All categories/Data and analysis
Level Advanced · Anomaly Debugging Data quality
Anomaly: hypotheses before investigation
A number jumps or drops and the first instinct is usually the wrong place to look.
Prompt
Something is off in the data; help me investigate.
Symptom: {{symptom}}
Started: {{when}}
Known facts: {{known}}
Requirements:
1. No conclusions yet. List 6 possible causes ordered by 'check this first', each with one line on how to verify it (which table, which dimension).
2. Separate real business change from collection or definition change; give rough likelihood and the basis.
3. Give an investigation order whose first three steps fit within {{budget}}.
4. If it turns out to be a collection issue, outline how to backfill, no code.What to state before it starts
Leave these unsaid and the model fills them with an assumption. Most 'wrong answers' start here.
- {{symptom}}: paste the numbers and the comparison period
- {{when}}: down to the day and any release near it
- {{known}}: what changed recently (tracking, definitions, upstream fields)
- {{budget}}: e.g. 'two hours'
Go through each of these
Where this kind of work most often goes wrong. Already in the prompt, so you do not have to remember it.
- A time budget forces an executable order instead of a list.
- Asking it to split real vs collection change beats asking 'why'.