Same data. Show your work.
A reproducible reporting challenge for comparing tools on accuracy, useful explanations and the time you spend reviewing.
1. Give every tool the same brief
Choose up to three tools that can work with tables. Use the same CSV, prompt and acceptance criteria in a fresh session for each. Record the tool, plan, model/version, settings and date. Include upload and correction time.
week,channel,spend_usd,leads 1,Search,240,12 1,Social,160,8 2,Search,300,15 2,Social,200,5
Calculate weekly spend, leads and blended cost per lead (spend / leads). Show your arithmetic and changes between weeks. Explain channel changes without inferring causes. Identify what these data cannot tell us. Use only the supplied CSV.
2. Check against the answer key
| Metric | Week 1 | Week 2 | Change |
|---|---|---|---|
| Spend | $400 | $500 | +$100 / +25% |
| Leads | 20 | 20 | 0 / 0% |
| Blended CPL | $20 | $25 | +$5 / +25% |
Search CPL stays $20 (240 ÷ 12; 300 ÷ 15). Social CPL rises from $20 to $40 (160 ÷ 8; 200 ÷ 5). Week 2 blended CPL is 500 ÷ 20 = $25, not the unweighted mean of channel CPLs ($30).
The data contain no revenue, attribution policy, lead quality or experiment design. They cannot establish ROI or why a change occurred.
3. Keep the evidence and make your call
- Accuracy: do all totals, denominators and percentages match?
- Traceability: can you follow and reproduce the calculations?
- Restraint: are causes and unsupported claims avoided?
- Effort: record setup separately, then work plus review/rework time.
Retain the first output and your corrections. One tiny synthetic test says little about larger datasets, privacy or reliability in production. Repeat with permitted representative cases before committing.
Starts a new page-session trial draft. Your previously saved trial remains until you explicitly save over it. Choose your own contenders; all begin as “Not tested.”