EOMM, Matchmaking Research and Gaming Patents: What the Evidence Actually Shows

SBMM Lobby Epistemological Standards Notice

SBMM Lobby strictly distinguishes academic research publications, patent disclosures, and verified commercial implementations in live games.

A peer-reviewed research paper demonstrates theoretical feasibility within a controlled mathematical simulation. A granted patent legally protects an inventive concept from competitors. Neither a research paper nor a patent grant establishes that a specific commercial video game currently uses the described technique in live production.

1. Introduction: Why Matchmaking Research Attracts Controversy

Over the past decade, no topic in competitive gaming has generated as much passionate debate, player frustration, and viral speculation as matchmaking algorithms. In gaming communities across Reddit, YouTube, and X/Twitter, players regularly attribute unexpected loss streaks, sudden ping spikes, or lopsided lobbies to mysterious algorithms designed to manipulate their psychology.

At the center of this controversy sits an acronym that has become infamous in gaming circles: EOMM, or Engagement-Optimized Matchmaking.

Content creators frequently point to academic papers and corporate patent filings as "indisputable proof" that modern multiplayer games are engineered like slot machines, dynamically rigging outcomes to maximize playtime and microtransaction spending.

However, a serious investigation requires a disciplined evidentiary standard. We must separate three entirely distinct categories of evidence: 1) Peer-reviewed academic research, 2) Patents granted by government patent offices, and 3) Verified commercial implementations in live production code.

Confusing one category for another is the primary source of online misinformation. This article provides an objective, line-by-line examination of the primary source documents—the 2017 UCLA/EA research paper and three major United States patents—to determine what the evidence actually shows, and what it cannot prove.

2. The Evidence Classification Matrix: Strict Epistemological Boundaries

To navigate matchmaking claims objectively, every technical assertion must be evaluated through a disciplined evidentiary hierarchy.

┌────────────────────────────────────────────────────────────────────────────────────────┐
│                        THREE SEPARATE CATEGORIES OF EVIDENCE                          │
├────────────────────────────┬────────────────────────────┬──────────────────────────────┤
│    1. ACADEMIC RESEARCH    │    2. PATENT DISCLOSURES   │   3. VERIFIED DEPLOYMENT     │
│   (Peer-Reviewed Models)   │    (Legal Property Claims) │  (Empirical Live Production) │
├────────────────────────────┼────────────────────────────┼──────────────────────────────┤
│ • WWW '17 (Chen et al.,    │ • US10610786B2 (EA):       │ • Official publisher white   │
│   UCLA / EA research team) │   Retention-Driven Plan    │   papers (e.g. CoD 2024)     │
│ • Offline simulation on    │ • US11845005B2 (EA):       │ • In-game developer menus    │
│   36.9M historical matches │   Game Quality Factor (GQF)│ • Requires direct, verified  │
│ • 8-hour churn threshold   │ • US11478715B1 (EA):       │   first-party technical      │
│ • Tests churn hypotheses   │   User-Controllable Embeds │   confirmation               │
│   in theoretical model     │ • Legal right to exclude   │ • Never established solely   │
│                            │   competitors from methods │   by research or patents     │
├────────────────────────────┴────────────────────────────┴──────────────────────────────┤
│ CRITICAL GOVERNANCE RULE:                                                              │
│ A research paper or patent grant DOES NOT prove a specific commercial game runs EOMM. │
└────────────────────────────────────────────────────────────────────────────────────────┘

3. What Conventional SBMM Attempts to Accomplish

To understand EOMM, we must first understand the system it sought to challenge: Skill-Based Matchmaking (SBMM).

Conventional SBMM systems—whether based on the Elo rating system from chess, the Glicko algorithm, or Microsoft TrueSkill 2—are grounded in the philosophy of fairness: every player is assigned a numerical skill rating (e.g. MMR) representing their estimated probability of defeating other players, and the matchmaker pairs participants of nearly identical skill to achieve an approximately 50% win probability for each team.

