Which schema details prevent ambiguity in generated adjustment records?

A generated release can provide JSONL for task episodes and Parquet for related transaction tables. Both should follow a versioned schema with stable identifiers and documented joins. An adjustment line might carry `adjustment_id`, `line_id`, `source_line_id`, `account_code`, `currency_code`, `amount_decimal` and `entry_side`.

Define monetary values using decimal semantics rather than binary floating-point arithmetic. State whether amounts are always non-negative with a separate debit or credit field, or whether signed values encode direction. Mixing those conventions can turn a valid correction into a scoring error.

Relationships also need explicit cardinality. An adjustment can affect several source lines, and one source line can receive several adjustments over its lifetime. A link table can represent those relationships without repeating or overwriting source records.

Document the difference between an unknown value, an inapplicable value and an omitted field. Validation hooks can reject malformed currency values, orphaned references and incompatible schema versions before an evaluation starts. This leaves business uncertainty inside the task while catching transport and representation problems at the ingestion boundary.