Matchmaking in online games is the system that pulls players from a queue, sorts them by a hidden skill number, and forms two teams whose totals roughly match. The system also weighs latency, party size, and how long you’ve been waiting, so a 90-second search usually turns into a balanced lobby. In Overwatch 2, that process runs millions of times a day across every region, and most players never notice the math at all.
Curious why some lobbies feel brutal while others are a breeze? This walkthrough explains the hidden MMR math, weighting factors, and rating shifts that decide who you queue with in competitive games like Overwatch 2.
The Core Purpose of a Matchmaking System
Every matchmaking system chases a small set of goals, and the trade-offs between them explain nearly every odd lobby you’ve queued into. Fairness, speed, and fun rarely align, so the system has to negotiate between them every time you press Play.
Fairness vs. Speed vs. Fun
First, the system pairs you with opponents whose estimated skill sits inside a narrow band, so a silver-ranked Counter-Strike player rarely meets a Global Elite on their first game of the night. A solid matchmaking system in multiplayer games hits four targets at once:
- Skill parity: both sides start with a similar chance to win.
- Low latency: nobody fights rubber-banding teleports caused by 200ms ping.
- Short queues: waits stay under a minute or two during peak hours.
- Engagement: matches feel close enough that you want one more game.
These goals fight each other. A perfect skill match can mean a 20-minute wait at 3 a.m. in a sparse region, so the system loosens its rules the longer you sit in queue. Studios like Riot Games and Valve tune the exact balance for each mode, which is why ranked and casual feel like different products.
How Player Skill Gets Measured and Tracked
Skill lives in a single number most players never see, and that number is the engine of the whole system. Everything else, from your rank badge to your win streak, is a downstream effect of how that number moves.
Hidden MMR and Visible Rank
Your Matchmaking Rating (MMR) is a floating-point value that moves up or down after every match. League of Legends, Valorant, Halo Infinite through Xbox Live, and most Steam-based shooters keep an MMR per queue and convert it into a visible tier only when you load into the client. The visible rank is the UI; the hidden number is the truth the system actually uses.
The Algorithms Behind the Number
Online game matchmaking explained through history starts with Arpad Elo’s chess formula from the 1960s, which the Elo rating system still anchors in many modern games. Studios have since added confidence intervals, team math, and faster convergence on top of it.
| System | Origin | Best At |
|---|---|---|
| Elo | Chess, 1960s | Simple 1v1 ratings |
| Glicko / Glicko-2 | Chess, 1990s–2000s | Tracking rating when a player goes inactive for a while |
| TrueSkill / TrueSkill 2 | Microsoft Research, 2005 | Team games with several players per side, including Halo on Xbox Live |
Glicko adds a “rating deviation” that says how sure the system is about you, so a new account gets wider swings than a player with 500 games on record. TrueSkill models every player on each team as a bell curve and figures out the most likely outcome, which is why it scales so well to 5v5 or larger modes. Microsoft Research originally built TrueSkill for Xbox Live, and it still ships inside Halo’s ranked playlist today.
The Ingredients That Shape Every Lobby
Before the system commits to a match, it weighs several inputs at once, and skipping any one of them turns a fair game into a frustrating one. The order of those inputs matters as much as the inputs themselves.
The Core Checklist
The system runs a real-time filter on the entire player pool with criteria like these:
- Estimated skill: pulled from hidden MMR, not the rank badge on your profile.
- Ping and region: filtered in first so latency never makes it into the lobby.
- Party size: solo, duo, and full squad each get their own matchmaking slice.
- Queue time: the longer you wait, the wider the skill band the system accepts.
- Population at your skill band: thin ranks force the net to widen, even at the start.
Party size is the silent one. A 5-stack in Valorant Ascendant gets matched against a coordinated 5-stack in Ascendant, not five solo Ascendants, and the system adds an “expected win rate” fudge so groups that queue together face slightly tougher opponents. Server regions sit behind all of this, and a player queuing in Tokyo rarely meets a player queuing in São Paulo unless the regional pool runs dry.
Those same regional pools determine how much a rating actually shifts after a result lands.
How Ratings Move After Each Match
After the match ends, the system replays it on paper and updates everyone’s number. The math is what produces the streaks everyone swears they notice, even when the underlying skill is steady.
The Rating Update in Plain Steps
After a match, the system follows a tight sequence that explains every gain or loss in your match history.
- Compare expected vs. actual outcome: the expected win rate comes from the MMR gap, not the scoreboard.
- Apply a multiplier: upsets move ratings further than expected wins.
- Scale by confidence: new accounts swing harder, veterans swing less.
- Apply penalties or decay: leavers, abandons, and seasonal inactivity all trim the result.
- Update visible rank: only when the new MMR crosses a tier threshold.
