Signature stats

Goal anatomy: what tracking data tells us about how goals are scored

Play-by-play tells us a goal happened, where the shot came from and who got the assists. The goal tracking clip tells us how: how fast the shot was, how many passes led to it, and whether the attackers came in with numbers. This page explains how we turn a clip into those numbers and how far to trust each one.

Updated · Method version goal-anatomy v1

More of the TL;DR below

How it's calculated

Each clip is a list of frames, 10 per second. Each frame gives the x/y position of the puck and every player on the ice. We convert the tracking grid (inches) to the same feet coordinates as the play-by-play, then run a few simple detectors.

Derived measures
MeasureHow it is detected
Goal and release framesThe frame the puck enters the net mouth; release is the scorer's last touch before it.
Shot speedDistance from release point to goal ÷ time between the two frames.
Passes and assistsPossession chain: the nearest skater within 6 ft of the puck, with hysteresis so a brief gap doesn't split a possession.
DeflectionA fast incoming puck that turns slightly after a brief touch.
ReboundA fast puck reaches the goalie within 3 s before the shot.
Entry typeCarry, dump or pass, from whether the puck carrier moves with the puck across the blue line.
Odd-man rushAttackers against goal-side defenders at the moment of entry.
Shot speed

speed (mph) = distance from release to goal (ft) ÷ time (s) × 3,600 / 5,280

Example data

Play DNA of an example goal

Example data: invented for illustration

Invented rush goal: a stretch pass, a 3-on-2 entry, a cross-ice pass and an 87 mph shot, 8 seconds from first touch to goal.

0 s1 s2 s3 s4 s5 s6 s7 s8 s9 s2972948 ft22 ftENTRY 3-ON-287 MPHGOALEXAMPLE SCORER · 3-ON-2 RUSHExample game

Play DNA for Example scorer (3-on-2 rush): 3-on-2 entry, 2 passes, 87 mph shot, 2.7 s from entry to goal. Tracking-derived via NHL EDGE.

Invented events drawn with the same glyph used for real goals. Real glyphs carry the credit line shown under the strip; this one is an example.
Research data

How the 77 test goals were scored

Research sample: 200 games from 2025-26 (2025020500–699)

Of 77 tracked test goals, 14 came off odd-man rushes and 14 off turnovers or the forecheck; one came straight off a faceoff.

77 goals from 10 games, 2023-24 to 2026-27. Tracking-derived via NHL EDGE.

Worked example

Timing a shot at 10 frames per second

Includes example data

The release point is 32 ft from the goal and the puck crosses the goal line 0.25 s later (between the 2nd and 3rd frames after release).

  1. 32 ft ÷ 0.25 s = 128 ft per second.
  2. 128 × 3,600 / 5,280 = 87.3 mph.
  3. Because frames are 0.1 s apart, one frame either way changes the time by up to 40% on a shot this short. That is why a single shot can be off by about ±25%, while averages over many goals are accurate.

87.3 mph. For comparison, three of our test goals read 92.2, 85.1 and 87.0 mph against NHL EDGE's official 92.32, 84.63 and 91.83 mph.

Validation on 77 goals

We tested every detector on 77 goals from 10 games between 2023-24 and 2026-27, against the play-by-play, NHL EDGE's official shot speeds and visual checks of each clip.

Accuracy by measure
MeasureResultConfidence
Goal and release frameFound on 75 of 77. Release point within 5 ft of the play-by-play location on 90% (median 1.9 ft).High
Shot speedMatches EDGE's official figures (e.g. 92.2 vs 92.32 mph); about ±25% on a single shot at 10 HzHigh
Primary assist87% correctMedium–high
Secondary assist67% correctMedium
Unassisted goals2 of 5 correctLow
DeflectionRecall 5/6, precision 5/6Medium–high
Rebound91% agreement with play-by-playHigh
Entry typeAbout 80% right on visual check (no official source)Medium
Odd-man rushAbout 10 of 12 right on visual checkMedium
Royal-road pass, screens, defender gap, goalie movementPlausible, not validatedNot shown yet
Power-play formationNot validated; clips are too short (6–9 s of set-up)Not shown
Play-type mix of the 77 test goals
Play typeGoals
Odd-man rush14
Turnover / forecheck14
Cycle / settled12
Rebound10
Rush9
Deflection7
Dump and chase5
Point shot5
Off a faceoff1

What we show, and what we don't

The tracking clips belong to the NHL, and its commercial tracking rights are licensed exclusively. We use the public goal clips to derive numbers and draw our own replays and summaries, and we credit the source on every one: "Tracking-derived via NHL EDGE".

