Fix a Problem That Keeps Coming Back

    You are an operations analyst. This problem has happened more than once. Help me fix the thing that keeps producing it, not just this instance of it.

The problem: {{what_happens}}
How often: {{frequency_and_dates_if_known}}
What we did about it each time: {{past_fixes}}
Who it affects: {{impact}}
What we think causes it: {{current_theory}}

Note first that the past fixes did not work, and that is data. Tell me what the pattern of failed fixes suggests: a fix aimed at the wrong layer, a fix that depended on someone remembering, or a fix that was never actually implemented.

1. Describe the problem precisely enough to argue about. What exactly happens, to what, how often, and how big. "Orders sometimes go out wrong" is not workable. "Roughly 4 of 200 weekly orders ship the wrong variant, mostly in the two hours after a stock update" is.

2. Find the pattern. Ask me what the incidents have in common and — just as useful — what makes them different from the times it works: time of day, person, product, customer type, volume, day of week, which system, whether someone was covering for someone else.

3. Ask why, layer by layer, and stop when you hit something we can actually change. Push past the first plausible answer. "The operator made a mistake" is never a root cause; the question is what made the mistake easy to make and hard to notice. Keep going until you reach a design, a rule, a tool, or an incentive — those are causes you can act on. But do not tunnel: if two independent chains both had to hold for this to happen, name both.

4. Sort the causes into what we control, what we influence, and what we absorb.

5. Give me three tiers, clearly labelled, because conflating them is why the last fix failed:
- Contain: what stops the bleeding now
- Correct: what removes the cause
- Prevent: what stops this class of problem, not just this problem

6. For each corrective action: owner, date, and how we will verify it worked. An action item with no owner and no verification date is a wish.

Finish with the one measurement that will tell us in a month whether this is actually fixed — and what number would mean it is not.