# Custom client-reporting application: reference brief

This is an implementation reference, not a deployed product claim.

## Job
Prepare a recurring client report from scoped business measures and relevant external evidence. Users need to inspect, revise, approve and export it.

## Boundaries
- Business systems/warehouse: records, approved views, measure definitions and customer identity.
- Evidence preparation: selected source retrieval, observation period, captured version and lineage.
- Analytical operation: typed inputs, documented grain, exact calculations and coverage checks.
- Execution agent/workflow: coordinate the defined task, surface missing inputs and carry results into review.
- Reporting application: client workspace, report state, review, export and issued versions.
- Operating service: refresh ownership, run history, failure recovery, support and agreed changes.

## Acceptance example
Use the synthetic revenue diagnostic. Its bridge must reproduce $10,000 to $9,000 with zero residual. Raw support tickets must not multiply revenue. Every displayed movement must open the input records. Keep the measured movement separate from the suggested explanation.

## Revision and failure behaviour
A source revision creates a new working report version. It does not silently replace an issued report. A failed required extract leaves the new run incomplete and keeps the prior approved report accessible. Display publication, observation and retrieval clocks separately.

## Decisions for the sprint
Users and permissions; source access; refresh frequency; report outputs; deployment; source-code handover; costs; support ownership; change process.

Reference: https://whiteboxx.ai/resources/ai-architecture/