Incremental transform runner

Implement an incremental transform runner that processes only new records, skips stale or duplicate inputs, and returns the next processing state.

Separate cursor filtering, transform logic, and state advancement so replay behavior is easy to reason about.

Sample input

records and state

Only records newer than the cursor should be processed.

{
  "state": {
    "last_successful_cursor": "2026-03-22T00:00:00Z"
  },
  "records": [
    {
      "record_id": "r-88",
      "updated_at": "2026-03-21T23:58:00Z",
      "account_id": "a-099",
      "amount": 300
    },
    {
      "record_id": "r-90",
      "updated_at": "2026-03-22T01:05:00Z",
      "account_id": "a-100",
      "amount": 1200
    },
    {
      "record_id": "r-91",
      "updated_at": "2026-03-23T02:10:00Z",
      "account_id": "a-102",
      "amount": 640
    }
  ]
}

Expected output

expected run result

{
  "processed": [
    {
      "account_id": "a-100",
      "metric_day": "2026-03-22",
      "daily_revenue": 1200
    },
    {
      "account_id": "a-102",
      "metric_day": "2026-03-23",
      "daily_revenue": 640
    }
  ],
  "skipped": [
    {
      "record_id": "r-88",
      "reason": "not newer than cursor"
    }
  ],
  "state": {
    "last_successful_cursor": "2026-03-23T02:10:00Z"
  }
}

Constraints

Respect the last successful cursor, keep natural-key idempotency visible, and advance state only from successfully processed records.

Expected skills

State handling, cursor boundaries, idempotent transforms, and pipeline-safe return shapes.

Python

AI evaluation

Run the transform to inspect normalized rows and error summaries.