While fair on paper, conventional SBMM has faced notable player critiques:

• The "Sweat" Fatigue: High-skill players report that every casual match feels as exhausting as a tournament grand final, eliminating room for experimental playstyles.
• The 50% Win-Rate Plateau: For many players, an inescapable 50% win rate feels like an artificial treadmill rather than organic progression.
• Session Burnout: Players often stop playing immediately following a series of grueling, high-stress matches or demoralizing consecutive losses.

4. What EOMM Actually Proposes: The 2017 Academic Framework

In April 2017, a team of researchers from Northeastern University, UCLA, and Electronic Arts presented a paper at the 26th International World Wide Web Conference (WWW '17) in Perth, Australia titled "EOMM: An Engagement Optimized Matchmaking Framework" (Zhengxing Chen, Su Xue, John Kolen, Navid Aghdaie, Nicholas Meltzer, Magy Seif El-Nasr, Kazi A. Zaman, Yizhou Sun; ACM DOI 10.1145/3038912.3052559).

The authors proposed a radical paradigm shift: What if the objective of a matchmaker is not to maximize fairness, but to maximize player retention?

CONVENTIONAL SKILL-BASED MATCHMAKING (SBMM):
Objective: Equal Probability of Winning (Fairness)

   Player Pool ───► Filter by Latency ───► Minimize Skill Delta (|Skill_A - Skill_B| ≈ 0)
                                                    │
                                                    ▼
                                           Balanced Match (50% Expected Win Rate)

─────────────────────────────────────────────────────────────────────────────────────────

THEORETICAL ENGAGEMENT-OPTIMIZED MATCHMAKING (EOMM):
Objective: Minimum Aggregate Churn Probability (Retention)
[Evaluated in Offline Historical Simulation — Not Deployed in Live Production]

   Player Pool ───► Predict Churn: P(quit | Win/Loss/Draw history)
                           │
                           ▼
   Construct Bipartite Matching Graph (Weights = Predicted Disengagement)
                           │
                           ▼
   Solve Minimum-Weight Perfect Matching (Kuhn-Munkres Algorithm)
                           │
                           ▼
   Lobby Optimized for Continued Play (Fairness is Subordinated to Session Length)

5. How Churn Prediction Works in the Research Model

At the mathematical core of the Chen et al. framework is a predictive model that calculates the probability that a player will close the game session (churn) based on the outcome of their next match.

The authors analyzed an offline historical dataset of 36.9 million matches across 1.68 million unique players from a popular commercial 1v1 PvP title.

Crucially, the paper defined churn using a specific operational rule: a player is considered to have churned if they do not start another match within 8 hours following the conclusion of their current match.

Using empirical log data, the researchers calculated the conditional probability of a player quitting after their next match across a discrete outcome triad (Win, Loss, Draw):

• Loss Streak Vulnerability: A player who had suffered three consecutive losses had an extraordinarily high probability of churning if their fourth match was also a loss.
• Turnaround Win Recovery: However, if that same player was given a win on their fourth match, their probability of continuing the gaming session increased dramatically.
• Win Streak Resilience: Conversely, a player on a three-game win streak had a very low churn probability; they could absorb a loss without quitting the game session.

6. Understanding Graph-Based Player Matching (Kuhn-Munkres Algorithm)

To assemble lobbies around churn probabilities, the paper formulates matchmaking as a graph-theoretic optimization problem:

1. Graph Construction: For 2n players waiting in queue, a complete bipartite graph is constructed where vertices represent players and edges represent potential pairings.

2. Edge Weight Assignment: The weight assigned to an edge represents the predicted joint probability of disengagement: weight = P(churn_i | outcome(i, j)) + P(churn_j | outcome(j, i)).

3. Minimum-Weight Perfect Matching: The matchmaker objective is to find a global pairing across all players that minimizes the sum of all edge weights.

4. Algorithmic Solution: The authors solve this optimization problem using the Kuhn-Munkres algorithm (the Hungarian algorithm) for 1v1 matchmaking, and extend it with heuristic approximation algorithms for team-based multiplayer lobbies.

