Reproducible technical scenario
The invoice sync completed, but zero records should not look healthy.
A process exit can be technically successful while its business outcome is empty. This automated product proof checks that LatidoFlow rejects that false green when a numeric output contract requires at least one exported row.
Last reviewed: August 1, 2026
Lab evidence, not a customer story
The invoice-sync label makes the failure concrete. The verified behavior comes from automated acceptance coverage and does not represent a named customer, production event, or measured revenue impact.
Direct answer
A zero-row success becomes a failed run.
The monitor requires exported_rows >= 1. The run starts normally and sends a success completion containing exported_rows: 0.
LatidoFlow evaluates the output before accepting the completion. The response and stored run are marked failed, one invalid_output incident opens, and the linked monitor, component, and overall status page become unhealthy.
The public incident stays generic while the actual metric, comparator, threshold, and observed value remain in private evidence. A later run whose output satisfies the contract can recover the occurrence.
Verified evidence chain
Nominal success is evaluated before it becomes operational truth.
exported_rows: 0 / required: gte 1
-
01
Start evidence
The run receives a unique identity and enters the running lifecycle.
-
02
Typed output
The completion carries a flat finite numeric metric rather than an untyped success message.
-
03
Failed assertion
Zero does not satisfy the configured lower bound, so the completion cannot remain successful.
-
04
Visible outage
The incident updates the linked component and overall public page to major outage.
What this proves
- Output assertions run before a success completion is accepted.
- A failed condition creates one distinct invalid-output occurrence.
- Private evidence retains the expected and actual numeric values.
- Linked public state changes in the same verified failure path.
- A subsequent valid completion can restore healthy state.
Current boundary
- This scenario does not measure customer impact, detection time, or money recovered.
- A monitor accepts up to 10 conditions using
gte,lte, orequals; every condition must pass. - Output is limited to at most 20 flat, finite numeric metrics. It is not arbitrary JSON or schema validation.
- Numeric strings are not treated as numeric proof.
- Public pages show on-page updates; email status-page subscriptions are not available.
Follow the output-validation path
Define what a useful completion looks like.
Start with one business-critical job and one bounded numeric outcome. LatidoFlow can keep a technically successful process from masking an empty result.