Deployment and context
How we detect lines, defence pairs and power-play units from shifts
Line combinations on most sites are typed in from practice reports. Ours are measured: we know who was on the ice for every second of every game, so we can tell who actually played together, and how consistently.
Updated · Method version lines v1
How it's calculated
- Build segments: split each period at every shift change and goal, so each segment has a fixed set of players on the ice.
- For each segment, add its length to every forward trio, forward pair, defence pair and five-skater power-play group on the ice.
- Forward lines: rank candidate trios by shared 5v5 time, then pick four disjoint trios greedily, repairing with a local swap when pairwise time says a different split fits better. Players under 10% of team 5v5 time are spares.
- Defence pairs: the same method with pairs, three per team.
- Power-play units: group the five-skater sets used on the power play, merging sets that share at least four skaters so a one-player swap does not split a unit. PP1 has the most power-play time and PP2 the next. Members are the players with at least 40% of the unit's time.
- Deployment tier: a player's tier is the most common one over his last five games (top six, middle six, bottom; top four or third pair), with the share of those games as its confidence.
confidence = unit TOI together ÷ average 5v5 TOI of its members (capped at 1)
- 1.0
- the three players only ever played together
- below 0.5
- the unit was often broken up
PP share = player power-play TOI ÷ team power-play TOI, same games
One team's forward lines from a single game
Example data: invented for illustration
Four detected forward lines: the first line took 31% of 5v5 time with confidence 0.82, the fourth 14% with confidence 0.41.
Worked example
A first line's confidence
Includes example dataA first line spent 11:40 (700 s) of 5v5 time together. Its three forwards played 14:10, 14:20 and 14:00 at 5v5 (850, 860 and 840 s).
- Average member 5v5 TOI: (850 + 860 + 840) ÷ 3 = 850 s.
- Confidence: 700 ÷ 850 = 0.82.
- The other 150 s per player were spent in blended combinations, such as after penalties or late in a game when the coach shortens the bench.
Confidence 0.82: a stable line. Third and fourth lines are usually lower because they absorb the blending. The TOI figures are illustrative.
Checks on the shift data
- On-ice skater counts from shifts match the strength code the NHL attaches to each event on 99.2% of events and 99.7% of shots.
- Shift charts round to whole seconds, so line changes can briefly show six skaters or four. Segments of two seconds or less with an impossible count take their strength from the nearest event; the players still get their ice time.
- Shift charts sometimes contain rows from another game (game 2025020565 had 667 rows from a Vegas–San Jose game). We drop any row whose team is not in the game and log it.
- Per-player ice time is checked against the boxscore after every game.
Reliability
Line detection is a measurement of what happened, so the question is accuracy, not stability. We plan to publish agreement with hand-labelled lines; for now each unit carries its confidence score.
| Measure | Split-half r | Sample | Verdict |
|---|---|---|---|
| On-ice counts vs NHL strength code99.2% of events, 99.7% of shots. | — | 200 games, 2025-26 | Stable |
| PP1 vs hand-labelled truthAgreement rate to be published. | — | Fixture games | Not measured yet |
| Deployment tierShown with its share of games. | — | Last 5 games | Moderate |
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
- Lines are measured after the game. Injuries, call-ups and practice changes can make tonight's lines different from the last game's.
- Coaches blend lines when trailing, after penalties and in overtime; low-confidence units often reflect that, not a data problem.
- A player who kills penalties and plays on the power play may look like a spare at 5v5 if his even-strength time is small.
- A small number of games have missing or broken shift charts. We fall back to the NHL's HTML time-on-ice reports, and when those fail too, the game gets no line data.
FAQ
Where do your NHL line combinations come from?
From the NHL's official shift charts. We rebuild each game second by second and find the trios, pairs and power-play groups that shared the most ice time.
How do you decide who is on PP1?
PP1 is the group of skaters with the most power-play time together in a game, after merging groups that differ by one player. A player's PP1 rate is the share of recent games in which he was on that unit.
What does a line's confidence number mean?
The share of its members' 5v5 ice time that the exact trio spent together. Near 1 means a set line; below 0.5 means the coach was mixing it up.
Why do your lines differ from other sites?
Most sites post projected lines from practice. Ours show what was actually used in games. When a coach changes lines mid-game, our numbers show the time each combination really got.