The scoreboard is a small sample
A hockey game produces a handful of goals. That is the whole difficulty. In a sport where a typical result rests on five or six scoring events, the outcome of any single night is dominated by which of those events happened to fall, not by which team spent the evening playing better.
This is not a claim that results do not matter — they are the competition — but a claim about what they tell you. If you want to know who is likely to win the next game, the last game's score is a weak signal, and analysts have spent two decades looking for stronger ones.
The search is complicated by the sport's continuity. Hockey has no discrete possessions to count the way basketball does, and no plays to chart the way American football does. It flows, and the analyst has to impose structure on something that does not naturally have any.
Shot-share, and what it actually says
The first useful answer was to count shot attempts. A team that consistently generates more attempts than it concedes is spending more time with the puck in the attacking end, and over enough games that territorial edge shows up as goals. It is a proxy for possession, and a surprisingly good one for something so crude.
It is crude, though. A shot from a bad angle counts the same as one from the slot, and a team can inflate its numbers by throwing pucks at the net from distance while creating nothing. It also flatters teams that trail often, because a trailing team attacks more and a leading team defends the lead.
This is why any honest use of shot-share adjusts for score and situation before drawing conclusions. Raw totals are the beginning of the analysis, not the end of it.
Expected goals and its limits
The refinement was to weight each attempt by how often shots like it become goals: location, type, what preceded it. Expected goals is that idea, and it improves on shot-share because it distinguishes between volume and danger. A team generating fewer but better chances is now visible.
Its limits should be stated as plainly as its uses. The model only knows what it was given, which typically does not include who was standing where, how tired the defenders were, or whether the goaltender's view was blocked. Two shots that look identical to the model can be entirely different events on the ice.
And then there is goaltending, which sits above all of it. A goaltender playing exceptionally can make a badly outplayed team win for weeks, and no measure of chance quality will predict that, because the whole point of an outstanding save is that it defies the base rate.
How to use the numbers honestly
The reasonable position is neither to dismiss possession metrics nor to treat them as the truth behind the scoreboard. They describe territory and chance quality, which are real and predictive over time. They do not describe goaltending, luck or the particular way one team matches up against another on a given night.
Used that way, they answer a specific question well: is this team's record supported by how it is playing? Arena's NHL guides cover the statistics in more detail, along with the points system that makes the standings themselves harder to read than they look.