How a Forensic Appraisal Review runs end-to-end

FAR accepts the appraisal alongside CU docs, title docs, purchase contract, and the borrower / loan package. Each document type lands as a first-class artifact in the review record.

The intake service classifies every artifact, runs the 5-gate intake security model, and routes each document to the right Section 15 extractor (Lex Parsewell).

The MISMO 3.6 parser runs on the appraisal; document-specific parsers run on title prelims, CU exports, purchase contracts, and the loan package. All extracted fields are provenance-preserving.

The reviewer sees how DeepV read every document — appraisal grid, title clauses, contract terms, AUS findings — in a structured form view so the package and the brain inputs stay in sync.

The canonical 21-step runtime executes against the wider evidence set. Brains can pull from any document via Lex Parsewell. Every brain still produces structured output: findings, "searched but not found", or "not applicable".

Reviewers ask follow-ups inside the review record. The chat resolves against the committed evidence + brain outputs (never the open model). Every answer is reproducible from the trail.

Prior FAR sessions on the same file are visible to later reviewers. Memory is scoped to the file and to the engagement — no cross-customer leakage.

Same RISC contract as AR: materiality weighting, severity bucketing, black-ball indicator. Findings carry repurchase_defense_mapping where applicable.

Every chat message, evidence pull, and brain output is pinned to the audit trail. PDF download is gated behind the portal payment / report-completeness gate; admin-authorized review credits cover investor and internal demos.