Random groups vs balanced groups vs constrained groups
Use the least complicated method that can produce an acceptable result. Random grouping needs the least data; balance adds a composition preference; constraints add rules whose violation would make the assignment unusable.
Guide visual
Choose the right grouping mode
From quick random splits to rule-aware schedules.
Combine only the layers the outcome needs
These are not mutually exclusive modes. A scenario can use valid capacities, hard requirements, balance preferences, and random variation together.
Select the smallest sufficient model
Define an acceptable result
Write one sentence describing what must be true. If it mentions only group sizes, start random.
Add one composition goal if needed
If the sentence mentions role, topic, or another measurable composition concern, add one balance preference.
Encode actual requirements
Add fixed, together, apart, attendance, or capacity rules only when violating them would make the assignment unusable.
Choose the interface
Quick Setup handles straightforward lists and common options. Use the Scenario Editor for per-session attendance, detailed capacities, numeric sums, soft relationship preferences, or objective tuning.
Review in the correct order
Check attendance and capacity, then requirements, then preferences. Do not use a good preference score to excuse an invalid schedule.
Simplify when possible
If random grouping would have been acceptable, remove unnecessary columns and rules. Less data and fewer assumptions are easier to explain.
The evidence changes with the grouping model
- For random groups, the useful evidence is the roster itself: assigned count, capacity, and who ended up together. Randomness is not evidence that the result is fair.
- For balanced groups, the Attribute Balance or Attribute Sum card shows where the requested count or range was missed. That detail is more informative than the aggregate penalty alone.
- For constrained groups, required fixed, together, and apart rules report “No violations” in a valid result. A violation is a solver defect, not an acceptable trade-off produced by optimization.
- For multi-round groups, the histogram describes the contact distribution and the matrix identifies the exact pairs behind it.
- Saved alternatives can be compared against the original outcome statement, but Results cannot decide whether the extra fields and rules were justified in the first place.
Compare the building blocks in GroupMixer
Product routes open the current saved scenario; the Help topic explains how the rule categories differ.
Worked example: three layers in one workshop
The preloaded example has 18 people in three groups of six, with six people from each track. A clean 2 Strategy / 2 Design / 2 Technical distribution is possible, but together, apart, and fixed-assignment requirements must be satisfied first.
Setup facts
- three groups of six
- one categorical track preference
- one together pair and two apart pairs
- one person fixed to each group
Review the result
- Base assignment: all groups have six people.
- Requirements: every hard relationship and fixed assignment holds.
- Preference: track counts approach 2/2/2 without invalidating the first two layers.
Try this setup in GroupMixer
This tool is preloaded with the example from this guide. You can edit the participants, constraints, sessions, and balance settings before generating groups.