01 — Project context
What problem are we solving?
Document the real intake process, actors, pain points, constraints and success criteria before implementation claims are added.
02 — Human understanding
Say it without node names.
The Ops concept describes a system that receives inbound leads, checks their quality, enriches and scores them, stores the result, and alerts the appropriate person. The precise business rules remain to be documented.
03 — Initial hypothesis
The first mental model.
04 — Prompt engineering
What language changed the design?
Add the prompt, intended behaviour, weak output, diagnosis, revision and resulting implementation change. Prompt text should earn its place by showing causality.
05 — Workflow architecture
Nodes with reasons, not decoration.
For each confirmed stage, explain its input, output, dependencies and failure behaviour. Relevant guide concepts can link directly to Triggers, data shaping and reliability.
06–09 — Break, diagnose, fix
Failures only when they happened.
Record the observed symptom, actual cause, evidence used during diagnosis, and the smallest effective fix. Never invent a dramatic failure to make the story interesting.
10 — Testing
Prove the important paths.
Add happy path, invalid input, duplicate submission, external-service failure and edge-case results when they have actually been run.
11 — Production thinking
What could hurt on a bad day?
Future evidence should address idempotency, credentials, retries, error workflows, logging, auditability and observability in proportion to the system’s actual risks.
12–14 — Final design and lessons
Close the loop when the build earns it.
The final architecture, lessons and repository/workflow evidence will be published only after they exist.