Plus/Minus measures the point differential of a team while a specific player is on the court. It's the simplest possible way to gauge on-court impact: how many more, or fewer, points the team scores than it allows during that player's minutes.
Plus/minus has zero context for who else is on the floor. A player who happens to share most of their minutes with a superstar teammate will inherit a lot of that value in their own +/-, even if their individual contribution is modest.
It's extremely sensitive to sample size and lineup luck. A single hot or cold stretch, a favorable bench-unit matchup, or a garbage-time run can swing a game's +/- by a lot, and over a season, who a player happens to share the floor with (and against) can matter as much as what that player actually does.
This is exactly why the site builds Simul RAPM: it's a statistical adjustment on top of raw plus/minus that isolates a player's impact from their teammates and opponents, instead of taking the raw on-court differential at face value.
Not necessarily. A very good bench player surrounded by four other bench players will often post a worse raw +/- than a mediocre player who shares the floor with stars, simply because of who else is on the court. Raw +/- is a team outcome, not a clean individual measurement.
Adjusted plus-minus models (like RAPM) use possession-level data and statistical regression to separate a player's own impact from the impact of their teammates and opponents, which raw +/- can't do on its own. That's a big part of why the site's Simul RAPM exists as a separate metric.
Both versions exist. A single game shows that game's +/- for the minutes a player was on the floor, while a season total (or per-game average) sums that same on-court differential across every game played.