Unit 13.03: Writing the evidence memo

Unit ID: SQL-M13-U04 - Estimated active time: 30-40 minutes Objective: produce a memo that lets a reader verify or challenge every number without asking you.

The memo is the deliverable

The queries are working material. What leaves your hands is a document a reader can act on - and challenge.

The structure

ANALYSIS: June 2026 order performance
Prepared: 2026-07-01 - Source: starter database (orders, order_items, customers, payments)

QUESTION
  How did orders and revenue perform in June 2026?

METHOD
  Population:     orders with placed_at >= 2026-06-01 and < 2026-07-01  (half-open, Module 8)
  Grain:          one row per order (verified 1,000 in, 1,000 out)
  Normalisation:  city lowercased and trimmed; country NULLs bucketed as UNKNOWN

FINDINGS
  1,000 orders placed; 988 completed, 12 still pending
  Completed revenue ₹26,85,905 - booked revenue ₹27,01,463
  Average order value ₹2,718.53 (completed orders)
  3,400 items sold across 1,000 orders
  3,812 of 4,812 customers have never ordered
  Payments received ₹26,85,905 - matches completed revenue exactly

RECONCILIATION
  988 + 12 = 1,000 orders
  ₹26,85,905 + ₹15,558 = ₹27,01,463
  Payments − completed revenue = ₹0.00
  Estimate (1,000 × ₹2,701.46 = ₹27,01,460) within ₹3 of actual

LIMITATIONS
  Country split covers 4,512 of 4,812 customers (93.8%); 300 unknown, shown separately
  City figures use lower+trim normalisation; does not merge genuine name variants
  12 pending orders excluded from completed revenue but included in booked
  No refunds present in this dataset; a live extract would need that treatment defined
  Timezone not specified in source data; dates treated as recorded

REPRODUCE
  queries/june_performance.sql - run against schema.sql + seed.sql
  Validation: SELECT COUNT(*), SUM(order_total) FROM orders WHERE status='completed'
              expect 988 | 2685905.00

What makes it good

Method before findings. A reader judges the numbers knowing how they were made.
Reconciliation is shown, not claimed. Arithmetic anyone can redo in their head.
Limitations are specific. "93.8% coverage" is useful; "some data may be missing" is not.
It is reproducible. A named query file and an expected validation result.

The limitations section is the credibility test

Five limitations here, and each names something real. A memo with no limitations section signals that nobody looked - which is a stronger negative signal than any single caveat.

Practice

Write the limitations section for question 5 (most populous city).

Check your answer

LIMITATIONS (Q5)
  City normalised with LOWER(TRIM(city)); 6 raw values collapse to 4 real categories
  New Delhi's 1,604 customers appear as 535 + 535 + 534 in the raw column
  Normalisation does not merge genuine alternative names (e.g. Bangalore / Bengaluru)
  Counts include customers who have never ordered

The third line is the honest one: it states what the cleaning does not fix, which is what stops a reader over-trusting the figure.

Takeaway

Method, findings, reconciliation, limitations, reproduction. The limitations section is what turns a number into evidence.

---