play API is coarser than the coordinate system used by the tracking data, I had to also pre-process non-deflection goals to find the location of the original shot in the tracking data.

To do so, I looked for the spot in the tracking data closest to the location given by the play-by-play data that fell under a distance threshold. Using a distance threshold helped filter out goals where the tracking data didn’t match the goal. For example, this goal by Brandt Clarke against the Golden Knights on October 8, 2025 has tracking data where the last part is incorrect. I excluded the goals where the tracking data and the play-by-play’s shot location were too far apart.

Results

With the shot locations found in the tracking data, I then calculated the lateral puck movement before the shots for the 2025-2026 regular season goals. Empty-net goals were omitted. I tried looking at windows of different lengths of time before the shot.

Below are histograms of the amount of lateral puck movement before goals:

Unsurprisingly, as the time window before the shot increases, the amount of lateral puck movement increases as well.

The median and mean amounts of lateral puck movement:

Seconds before Goal	Median Lateral Puck Movement (Feet)	Mean Lateral Puck Movement (Feet)
0.88	9.70	12.77
1.25	15.08	17.93
1.62	20.65	22.97

Example Goals

An example of a goal that has 4 feet of lateral puck movement in the 1.25 seconds before the goal:

Jesper Boqvist against the Blackhawks on October 7, 2025

An example of a goal that has 15 feet of lateral puck movement in the 1.25 seconds before the goal:

Nick Schmaltz against the Jets on April 14, 2026

An example of a goal that has 65 feet of lateral puck movement in the 1.25 seconds before the goal:

Matt Boldy against the Mammoth on February 27, 2026

Conclusion

I was expecting goals to have more lateral puck movement than they actually do since I often heard how moving the puck east-to-west is an effective way to score goals. Perhaps I just have a bad sense of scale.

One follow-up analysis I might do in the future is to see how much vertical puck movement there is before goals.

Defining Rush Goals with Puck Tracking Data

2026-05-04T12:07:18+00:00

Introduction

One of the analyses I wanted to do with the publicly available goal tracking data is to see how many goals each team scores off the rush. In order to do that, I had to define criteria for rush goals. That process involved iteratively setting some tentative criteria, reviewing the results, and updating the criteria. At the end of it, I ended up with three criteria that I used to calculate how many rush goals each team scored in the 2025-2026 regular season.

You can find the code I created for this analysis here.

Data

The data I used was the public goal tracking data for the 2025-2026 regular season. This data is the same data that is used for NHL.com’s goal visualizations. Here is one example goal visualization.

For this analysis, I did the same pre-processing steps I had used when looking at goal similarity. This includes trimming the starts and ends of the tracking data to account for things like faceoffs.

I filtered out empty net goals from the data using NHL.com’s API to get empty net goal data.

One caveat: some goals are missing data, but this is a very small percentage of goals (< 0.1% of all goals).

To keep it simple, I only used puck tracking data and not the player tracking data. In the future, I may revisit the criteria to include the player tracking data.

Process

The process I used to come up with criteria to define rush goals was:

Set tentative criteria
Review a sample of goals that were close to being considered rush or non-rush
Update the criteria based on that review and repeat the process

Throughout this process, I reviewed around 60-80 goals in total.

Criteria

The final criteria are:

The puck traveled at least 67 feet along the length of the ice
The puck spent less than 40 timesteps in the offensive zone
- A timestep is the interval in between data points in the tracking data.
- In a future analysis, I hope to calculate how much time this interval is.
The puck spent less than 9 timesteps beyond the goal line

Results

Using the above criteria, about 29% of non-empty net goals were off the rush in the 2025-2026 regular season.

Some example goals defined as rush goals by the criteria:

Here are the number of rush goals by team:

Team	Number of Rush Goals
CAR	89
PIT	86
UTA	85
BUF	84
EDM	80
LAK	77
MTL	75
ANA	73
STL	73
TBL	73
CBJ	71
NYR	71
CGY	70
MIN	69
TOR	69
NYI	68
PHI	68
VGK	68
NJD	66
WPG	66
COL	65
SJS	65
DET	64
FLA	62
VAN	61
SEA	59
OTT	58
BOS	58
WSH	58
CHI	57
DAL	56
NSH	45

Here are the percentages of goals that were rush goals by team (excluding empty net goals):

Team	% of Goals that are Rush Goals
LAK	37%
CGY	36%
STL	34%
UTA	34%
BUF	33%
NYR	32%
NJD	32%
CAR	32%
PHI	31%
CBJ	31%
NYI	31%
WPG	31%
PIT	31%
EDM	30%
TOR	30%
VAN	30%
SEA	29%
ANA	29%
DET	29%
CHI	28%
MTL	28%
TBL	28%
MIN	27%
FLA	27%
SJS	27%
VGK	27%
WSH	25%
COL	24%
OTT	23%
BOS	23%
DAL	21%
NSH	20%

Conclusion

While the criteria I developed seem to work well, I may in a future analysis try to refine them by including player tracking data in addition to the puck tracking data. I would also like to dive deeper into rush goals as well.

