Trust & methodology
Figures you can trace. Decisions you can reconstruct.
ForensicTrail is built so a reviewer, and later a court, can trace a number back to its source and a decision back to the person who made it.
How it holds up
Six controls behind every case.
Evidence integrity
Every page extracted from a statement receives a sequential Bates label for the case, such as MARLO_000147, rendered on the ledger's source link, with an ordered manifest available for discovery production. Manually entered transactions and CSV imports have no source page, so they carry no label.
Reconciliation
Extraction only accepts a file once its transactions reconcile against the statement's printed balances; a mismatch — or activity the system can't fully account for — holds the file for a reviewer instead of passing it on to classification.
Human review
Transactions the system isn't confident classifying wait in one case-wide human review queue, where a reviewer approves, corrects, or excludes each one — an exclusion carries a reason — before it reaches analysis or a report.
Decision audit trail
Rule changes, review decisions, and exclusions are logged with who acted, when, and the before-and-after values — a decision audit trail kept separate from the file-handling chain of custody.
AI boundaries
AI assists in bounded places. In extraction, it can work out how to read an unfamiliar statement layout, and that reading is used only once it reconciles the document against its printed balances; it can also itemize a paystub's lines, kept only when they add up to the printed totals. In classifying transactions, it handles what the rules, the known-merchant lookup, and earlier classifications couldn't place, and it takes a second look at large, flagged transactions it categorized, sending any disagreement to review. And the read-only case assistant answers questions about one case, citing the transaction or case data it retrieves. Reconciliation, transfer matching, Bates labeling, and the digital fingerprint recorded when extraction first reads a file don't depend on any AI model. A reviewer decides what a low-confidence classification means, and an investigator decides what goes into a saved Finding; the assistant does not create Findings on its own.
Security and firm isolation
Each firm's data is kept separate from every other firm's. A firm's users can reach only the firms they currently belong to, and our servers check that membership rather than trusting the user's browser. This isolation is between firms, not cases: anyone signed in to your firm can open every case in it, since there is no separate access control between cases in the same firm yet.