Plate E.1 · Progressive introduction
A route into AI that does not require abandoning engineering judgment.
The visitor chooses the depth. Each stage begins with a familiar discipline and reveals the model layer only where it adds value.
The model matters, but it does not define the system. Select it only after the work, authority, evidence, control and trace are clear.
Technical artifact
E.1- 01 Work What is the work?
- 02 Boundary Where is the boundary?
- 03 Evidence What evidence is enough?
- 04 Control How is control retained?
- 05 Trace What must remain traceable?
- 06 Model Which model fits this station?
- 07 Autonomy What has the evidence earned?
What is the work?
Define the workflow, failure cost and observable outcome before choosing a model.
Where is the boundary?
Decide what the system may read, change, call or publish. Treat external content as untrusted.
What evidence is enough?
Write success criteria, representative cases and failure tests before expanding capability.
How is control retained?
Human command must correspond to real pause, review, confirmation and override mechanisms.
What must remain traceable?
A transcript is insufficient. Preserve identity, evidence, transformations, authorization and result.
Which model fits this station?
Use the model that fits the task, risk, cost and operating conditions. Preserve the option to change it.
What has the evidence earned?
Expand authority only when observed performance supports it. Otherwise narrow the boundary, revise the workflow or replace the model.