Follow-Up to 24-25 Winnipeg Jets Power Play Goals: Finding Goals That Follow the Same Play

2025-11-19T21:20:18+00:00

Introduction

After writing the previous post about clustering 24-25 Winnipeg Jets power play goals, I saw a Dallas power play goal that was similar to the largest cluster of Jets power play goals.

That made me wonder how common this kind of power play goal is, where the puck gets moved from the half-wall to the goal line and then to the slot. So I did a quick analysis to find those kinds of goals in the 24-25 season.

Results

The Winnipeg power play goals I wanted to find similar goals for are:

I was able to find 13 power play goals from the past season similar to the Winnipeg goals:

Approach

To find power play goals similar to the Winnipeg ones, I used the same hyperdimensional computing encodings of the goal tracking data used to cluster Winnipeg power play goals.
That let me sort goals by similarity to the Winnipeg goals. Then I used the NHL API to filter on just the power play goals. After that, I manually reviewed a handful goals to set a similarity threshold, and those that met that threshold were all reviewed. In all, I reviewed around 30 or so goals. You can find a reference for the NHL API here.

Example Python code:

from dataclasses import dataclass
import json
import polars as pl
import numpy as np
import requests

@dataclass
class Goal:
    game: int
    goal: int
        
@dataclass
class GoalDist:
    game_id: int
    goal_id: int
    dist: float


def dist(x: np.ndarray, y: np.ndarray) -> float:
    """
    Calculates the distance between 2 binary hypervectors
    
    PARAMETERS:
        - x (numpy array): binary hypervector of shape (d,)
        - y (numpy array): binary hypervector of shape (d,)
    
    RETURNS:
        - dist (float): Hamming distance between x and y
    """
    if x.shape != y.shape:
        raise ValueError(f'x and y have different dimensions: {x.shape} and {y.shape}')
    
    dim = x.shape[0]
    return np.count_nonzero(x!=y)/dim


# read in exported goal hypervectors that have been created by trimming
# the tracking data to the last 17 instances
df_goal_hvs_last_17 = pl.read_parquet('goal_hvs.parquet')

# sort goals by similarity to this Winnipeg one
specified_game_id = 2024020587
specified_goal_id = 1063

specified_goal_hv = df_goal_hvs_last_17.filter(
    (pl.col('game_id')==specified_game_id) &
    (pl.col('goal_id')==specified_goal_id)
).select(
    pl.col('goal_hv')
).to_series().to_list()[0]
specified_goal_hv = np.array(specified_goal_hv)

goal_dists = []

for row in df_goal_hvs_last_17.filter(
    (pl.col('game_id')!=specified_game_id) &
    (pl.col('goal_id')!=specified_goal_id)
).rows(named=True):
    other_game_id = row['game_id']
    other_goal_id = row['goal_id']
    other_goal_hv = np.array(row['goal_hv'])
    
    goal_dist = dist(specified_goal_hv, other_goal_hv)
    
    goal_dists.append(GoalDist(game_id=other_game_id, goal_id=other_goal_id, dist=goal_dist))

l_sim_goals = sorted(goal_dists, key=lambda item: item.dist)

# print out info for the most similar ones
NUM_PDS = 3

threshold = 0.141

for goal_dist in l_sim_goals:
    if (goal_dist.dist <= threshold):
        r = requests.get(f'https://api-web.nhle.com/v1/gamecenter/{goal_dist.game_id}/landing')
        landing_str = r.content.decode()
        dict_landing = json.loads(landing_str)

        for i in range(NUM_PDS): # ignoring OT pp goals   
            l_dict_goals = dict_landing['summary']['scoring'][i]['goals']

            # go through each goal in the game to find the similar one
            for dict_goal in l_dict_goals:
                event_id = dict_goal['eventId']
                strength = dict_goal['strength']
                scoring_team = dict_goal['teamAbbrev']['default']

                if (strength == 'pp') and (event_id == goal_dist.goal_id):
                    print(f'Team: {scoring_team}')                    
                    url = f'https://www.nhl.com/ppt-replay/goal/{goal_dist.game_id}/{goal_dist.goal_id}'
                    print(f'Game: {goal_dist.game_id}, goal: {goal_dist.goal_id}, dist: {goal_dist.dist}, url: {url}')

Here is a visualization of the paths the puck took in the similar goals:

Each separate line represents a different goal. The arrows show the direction the puck moves in. The large overlap between the different goals highlights the similarity between the goals.