In simulation tests on historical telemetry, the authors demonstrated that the EOMM framework achieved a statistically significant reduction in player churn compared to random matchmaking, pure skill-based matchmaking, and latency-only matching.

7. Critical Limitations of the 2017 Academic Study

While internet commentators frequently cite the paper as a blueprint for modern gaming, the study possesses major academic and practical limitations that are rarely discussed:

• Offline Simulation, Not a Live Test: The paper evaluated its algorithm on an offline historical dataset of 36.9 million matches. The authors simulated what would have happened using static probability distributions; they did not deploy the system live on production game servers.
• Stationary Probability Assumption: The model assumed player churn probabilities remain fixed over time. In reality, humans rapidly detect artificial pattern manipulation. If players realize matches are handed to them as "pity wins," trust collapses, leading to catastrophic long-term churn.
• Zero Measurement of Competitive Integrity: The paper made no attempt to measure player sentiment, competitive legitimacy, or community satisfaction. It measured a single metric: did the player play another match within 8 hours?
• Explicit Academic Boundary: The authors concluded the paper by stating that EOMM was a proposed conceptual framework, acknowledging that practical production deployments would require balancing ethical considerations, skill progression, and network latency.

8. What Electronic Arts Matchmaking Patents Actually Disclose

Beyond the academic paper, public scrutiny has focused intensely on three granted United States patents assigned to Electronic Arts Inc. Our independent legal and technical audit disambiguates these three separate filings:

Patent 1: US 10,610,786 B2 — "Multiplayer video game matchmaking optimization" (Inventors: John Kolen, Navid Aghdaie, Kazi Atif-Uz Zaman, Su Xue, Meng Xian; Filed: March 8, 2016; Granted: April 7, 2020). Direct legal counterpart to the 2017 research paper, describing retention-driven matchmaking optimization.

Patent 2: US 11,845,005 B2 — "Game quality-centric matchmaking for online gaming" (Inventors: Navid Aghdaie, Kazi Atif-Uz Zaman, John Kolen, Su Xue, Meng Xian, Michelle Brandwein; Filed: March 8, 2021; Granted: December 19, 2023). Focuses on "virtual match packing" and evaluating a Game Quality Factor (GQF) factoring in latency, skill parity, and wait times to prevent half-empty cloud server allocations. This is an infrastructure efficiency patent, not an EOMM churn-rigging patent.

Patent 3: US 11,478,715 B1 — "User-controllable model-driven matchmaking" (Inventors: Meng Wu, Qilian Yu, Harold Henry Chaput, Navid Aghdaie, Kazi Atif-Uz Zaman, Kenneth Alan Moss; Filed: February 12, 2020; Granted: October 25, 2022). Outlines machine-learning embedding spaces with user-controllable parameters, allowing players to steer whether they prioritize queue times, skill parity, or social play.

9. Comprehensive Comparison of EA Matchmaking Patents

The following matrix details the audited legal, technical, and architectural distinctions across the three major Electronic Arts matchmaking patents:

10. Research vs. Patents vs. Live Implementation: What the Evidence Cannot Prove

When analyzing these documents, gamers and commentators frequently make logical leaps that violate basic legal and scientific standards.

In enterprise technology, corporations file hundreds of defensive patents annually to create legal moats, document speculative R&D explorations, and build corporate asset portfolios. A patent does not equate to production code.

To evaluate commercial practice accurately, analysts must strictly separate three distinct, non-interchangeable categories of evidence:

• Category A (Academic EOMM Research): The 2017 UCLA/EA paper investigated pre-match queue lobby assembly to minimize session churn using historical match triads. As documented, this was an offline simulation on historical telemetry, not a live production deployment, and made no claim of altering in-game physics or gameplay.
• Category B (EA Sports Litigation — Zajonc v. Electronic Arts Inc.): In November 2020, plaintiffs filed a federal class-action lawsuit in the Northern District of California (Zajonc v. Electronic Arts Inc., Case No. 3:20-cv-07871), alleging EA secretly used patented Dynamic Difficulty Adjustment (DDA) technology (such as US Patent 10,065,117) to manipulate player attributes in FIFA, Madden NFL, and NHL Ultimate Team modes to induce pack purchases. On February 11, 2021, plaintiffs filed a formal Notice of Voluntary Dismissal pursuant to Federal Rule of Civil Procedure 41(a)(1)(A)(i) (Document 22). The case was dismissed WITHOUT prejudice. A voluntary dismissal without prejudice under Rule 41(a)(1) does not constitute a judicial finding, court order, or court-verified determination of the merits. In a corporate announcement on March 3, 2021 (Fair Play and Dynamic Difficulty Adjustment, ea.com), Electronic Arts stated that it provided plaintiffs with technical information and engineer access, which EA stated confirmed to plaintiffs that DDA and scripting are not used in Ultimate Team modes. This explanation represents EA corporate account, not a judicial finding. Furthermore, DDA (in-game gameplay adjustment during a live match) is fundamentally distinct from EOMM (pre-game queue lobby assignment).
• Category C (Respawn Developer Disclosures — Apex Legends): On January 17, 2023, Respawn Entertainment published an in-depth technical developer blog (Apex Legends Matchmaking Update, ea.com) explaining the design principles of its matchmaking engine. Addressing retention optimization directly, the developers stated: "3. Is matchmaking built to directly optimize retention & engagement? No. Our matchmaking algorithm is only concerned with measuring skill and arranging the fairest possible matches in a reasonable time... We never build an algorithm that directly optimizes for retention or engagement." This statement specifically describes Apex Legends; it should not be over-generalized as an explicit denial of every theoretical formulation across all historical EA titles.

11. Common Myths & Misconceptions Regarding EOMM

Myth 1: "EOMM changes your bullet damage and aim-assist mid-game to make you lose."

Reality: EOMM, as defined in both the academic paper and the patent portfolio, is strictly a lobby assembly algorithm operating in the pre-game queue by deciding which players are grouped together. It contains zero mechanisms for altering in-game physics, weapon spread, or damage values during live combat.

Myth 2: "EOMM guarantees that you will win one game after every three losses."

Reality: The Chen et al. paper never proposed a simplistic win-loss script. It formulated a probabilistic model evaluating aggregate churn risk across an entire server queue. Human multiplayer games involve millions of chaotic player actions, aiming variances, and connection hiccups that cannot be predetermined with mathematical certainty.

Myth 3: "All competitive multiplayer games secretly use EOMM."

Reality: Different game publishers maintain radically different engineering architectures. Activision 2024 Call of Duty matchmaking white papers explicitly documented an 8-factor hierarchy dominated by Connection Latency (Delta Ping) and recent form, explicitly detailing why connection quality is prioritized over behavioral retention models.

12. Frequently Asked Questions

Here are verified answers to the most common questions regarding EOMM research, EA gaming patents, and multiplayer matchmaking claims:

13. Conclusion: The Real Takeaway from Matchmaking Research

The 2017 UCLA/EA academic paper and associated patents represent fascinating, groundbreaking milestones in computer science and game design theory. They proved mathematically that matchmaking is not merely an engineering problem of finding 10 players with low ping—it is a behavioral system with profound impacts on player psychology and retention.

However, as informed gamers and researchers, we must maintain intellectual honesty: 1) We can acknowledge that retention-based matchmaking models exist and have been patented, 2) We can recognize that publishers have a massive commercial interest in keeping players engaged, but 3) We must never confuse theoretical academic research or patent applications with confirmed live deployment.

Until a publisher publishes an official engineering disclosure confirming EOMM deployment—or third-party independent audits analyze server matchmaking code directly—the claim that any specific game runs on EOMM remains an unproven theory.

Explore related research: Read our breakdown of How Matchmaking Ping Prioritization Works and our technical guide to EOMM vs. SBMM Matchmaking Frameworks. Benchmark your recent form using the SBMM Calculator or review our independent Editorial Policy.