Streaks, Decay, and Placement
Wins against stronger opponents raise your MMR more than wins against weaker ones, and losses against higher-rated players barely dent it. Streaks are the natural fallout of that math: a few upset wins push your rating up, so the next game expects more of you, and the system hands you a harder opponent. Seasonal ladders also run placement matches at the start of a new act or split, using a short series (often 5 or 10 games) to seed a starting rating rather than resetting everyone to zero.
Decay is rarer than people think. Most shooters only trim MMR after 14 to 30 days of inactivity, and only at the top of the ladder.
Ranked, Casual, and Party Play Compared
Same MMR engine, different dials, and that is the cleanest way to think about the split between modes. Ranked and unranked modes are not separate systems, they are the same engine with different thresholds.
Mode-by-Mode Behavior
| Mode | Skill Band | Queue Strictness | Visible Consequences |
|---|---|---|---|
| Ranked solo | Tight | Strict, solo-only pools | Rank changes, seasonal rewards |
| Ranked party | Tight, blended by group average | Strict, scaled against tougher lobbies | Rank changes, group-based penalties |
| Casual / unranked | Loose | Fast, often cross-region | None, hidden MMR still moves |
| Bot-filled lobbies | Wide | Forced, when player count is too low | None, bots replace missing humans |
Casual matchmaking relaxes the rules on purpose. Studios like Valve and Riot know that a 90-second wait in quick play pushes players back to single-player, so they trade a bit of fairness for throughput. Solo queue is the cleanest test of skill because the system refuses to mix you with parties, while duo or squad queues blend ratings across the group and inflate the expected skill of the opposing team. How players get matched in online games depends almost entirely on which dial the studio tightens for that mode.
That dial-tuning also explains why some lobbies feel lopsided the moment they start.
Why Matches Sometimes Feel Broken
Even a tuned system produces bad nights, and the reasons are usually a mix of population, psychology, and product decisions. Most “broken” matches are actually the system working exactly as designed, just against goals you cannot see.
Five Common Causes
When a match feels off, one of these is almost always the culprit, and most of them are working as intended.
- Smurfs and banned players returning on new accounts slip past skill filters until they climb enough games to settle.
- Low off-hour population forces the system to widen its net, especially above Diamond or below Bronze.
- Engagement-optimized matchmaking (EOMM) prioritizes churn and session length over pure 50/50 fairness, and several large publishers have published research on it.
- Perception bias makes close losses feel unfair while close wins feel deserved, even when the math is identical.
- A high-ping teammate or opponent turns an otherwise balanced match into a stutter-fest.
EOMM deserves its own line because it is the most controversial. The idea is that a player who just lost five straight will stick around longer if their next match is winnable, and a player on a hot streak will tolerate a tougher game if it stays close. The math optimizes for retention, not ledger fairness, and that is why your ranked games can feel like a roller coaster even when your MMR is steady. Skill based matchmaking (SBMM) sits underneath all of this as the baseline, and EOMM layers on top of it.
What to Take Away From the System
Matchmaking is a single decision made from skill, latency, party size, and queue time, then sanity-checked against engagement metrics. Hidden MMR is the real engine, with visible ranks acting as a friendly summary, and small rating swings after each game are the math doing its job. A few practical moves, like checking your region, playing during peak hours, and queuing as a consistent party size, can noticeably improve match quality. Mostly the way the inputs say it should, and the way you queue decides which inputs dominate. Once you see the inputs, the queue stops feeling random and starts feeling like a negotiation you can actually win.
FAQ
What algorithm do games use for matchmaking?
Most studios build matchmaking on top of Elo, Glicko-2, or Microsoft’s TrueSkill, then layer in ping, party size, and queue time. The skill algorithm is just one input; the final match is the output of all four working together.
How is matchmaking rating calculated?
A baseline number kicks off every player’s matchmaking rating, shifting after each match according to how far the real result drifts from what the system predicted. Upsets swing the number more than expected results, and a confidence value keeps new accounts volatile while veterans stay steady.
Why does matchmaking take so long?
Long queues usually mean a thin player pool at your skill band, a strict region filter, or a ranked mode with tight MMR bands. The system widens its search the longer you wait, so a slow queue almost always ends in a match that feels slightly off.
Is matchmaking based on skill or connection?
Both, but connection is usually the hard filter. Studios drop any candidate whose ping would make the match unplayable, then pick the closest skill match from whoever survives the cut.
How do games prevent smurfing in matchmaking?
Studios combine phone or email verification, behavior detection, hardware bans, and a hidden MMR that takes a few games to settle into a stable rank. Smurfing still happens, but the rating math usually catches up within five to ten matches.
What is skill based matchmaking?
Hidden MMR numbers act as the main ingredient that skill based matchmaking stirs into every lobby it builds. It shows up in nearly every modern shooter and ranked game, including Overwatch 2, Counter-Strike 2, and Halo Infinite.
