Rent, adapt or build: the model census on one page

Every model decision comes down to three choices. Rent a model through an API. Adapt one with your own data. Or build one from scratch. The Foundation Studio Playbook's rule is short: rent first. Adapt when your evaluation fails and you own the data. Pre-train only when sovereignty or a real moat demands it.

The decision does not end at launch. Models are retired, repriced and replaced. That is why the playbook asks for a model census: every model has an owner, a card, a licence, an eval record and an expiry date.

The census is a practice, not a tool. A spreadsheet kept honestly beats a platform nobody updates. What matters is that someone can answer, for each model in production: who owns it, what it is allowed to do, how it last scored, and when it must be reviewed.

This is the floor every other stage stands on. Generative pipelines, agents and robots all call models. If you cannot see your models, you cannot govern what is built on them.

Model Engineer learners run one model through the MDLC and leave with a census entry as part of their portfolio.