the joint distribution
Most interesting questions involve more than one quantity at once. How tall AND how heavy is a person? How much rain fell today AND how much yesterday? Knowing each quantity on its own is not enough — you also want to know how they move together. The joint distribution is the single object that answers every question about a whole collection of random variables at the same time.
Formally, for random variables X and Y, the joint distribution assigns a probability to every region of the (x, y) plane: P((X, Y) is in some set A). It tells you P(X is in some range AND Y is in some range) for all ranges at once. Think of it as a probability spread over a two-dimensional map instead of along a single line: more probability sits where the pair is more likely to land. From this one map you can recover everything — the behaviour of X alone, of Y alone, and of one given the other.
The joint distribution carries strictly more information than the two separate distributions. Two pairs of variables can have the very same individual behaviour yet completely different joint behaviour — perfectly tied together, or moving independently, or anything between. That extra information about how the variables relate is exactly what dependence, covariance, and correlation try to summarise.
Roll two fair dice. The pair (first die, second die) has a joint distribution: 36 equally likely outcomes, each with probability 1/36. P(first = 2 AND second = 5) = 1/36. The joint distribution of (first, sum) is different: knowing the first die constrains which sums are possible, so the two coordinates are NOT independent there.
The same single die can take part in two pairs with very different joint behaviour.
The two individual (marginal) distributions do not pin down the joint distribution — many different joints share the same margins. You cannot recover the joint from the margins alone.