back to labs
lab experiment LAB-002 feedback design · 2025 ai prototype · supabase

emotion-aware
feedback.

Same response. Different emotional state. Completely different coaching. An interactive prototype that adapts what good feedback looks like based on how the learner is actually showing up.

the mechanic

three steps. one insight.

Respond to a real workplace scenario, report how you're feeling, and watch the same response generate completely different coaching across three emotional states.

1

respond

Answer a workplace scenario in your own words — no right or wrong answer.

2

set your state

Slide to where you actually are right now — distressed, okay, or confident.

3

see the difference

Get coaching tuned to all three emotional states simultaneously.

try it

the interactive demo

distressed okay confident
showing up okay
desperation detected. You've submitted this scenario multiple times. The slider has been overridden — this is what the feedback looks like when the system reads repeated struggle as distress, regardless of what you report.
submit your response to see emotion-aware feedback
the thesis

accuracy isn't the only variable.

Most feedback systems optimize for accuracy — is the feedback correct? But accurate feedback delivered to someone in distress lands differently than accurate feedback delivered to someone confident. The emotional state of the learner isn't noise. It's signal. This prototype asks: what if the feedback system read that signal and responded to it?

self-reporting limits

You pick your emotional state manually. A production system would infer it from behavioral signals — response time, revision patterns, session history — without asking.

three states is simplified

Distressed / okay / confident is a useful starting framework, not a complete emotional taxonomy. Real learner states are messier and more contextual.

tone vs. substance

This demo primarily adjusts tone and emphasis. A full system would also modulate content depth, pacing, and what it chooses not to say.

the desperation signal

Submit the same scenario three times and watch the system override your reported state. Repeated struggle is a behavioral signal the slider can't capture.

next in the lab

content ops
dashboard.

What happens when you apply instructional systems thinking to your own workflow — and automate the entire stack with Claude Code.

explore the experiment