Compatibility and support policy#
Level: Expert
You will: decide which APIs, schemas, manifests, and artifacts are safe to
carry across an upgrade.
Before you start: a released FraudTwin run or package version.
Services: None.
FraudTwin keeps generated artifacts reproducible, but not every interface has the same stability promise. Pin the documentation version when a pipeline or benchmark must be replayed exactly.
Surface |
Compatibility rule |
Status |
|---|---|---|
Supported public API exports |
Public names are reviewed for backward compatibility; removals are announced in the changelog. |
Stable |
Configuration YAML |
Validate with the matching package version. Unknown fields fail validation instead of being ignored. |
Stable |
Parquet/JSON/Avro contracts |
Schema versions and manifest fingerprints identify the producer and contract. Read with the corresponding contract version. |
Stable |
Public benchmark packs |
Packs declare generator compatibility ranges and frozen fingerprints. A mismatch is an error, not a warning. |
Versioned |
|
Optional integrations require their matching extra and external service versions. |
Optional |
Calibration, campaign dynamics, learned generation |
APIs may evolve between minor releases; persist profile/configuration fingerprints with results. |
Experimental |
|
|
Experimental |
|
|
Experimental |
Model-service HTTP schemas |
Request/response version |
Reference example |
Upgrade checklist#
Install the target version in an isolated environment.
Run
fraudtwin config validateagainst each configuration.Generate a bounded 1,000-payment fixture and compare manifest, schema, and logical fingerprints.
Re-run representative ML, graph, and integration checks before promoting the new version.
See migration notes and the release notes for version-specific changes.
Next#
Use Release evidence before publishing a package or benchmark result.