How NoteGage turns claims into a reviewable analysis.

A report starts by capturing what the listing actually says — the stated figures, the seller's comments verbatim, and every attached document. Files that can't be analyzed are listed with the reason rather than silently dropped, because an evidence file with quiet gaps is worse than none.

A stated property value is checked against independent sources — the county assessor's market value, an independent automated valuation, the most recent recorded sale, and any attached appraisal or BPO. When independent sources agree within tolerance, the value is corroborated and the deal is priced from that basis. When they disagree materially, the value is flagged as unconfirmed — the report never resolves a disagreement in the seller's favor, and never simply picks the highest number.

The supported basis, exactly

The rule is deterministic and worth spelling out. Independent sources — the county assessor, the vendor's automated valuation and comparable-sales estimate, a recent arm's-length recorded sale — are first qualified (same collateral, a comparable date, an assessed value that actually describes the whole property). The qualified independent sources form a supported range from their lowest to their highest figure; a strength-weighted center is reported alongside it. When they agree within ±20% the value is corroborated. The underwriting basis is the conservative end of that range — every collateral-dependent figure is priced from the low end, and sensitivity is shown across the range. On the sample deal: assessor $154,440, AVM $158,837, comparable-sales estimate $163,534 → supported range $154,440–$163,534, center ~$156,439, underwriting basis $154,440. Seller-side figures — the stated value, an attached appraisal, its comps — are compared against the range as claims but never blended into it. A spread inside the range is treated as decision-material only when it would change the recommended position; otherwise it is sensitivity, not a blocker.

Every value a report shows is either something the source literally stated, something read from a document (labeled as an auto-read to confirm), independent record data, or our own calculation — and it is labeled as which. A seller's claim is never displayed as an established fact, and our inference is never attributed to the seller.

When stated terms contradict each other, the yield is refused with the reason shown — not computed from a silent assumption. When the value is unconfirmed, the ratios that depend on it are withheld rather than printed with false confidence. A calculation that didn't change because its input didn't change is never presented as independently verified.

A payment ledger substantiates the months it covers and nothing after its final entry. A county record corroborates a lien claim; it does not prove title. Absence of a public record is treated as absence of evidence — never as confirmation that a claim is true.

When independent evidence contradicts a stated figure, the figure stays visible — struck through, with both numbers shown — and everything computed from it is recalculated in place.

AI is used to read — to extract terms from listing text, to read attached documents, and to write the explanation of a finished report — and every extraction is validated against a strict schema before it is stored. The deal math is deterministic code, so the same inputs always produce the same numbers, and a financial fact the sources never stated is reported as missing rather than generated. The explanation is constrained to the report's own facts: it may not compute new figures, dispute the deterministic result, or reinterpret which source is authoritative.

NoteGage is pre-purchase triage and verification. It does not order title insurance or an O&E report, review the collateral file for original documents, verify the servicer's ledger against the borrower, commission an appraisal, or provide legal, tax or investment advice. Its job is to make sure the diligence money you do spend goes to deals whose numbers survived the first pass — and to hand you the open questions, with the evidence that would resolve each one.