The classic Math.floor(Math.random() * (hi - lo + 1)) + lo is inclusive on both ends; the + 1 is the part beginners often forget, which silently drops the upper bound from the distribution. Math.ceil(min) and Math.floor(max) normalise non-integer inputs so a caller passing 1.5, 6.5 still draws from {2, 3, 4, 5, 6}. The swap guards against min > max so the function does not silently return only min. Use this for any non-cryptographic case (UI demos, retry jitter, sampling).

Math.random() is fast but not cryptographically uniform: a % range shortcut introduces tiny bias when 2^32 is not a multiple of range. Rejection sampling fixes that by drawing fresh 32-bit integers until one falls inside the largest exact multiple of range (cap), then taking the remainder. The loop iterates more than once with vanishingly small probability, so throughput is essentially identical to a single draw. Use this whenever the value is security-relevant (CSRF tokens, nonces, password resets, raffle picks); use the cheaper randomInt everywhere else.

randomChoice is a one-liner that benefits from an explicit empty-array guard so callers do not silently get undefined from arr[NaN]. sample returns n distinct items via a partial Fisher-Yates shuffle on an index array, which avoids mutating the input and runs in O(n) instead of the naive O(n^2) reject-on-collision approach. Operating on indices keeps the helper non-destructive even when the input is large or shared. For weighted sampling, swap the index pick for a cumulative-weight binary search; the structure stays the same.