Team Randomizer

No. There are no usage limits on GroupMixer.

No. You do not need an account or sign-up. Paste names, adjust the setup, and generate groups right away.

In your browser on this device. Your participant list and all other data you enter stays on your device while you work.

Yes. After the page has loaded, you can keep using GroupMixer offline in your browser.

Yes. Add attributes such as role, skill, gender, or department to your participants, then choose which attribute to balance across groups.

Yes. Use Pinned people in the advanced options to pin specific people to a specific group across all sessions. This is useful for leaders, presenters, or anyone who must stay in a known group.

Yes. Set the number of sessions and enable "Avoid repeat pairings" to cut down on repeats across rounds.

Import a Zoom pre-assign CSV or a spreadsheet with participant emails, choose the number of rooms and sessions, keep "Minimize repeat pairings" enabled, then generate groups. Use Share & Export → Download Zoom breakout CSV. For multiple sessions, GroupMixer downloads a ZIP with one Zoom CSV per round; upload the right CSV in Zoom’s Breakout Room pre-assign screen before that round. Zoom matches people by email address, so make sure participants join with the same email.

Yes. Use the scenario editor for partial attendance. You can set which participants attend each session before generating the schedule.

Yes. Use the scenario editor to set custom group capacities, including different capacities for different groups or sessions.

Yes. Use Group Visits in the scenario editor to set exact, minimum, or maximum visit counts for selected people and selected groups across selected sessions. This is useful for station rotations, task exposure, and visit caps.

Use the scenario editor for controls this page does not expose, such as partial attendance, custom capacities by group and session, session-specific rules, weighted preferences, Pair Encounters, Group Visits, advanced tuning, solver settings, previous results, and result analysis.

The solver minimizes a weighted score. The main terms are repeat-encounter penalties, attribute-balance penalties, weighted soft-constraint penalties, and a reward for unique contacts. Lower is better: repeat and constraint violations increase the score, while new unique pairings reduce it.

GroupMixer does not start from a purely random grouping. It first tries a bounded constraint-scenario/oracle-guided constructor. That constructor builds a hard-valid scaffold, uses pure Social Golfer structure where it helps reduce repeats, projects that structure back into the real scenario, and validates the result before search begins.

The dense search engine repeatedly proposes local changes to the current schedule. Each candidate change is previewed against the scoring model before the optimizer decides whether to keep it, so the search can improve the schedule while preserving hard validity.

Repeat reduction starts before local search. GroupMixer tries to construct a strong social-golfer-style contact pattern, using exact or near-exact zero-repeat structures when they match the scenario. After that, dense search tracks pair-contact counts, rewards first-time contacts, penalizes repeat excess, and improves the schedule without breaking requirements.

Requirements are rules GroupMixer must satisfy to return a schedule. These include attendance, group capacity, Fixed Placements, Keep Together, and Keep Apart. Preferences are scored goals: violating them is allowed, but it increases the score. These include Repeat Limit, Balance Attributes, Attribute Totals, Prefer Together, Prefer Apart, Pair Encounters, Group Visits, and the unique-contact objective.

If GroupMixer finds a score of 0, that indicates optimality for the active scoring model: there are no remaining scored violations or repeat penalties to improve. For many constrained scenarios, however, the true optimum may be greater than 0 because the requirements conflict or the design is mathematically impossible. GroupMixer is not designed to prove that a positive score is globally optimal, and for many complex scenarios that proof is completely infeasible in practice.

No. Requirements are treated as feasibility rules, not suggestions. GroupMixer first tries quality-oriented constructors; if those cannot build a schedule, it can fall back to a SAT-based feasibility constructor. If the SAT solver proves the requirements unsatisfiable, GroupMixer fails instead of relaxing them. If a satisfying schedule is found, GroupMixer uses it as the starting point and continues optimizing preferences and other quality objectives.

Please report it through the feature request page. Include the scenario shape, the GroupMixer result, the better schedule, and how you scored it. Cases where another method beats GroupMixer are especially useful because they can become regression tests or benchmark cases for improving the solver. Send a feature request.

The search uses randomized exploration unless a deterministic seed is supplied. Different runs can therefore find different high-quality schedules with the same score or similar scores. In the scenario editor, advanced users can set a seed when reproducibility matters.