Flagline is a working v1 implementation, not a production-ready service. The dashboard reads real persisted data and shows honest empty, setup, permission, and failure states.

## Implemented [#implemented]

- Workspace setup and switching, with owner, admin, and member roles.
- System presets and team presets, immutable versions, publishing, history, and rollback.
- A required real-inference test before creating, forking, or publishing a preset.
- Public GitHub preset import from one JSON file pinned to a full commit SHA.
- API key creation and revocation, with allowed levels, rate and concurrency limits, and an optional daily spend ceiling.
- The `POST /v1/moderations` route with saved and inline presets, idempotency, hard limits, auditing, and metered charges.
- The playground, decision logs, usage views, and workspace settings.
- Platform operator tools for tenant lookup, ledger history, manual credits or deductions, and suspension or reactivation.

## Not production-ready [#not-production-ready]

The v1 HTTP surface has not been live-tested against Baseten, prepaid billing, shared caching, or production load. The repository's own API notes call it an incomplete v1 implementation. Do not treat a passing local build or test suite as a production readiness claim.

- **No hosted CI.** The checked-in test commands are the reproducible local contract.
- **Stage 0 infrastructure.** The Compose deployment has not been load-tested or hardened for production.
- **Legal copy is unfinished.** Legal pages are placeholders pending approval.

## Deliberately unavailable [#deliberately-unavailable]

- **Stripe and checkout.** No self-serve payment, prepaid balance, or refund flow is implemented. Usage and Billing show recorded ledger totals, not a payable balance.
- **Shared decision cache.** It is not active. The parked migration must not be applied.
- **Tenant deletion.** Destructive deletion is blocked pending the retention decision.
- **Public moderation demo.** Disabled by default. It needs a dedicated server-only credential and explicit operational checks before enabling.
- **Account registration.** Closed by default. Existing users can still sign in.
- **Repository sync.** Imports are one public file at a time. There are no private repositories, bulk discovery, branch tracking, or automatic updates.
- **AI assist without configuration.** It is optional and returns a clear not-configured message when the deployment has no model configured.

## Safety notes [#safety-notes]

- Flagline recommends a verdict. Your application owns enforcement and user communication.
- A high score can produce a block, not just a flag. Never treat an `allow` as proof that content is safe.
- A block is not a crisis response. Flagged or blocked crisis-related content needs a human review process that your organization owns.
- System presets are starter templates. Their thresholds, examples, and severities require review and calibration for your community.
- A daily spend ceiling bounds committed Flagline charges, not provider spend from failed or rolled-back inference.
- An imported policy and its version summary are not authoritative provenance. Review the policy and test the exact draft before publishing.

## Operate it [#operate-it]

The deployment requires Postgres, a non-evicting Redis instance for hard limits, and pinned inference models. Migrations and the system preset seed are explicit operator steps, not boot-time work. See the repository's `README.md` and `infra/README.md` for the exact commands and environment variables.