Journal

Experiments · 23 May 2025 · 6 min read

Variable Reward Loops We Refused to Ship

We prototyped engagement patterns common in consumer apps and tested them against our values commitments. Three stayed in the lab, deliberately.

To argue against extractive patterns credibly, we prototype them. Not to ship, to measure. This experiment compared standard growth mechanics against alternatives aligned with voluntary engagement.

We built functional prototypes of streak mechanics with loss framing, notification cascades on inactivity, and social comparison feeds, then ran moderated sessions and seven-day diaries.

Findings

Short-term return behaviour increased in variants A and B. Stated trust and willingness to recommend dropped. Participants described feeling managed, not supported, even when they complied with the behaviour we measured.

Variant C, clear goals, completion celebration without loss framing, opt-in social sharing, showed lower spike metrics and higher seven-day voluntary return without prompts.

Metrics that rise while trust falls are borrowing from tomorrow’s retention to pay for today’s dashboard.

Why we publish failures-to-ship

We will not optimize for engagement at the expense of wellbeing. Publishing what we refuse to build is as important as publishing what we ship, it sets expectations for practitioners who follow our work.

Experiments like this are how research becomes practice: evidence teams can cite when pressure arrives to add just one more nudge.

Published 23 May 2025

All journal