Steller: The Complete Story — What actually happened
The earlier essay got the facts right and the story wrong. The rebuild was a redesign, the onboarding was a philosophy, and this is the complete account.
Read post →Steller
The Steller Story, and operating notes on agents, jobs, onboarding, and cost.
The earlier essay got the facts right and the story wrong. The rebuild was a redesign, the onboarding was a philosophy, and this is the complete account.
Read post →The published pivot cost is $672 over 27 days. The weekly buckets are not a second invoice, and zero developers is not free.
Read post →The published path from Sudan business development to Steller’s 60-second onboarding, without unpublished deals.
Read post →referenceId is the partner replay field. Bamboo’s Idempotency-Key is a separate operator lock.
Read post →202 is the value signal, not the PIN. The essays and the developer hub do not publish the same poll interval.
Read post →Invalid key, empty catalog, unfunded wallet — the three failures both onboarding essays surface, in each essay’s own step order.
Read post →RabbitMQ hid the next file. PlaceBambooOrderJob is a named job with a no-retry and refund lock.
Read post →Fat controllers fixed a context-window failure. The same essay already states the three conditions where they stop.
Read post →Zero full-time developers is Steller’s published operating claim. It depends on a visible flow, locks, and a human approval on production.
Read post →The hardening week named four protocol files. One can be quoted. Three cannot, and this essay does not invent them.
Read post →What Steller’s AGENTS.md files actually contain: a short router, backend locks, and the commands that close a session.
Read post →PLG used to mean reducing friction. The real unlock is removing the work entirely — how we applied that to Steller onboarding.
Read post →A development team displaced by war, a codebase full of invisible debt, and 27 days to fix Steller before the first partner integration deadline.
Read post →