# Just A Bill AI

Just A Bill AI is a public legislative red-team dashboard. It accepts Congress.gov bill URLs and public archive manifests, produces a government-facing citizen summary, checks bill-title claims against operative text, and conditionally runs a private-company loophole analysis when the bill appears to affect companies.

## Public URLs

- Dashboard: /
- Workflow graph: /process.html
- Run stats: /stats.html
- Terms: /terms.html
- Privacy: /privacy.html
- MCP endpoint: /mcp

## Public API Surfaces

- GET /api/runs lists public run summaries.
- GET /api/runs/{run_id} returns a sanitized run read model.
- GET /api/runs/{run_id}/steps returns sanitized workflow status and final-step previews.
- GET /api/runs/{run_id}/steps/{step_key} returns step detail when available.
- GET /api/runs/{run_id}/summary-details returns bill-summary detail.
- GET /api/runs/{run_id}/claim-check-details returns bill-claim-check detail.
- GET /api/runs/{run_id}/verification-bundle returns replay-bundle metadata.
- GET /api/runs/{run_id}/verification-bundle.zip downloads the replay bundle when the run has one.

Hosted deployments expose the same run routes under the configured Supabase Edge API base, usually /functions/v1/api.

## MCP

The MCP server uses Streamable HTTP at /mcp. It supports JSON-RPC POST for initialize, tools/list, and tools/call. Public read tools expose site information, policy links, run lists, run details, step details, and verification-bundle metadata. Write, payment, and account tools require a signed-in Clerk user or a Bearer personal access token created from account controls.

## Verification Bundles

Completed hosted runs can publish replay bundles. A bundle includes a manifest with SHA-256 hashes, public read models, bill text, bill metadata, workflow status, step details that were available at publication time, final step outputs, and replay instructions. If runtime model attribution was not captured, the bundle reports the attribution as unknown rather than substituting static configuration.

## Feedback Policy

Feedback is for factual corrections, UI defects, replay or attribution problems, and model-output quality issues. Logged-in feedback submissions debit $0.01 in credits through the same validation and billing path used by MCP feedback. Do not submit private personal data, secrets, or privileged legal advice.

## Data And Providers

- Clerk is used for sign-in and account identity.
- Stripe is used for credit purchases.
- Supabase stores hosted run control-plane data and sanitized public read models.
- Cloudflare Pages serves the static dashboard.
- Cloudflare R2 or compatible S3 storage may hold replay bundles and large artifacts.
- Resend may send completion notifications when enabled.

## Data Retention

Public run data, sanitized artifacts, replay bundles, and feedback may be retained to support comparison, auditability, and public-interest verification. Account, payment, and administrative data are retained only as needed for operating the service, fraud prevention, support, accounting, and legal obligations.