Prompt moderation

When you post a prompt, the planner drafts a plan; you curate the milestones and submit it for human review. A project goes live and opens for funding only after an admin approves it — nothing is public or spends until then.

• Projects rejected at review are hidden. A rejected project holds no pooled credits (funding only opens after approval), so there is nothing to refund; the plan-generation fee is non-refundable once the plan was produced.
• Illegal content, malware, harassment, and deceptive impersonation prompts are rejected. The model itself may also refuse to plan such prompts.
• Impossible prompts are welcome — the planner stages them into ambitious-but-feasible milestones (prototype, design, demo, builds).
• A plan must be substantial to submit: its milestone funding targets must total at least $30. If your draft is below that, broaden the milestones or raise their token estimates before submitting.
• The model itself may decline a milestone on safety grounds; declined runs are never charged.
• Repeat abuse leads to account suspension. One account per Google identity.

Build decisions & who decides

• Every project is classified up front by what it can actually deliver (an app, a service, a research/document deliverable, or "not buildable" by an AI agent). A prompt classified as not-buildable will not provision a build.
• When the AI build agent needs a product or scope decision, its question is surfaced on the project page. The project owner decides; backers may cast advisory votes to guide them. A decision the owner leaves unanswered past its deadline (24 hours) is resolved automatically to the most-voted option (or a sensible default if there are no votes). Locked decisions cannot be changed once the build consumes them.