FAR — active build lane
Forensic Appraisal Review — the expanded DeepV review workflow
FAR widens the DeepV review beyond the appraisal payload. CU docs, title, purchase contract, borrower / loan package evidence, reviewer chat, chat history, memory, and the full audit trace land in one review record. The canonical 20-brain runtime stays the same; the evidence the brains reason over expands. FAR is an active build lane — the AR review demo is the current live path.
What FAR adds on top of AR
Beyond the appraisal payload
FAR ingests the appraisal alongside CU docs, title docs, purchase contract, and the broader borrower / loan package — every artifact that supports or contradicts the appraisal.
Same 20 canonical brains
Prime Brain still orchestrates the canonical 20-brain runtime; FAR widens the evidence set each brain reasons over and lets reviewers pull cross-document context into a single review.
Chat with the review
FAR adds a reviewer chat surface to the review record — the brains return structured outputs as in AR, and reviewers can ask follow-ups against the captured evidence rather than the raw model.
Chat history + memory
Every reviewer interaction is captured and traceable. Memory carries across reviews on the same file so a later QC pass sees what was already asked, answered, and decided.
Full audit trace
Every chat message, evidence pull, and brain output is pinned to the audit trail with rule_pack_version, brain_version_ids, prompt_version, model_id, seed, and LangSmith trace_id.
AR-grade RISC scoring
Findings still aggregate into the RISC bar with materiality weighting and a black-ball indicator. The wider evidence set produces denser findings without changing the scoring contract.
Evidence the review reasons over
Every artifact below becomes a first-class member of the review record.
Appraisal payload
1004 / 1004UAD / 1073 / MISMO 3.x
CU / SSR documents
Collateral Underwriter messages, SSR exports
Title documents
Preliminary title reports, vesting deeds, ALTA endorsements
Purchase contract
Signed contract, addenda, escrow instructions
Borrower / loan package
Income, asset, identity, credit, AUS findings
Reviewer notes + chat history
In-review questions, prior decisions, memory
Reviewer surface
Prime Brain fan-out across the package
The 21-step runtime runs on the expanded payload; brains can pull from any committed evidence document via Lex Parsewell.
Structured findings + repurchase-defense mapping
Same findings schema as AR. Findings that map to repurchase categories surface the mapping inline.
Reviewer chat over the captured evidence
Chat questions resolve against the committed evidence + brain outputs, never the open model. Every answer is reproducible from the audit trail.
Chat and memory in FAR live on the reviewer surface and are part of the audit record. They are not exposed in the AR review-only product, which remains chat-free.
Scope a FAR engagement
FAR is an active build lane and ships behind the customer portal. Scoping calls cover evidence sources, reviewer access, audit-trail requirements, and integration timing.