Common Ground

Common Ground turns local concerns into safer civic action, stress-tested for privacy, clarity, and human review.

Comment

Inspiration

A civic problem is often ignored not because it is small, but because each person experiences it alone. A blocked ramp, an unsafe bus stop, or an unlit school route may affect hundreds of people, yet every report reaches an institution as an isolated anecdote.

At the same time, public bodies cannot responsibly act on outrage, accusations, or an unstructured complaints feed. Common Ground creates the missing middle: human experience, independent corroboration, and one concrete request for action.

What it does

Common Ground follows a careful three-step loop:

  1. Report — a resident privately records what happened, where, and when.
  2. Confirm — other people add only what they directly experienced, with dated evidence where possible.
  3. Act — once a credible pattern emerges, Common Ground produces a concise civic brief containing the known facts, remaining uncertainty, public impact, and one measurable requested action.

It is deliberately not a petition platform or a popularity contest. Five independent observations with provenance can be more useful than a thousand anonymous clicks.

The interactive demo contains three realistic civic cases. Visitors can explore a case, add a private confirmation, examine the evidence trail, open the civic brief, and turn a new report into a neutral question that others can independently verify.

How we built it

We used Codex and GPT-5.6 throughout the build: to design the evidence model and safety boundaries, shape the product language, create the interaction system and realistic test cases, implement the responsive interface, and verify the end-to-end flow.

The GPT-5.6 integration is intentionally bounded. A server-side Responses API contract converts a messy report into a strict structured result:

  • a neutral issue summary;
  • a confirmation question for other residents;
  • the next useful piece of evidence;
  • an explicit uncertainty note;
  • a recommended civic route.

GPT-5.6 never decides whether an allegation is true, assigns blame, identifies a responsible party, or publishes a claim automatically. The public prototype includes an explicit deterministic fallback so judges can test the full product flow without exposing an API key in the browser. The repository README documents the architecture, setup, safeguards, and the exact roles of Codex and GPT-5.6.

Challenges we ran into

The hardest design decision was not what the model could generate, but what it must never do. Civic technology can amplify harm when it rewards outrage, exposes vulnerable reporters, or converts weak signals into public accusations. We therefore treated privacy, provenance, uncertainty, consent, and the right to respond as product requirements rather than disclaimers.

A second challenge was making heterogeneous evidence — lived experiences, dates, locations, access needs, photographs, and institutional responses — understandable in one short brief without pretending certainty.

Accomplishments we're proud of

  • A working interactive prototype rather than a slide-only concept.
  • A visible Report → Confirm → Act journey that can be understood in seconds.
  • An evidence model that rewards independence and provenance instead of virality.
  • A private confirmation flow that strengthens the aggregate without exposing identities.
  • A bounded GPT-5.6 role designed for translation and constraint, not autonomous judgment.

What we learned

The highest-value use of an advanced model in civic systems is often careful translation. GPT-5.6 is most useful here when it turns emotional or ambiguous input into a neutral, falsifiable verification task while preserving what remains unknown.

We also learned that collective action becomes more credible when every participant contributes a small observation they can stand behind instead of merely signalling agreement.

What's next

We would pilot Common Ground with one municipality and one disability-access organisation, add cryptographic provenance for submitted media, build an institutional response workspace, and measure time to acknowledgement and time to remedy.

One person sees a problem. A community makes it impossible to ignore.

Built With

Share this project:

Updates