Detect a quiet carrier feed independently of queue health
Every worker is healthy, but Carrier Elm has not sent a batch since morning. No alert fires because all dashboards measure processing failures.
- Focused work estimate
- 2h + prerequisites
- Priority in the scenario
- Medium
- Engineering practice
- Data observability · Schedules · Alert semantics
Estimated field mix
- Data engineering50%
- Site reliability50%
Field percentages are editorial estimates of the ticket's engineering focus. They total 100%; they are not measured time, proficiency scores, or ownership evidence.
Review it, then add it to your workspace.
The board opens an editable draft; nothing is saved until you confirm it. Sign-in and workspace permissions apply, and Demo boards remain ephemeral.
Project context
A fictional delivery marketplace receives JSON batches from three carriers. One uses local timestamps, another retries whole batches, and a third corrects delivery scans. Customer support needs a stable timeline rather than the last payload received.
Setup prerequisites
- JSON schema validation
- SQL queries
- Event-time concepts
Preceding work
Complete these dependencies, or supply their agreed outputs before taking this ticket.
Acceptance criteria
- Track last accepted receipt separately from event-time freshness.
- Apply each carrier's configured operating schedule and lateness budget.
- Feed-gap alerts name the integration and diagnostic step without parcel data.
Implementation constraints
- Empty valid batches count as receipts but do not imply fresh parcel scans.
Verification to include
- Advance the clock during operating hours to trigger a feed-gap alert.
- Suppress quiet-window paging and distinguish stale event times.
Deliverables
- Freshness metrics, alert rule, and escalation notes
Rollout and recovery
Review a simulated week in report-only mode before enabling paging.
Value of the work
For the engineer: Practice ingestion contracts, deduplication, event-time reconciliation, and replayable data repair.
For the team: Examine how an engineer preserves source lineage and handles bad partner data without losing valid business events.
Evidence boundaries
Outcome Evidence: Tests, patches, and runbooks are requested deliverables. They become Outcome Evidence only through a qualified Mission and immutable Evidence IDs.
Ownership Evidence: Independent adaptation must be observed under a declared verification policy and cite immutable Evidence IDs. Completing a planning ticket establishes no Ownership Evidence.