Scope → vertical slice → ground → evaluate → harden → deploy → monitor

Run a visible loop for a personal knowledge assistant that answers questions from approved notes with citations: prepare the approved inputs, request a bounded output, validate it against the evidence, revise one issue, and send it to the named human gate. Each pass should leave a small record: input version, prompt version, output, reviewer decision, and unresolved question.

Do not turn a draft step into an irreversible action. The workflow stops when a citation cannot point to a source span, the evaluation set is missing, or a proposed feature has no user need. A stop is useful information: it tells you which approval, source, or test is missing. Once the gate approves, archive only what policy permits and note what should be monitored after release.

Write a short record before you move on. For a personal knowledge assistant that answers questions from approved notes with citations, state the claim or choice under review, then name the evidence that supports it: a source-linked answer, a small held-out question set, trace records, and a milestone demo. Next, name what the evidence does not establish. This last line prevents a narrow test from becoming a broad promise. If another team member opened your record next month, they should be able to reproduce the review without trusting your memory.

Now make the trade-off visible. A mentor reviewing whether a learner can make sound engineering trade-offs may value speed, clarity, cost, control, or reassurance differently. Explain which of those mattered in the current version and why. Do not let a model choose the trade-off simply because it can produce a confident answer. The responsible owner decides whether the upside justifies the remaining uncertainty.

Finally, connect the decision to a next action. If the current evidence is enough, identify the smallest safe step forward. If it is not, request a specific source, approval, or test. The stop condition remains concrete: a citation cannot point to a source span, the evaluation set is missing, or a proposed feature has no user need. A documented pause is a successful outcome when it keeps a weak result from becoming a consequential one.