How the rating works
What the number on the rankings board means, and what it does not.
Your rating estimates how much you add to your team's chance of winning. It is worked out from results alone: who was on each side, and who won.
Every game is a piece of evidence. Beating a strong team is stronger evidence than beating a weak one, and losing to a strong team is weaker evidence against you than losing to a weak one. The rating is the number for each player that best explains every result at once.
Roughly 0 is an average rated player. Higher is better. The scale has no ceiling, but in practice almost everyone sits between −1 and +1.5.
Because win percentage cannot tell the difference between a good player and a player with good teammates. Someone who plays every game alongside the strongest people at their center will have a high win rate whether or not they contribute.
Solving for every player simultaneously fixes that. If you consistently win more than your teammates alone would explain, the credit lands on you; if your side wins despite you, it does not. We tested this directly against win percentage, and win percentage was by far the worst method at telling a player apart from their team.
| Factor | Effect |
|---|---|
| Who you beat | The main input. Results against strong opposition move your rating most. |
| Margin of victory | A comfortable win is slightly stronger evidence than a narrow one. The effect is deliberately damped so one blowout cannot define a season. |
| Game type | Competition games count in full, weighted by the round's own multiplier. Social and nightly games count for a quarter — casual rosters make them noisy. |
| Position | Not used. Commanders and Medics are rated on the same scale, since the rating only asks whether your side won. |
| MVP points | Not used — see below. |
This is the most common objection, and we tested it thoroughly rather than assuming. Several versions of the rating were built that shared out credit within a team according to each player's MVP, so a strong individual performance in a loss would cost you less.
Every one of them predicted future results worse than the version that ignores MVP, at every strength we tried. Pushed harder, they became unstable — the ratings drifted apart without limit, because a player who reliably scores well gets pushed up game after game whether or not their team is actually winning.
The fairness concern behind the objection is real, though, and it is handled a different way. Because every player is solved for at once, the rating already accounts for who your teammates were: if you lose alongside weak teammates, the model expected you to lose, and the result barely moves you. That mechanism turned out to protect players from bad teammates better than MVP-sharing did.
MVP remains the right stat for what it was designed for — measuring individual contribution within a game. It is on every scorecard and drives the leaderboards. It is just not a good ingredient for a win-based rating.
The board covers SM5 games from the last 24 months, and you need 50 games in that window to be rated. Older results drop out over time, so the board reflects current form rather than career history.
Guest players — anyone without a Laserforce member ID — cannot be tracked between games, so they are not rated. They are still counted as part of their team's strength, so their presence does not distort anyone else's number.
Laserball is not rated. It is a different game with different stats and would need its own model.
The labelled rules near the top of the board mark slices of the rating scale. They are there to make the scale readable — they are not tiers, and crossing one is not evidence that two players are meaningfully different.
This matters more than it sounds. Ratings are estimates with uncertainty attached, and across the top of the board the typical gap between neighbouring players is several times smaller than that uncertainty. Ranks are printed exactly because a leaderboard needs an order, but the honest reading of #14 and #17 is that they are about the same. Only clear separations — a player well ahead of the field, or a large jump between bands — are real.
- Ratings are only comparable between players connected by a chain of games. Almost everyone is connected through travelling players and tournaments, but a center whose players never face outsiders is weakly anchored to the rest.
- A player with exactly 50 games has a much less certain rating than one with 500. Both appear; the first should be read with more caution.
- The rating says nothing about how you win. A player who wins through positioning and a player who wins through accuracy look identical to it. The scorecards and leaderboards are where that lives.
- It is a measure of results, not of skill in the abstract. Playing well and losing anyway will, over enough games, still show up — but not from any single night.
The method was not picked on preference. Thirty-four candidate models — including Elo, TrueSkill-style ratings, and the MVP-weighted variants above — were run against every recorded SM5 game and scored on how well they predicted the results of games they had not seen.
The model in use won that comparison, and separately did best at telling a player apart from their teammates. It is versioned, so if it is ever replaced, historical numbers stay attributable to the model that produced them.