Small fixes for A/B testing from feedback

[sethbern]

I tested this with the peer instruction grant group and everything worked as expected. The only change is that students who didn't vote still stay in the same group as the people they talked to. Before, they could accidentally get split up which would not be the desired outcome for the experiment.

[Copilot]

Pull request overview

Updates the A/B assignment logic in the peer-instruction pairing endpoint so students remain in the same experimental condition as their recorded in-person (verbal) partners, even when some partners did not submit a vote for the current question.

Changes:

• Build verbal-discussion clusters using the full recorded peergroup membership (not just current answerers).
• Persist experiment assignments for all clustered students (including non-voters), so partners always share a condition.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[bnmnetp]

The grammar fixes are an easy accept I think. The only one I have curiosity about is the double call to grp.add(p ) which just seems like an easy fix also?

[sethbern]

Yeah I agree these are all easy accepts.

[bnmnetp]

Thanks. Will merge now.

The only improvement Claude recommended was to refactor the make_pairs endpoint to use a helper function to do the clustering so that we could write unit tests against it. That may be something to do if we have further issues or the next time we need to refine the clustering.

[bnmnetp]

Or maybe I'll have claude do that and see what it looks like...