Conclusion

When the question, “How common are power play goals like this one?” popped into my head, the goal encodings I created using the tracking data made it simple to find out. They helped narrow down the goals I had to review to just a few dozen goals.

Clustering Winnipeg 24-25 Power Play Goals

2025-11-15T20:48:18+00:00

Introduction

I was interested in seeing if I could cluster NHL goals using goal tracking data. Clustering is a technique to group similar items together. For a first experiment, I decided to look at a team’s power play goals since that is a smaller number of goals to work with and because I thought there were be more similar plays on the power play.

So I took a look at Winnipeg’s 2024-2025 power play goals since they had the number one power play last regular season.

Data

Last season, the Winnipeg Jets had 62 regulation power play goals and one in overtime. I focused on just the regulation power play goals. To get the game and goal id’s of their power play goals, I used the NHL’s API as documented by Zmalski. Specifically, I used the club schedule season and game landing endpoints.

I already had the 24-25 regular season goal tracking data downloaded, so I could just use the game and goal id’s of the Jets power play goals to filter the tracking data to just those goals.

Then I encoded the goal tracking data into vectors using the approach I took to find similar goals. If you’re interested in the details, check out the repo. This encoding only uses the puck tracking data, not the player tracking data.

Results

The resulting clusters frequently have mostly similar goals. However, there are some clusters where the goals are pretty different from each other. In all, 22 clusters were found. Below is a graph of the cluster sizes:

Each bar represents one cluster.

Here is a visualization of the biggest cluster found, which had seven goals:

The lines are the paths the puck took, and the arrows show the direction the puck headed in. Arrows that are farther apart show the puck moving faster.

With many goals shown, it’s a little cluttered, but it’s possible to see the puck moving from the dot to near the goal line. In some goals, a Winnipeg player passes to the slot, while in others, the Jet player attacks the net from the goal line. One goal that differs from the others is the goal shown in blue, which is Scheifele’s goal against the Sharks.

Update: I made a follow-up post that searched for power play goals similar to the first four goals in this cluster.

Links to replays of the goals in this cluster:

Here is a visualization for the second biggest cluster, which had six goals:

Most of the goals in this group come off a rebound. However, Vilardi’s goal against the Flames and Ehlers’s against the Sharks aren’t. The Ehlers goal, in blue, especially stands out as being different against the others.

Links to replays of the goals in this cluster:

Below is one example of a smaller cluster that has dissimilar goals:

None of the goals in this cluster are particularly similar to each other, showing that the clustering doesn’t always produce great results.

Links to replays of the goals in this cluster:

To check out the goals in the other clusters, please take a look at the appendix.

Approach

In order to cluster the power play goals, I had to first create vector representations for them. To do so, I used hyperdimensional computing, which uses high-dimensional random vectors to represent data. You can find more details on this vector representation of goals here. The repo also has the code to create those representations.

One slight tweak I had to make here is to use just the last 17 instances of each goal’s tracking data. During some early experiments, I noticed that the earlier instances in the tracking data were throwing off the results. Most goals had around 90-100 instances in their tracking data after trimming some instances to account for things like data before faceoffs or data after the puck enters the net.

I tried several different numbers of instances to take from the end of the data: 10, 15, 17, and 20, with 17 working the best.

For the clustering algorithm itself, I chose to use agglomerative clustering. One main reason is because I didn’t want to force goals that were dissimilar from all the other goals into a cluster. Agglomerative clustering starts with each goal in its own cluster. Similar clusters then get combined into bigger ones.

Since I used agglomerative clustering, I had to choose the linkage used. The linkage determines how clusters get combined into bigger ones. I tried single, complete, and average linkages, with complete ending up doing the best. Complete linkage looks at two clusters and then takes all the possible pairs where one goal is from one cluster and one is from the other. The distance between these two clusters is defined as the largest distance between two goals in those pairs. Then the two clusters with the smallest distance between them gets merged if that distance falls under a threshold. After trying several different thresholds, I found that a threshold of 0.3 worked the best. For details on the other linkages, refer to this page.

