Verity

At closing, the signed SPA (Share/Stock Purchase Agreement) is the single source of truth for an M&A deal but the disclosure schedules and board consent are drafted separately, often by different people at different times, and routinely drift out of sync with it: a purchase price that doesn't match, a schedule referencing a section that doesn't exist, a board consent authorizing the wrong signatory. Today this is caught, if at all, by a senior lawyer manually cross-reading three documents side by side before they can even start applying judgment.

Verity takes the signed SPA as ground truth and checks the Disclosure Schedules and Board Consent against it across four checks Term Consistency, Cross-Reference Validity, Authority Match, and Completeness flagging every mismatch with its exact location in both documents and a plain-English explanation. An LLM extracts each document's key terms once; the actual comparison then runs as fast, deterministic code, with a rule-based judgment layer for the two checks that need semantic matching (e.g. recognizing "CFO" and "Chief Financial Officer" as the same office) so results are fast and consistent instead of one large prompt guessing at differences. Verity checks consistency, not legal correctness: it flags, and the senior lawyer stays the decision-maker.

Four automated consistency checks — Term Consistency, Cross-Reference Validity, Authority Match, and Completeness, run against the signed SPA as ground truth Flexible document intake — required SPA + Disclosure Schedules + Board Consent, plus unlimited optional documents (Officer's Certificate, Escrow Agreement, IP Assignment, or anything else you label) automatically checked for consistency Two-stage AI pipeline — LLM extraction once per document, then mostly-deterministic comparison logic; every finding is tagged [rule], [LLM judgment], or [heuristic] so you know exactly how it was decided Offline fallback — if no LLM is configured or a call fails, a rule-based extractor takes over automatically; the tool never goes down mid-demo Severity-ranked flags — every inconsistency is marked Critical or Moderate so a long report can be triaged at a glance Click-to-highlight source verification — jump from any flagged finding straight to the exact passage in the original document it's based on Mark as reviewed — dismiss confirmed flags with a note to track progress through a report Extraction confidence warnings — surfaces when a document's structure kept a field from being read, instead of silently under-reporting Export & history — download the full report as Markdown or a printable PDF; past checks are saved locally for later reference Privacy by design — documents are processed in memory only and never stored server-side