Select the statement that best describes the organization today, one per
chapter. The result shows a level for each chapter and the chapters to
read first. Answers are stored in this browser only.
1. When How is the model for a task chosen?Not yet One model for everything, or whatever a tool uses. Foundational A written rule gives a first choice per task from its properties; confidential tasks are local. AI-ready Every first choice is tested against its comparison on the suite and recorded in the register. AI-native Choices are revisited on releases and volume changes, and routed by the frontier. 2. Candidates How are candidate models selected?Not yet By reputation. Foundational Two or three per size class from named sources, with licences read and classified. AI-ready Model cards checked, digests recorded, shortlist kept in a registry with licence classes. AI-native The shortlist is reviewed on each notable release; the organization's own models carry cards. 3. Serving How are models run?Not yet On someone's laptop, or not at all. Foundational A small quantized model on an existing machine, with throughput measured. AI-ready A sized, isolated server inside the network with a batching stack; measurements recorded per model. AI-native A shared service with quotas, monitoring, and capacity planning from demand. 4. Adaptation How are models adapted?Not yet Ad hoc prompts. Foundational Versioned prompts with the task's context, tested on the suite. AI-ready Retrieval over the organization's material scored separately; fine-tuning on labelled data with cards and gates. AI-native Retraining from recorded decisions; adaptations compared on the frontier at each release. 5. Comparison How are models compared?Not yet On public benchmarks, or not at all. Foundational The first choice and its comparison on the suite; chosen by the required score. AI-ready Every candidate on the suite per language; cost per query measured; the frontier drawn; the cheapest passing model chosen. AI-native The frontier is redrawn on each release and published on the report card; tasks routed by it. 6. Security How are models kept secure and current?Not yet Downloaded when needed. Foundational Source and digests recorded; verified before loading; safe formats only. AI-ready Lock files under version control; signatures verified where available; updates gated; server isolated. AI-native Verification automated in deployment; an internal registry reviewed on schedule. 7. Cost Is the capacity costed and staffed?Not yet Nobody has done the arithmetic. Foundational A three-year estimate against hosted use; one named person runs the deployment. AI-ready Cost per query measured per task; roles staffed with runbooks and a second person; renewals planned; one shared server. AI-native Infrastructure, test sets, and models shared across organizations; cost and capacity reported yearly. 0 of 7 answered Show reading path Reset
Interpretation of the result
Chapters 1 and 2 first: the rule and the shortlist are a week of
work and decide where every other chapter spends effort.
Chapter 6 before any production use: a verified, isolated server is
the difference between an open model and an open door.
Chapter 7 is where the decision to go local is made or unmade; the
arithmetic is short and the staffing question is the real one.