The questions we get from
buyers, before they buy.
If a question you have isn’t here, ask via the contact page and we’ll add it (with the answer).
DATA OWNERSHIP & RECORDS LAW
Who owns the interview data?
The participant owns their interview — and it stays with them: transcripts are kept only in the participant's browser and are not stored on our servers. The deploying organization owns the aggregate LRD outputs. Civic Filament processes the extracted perspective summaries, not as an owner. The pilot / production contract names all three roles explicitly.
Is a resident's interview subject to public records law?
It depends on your jurisdiction and how you structure the deployment. See the Records-Law Posture section on the Participant Rights page. The short version: every governmental deployment starts with a written Data Handling Addendum that names the retention schedule, the FOIA posture, and what participants are told before consenting.
Can you export the data if we leave?
Yes. Machine-readable JSON (and Markdown) export of every published LRD version, including its provenance hashes. Raw interview transcripts are not stored on our servers, so there is no transcript corpus to export; participants can export their own data from the Your Data page. No lock-in.
AI METHODOLOGY
Is the AI mediator a fine-tuned MI/NVC model?
No. It is a general-purpose LLM (Anthropic Claude Sonnet 4.6 for the interview and extraction, Haiku 4.5 for value-profile ratings) operating under a versioned system prompt using Motivational Interviewing (OARS) and Nonviolent Communication (observation → feeling → need → request) techniques. Full mediator rules are published on the Methodology page.
How do you prevent the AI from steering participants?
The mediator is instructed never to argue, debate, introduce counter-arguments, or represent other participants' views mid-interview. This is set in the system prompt; when a participant's wording shows loaded language, a debiasing instruction asks the mediator to explore other angles neutrally rather than push back. Deviations are treated as bugs.
How is a "consensus" statement generated?
Claude groups the needs and values participants named into concerns that express the same underlying point; support is then counted from the participants in each group — the model never asserts a percentage. A concern raised by at least 70% of participants and at least 5 people is Strong Consensus. Each figure carries a Beta-Binomial 95% credible interval. Nothing is published for an issue until it has at least 10 perspectives, and no statement or group with fewer than 5 supporters is published. Full detail on the Methodology page.
Is participant data used to train models?
No. Anthropic does not train on API data and retains it only under its commercial API data-retention terms. We do not send data to any other model provider. Structured identifiers are redacted before text reaches the model.
ACCESSIBILITY & LEGAL
What accessibility conformance target do you meet?
Our target is WCAG 2.1 Level AA and US Section 508; we have not yet had a third-party audit. Full statement, including known gaps, on the Accessibility page. A formal VPAT is on the roadmap once that audit is done.
Are you SOC 2 / ISO 27001 / HIPAA?
Not certified today. What we do: TLS in transit, AES-256 at rest, daily backups, row-level security on all tables with no public policies (all access via server routes), rate limits on AI and write endpoints, and no transcripts stored server-side — but we have not been through a SOC 2 or ISO audit. The Security page lists the exact posture.
What jurisdictions are your sub-processors in?
Vercel (US), Supabase (us-east-2) and Anthropic (US) process data today. PostHog (US/EU) and Sentry (US) are configured but inactive unless enabled; Resend email is pending activation. If you need EU-only data residency, that's an Enterprise conversation.
PRICING & CONTRACTING
How does the free Community tier work?
Community deployments under 500 participants per issue run on the free tier — full protocol access, LRD generation, and dashboard. Above 500 participants you're in Municipal territory.
What does $2,500/mo Municipal cover?
Unlimited issues with up to 3,000 interviews per month included (higher volume on request), admin panel, pilot onboarding, dedicated support. Annual invoicing available for procurement calendars. Reference class: a single consultant-run community engagement study for one issue typically costs $20K–$50K — Municipal runs unlimited issues under that.
What is Enterprise?
For state/regional bodies and large organizations: multi-jurisdiction, custom integrations, dedicated account management; EU hosting can be discussed (on-premise / air-gapped is not offered today). Contact for pricing — genuinely custom, not a hidden-menu tier.
Do you accept POs / net-30 / net-60?
Yes for Municipal and Enterprise. We know how procurement calendars work.
Related: Methodology · Participant Rights · Security · Accessibility · Pilot