DocAI contract scan is part of BizLegal. 24/7 compliance ops, $2,500/mo managed.See the offer →
Skip to main content

Disclosure v1.0.0-p4

DocAI methodology.

How DocAI produces document intelligence: source handling, knowledge-base retrieval, confidence scoring, human review, and the explicit limits of what we do and do not do with your documents.

Source handling.

When you submit a document, the text is processed for analysis and the result is made available to you. Submitted documents are retained for 30 days unless you request earlier deletion, then purged. We do not use customer documents to train shared models. The privacy page sets out transfer and retention particulars.

Knowledge-base retrieval.

For questions DocAI can answer without a submitted document, retrieval runs against a vetted knowledge base of public regulatory texts and guidance. Every answer cites the source passages it drew from. If the knowledge base does not cover a question, the system says so rather than hallucinate.

Confidence scoring.

Every response carries a confidence score in [0, 1] composed of four weighted components, published as part of this methodology and versioned with the disclosure stamp:

  • Retrieval signal (40%) — top-hit relevance from the knowledge base.
  • Model certainty (35%) — 1 minus the model's reported uncertainty.
  • Citation integrity (15%) — 1 if every claim in the draft ties to a retrieved item, else 0.
  • Limits admitted (10%) — trust signal when the draft admits what it does not know.

Tiers at disclosure v1.0.0-p4: score ≥ 0.75 auto-delivers on free tiers; 0.45 ≤ score < 0.75 routes to a named human analyst; score < 0.45 holds with a request for more information from the submitter. Retrieval hits below 0.15 trigger the canonical "out of scope" response automatically — no fabrication path.

Human review.

On paid tiers, a named analyst reviews every generated answer or drafted clause before delivery. The reviewer verifies that the cited sources actually support the claim, flags uncertainty the model did not, and can hold delivery if any assertion cannot be substantiated.

What we do not do.

We do not practice law. We do not send filings on your behalf. We do not represent you in any transaction or proceeding. A generated clause is a starting point for your attorney, not a finished product for execution. The seven-clause disclosure on every page governs.

Versioning.

Every generated response carries a disclosure version stamp (v1.0.0-p4 at the time this page was generated) and an issued_at timestamp. If a response is ever challenged, the stamp lets us reproduce exactly which disclosure was in force and which knowledge-base version was active when the response was produced.

Incident response.

If a response is challenged — by a customer, opposing counsel, or a regulator — we follow the Liability Incident Response SOP on the hub workflows library. Preservation first, notification to counsel next, no fault admission prior to review.

Disclosure v1.0.0-p4

What we are. What we are not.

We are not a law firm.

BizLegal AI is a regulatory intelligence platform. We do not practice law in any jurisdiction. Accessing this site or using our tools does not create an attorney-client relationship.

Every output passes human review.

No intelligence ships to a customer without a named human reviewer signing off. The reviewer verifies sources, flags uncertainty, and can hold delivery if any assertion cannot be substantiated.

We do not file on your behalf.

We do not submit filings to FinCEN, the SEC, a court, a registry, or any other regulator for you. Where our output references a filing, you remain the filer of record and are responsible for submission, timing, accuracy, and consequences.

Jurisdictional scope is finite.

Our intelligence covers specified jurisdictions and frameworks. Gaps exist. If a question sits outside our coverage, we say so rather than infer. The pages listing covered jurisdictions and frameworks control.

Regulation changes. We do not warrant currency.

Regulatory texts, enforcement postures, and guidance evolve continuously. We make reasonable efforts to keep intelligence current, but we do not warrant that any output is complete or up-to-date at the moment you act on it. The disclosure version and issue timestamp on each output pin what was in force when the intelligence was produced.

Limits of liability.

Use of our intelligence is at your own risk and, where applicable, subject to the liability limits in our Terms. We are not liable for outcomes of decisions you make after reading our intelligence, including filings you submit, contracts you sign, or transactions you execute. If you have a complaint or a concern about a specific output, our incident response procedure is published on the Trust page.

BizLegal AI publishes regulatory intelligence. We are not a law firm. This is not legal advice. All output is reviewed by a human analyst before release and is subject to the version-stamped disclosure in force at the moment of issue.