The review method

Every finding needs a source and an unresolved claim.

The unit of work is a draft credit memo, its existing issue log and a frozen set of source documents.

Reconcile before interpreting.

Check source periods, units, currencies, consolidation scope and the lender's definition of cash available for debt service. Unknown inputs remain unknown. A second source check on decision-critical fields is separate from confirming that a document is authentic.

Make the memo the reference point.

A useful finding identifies the exact claim, the contradictory or missing support, why the stated mitigation does not resolve it, and what evidence could disconfirm the finding. Repeating a risk the analyst already addressed adds work without adding information.

Keep arithmetic explicit.

Financial ratios use decimal calculations under an approved definition version. Each numerator and denominator adjustment retains its source. A different definition needs a matching threshold. Model prose cannot silently redefine the calculation.

Separate evidence from disposition.

Validity, novelty, materiality and analyst action are distinct fields. Each material finding has an individual human disposition and rationale. Disputed material closures and overrides require a different reviewer.

Test the mechanism.

Evaluation compares rules and reconciliation, one model with specialized passes, two independent providers and a bounded critique round. Prospective shadow files carry more weight than historical demonstrations. The baseline includes the analyst's actual existing tools.

No finding, model vote or report status approves a loan. An empty finding list does not establish that a file is complete or free of issues.