Episdemos Product Requirements Document (PRD)

Product Name: TRM Swarm Deliberation Platform (Codename: Episdemos)

Target Pilot Launch: Q3 2025

Status: POC Development

To create an anonymous, real-time, AI-assisted platform for collective deliberation among a representative micropublic ("the Swarm"), enabling citizens to express, explore, and revise their views free from identity-based influence or performative distortion. The platform is grounded in political theory and designed to avoid capture, ensure fairness, sustain long-term legitimacy, and promote cognitive participation by ordinary citizens. If a TRM candidate running in the 2026 US Congressional elections were to be elected, they would use Episdemos, powered by a swarm of 500 randomly selected members of their district electorate, to make their voting decisions.  As per the TRM platform, the Swarm is not a consultative or an advisory entity, but the Sovereign whose final decision is binding and non-negotiable, with the Representative possessing no agency of their own.   

• Anonymous, freeform text input by swarm members.
• AI-powered language normalizer removes stylistic cues (slang, idioms, punctuation quirks).
• No persistent handles or identifiers.
• Edge-case prompts and entropy injections introduced periodically to prevent ideological stagnation.

• Real-time generation of:
  • Key arguments for and against the issue.
  • Emerging nuances and edge cases.
  • Shifts in reasoning and sentiment clusters.
• Multiple competing AI summarizers generate diverse interpretations, with divergence surfaced to the swarm.

• Each member of the 500-person swarm is represented by an anonymous point.
• Colors indicate stance: Red = Opposed, Green = Supportive, Blue = Neutral/Undecided.
• Dynamic overlays show dissent density, underrepresented clusters, and temporal vote shifts.

• Users can change their stance at any point.
• Voting UI includes confidence sliders.
• Late-cycle resonance checks using fresh samples to validate swarm decisions before publication.

• System detects and highlights arguments that precede vote shifts.
• Prominently features rare-but-insightful minority positions that exhibit downstream influence.

• Cryptographically enforced non-linkability.
• No persistent identities or behavioral tracking.
• Differential privacy layer masks indirect identity signals (e.g., syntax, cadence).

• Token-based entry for verified swarm members.
• Minimum exposure to arguments before voting.
• Time-weighted influence balancing—frequent participants get reduced per-action weight.

• Longitudinal tracking of past swarm decisions.
• Highlights contradictions or deviations from prior judgments.
• Allows current swarm to reflect on institutional memory without being bound by it.

• Before final submission, users answer prompts like:
  • "Would I accept this policy if I didn't know my social status?"
  • "What might someone very different from me think about this?"

• Measures and visualizes diversity of language, position, and reasoning across swarm.
• Low-diversity scores trigger entropy prompts and contra-exposure mechanisms.

• Voice-based input
• Low-literacy mode with simplified summaries
• Multi-language support with consistent tone normalization

• Personalized 2-minute summary of what's changed since the user’s last visit.
• AI suggestions on where the user’s input is most needed.

• "Statement mode": full deliberation
• "Quick take mode": read summary + vote
• Adaptive routing to under-covered areas of the issue