DociCA

How DociCA verifies every legal citation against real government text

Last updated: June 2026 (Beta)

Beta. DociCA is in invite-only beta. These terms apply to beta use. DociCA is document-automation software, not a law firm, and does not provide legal advice. See our Beta Disclaimer.

DociCA's document review verifies every statutory citation against real government text and removes anything it cannot confirm. Instead of trusting a language model to remember the law, DociCA checks each citation against authoritative published statutes and only keeps what it can prove. This page explains the process and its limits.

The problem: language models can invent law

General-purpose AI can produce citations that look authoritative but do not exist, or quote statute text that was never written. For a legal document that is unacceptable. DociCA is built so that the parts of a review that assert law are checked against real sources, and unverified assertions are stripped rather than shown.

Three grounding tiers

Every point in a DociCA review carries a grounding tier that tells you how well it is supported:

  • Verified against government text — the review quotes statute language that has been confirmed against the real published text.
  • Verified citation — the statute reference has been confirmed against an authoritative legal database as real and in force.
  • Plain explanation — general information with no asserted law. Many useful findings sit here, and that is expected.

How the verification works

DociCA separates finding issues from citing law, then checks the citations before anything is shown:

  • First, the model reads the document and flags candidate issues, referencing statutes only by a curated key — never by inventing a URL or reproducing statute text.
  • Each referenced statute is then confirmed against an authoritative legal database. References that cannot be confirmed — or that point to a repealed statute — are dropped.
  • When the review quotes statute language, that quote is accepted only if it is a literal substring of the real government text (compared after normalizing spacing and letter case, above a minimum length). A quote that does not match is removed, and the finding falls back to a plain explanation with no asserted law.
  • Finally, the report is assembled deterministically: it attaches only the citations the verification step confirmed and removes the rest. The model never supplies the citation text itself.

Curated statutes, not invented links

DociCA references statutes from a curated registry mapped to government sources — federal Justice Laws, the Alberta King's Printer, and Ontario e-Laws. This is deliberate: letting a model emit its own citation links is exactly what invites hallucinated law. The registry is the single source of truth for what the review is allowed to cite.

What the review flags

Findings are sorted into objective, checkable categories rather than matters of style or preference — for example a missing standard protection, a citation that does not match the document's jurisdiction or type, a term that falls short of a statutory standard, or a drafting defect such as an unfilled placeholder or broken cross-reference.

What this does — and does not — mean

Verified grounding makes the law DociCA cites trustworthy. It does not turn the review into legal advice. DociCA's review is issue spotting: a plain-language issue summary and a review-priority score. It can miss issues and can be wrong. It is not a legal opinion, a risk rating, or a guarantee, and it does not replace a lawyer. Generated documents are drafts for review; every template is currently under lawyer review. Always confirm anything important with a qualified lawyer.

See also our FAQ and Beta Disclaimer.