Real runs
The proof is a log, not a logo wall.
Every real run gets one row: the idea, hours, measured API cost and the outcome, including failures.
No public runs yet.
The factory has built two internal apps on its maker's own Claude plan. We do not publish those as proof: no API cost was measured in dollars, and a subscription run is not what you will do.
The first logged run will be an instrumented run on a plain API key, with tokens and cost recorded. Until then this page stays empty, and every cost figure on this site stays labelled "estimate".
What will be logged
- The idea, as written.
- Wall-clock hours and the preset used.
- Anthropic spend as recorded by the run, not modelled.
- The outcome, with failed runs shown and the stage they failed at.
- A link to the pull requests and logs.
Not shown, ever: invented quotes, customer logos we don't have, or revenue claims.