To perform clustering, I used Scikit-Learn’s implementation. For those unfamiliar, Scikit-Learn is a Python machine learning library. You can find the Python script I wrote for clustering here. If you run the clustering, you may end up with slightly different results due to the randomness used when encoding goals as vectors.

Appendix

Here are all the clusters found by running clustering on Winnipeg’s 2024-2025 regular season regulation power play goals. Note that the cluster labels are arbitrary.

Cluster 0

Cluster 1

Cluster 2

Cluster 3

Cluster 4

Cluster 5

Cluster 6

Cluster 7

Cluster 8

Cluster 9

Cluster 10

Cluster 11

Cluster 12

https://www.nhl.com/ppt-replay/goal/2024020143/619

Cluster 13

Cluster 14

Cluster 15

Cluster 16

Cluster 17

https://www.nhl.com/ppt-replay/goal/2024020647/953

Cluster 18

Cluster 19

https://www.nhl.com/ppt-replay/goal/2024020099/683

Cluster 20

Cluster 21

https://www.nhl.com/ppt-replay/goal/2024020452/967

Team	Number of Rush Goals
CAR	89
PIT	86
UTA	85
BUF	84
EDM	80
LAK	77
MTL	75
ANA	73
STL	73
TBL	73
CBJ	71
NYR	71
CGY	70
MIN	69
TOR	69
NYI	68
PHI	68
VGK	68
NJD	66
WPG	66
COL	65
SJS	65
DET	64
FLA	62
VAN	61
SEA	59
OTT	58
BOS	58
WSH	58
CHI	57
DAL	56
NSH	45

Team	% of Goals that are Rush Goals
LAK	37%
CGY	36%
STL	34%
UTA	34%
BUF	33%
NYR	32%
NJD	32%
CAR	32%
PHI	31%
CBJ	31%
NYI	31%
WPG	31%
PIT	31%
EDM	30%
TOR	30%
VAN	30%
SEA	29%
ANA	29%
DET	29%
CHI	28%
MTL	28%
TBL	28%
MIN	27%
FLA	27%
SJS	27%
VGK	27%
WSH	25%
COL	24%
OTT	23%
BOS	23%
DAL	21%
NSH	20%

Team	Number of Rush Goals
CAR	89
PIT	86
UTA	85
BUF	84
EDM	80
LAK	77
MTL	75
ANA	73
STL	73
TBL	73
CBJ	71
NYR	71
CGY	70
MIN	69
TOR	69
NYI	68
PHI	68
VGK	68
NJD	66
WPG	66
COL	65
SJS	65
DET	64
FLA	62
VAN	61
SEA	59
OTT	58
BOS	58
WSH	58
CHI	57
DAL	56
NSH	45

Team	% of Goals that are Rush Goals
LAK	37%
CGY	36%
STL	34%
UTA	34%
BUF	33%
NYR	32%
NJD	32%
CAR	32%
PHI	31%
CBJ	31%
NYI	31%
WPG	31%
PIT	31%
EDM	30%
TOR	30%
VAN	30%
SEA	29%
ANA	29%
DET	29%
CHI	28%
MTL	28%
TBL	28%
MIN	27%
FLA	27%
SJS	27%
VGK	27%
WSH	25%
COL	24%
OTT	23%
BOS	23%
DAL	21%
NSH	20%

Team	Number of Rush Goals
CAR	89
PIT	86
UTA	85
BUF	84
EDM	80
LAK	77
MTL	75
ANA	73
STL	73
TBL	73
CBJ	71
NYR	71
CGY	70
MIN	69
TOR	69
NYI	68
PHI	68
VGK	68
NJD	66
WPG	66
COL	65
SJS	65
DET	64
FLA	62
VAN	61
SEA	59
OTT	58
BOS	58
WSH	58
CHI	57
DAL	56
NSH	45

Team	% of Goals that are Rush Goals
LAK	37%
CGY	36%
STL	34%
UTA	34%
BUF	33%
NYR	32%
NJD	32%
CAR	32%
PHI	31%
CBJ	31%
NYI	31%
WPG	31%
PIT	31%
EDM	30%
TOR	30%
VAN	30%
SEA	29%
ANA	29%
DET	29%
CHI	28%
MTL	28%
TBL	28%
MIN	27%
FLA	27%
SJS	27%
VGK	27%
WSH	25%
COL	24%
OTT	23%
BOS	23%
DAL	21%
NSH	20%