Output validation

A zero-exit process can still be a failed business run.

LatidoFlow evaluates strict numeric output against a monitor-level contract. If a required metric is missing or fails its comparison, the reported success becomes failed evidence and opens an invalid-output incident.

Check the result, not only the exit code.

A billing export can finish cleanly while producing zero rows. A synchronization can return HTTP 200 after writing fewer records than required. Output assertions let the workload report a bounded numeric map and make every configured rule part of the success decision.

The supported operators are gte, lte, and equals. Every assertion must pass. A missing metric or failed comparison is retained as invalid-output evidence and fails the run. A malformed output payload is rejected without terminalizing the run.

Example contract

"output_assertions": [
  {
    "metric": "records_processed",
    "operator": "gte",
    "value": 1
  }
]

A reported value of 0 fails this contract even if the process reached its success endpoint.

Send typed output from the real completion path

POST /api/v1/runs/{run_uuid}/success
Content-Type: application/json

{
  "output": {
    "records_processed": 42,
    "amount_exported": 1804.75
  }
}

A POST heartbeat can carry the same typed output for a completion-only integration. Direct GET success and GET heartbeat requests cannot supply typed output.

A deliberately bounded contract

Assertions maximum

Each monitor can define no more than ten output rules.

Metrics maximum

Each completion accepts no more than twenty output metrics.

Flat

Numeric map

Nested values are rejected; metrics must be finite JSON numbers.

All

Rules must pass

The contract does not average away a failed condition.

Current limitations

Numeric strings such as "42", nested JSON, non-finite numbers, and arbitrary text values are not accepted output metrics. Assertions currently apply to heartbeat monitors. This mechanism validates the fields your workload reports; it does not independently query the source database to confirm them.

Keep exploring

Zero-output invoice sync scenario See a test-backed invalid-output incident and recovery path. Cron job monitoring Detect missing completion before checking what a run produced. Run lifecycle API Review the POST transports that accept typed output.

Define one minimum acceptable result

Start with a metric whose zero or missing value would make a technically successful run useless.

Create an account Review output reporting