Notes
Every post follows the same shape: the workflow in the owner's words, what the audit found, how the work was split between code, model and person, the evidence, and what is still wrong. The filters match the six stages of the loop.
Why two weeks of audit comes before any code, and what to collect: interviews, exception paths, the spreadsheets people built to get around the system, and where colleagues already use AI on their own.
There is one way work goes right and a thousand ways it goes wrong. How to list the failure modes, and why a disagreement between people is a policy question and not an exception.
Three gates decide where a model belongs: does the workflow need changing, can plain code do it, and which steps are truly fuzzy. Most steps land in code.
Proving one workflow as a standalone script before building shared infrastructure, and what to leave undecided until a second workflow exists.
Five case types, four criteria, and why escalating to a person counts as a pass. Includes how to break a subjective judgment into parts you can check.
How to compare a new system's output with the real one before it is allowed to act, and what every run needs to log.
Four layers of observability, and the two numbers that show early whether the time saving is real.
Nothing in this stage yet.