CHANGELOG
What changed, when.
Civic Filament is in active development. This is the running record of what shipped. If a claim on the site depends on a specific behavior, its provenance is here.
Launch-hardening pass
- ◇Interview transcripts are no longer stored on our servers. They stay only in the participant's browser and are auto-deleted there after 30 days.
- ◇Pattern redaction of emails, phone numbers, SSN-style and card numbers, street addresses and URLs before any text reaches the AI or is stored. Anonymous session IDs are stored only as SHA-256 hashes.
- ◇Synthesis now runs server-side automatically (within ~15 minutes of new perspectives, plus daily). Consensus percentages are counted from participants, not estimated by the model. Minimums: 10 perspectives before publishing; 5 supporters per statement or group.
- ◇Real Merkle-tree provenance: every perspective gets a leaf hash, every version publishes its root and leaves at /api/verify, and participants get a receipt they can check on the LRD page.
- ◇Your Data page: export, withdraw (and rejoin), or delete; deletions trigger regeneration of affected syntheses. In-app account deletion now removes all rows linked to the account.
- ◇Shared database-backed rate limits on all AI and write endpoints plus a global daily AI cap; admin endpoints and scheduled jobs refuse access unless their secrets are configured.
- ◇Dashboard now refreshes every 60 seconds (the SSE stream did not work on serverless hosting). Real pages never fall back to demo data.
- ◇Corrected site copy: removed the 3-run Cohen's-kappa claim, AI-scored DQI claim, "open-source" claim and compliance badges; features that exist only as code are now labeled In Development.
Launch-readiness pass
- ◇Standardized on Civic Filament as the canonical name across the app (retired the Filament: Civics and Consensual Reality Protocol variants).
- ◇Removed illustrative homepage metrics that could have read as adoption evidence; replaced with capability descriptions.
- ◇LiveConsensus visualization now labeled as an illustrative synthesis of the bike-lane demo issue, not a live counter.
- ◇Removed the Google AdSense integration site-wide. Civic Filament is a B2G/trust-first product; ad placements were self-sabotaging on a page selling to procurement.
- ◇Published new pages: /methodology, /participant-rights, /prior-art, /security, /accessibility, /pilot, /changelog, /faq.
- ◇Restructured footer into Protocol / Trust / Deploy / Legal & Connect columns.
- ◇Removed the fabricated aggregateRating from the SoftwareApplication JSON-LD schema.
v0.1.0
2026-05-09Supabase auth + Federal Register cron save
- ◇Sign-up, sign-in, forgot/reset, and email-verification callback wired against Supabase Auth.
- ◇In-app account deletion (at the time, nulled user_id on historical rows; since 2026-09-29 it deletes all linked rows).
- ◇Federal Register cron now persists Executive Orders to Supabase for tier `auto` deployments.
- ◇Header shows SIGN IN vs. ACCOUNT based on auth state.
Supabase backend + civicfilament.org domain
- ◇Provisioned Supabase project (Pro Micro, us-east-2). 16-table schema mirror with RLS, FKs, indexes, updated_at triggers.
- ◇Domain moved to civicfilament.org (apex + www).
- ◇DB-backed route wiring: /api/public/issues, /api/synthesis, /api/lrd, /api/query, /api/verify, plus sync and pipeline routes (since removed).
- ◇/issues page now fetches from /api/public/issues on mount.
Pre-flight pass + Round 2 features complete
- ◇Round-2 modules built: Policy Simulator, NL Synthesis Querying, Cognitive Debiasing and Structured Argumentation are in the product; Spatial Intelligence, Outcome Tracking, Knowledge Gap Steering and Cross-Community Transfer exist as code but are not wired into a user-facing flow; the real-time SSE dashboard did not work on serverless hosting; a Multi-Language Mediator was planned but never built.
- ◇AI optimization pass: 14 call sites on Sonnet 4.6 / Haiku 4.5 with prompt caching. All 10 JSON-parse paths converted to Anthropic tool use.
- ◇PostHog analytics (silent no-op until key set).
- ◇Kokoro TTS read-aloud on blog, mediator, and LRD.
- ◇Three-theme system (Void / Lumina / Cipher).
- ◇Capacitor scaffolding for iOS + Android.
Original 8-phase build complete
- ◇Phase 1 — Foundation: IndexedDB via Dexie, offline-first service layer, service-worker cached app shell.
- ◇Phase 2 — Synthesis engine: 16-D value vectors, DBSCAN clustering, Bayesian Beta-Binomial confidence.
- ◇Phase 3 — Scientific frameworks: Schwartz value tags, Discourse Quality Index heuristic (a multi-run Cohen's-kappa reliability check was prototyped but is not used).
- ◇Phase 4 — AI integration depth: multi-model architecture, adaptive interview length, bias detection.
- ◇Phase 5 — Novel feature modules (not yet wired into user-facing flows): perspective diff, counterfactual synthesis, empathy engine, issue genealogy.
- ◇Phase 6 — Dashboard + LRD: issue comparison, community value fingerprint, versioned LRD with Merkle-root provenance.
- ◇Phase 7 — UX: WCAG AA baseline, mobile voice input, PWA manifest.
- ◇Phase 8 — Production hardening: token accounting, deliberative loop, admin auth.
Related: Feedback & requests · Roadmap