Avro

The Avro adapter accepts both single‑schema `.avsc` files and multi‑schema `.avpr` protocol files. Records, enums, fixed types, arrays, maps, and unions are all preserved in the MetaModel without lossy flattening.

Avro logical types — `decimal`, `uuid`, `date`, `time-millis`, `timestamp-millis`, `local-timestamp-millis`, and `duration` — are retained as first-class annotations on the corresponding MetaModel attribute. Downstream Semantic Match passes use these annotations to drive the `domain_equivalence` algorithm class instead of treating logical types as plain strings.

Default values, aliases, and namespace qualifiers are preserved verbatim. When a downstream consumer reads back the schema through the export pipeline, the round-trip produces a byte-identical artifact — which is what makes Avro safe to use as a wire format under a hash-chain audit regime.

Each ingest writes a `SchemaIngested` entry to the ULID-keyed, SHA-256 hash-chained ledger. The entry records the source fingerprint (sha256 of the raw bytes), the importer module identifier and version (`AvroImporter@x.y.z`), the actor's role at time of action, and the assigned schema ULID. A reviewer can replay the chain to prove the MetaModel state was derived from a specific `.avsc` artifact.

Schema evolution rules — backward, forward, full — are computed against the prior `SchemaVersionPublished` entry's payload hash and recorded as `SchemaEvolutionEvaluated`. A breaking change cannot reach a Published state without an explicit Approver acknowledgment, satisfying SOX §404 change-management ITGCs and 21 CFR Part 11 §11.10(c).

Avro support is paired with the Predictive Mapping Semantic Match engine: when a target field's logical type is `decimal(precision=18, scale=4)`, candidate sources with matching precision and scale are surfaced first, instead of being hidden behind name-similarity heuristics.