Lifecycle and ordering

Three events, three states

A queued event creates a batch at its source station. A dispatched event moves the state to in_transit and still references the source. A delivered event changes the state to delivered and references the destination.

The fixture snapshot contains eight three-event delivered lifecycles, three two-event in-transit lifecycles, and one single-event queued lifecycle: 31 events in total. Missing later stages in this snapshot are deliberate, not malformed records.

queued      → state: queued      · source station
dispatched  → state: in_transit  · source station
delivered   → state: delivered   · destination station

Order within a batch

Group by batch_id, deduplicate by id, and sort by sequence within each group. sequence is local to one batch, so it must never be used as a global cursor. Check timestamps as a consistency constraint, not as an identity key.

If two events reuse a sequence with different IDs or content, report a conflict. If an event is absent in the middle of a sequence, report a gap rather than inventing it.

Replay a known example

Batch batch-001 queues at 08:00 UTC, dispatches at 08:02, and delivers at 08:10 on October 1, 2026. It carries 42 units from wp-001 to wp-003. The authored end-to-end duration is 10 minutes.

const response = await fetch('/resources/datasets/relay-events.json');
if (!response.ok) throw new Error(`HTTP ${response.status}`);
const fixture = await response.json();
const events = fixture.records
  .filter(row => row.batch_id === 'batch-001')
  .sort((a, b) => a.sequence - b.sequence);
// queued, dispatched, delivered
const elapsedMs = Date.parse(events.at(-1).timestamp)
  - Date.parse(events[0].timestamp);
console.log(elapsedMs / 60_000);
// 10

Treat the snapshot as a fixture

All timestamps and IDs are stable within a catalog version. There is no scheduler advancing these batches, no webhook endpoint, and no polling stream. Re-fetching does not turn a queued sample into a delivered one.

Use the fixture to test deterministic reconstruction and error handling. It does not model network delays, delivery guarantees, or the complete behavior of a production event system.