Working papers and notes.

• Working paper · WP-2026-01 · July 2026 · v1.1

  The Matchmaker’s Fixed Point: Engagement-Optimized Matchmaking as Performative Prediction

  Adnan Selimović, Ph.D. · Working paper, not peer reviewed.

  Engagement-optimized matchmaking (EOMM) estimates each player’s churn risk, assembles matches that minimize predicted churn, and retrains its churn model on the behavior that follows. The behavior that follows is behavior the matchmaker itself produced. We show that this pipeline is a performative predictor in the sense of Perdomo, Zrnic, Mendler-Dünner & Hardt (2020): the deployed model shifts the distribution it is trained on, and naïve retraining converges to a performatively stable point — a model that is optimal on the data it induces. Stable points and performative optima are distinct solution concepts, and the gap between them can be large. A studio can run a technically sound retraining loop, watch its metrics agree with its model, and be parked at an equilibrium that is suboptimal for engagement itself — and further still from player welfare. We state the mapping, mark its load-bearing assumption, and derive falsifiable consequences a telemetry team could check against its own retraining cadence.

  Read the paper (PDF) ↓ The retraining trap — an interactive companion →

• Latest note · Sep 15, 2026 · 8 min

  Who a Player Is Placed With

  Group constitution, durable retention, and spend that holds — a simulated study design for live-service shooters.

  Read
• Note · Jul 23, 2026 · 5 min

  The Experiment Ends Before the Player Does

  Your A/B calendar is an instrument with a shutter speed — and the churn that costs you most is the one too slow for it to photograph.

  Read
• Note · Jul 22, 2026 · 5 min

  Wrapped

  The year, itemized and handed back.

  Read
• Note · Jul 19, 2026 · 5 min

  The Churn You Can’t See

  A satisfied player and a satiated one look identical on your dashboard — until one of them quietly leaves.

  Read

or write: hello@dividual.studio