On this page
In short. An operator uploads the customer's own raw export for a date range, and the platform reports which orders and which fields differ from what the system already holds. The file is recomputed with the platform's own ingest arithmetic, so a difference shown is a data divergence, not a formula mismatch.
How it works#
Customers send raw exports taken at different times, and the same window shows different numbers in each snapshot: one order whose status changed between snapshots moves a whole period's totals. Validation finds that divergence automatically.
- On Data validation ("Đối soát dữ liệu"), choose New validation run ("Tạo lần kiểm tra mới"), a data type (Shopee, Lazada or TikTok Shop orders; Shopee or TikTok Shop creator/affiliate orders) and the file. The screen reads the file, detects the type and fills the date range; both can be edited.
- The file (CSV, XLSX or XLS, up to 8 MiB) is parsed with the organization's own import mapping template, the one a real ingest uses. A data type with no template is refused until one exists in CSV/Excel mapping.
- The worker recomputes the file with the ingest arithmetic and reads the system side from the same figures a dashboard shows: order count, gross value, net value, status.
- The run page shows one verdict per KPI (Orders, GMV, NMV), the same figures per day, and one row per diverging order with a filter by difference type.
Status comparison only checks the cancelled-versus-active class, because each marketplace names the states in between differently. A cancelled order keeps its gross value and contributes zero to net value on both sides.
Every run is kept in Validation history ("Lịch sử kiểm tra"). The raw file and per-order detail are purged after the retention window set by a platform administrator for the organization (30 days by default); the summary stays.
Where it stops#
- Creator and affiliate data compares at day level only: it is stored as a daily rollup, so there is no per-order drill-down. Creator GMV (the commission base) is a different measure from Sales Order GMV and can differ from it.
- A very large run keeps the first 5,000 per-order differences and states how many were kept out of how many found.
- A file that ingest cannot parse cannot be validated either; both paths share one parser.
- Net value is read as the system stored it, the same figure every other report reads.
- Rollout state is on Project status.
Reference#
Divergence types
| Divergence type | Meaning |
|---|---|
| Missing in system | Order in the customer file, absent from the system |
| Missing in file | Order in the system, absent from the customer file |
| Cancellation mismatch | Cancelled-versus-active class differs between snapshots |
| Metric mismatch | Gross or net value differs for an order both sides hold |
What it reads and never reads
| Reads | Never reads |
|---|---|
| The organization's own uploaded file for one marketplace or creator dataset | Another organization's file or orders |
| The system's own order and creator-performance figures for the same window | A net value recomputed from raw line items |
| Organization-wide rows with no brand, on a brand-scoped run | A file the organization's own ingest could not parse |