Online matchmaking looks like it is answering one question about skill. It is actually solving three competing problems simultaneously, and the compromises show up in every match.
Skill is an estimate with uncertainty
Rating systems hold both a value and a confidence level for each player. New accounts have wide uncertainty, so early results move the number sharply.
As more games are recorded, confidence tightens and the rating becomes stable. This is why placement matches feel volatile and later ones feel static.
The rating describes expected outcomes against other rated players, not absolute ability. It only means anything relative to the population currently queueing.
Waiting time competes with match quality
A perfect match requires finding players with closely matched ratings, and the closer the requirement the smaller the eligible pool.
Systems therefore widen their search over time. A player who waits long enough will be matched with a broader spread than the system would ideally choose.
The widening is why matches at unpopular hours or in declining regions feel less balanced. The pool, not the algorithm, changed.
Connection quality is a hard constraint
Distance between players determines latency, and latency affects outcomes more than a moderate skill gap does. A geographically ideal match may be a poor skill match.
Servers are placed in fixed regions, so the system must choose a host location that is tolerable for everyone rather than optimal for anyone.
Parties distort the arithmetic
Groups queueing together are rated as a unit, but coordination gives them an advantage that individual ratings do not capture.
Systems apply adjustments to compensate, which can place a mid-rated group against higher-rated solo players. Both sides then experience the match as mismatched.
Mixed lobbies of parties and individuals are the hardest case, because the same numeric rating means different things depending on who a player is with.
Perceived fairness diverges from measured fairness
A well-calibrated system produces close matches, and close matches are decided by small margins. Losing narrowly feels worse than losing to an obviously stronger opponent.
Players also remember lopsided results more clearly than even ones, so recollection skews toward unfairness even when the distribution is correct.
Systems that were tuned to feel generous instead of accurate tend to inflate ratings over time, which eventually makes every number less informative.