CAPSTONE: FICTIONAL RETAIL ANALYSIS

QUESTION
Describe accepted January–March 2026 net merchandise revenue by channel, calendar
month and customer region. Describe repeat purchasing among identified customers.
Recommend the next evidence to collect before making a commercial decision.

DELIVER
1. Immutable raw CSVs, data dictionary, hashes, and environment requirements.
2. A reconciled audit and review of the three invalid source records.
3. Accepted orders, checked customer join, channel/month/region KPI tables.
4. A cohort calculation with explicit identity and observation-window limits.
5. At least one chart with clear units, population, accessible explanation and data.
6. Five-sentence decision memo separating finding, limitation and next investigation.
7. Rebuild command, test suite, and independent rerun into a new output folder.

KNOWN FIXTURE BASELINES
244 raw rows; 2 exact duplicate extras; 3 rejected; 239 accepted.
Net merchandise revenue: 24,112,420 paise (241,124.20 INR).
Channel, month and region totals must reconcile exactly.
Do not force new real exports to match these fixed teaching-data totals.

RUBRIC (100 POINTS)
25: correct metric definitions, units, denominators and reconciled amounts.
20: auditable cleaning and preservation of source evidence.
20: reproducibility and meaningful tests, including failure paths.
15: interpretation, sampling assumptions and honest limitations.
10: readable charts and supporting tables.
10: clear, minimal, reviewable delivery bundle.

REVIEW GATES
An unrebuildable project needs correction. An unsupported causal claim needs
correction. A project that hides invalid records or drops unknown revenue needs
correction. Visual polish alone cannot compensate for those defects.

EXTENSION
Add a permitted, genuinely new data source. Write its own schema and assumptions,
then add one new test that protects a conclusion. Keep real personal data private.
