How public spend reports work

Credits are pegged at 1 credit = $0.01 of list-price AI inference. They are a prepaid balance — non-transferable, and redeemable only for funding projects on FablePool.

• Buying credits: the platform keeps a fixed 20% margin at purchase time; the rest lands in your balance.
• Revenue is the 20% purchase margin plus the list-vs-batch inference difference (which also absorbs token overruns — backers are never asked mid-build for more), plus plan-generation fees and any credits that expire after 12 months of inactivity.
• Generating a plan costs a small credit fee (waived for subscribers); the platform absorbs the actual planning inference cost, so planning never touches a project pool.
• Funding moves credits from your balance into one project's pool, visible on the project page.
• Funding targets start from the AI planner's per-milestone token estimates, times an admin-set safety cushion, priced at list. The project's maker can adjust each milestone's token estimate before submitting the plan for review, but targets are always token-based estimates — never an arbitrary price — and a plan must total at least $30 to be submitted.
• A milestone runs when the pool covers its target. Each run records exact input/output token counts in the public build log and decrements the pool by exactly what the run cost — capped at the pool, with overruns absorbed by the platform.
• The ledger is double-entry and append-only: corrections are new entries, never edits. Every transaction balances and no balance can go negative; an automated invariant check enforces this.
• The build log and credit ledger are always public — including on projects whose artifact downloads are backers-only. The money trail is never private.
• For projects built on our VM build engine, the funded pool is a hard cap: the build agent reaches AI models only through our metering gateway, which charges each call to the pool at list price and refuses further calls once the pool is spent. Every charge lands in the same public ledger.
• When a finished project is hosted in "demo mode," an ongoing per-hour hosting meter decrements the pool; the rate and remaining runway are shown on the project page, and each charge is a public ledger entry. Hosting pauses automatically when the pool can no longer cover it.