Pre-funding QC
Catch defects before delivery. Pair DeepV reviews with your existing QC sampling for second-look coverage on every file.
← Back to HomeInside the RISC engine → Meet the brains →
Appraisal Review
Deterministic UAD 3.6 appraisal review at file, batch, and portfolio scale
DeepV reviews every UAD 3.6 appraisal with the same 20 canonical brains, the same 21-step Prime Brain runtime, and the same evidence-linked output every time. Reports are reconcilable months later; every finding traces back to the page, field, image, comparable, or supplement that drove it.
What you get on every review
Deterministic by contract
Same appraisal payload + same pinned configuration (rule pack, brain versions, prompt version, seed) produces the same review. Reports are designed for re-run consistency months after the original review.
20 canonical review brains
Every brain runs on every review. Each brain reports findings emitted, "searched but not found" with attempted sources, or "not applicable". No silent passes.
RISC scoring with materiality weights
Findings roll up to a 0–100 composite (lower is better) plus a black-ball indicator for hard fails. Materiality channels separate value-impacting, compliance-only, and cosmetic outcomes.
Glass-box audit trail
Each review pins rule_pack_version, brain_version_ids, prompt_version, model_id, temperature, seed, and the LangSmith trace_id. Findings carry evidence references back to the source field, image, or supplement.
Render-back parity
The parser output renders back as a structured form view so reviewers see exactly how DeepV read each UAD field. The render and the brain inputs stay in lock-step.
Pay per review
Priced per review. Investor and internal demos run through admin-authorized review credits inside the DeepV portal.
RISC palette
Six levels, lower is better. The black ball surfaces hard fails on any review, even when the composite score is otherwise clean.
- Low Risk
- Mild Risk
- Moderate Risk
- Elevated Risk
- Critical Risk
- Unknown / Unverified
Where DeepV review fits
Post-close audit
Run reviews against closed loans for repurchase-risk segmentation and process-quality monitoring.
Repurchase-demand defense
Reconcile demanded defects against the original payload and produce evidence packets your team and counsel can use.
Learn moreSecondary-market QC
Run portfolio-scale review pre-delivery; surface RISC distribution, defect categories, and repurchase-risk mapping.
Learn moreMI rescission support
Reconcile mortgage-insurance rescission claims against the underlying appraisal payload and structured findings record.
Learn moreConsulting + forensic audit
Independent, deterministic review records for consulting engagements, internal audit, and policy reviews.
21-section coverage
Each canonical section ties to a primary brain. Sections 17 (FHA / VA review) and 18 (ROV / rebuttal) are review-team functions that wrap multiple brains.
Sample review report contract
A real, deterministic R81 sample is committed in the repo and served from/sample-reports/r81-sample-review.pdf. It demonstrates the brain-by-brain output, RISC roll-up, evidence references, and audit-trail block.
- Brains executed per review
- 20 / 20
- Sections covered
- 21 (incl. S12A)
- Output format
- Structured JSON + audit-trail PDF
- Repeatability
- Format-stable always; content-stable when configuration is pinned
- Trace integration
- LangSmith trace_id pinned per review
What runs on every review
Twenty canonical brains across the 21-step Prime Brain runtime. Click any name to see the brain's full backstory, capabilities, and routing graph.
Run a real appraisal through DeepV
Pay-per-review pilot on a real file from your portfolio. We share the full review output, the audit trail, and the source/evidence map. Priced per review.