Canonical Synthetic Dataset Blueprint
Meridian’s story can operate at enterprise scale, but the lab begins with small deterministic synthetic datasets so the learner can inspect, reason about, break, repair, and reproduce the work.
| Dataset | Shape | Starting rows | Purpose |
|---|---|---|---|
customers_crm |
CSV | 500 | CRM-shaped customer source; intentionally contains some missing/stale contact data. |
customers_pos |
CSV | 350 | Store loyalty/customer source with different identifiers and name/address formatting. |
customers_service |
JSON | 220 | Service identities/cases with nested interaction context. |
products |
CSV / PostgreSQL | 150 | Product/SKU attributes, categories and lifecycle status. |
suppliers |
CSV / PostgreSQL | 30 | Supplier master candidates. |
supplier_products |
PostgreSQL | 250 | Many-to-many supplier/product relationships. |
orders |
PostgreSQL | 2,000 | Order headers; mix of known customers and synthetic guest orders. |
order_items |
PostgreSQL | 5,000 | Order lines at item grain. |
locations |
PostgreSQL | 8 | Stores/fulfillment centers for lab-scale use. |
inventory_snapshot |
CSV / PostgreSQL | 1,200 | Product/location/time inventory observations. |
service_cases |
JSON / PostgreSQL | 200 | Structured case fields plus text notes. |
reference_* |
CSV / PostgreSQL | Small | Country/state/order status/return reason/product category code sets. |
documents |
DOCX/PDF/TXT | 20–40 | Synthetic policies, supplier docs, invoices and service attachments. |
click_events |
JSON/Parquet | 100k+ later | Chapter 14 scale/format-awareness dataset; generated only when needed. |
Dataset rules
- Use synthetic data only.
- Preserve canonical raw source files; do not overwrite them during cleaning/transformation exercises.
- When exact answer keys depend on the data, use deterministic generation so results can be reproduced.
- Document intentional defects rather than allowing accidental corruption to become the lab state.
- Distinguish story scale from lab scale; the smaller sample is deliberate for learning and does not redefine Meridian’s business size.
- Every dataset should identify its producing business process and source system.
- Sensitive-looking attributes are synthetic and exist only to practice governance/security/ethics concepts safely.
Related areas
Source: Meridian Commerce Group Enterprise Data Management Casebook, Section 8.