The source-to-report chain

Map the eight boundaries

The source-to-report story is easier to review when you name its boundaries. For this course, follow source, meaning, Bronze, Silver, Gold, semantic, report and release. These are places to ask for evidence, rather than eight permissions granted by starting a workflow. Each transition has its own input, operation and observation. An earlier success cannot silently supply a later missing result.

Begin with source. Profiling SQL Server or PostgreSQL can describe structure, keys, counts and declared coverage in a value-free source index. That index is useful for design. Raw extraction and independent oracle capture are distinct operations. Source-read consent must cover the actual request; installing tools does not approve database access. Write that distinction beside the source node in your worksheet.

Put meaning before compilation

The meaning boundary connects source structure to a business process. A draft seed helps an engineer propose a model. It does not ratify the grain. A human needs to confirm what one fact row means before authoritative Gold compilation. Retain the decision as a decision, with the relevant review record, instead of treating a generated star as implicit approval.

Bronze is the captured landing layer, bound to a declared batch and source observation. Silver is the prepared layer: types, deduplication, accounting and quarantine observations matter because a clean-looking table can conceal discarded rows. Gold contains the ratified business-facing structures. Reconciliation with independent expectations is distinct from successfully compiling notebooks for these layers.

Distinguish serving from reporting

The semantic boundary concerns the model used to serve tables, relationships and measures. Generating TMDL does not deploy that model or refresh it. Measure probes, publication, definition read-back, refresh or framing, and value checks have separate roles. A model file can be structurally readable while the serving model still requires current observations.

The report boundary adds another distinction. Layout checks can inspect geometry. Emitted visual queries can be executed and compared. A rendered-value check observes what was displayed in a declared state. One of these does not automatically establish the others. Bounded Windows OCR is a preview capability for authored numeric-label regions, and the documented empty-state mismatch still refuses qualification. Keep unobserved states visible.

Release is where you defend the selected scope and its evidence. It does not mean every upstream boundary has been proved merely because a report exists. Existing assets can enter through evidence-bound admission routes, leaving unproved upstream claims explicit. A release review needs the actual scope and retained results, plus human acceptance where required.

Trace evidence rather than arrows

An arrow on a diagram describes intended flow. A receipt records a producer result and binds declared inputs, outputs and upstream dependencies. Gates inspect those bindings and recompute required comparisons against frozen expectations. If an observed dependency changes, dependent evidence can become stale. The local ledger does not continuously poll every remote system, so freshness must be described using the observations actually available.

Boundary group	Ask for
Source and meaning	Read scope, source coverage, oracle provenance and grain decision
Bronze, Silver and Gold	Batch bindings, execution results, accounting and reconciliation
Semantic and report	Serving observations, checked queries and declared render states
Release	Current scoped gate, retained evidence and separate human decisions

Completion check

Draw the eight boundaries and place one required observation beside each. Mark a missing notebook result, an unrefreshed model and an unobserved report state as three different gaps. Explain why a receipt or Hosted acknowledgement cannot fill them automatically. Your diagram is complete when a reviewer can identify both the intended engineering flow and the limits of the evidence available today.