Forensic Accounting

A fraud examination collects material that a subject would like to see. It also produces findings that must survive challenge from opposing counsel. A model can help with both tasks and can weaken both if the traffic is uncontrolled.

A fraud examiner reaches for this starter when interview notes, ledger extracts and tip records already reach a model. Requests that mention wire instructions, a client account export or a secret never reach the model. Redaction covers national identifier numbers, which often appear in interview notes. Injection screening runs first, so a planted instruction inside a collected document cannot disable the checks that follow.

Two screens read the returned text. A returned finding that uses the words insider trading is blocked, as is text that names material non-public information. Promise language is refused as well, including guaranteed returns and a direct call to buy this stock. Neither screen judges whether a finding is correct.

Scoring rejects a findings draft below 0.86 overall, below 0.78 on relevance, or below 0.80 on accuracy. Fewer than two sentences fails, and so does text under 120 characters. The two measures count equally toward the combined score.

Evidence handling stays with the examination team. The final marker records that a call was allowed. It builds no evidence store, sets no retention period, and makes no record immutable. The oversight step holds no settings yet, which makes it a review point to enable rather than an active handoff. Chain of custody, work paper support and any expert opinion remain outside what this starter provides.