Investigate the headline.

01 / Establish the contract

Produce a one-page decision memo for the simulated operations manager. Compare April and May 2026, reconcile channel contributions and explain the limits of the delivery evidence. Use SQLite locally; no query runs against this website.

The supplied raw_orders retains repeated keys and inconsistent text. The separate orders table is an authored verified reference, not an automatic cleaning of the raw input. Record corrections before comparing your work with it.

Grain: one verified order per order_id. Values are EUR major units. Completed order value excludes returned and cancelled records. Delivery coverage is measured durations divided by eligible completed orders. Missing duration stays missing.

Limits: no costs, spend, customer IDs, seasonality or experiment are supplied. You cannot establish profit, customer retention, a stable forecast or causation.

02 / Inspect and calculate

1. Download the kit and read the metric contract. Run analysis_setup.sql in a new, empty local SQLite database only.
2. Find repeated keys and distinguish exact repetition from conflicting evidence. Preserve the raw input.
3. Inspect invalid dates, inconsistent channel names, missing amounts and nonnumeric durations. Write a correction log with reasons.
4. Calculate completed order counts, value and AOV for each month. Reconcile channel totals to the monthly totals.
5. Report measured and missing delivery counts, coverage and observed mean. Do not substitute zero for missing.
6. Compare May with April. Keep the high-value verified A020 order in the baseline; report any exclusion as a separate sensitivity scenario.
7. Create one clear visual and write a recommendation with the next evidence needed.

A starting query to extend

SELECT substr(order_date,1,7) AS month,
       COUNT(*) AS completed_orders,
       SUM(order_value_eur) AS completed_value_eur
FROM orders
WHERE status = 'completed'
GROUP BY substr(order_date,1,7)
ORDER BY month;

Common mistakes: join inflation, equal-weighted channel averages, changing the status filter between periods, treating percentage points as percentage growth and claiming missing delivery observations show an improvement.

03 / Verify five reference outputs

Enter your results before opening the worked solution. This checks five numbers against the fixed dataset, with a tolerance of 0.011. It does not assess your SQL, query efficiency, originality or business reasoning.

04 / Challenge your conclusion

In your local copy, change the value of one completed order. Predict which totals should change, rerun the analysis and reconcile the difference. Then make a known delivery duration zero: it should remain measured. Compare that with a missing duration, which reduces coverage.

April and May each have 20 completed orders. A shift in channel mix can change overall AOV while within-channel patterns differ. A small fall in observed delivery mean is not reliable evidence of improved delivery for all orders when coverage changes from 70% to 95%.

Deliverables and self-review

• SQL, exact run instructions and source/version note
• Correction log retaining rejected and conflicting evidence
• Monthly/channel reconciliation and labelled visual
• Decision memo: finding, limitation, recommendation, next evidence

Use the supplied rubric to review correctness, validation, interpretation, reproducibility and communication. Human judgment is needed for the memo and recommendations; the five numeric checks do not grade these.