Produce a verified report

Outcome and preparation

Day 1 · 14:00–15:00 · 60 minutes. Produce a one-page report with an inspectable calculation appendix. Use the Project and all three CSVs. Spend 5 minutes orienting, 45 analysing/checking and 10 reviewing.

Today is a historical rehearsal as of 2026-10-12 UTC. The current week starts 2026-10-05 and ends 2026-10-11. Do not describe these files as today's live company data.

Step by step

1. Step 1

   Open a Project chat and attach the three CSVs. Ask Claude to list filenames, row counts, columns and date coverage before analysis. There are two business rows, two service rows and three incident rows. Stop if any file is missing.

2. Step 2

   Use the analysis prompt from Module 3. Require the source register's definitions, particularly the churn denominator and availability percentage points. Request a separate calculation table.

   Prompt / exampleCopy

   Analyse only the attached fictional sources as of 2026-10-12 UTC.
   Compare weeks starting 2026-09-28 and 2026-10-05.
   Calculate revenue change, churn for each period, backlog change, error rates,
   p95 change and availability change. Count incidents in the current period.
   Use an execution tool for arithmetic. Return formulas and input values with exact source locators.
   Check missing/duplicate rows and zero denominators before calculating.
   Do not claim a cause for any observed change.
   

3. Step 3

   Independently calculate revenue change, current churn and error rate in a spreadsheet. Check: +10%, 12/510 ×100 = 2.35% rounded, and 0.30%. Inspect INC-102 and INC-103 in incidents.csv; distinguish resolved from investigating.

4. Step 4

   Draft a brief using templates/executive-report.md. Include a short decision headline, verified facts, options, proposed actions and unknowns. Suggest one suitable chart: two-period error rate with the 0.20% target. Label axes and dates; do not hide the target breach with a truncated scale.

5. Step 5

   Attach failures/conflicting/revenue-note.md and run the review prompt. Expected response: the note says USD 115000 while the registered Finance CSV says USD 110000; ask Finance to resolve the discrepancy and disclose it. Do not average the values or quietly pick the larger one. Save the conflict version separately.

6. Step 6

   Restore the clean baseline context for your final exercise brief. Have a partner check citations, numbers, period and proposed actions. Save handover/executive-brief.md, handover/calculations.md and a review note listing corrections, reviewer and actual review time.

Expected output and verification

The brief reports revenue growth alongside increased backlog and breached service targets. It does not say reliability caused churn or estimate lost revenue without supporting evidence. The appendix makes every number reproducible.

A citation such as data/service-health.csv; period=2026-10-05; service=customer-api; failed_requests/requests is actionable. “According to the data” is not. In the conflict exercise, a visible unresolved discrepancy is a successful result.

Troubleshooting

If the assistant estimates arithmetic mentally, require code execution or verify in a spreadsheet. If it averages p95 values, point to the KPI definitions. If it lists “three incidents this week,” check dates: only two are in the current period. If it invents a chart it cannot produce, use a chart specification and label it accordingly.

Completion checklist

• Main brief and calculation appendix saved.
• Three calculations and two incident locators checked.
• Conflicting evidence is visible in its separate test output.
• Review time and corrections recorded.