Methodology

How the intelligence is made.

Claims on this site are architectural decisions you can inspect. This page documents how LegisGate produces regulatory findings, why no language model authors them, and where our role ends and your counsel's begins.

  1. 01The sourceHow Meridian is curated and verified
  2. 02The engineWhy findings are never model-authored
  3. 03The boundaryWhere LegisGate ends and counsel begins
01 · The source

The LegisGate Meridian™

The Meridian is LegisGate's curated regulatory intelligence source — the reference each finding is verified against. It is built and maintained by human curation: statutes, regulations, and supervisory authority guidance are read, mapped into obligations, and validated with effective dates and citation records.

“Verified to the Meridian™” means each citation in a report is resolved against this curated source. It is a machine-enforced validation of sourcing and structure — not a substantive legal review, which remains counsel's role.

The Meridian is read-only from the assessment pipeline. No report run can write to it. Vendor-specific intelligence flows through a separate per-assessment path and is not written into the regulatory source — preventing one customer's assessment from contaminating the regulatory intelligence used by another.

284 binding laws and regulations Statutes and regulations that create enforceable duties for organizations that deploy AI — the instruments counsel can cite.
92 supervisory authority guidance documents How regulators explain and apply those laws in practice — opinions, guidelines, and agency publications used alongside the statute.
242 jurisdictions with mapped obligations Countries and US states where Meridian has mapped what can apply once you declare footprint and use.
3,032 obligations mapped across jurisdictions The concrete duties those laws create — assessments, notices, documentation, monitoring, and related requirements — tied to place and use.
As of August 12, 2026 · Snapshot · Counts update as the Meridian is curated
02 · The rule

Findings are assembled, not generated.

Regulatory findings are assembled deterministically from Meridian-validated templates and rule evaluations. A language model does not author the statute identified, the citation, the obligation description, the severity, or the required actions.

Regulatory citation

12 CFR Part 748 Appendix A

Verified · LegisGate Meridian™

View in Intelligence Library

Meridian record ID: glba:ncua-12-cfr-748-app-a

Verified in Meridian: 2026-07-31

Potential penalty exposure

No statutory maximum mapped for this finding in the report penalty model.

Provenance Meridian template — validated 2026-07-31 View obligation instances in Part 2
Compass finding provenance — same fields as a delivered report, rendered for screen clarity

If a finding cannot be traced to a validated Meridian record or a deterministic rule, it does not ship.

03 · The gate

Reports pass a verification gate before delivery.

Assembled findings are checked before a report is delivered. The gate applies quality checks across sequential batches so each report meets the verification bar before it ships.

CheckWhat it tests
DeduplicationNo finding appears twice under different framings
Citation accuracyEach citation resolves to its Meridian record
Severity consistencyRatings are consistent with the obligation's statutory basis
CompletenessRequired fields are present for each finding
Jurisdiction validityEach finding is valid for the jurisdictions in scope
Count reconciliationFindings, obligations, and assessment determinations reconcile

Quality metrics from the gate are logged for each report.

04 · The triggers

Obligated when thresholds are met — not “assessment just in case.”

Mandatory assessments are determined by evaluating a deployment's facts against statutory threshold criteria for each jurisdiction in scope. A mandated assessment is identified when the statute's thresholds are met, with the statutory basis recorded in the determination.

01

No padding for coverage.

If a deployment does not trigger an assessment in a jurisdiction, none is recommended for that jurisdiction. Silence is more honest than a fabricated obligation.

05 · The division of labor

We bring the regulatory intelligence. You bring the judgment.

LegisGate

LegisGate prepares the regulatory substance of an assessment: which laws apply, which citations anchor each obligation, which risks the statute contemplates, and which thresholds triggered which mandated documents.

Your counsel

Your counsel supplies the organizational substance: measures implemented, data flows, internal controls, personnel and contracts — the facts only your organization knows, and the judgment only counsel can apply.

Sections requiring organizational input are explicitly identified in each prepared document. They are never pre-populated with generative content.

Counsel reviews, completes, and determines sufficiency. LegisGate prepares the file. We do not provide legal advice, certify compliance, or promise an enforcement outcome.

06 · The record

Timestamps that cannot be rewritten.

Each report carries a SHA-256 digest computed over its contents at generation. The digest appears on the report cover and allows counsel, auditors, or regulators to confirm the document matches the report as issued.

The Regulatory Review Initiation Record documents when an organization initiated its regulatory review of a deployment. It records the initiation of a review; it is not a certification of compliance or of the sufficiency of subsequent action.

Monitored vendor documents are baselined by cryptographic hash, and detected changes are evaluated against the assessments they affect.

Digest
SHA-256 computed at generation
Timestamp
Generation date and time recorded
Initiation
Regulatory review initiation record
Baseline
Vendor document hash
History
Change evaluation record
07 · The limits

Stated limits, on the record.

We do not provide legal advice. Counsel determines the sufficiency of any review, assessment, or remediation.

We do not certify compliance, and no output promises an enforcement outcome.

Coverage is curated, not exhaustive. The Meridian is maintained on an ongoing basis; counsel supplements where their judgment requires.

We do not author regulatory findings with a language model, and we do not let assessment runs write to the regulatory source.

We do not contact regulators about any client, and we do not participate in a client's regulatory matter.

A methodology you can inspect is the only kind worth trusting.

See the methodology in the artifact it produces.

See a sample report →