Where models contribute
AI updates leader news, aggregates leader feedback and supports comment moderation. Opinion classification and organization remain planned rather than presented as live behavior.
Experiment 04 / Demopulse / 2024–26
A civic-feedback product tested whether people could follow public leaders, receive current context and express a structured view without turning the product into another news feed.
01 / Hypothesis
The starting question was whether a serious direct-democracy MVP was technically feasible. The product moved from leaders to news discussion and back toward leaders, exposing a harder issue: flexibility without a clear anchor can weaken the product’s purpose.
02 / Smallest loop
Follow a public leader.
Receive a weekly digest of current public facts.
Rate the leader and add qualitative feedback.
Aggregate feedback without enabling direct reactions to news.
Current context enters. Structured civic feedback leaves.
Core mechanismNews provides context; the leader remains the feedback anchor.
Control boundaryIdentity and moderation improve accountability while increasing adoption cost.
03 / Division of labour
AI updates leader news, aggregates leader feedback and supports comment moderation. Opinion classification and organization remain planned rather than presented as live behavior.
Users choose whom to follow and what feedback to submit. I remain accountable for the product boundary, identity requirements, moderation rules, privacy trade-offs and production decisions.
04 / Product evidence
These production screens show the feedback proposition and a leader-centered news digest. They demonstrate the live interface boundary without claiming adoption or engagement.


Reality check / governance narrowed the design space
By July 2026, Apple had rejected the app eight times. Review requirements pushed the product toward accounts plus application-wide blocking and hiding. That improved moderation robustness while increasing adoption friction and the amount of sensitive political data the service had to hold.
Whether leader-centered feedback is a strong enough anchor to justify the identity and moderation cost.