How We Group

Grouping is a judgment call. The algorithm makes it a defensible one — and it runs entirely on your own machine.

The method

Three dimensions, weighted together

Most tools balance one axis at a time. Groups fail for more reasons than that.

Comparative advantage and complementary specialization

Each student is placed where their relative strengths raise the group’s collective output. A student doesn’t have to be the best writer in the class to be the one who should write — only the best writer relative to the others in their group.

Pragmatic and communicative compatibility

Members who can establish common ground and coordinate with each other. A group that can’t talk to itself doesn’t finish, however capable its individual parts look on paper.

Role differentiation and reciprocity

Teams form with balanced, interdependent roles rather than overlapping ones — so the work divides, instead of four students doing the same job and nobody doing the rest.

Two outcomes

Built for growth as well as results

A tool that optimized only for finished projects would have an obvious move: stack the strongest students together and let the rest fend for themselves. The projects at the top would look excellent, and the class would learn less.

SimpleGroup optimizes for both at once — student growth and stronger project results. Groups are composed so skill sets complement each other and each student lands somewhere they can both contribute something and develop something. That balance is the whole design problem, and it is why the grouping can’t reduce to sorting by grade.

Honest about it

What’s proprietary, and what isn’t

The weightings are proprietary

How much each dimension counts, and how they trade against one another, is the part we don’t publish. That is the only thing held back.

The data handling is fully open

What goes in, where it is stored, and what is written back is documented in full and available for a district to review. A closed algorithm is not a reason for closed data practices.

It computes locally

The grouping runs on the teacher’s own device, in code that makes no network access at all. No roster is sent anywhere to be scored.

Every run is a draft

The output is a proposal. Drag students between groups, re-roll with a new seed, or ignore it entirely — nothing reaches Canvas until you confirm the exact plan in front of you.

The input

Your judgment, not a student survey

The algorithm is only as good as what it’s given, and what it’s given is the teacher’s read on their own students — 1–9 trait sliders for things like writing, diligence and leadership preference, alongside the course grade Canvas already holds. It is the same professional judgment you apply forming groups by hand, written down once and reused.

Ratings persist across the year and can be edited as students change, so the second project costs a fraction of the first. And no protected characteristic is collected or used as a grouping criterion — not as a default, and not as a suggestion.

Two minutes, or fifteen

Group off the Canvas grade alone and you have a set in about two minutes. Score a class of thirty on the full trait library and you are still finished inside fifteen — once, for the whole year.

Grouping you can explain to a parent.

If you want to walk through the method in more detail, we’re happy to.