Which schemas and tool contracts can represent a generated financial adjustment?
A generated package can separate journal headers, journal lines and approval references, with typed posting actions that enforce accounting and relationship constraints.
A financial environment can separate `adjustment_header`, `journal_line` and `approval_reference` records. Header fields could include `adjustment_id`, `source_entry_id`, `reason_code`, `currency` and `status`. Each line links through `adjustment_id` and carries `account_id`, `debit_amount` and `credit_amount`.
Specify decimal precision and rounding rules in the schema. Avoid binary floating-point amounts for accounting validation. Within the scoped journal, total debits and credits must agree, and every referenced account must exist and accept the proposed posting type.
JSONL can carry individual records and action events; Parquet can carry typed tables for analysis. A schema manifest should describe required fields, permitted nulls and relationship constraints rather than leaving buyers to infer them from a sample.
Tool contracts could expose `draft_adjustment`, `validate_journal`, `submit_for_approval` and `post_adjustment`. Validation returns structured violations without changing balances. Posting returns the committed journal identifiers. An evaluation hook can compare the committed lines with the validated proposal, detecting a tool implementation that writes different values from those the agent submitted.
Make the adjustment schema precise enough to validate before an agent can commit a financial state change.
- Audience:
- Environment builders
- Content type:
- Integration
- Buyer question:
- What can be generated
Content release: r-a220eb6e099b243b
Related questions
- Which linked records could a generated correction dataset contain?
- Can generated tasks include policies that changed between the original entry and its correction?
- What files and environment state could a generated correction release include?
- What evidence conditions should a generated correction pack vary?
- How could a generated task represent a correction that must be split across destinations?
- What generated evidence can support an adjustment decision?