Source capture

A source snapshot records when source data was captured, its source system, coverage start and end, record count, freshness status and source hash. These fields give the Data Steward a practical way to explain the input behind an observed retail metric. A metric observations record then links a reading to both the retail snapshot and the source snapshot. Its period, numerator, denominator, unit and value can be checked against the intended measurement window.

Source mappings describe how a source field becomes a target field for a store definition. The mapping keeps a version, source and target types, transform rule and effective date. That is useful when a retailer changes an upstream event label or data type: the analyst can compare the dated mapping with the reading period rather than assuming every historical reading used today’s rule.

• Capture. The source snapshot records the observed coverage and capture time.
• Mapping. A versioned mapping identifies how source fields reach the store definition.
• Observation. A reading joins a retail snapshot and a captured source window.

Agree the source system name, the meaning of its coverage window and the acceptable capture delay for the reporting cycle. Decide who marks a capture fresh, stale or failed. A stale source should be examined before the analyst treats the period as complete. Also agree how mapping versions become effective, and keep the prior transform rule available for a historical result.

The team should identify which upstream data connection supplies these captures. Its credentials, schedule and retry policy are implementation choices. Keep source hashes and references with the resulting records so that an analyst can compare repeated captures of the same window.

Suppose a store snapshot is dated at month end. The Data Steward captures a source window ending that day and records its hash and row count. The analyst creates a metric observation tied to that capture and the retail snapshot. Before review, compare the observation period with the coverage start and end. If the capture is stale, inspect the reason and determine whether a newer capture is needed.

Now change a source field mapping for the following period. Reopen the earlier reading and compare its source capture and period with the effective dates on the earlier mapping. This avoids a silent change in interpretation when the next month is reported.

• Source coverage schedule — Compare capture windows and freshness across source snapshots.
• Result windows — Read the period of a calculated result beside its source context.

Review source coverage and see connections.