How can customers evaluate corrections beyond matching the final balance?
Invariant-based evaluation checks what must remain true throughout a correction, exposing unauthorised side effects that a final-balance comparison can miss.
A correct final balance does not establish that a correction was performed correctly. An agent could reach that balance by deleting an unrelated item, changing a source amount or posting an offset to an inappropriate account. Evaluation needs constraints on both the intended change and the surrounding records.
A generated evaluation pack can define case-specific invariants. These might require that unrelated account balances remain unchanged, the original source document remains intact and no additional payment instruction is created. Some invariants apply throughout execution; others apply only when the case reaches its terminal state.
Customers can compare the initial and final state against an allowed change set. The evaluation report should identify unexpected mutations by record and field, rather than return only a pass or fail result.
For resellers, include a readable explanation of each invariant’s business purpose. This lets customers determine whether the checks suit their intended use. Avoid presenting a narrow set of supplied invariants as proof that every accounting or operational consequence has been covered.
Provide customers with outcome checks that detect damage outside the target record.
- Audience:
- Licensed resellers
- Content type:
- Explanation
- Buyer question:
- How to evaluate it
Content release: r-a220eb6e099b243b
Related questions
- How can developers evaluate a correction beyond checking the final balance?
- What should buyers inspect before trusting a generated correction pack?
- How should an agent behave when a posting request times out?
- How do we distinguish a justified pause from an agent avoiding the task?
- How should evaluation hooks avoid leaking the expected correction to the agent?
- Which hooks reveal duplicate posting after a tool timeout?