  • Shown: shot speed, pass count, entry type, odd-man count, play type, entry-to-goal time, and the Play DNA strip we draw ourselves from those numbers. Each goal links to the NHL.com video.
  • Animated replays: for every tracked goal since 2023-24 we animate the clip on our own rink at its native 10 frames per second. It shows the ten skaters, both goalies and the puck, with fading trails, a long-exposure still of the puck path, and our derived annotations: entry and odd-man count, pass arcs, the release point, shot speed and the closest defender. A scrubber steps through the roughly 14 s clip frame by frame. Positions between frames are interpolated, so the motion is smoothed, not measured.
  • Where to find them: the goal replays page (/nhl/goals) has the latest night's goals and the season's fastest shots, every game page has a Goal anatomy section with a Replay button per goal, and player pages list each player's goals with replays.
  • Not shown: the raw frame files, and any tracking for shots that did not score. The replays are drawings of the tracked positions, not video; for the broadcast video each goal links to NHL.com.

Tracking also lets us check our play-by-play inferences. For goals we know from tracking whether the play was a rush, a rebound, or a settled cycle, so we can measure how often our transition and rebound flags agree, and publish that agreement rate here.

Reliability

These are measurement accuracies against a reference, not split-half reliabilities: they tell you how often a single goal is labelled correctly.

MeasureSplit-half rSampleVerdict
Rebound flag vs play-by-play91% agreement.—77 goalsStable
Primary assist87% correct.—77 goalsStable
Entry typeAbout 80%.—Visual check, 77 goalsModerate
Unassisted detection2 of 5; not relied on.—5 goalsNoise

Split-half r: we split each sample in two (for example odd and even games) and correlate the two halves across players or teams. Near 0 means the stat is mostly noise; near 1 means it is a stable trait.

Limitations

  • Only goals are tracked publicly, and only since 2023-24. Anything learned from goals describes scoring plays, not all shots.
  • The puck is missing at the start of 8 of 77 clips (up to 3 s), so we interpolate. The tracked point is a player's body, so tips and carries read 5–7 ft off.
  • Clips start about 9 s before the shot, so time in the zone is capped. Line changes break the possession chain. One wrap-around goal read 0 mph.
  • Detectors were tested on 77 goals from 10 games. We will expand the test set and update these numbers as the full archive is processed.

FAQ

Where does NHL goal tracking data come from?

The NHL publishes a short puck and player tracking clip for every goal since the 2023-24 season, as part of NHL EDGE. We derive numbers from those clips and credit them as "Tracking-derived via NHL EDGE".

Is there tracking data for every shot?

Not publicly. Full-game tracking is licensed exclusively and not available to public sites. Only goals have public clips.

How accurate is your shot speed?

It matches NHL EDGE's official readings closely (for example 92.2 vs 92.32 mph and 85.1 vs 84.63 mph), but a single shot can be off by about ±25% because the clip has only 10 frames per second.

Can I watch the tracking replay of a goal?

Yes. Every tracked goal since 2023-24 has an animated replay drawn from the NHL EDGE clip at 10 frames per second: the skaters, goalies and puck on our own rink, with pass arcs, the release point and the shot speed. Find them on the goal replays page (/nhl/goals), in the Goal anatomy section of each game page, and on player pages. The usual caveats apply: one frame either way can move a single shot speed by about ±25%, the puck is sometimes missing for the first seconds of a clip, and positions between frames are interpolated. For the broadcast video, each goal links to NHL.com.