Skip to main content
AGENCYSCRIPT
CoursesEnterpriseBlog
👑FoundersSign inJoin Waitlist
AGENCYSCRIPT

Governed Certification Framework

The operating system for AI-enabled agency building. Certify judgment under constraint. Standards over scale. Governance over shortcuts.

Stay informed

Governance updates, certification insights, and industry standards.

Products

  • Platform
  • AI Scripts
  • Certification
  • Launch Program
  • Vault
  • The Book

Certification

  • Foundation (AS-F)
  • Operator (AS-O)
  • Architect (AS-A)
  • Principal (AS-P)

Resources

  • Blog
  • Agency Archetype Quiz
  • Free Live Training
  • Build AI Agents Masterclass
  • Build with AI Challenge
  • OS Plugin Install
  • Verify Credential
  • Enterprise
  • Partners
  • Pricing

Company

  • About
  • Contact
  • Careers
  • Press
© 2026 Agency Script, Inc.·
Privacy PolicyTerms of ServiceCertification AgreementSecurityCookies

Standards over scale. Judgment over volume. Governance over shortcuts.

On This Page

R — Retrieval: Confirm the FoundationWhat this stage coversWhat it protects againstI — Inspection: Read the Output CriticallyWhat this stage coversWhat it protects againstG — Good-law Check: Verify Authority Still HoldsWhat this stage coversWhat it protects againstO — Outside Verification: Open the Source YourselfWhat this stage coversWhat it protects againstWhen it appliesR — Review: Close the LoopWhat this stage coversWhat it protects againstApplying the Model in PracticeCalibrating to stakesMaking it a shared standardWhere Teams Tend to Break the ModelCollapsing Inspection and Outside verificationTreating Retrieval as a one-time concernSkipping the Good-law check after a satisfying findEmbedding the Model Into Daily WorkMake the stages visibleUse the model for training and auditLet it evolveApplying RIGOR to a Single MatterFrom question to filingWhy the order mattersComparing RIGOR to Working Without a ModelWhat ad hoc research missesWhat the model addsFrequently Asked QuestionsWhat does RIGOR stand for?Which stage is the most important?Do I have to run all five stages every time?How is this different from a simple checklist?What does the Review stage actually involve?Key Takeaways
Home/Blog/The RIGOR Model for Evaluating Legal Research Engines
General

The RIGOR Model for Evaluating Legal Research Engines

A

Agency Script Editorial

Editorial Team

·February 7, 2017·8 min read
ai legal research platformsai legal research platforms frameworkai legal research platforms guideai tools

Ad hoc approaches to AI legal research tend to fail in the same places, because they have no structure to catch the predictable problems. A reusable model fixes that by giving each recurring risk a named stage where it gets addressed. The RIGOR model organizes the work into five components—Retrieval, Inspection, Good-law check, Outside verification, and Review—each mapping to a specific failure these tools produce.

The point of naming the stages is repeatability. When the process has a shape, you can train people on it, audit against it, and notice when a step was skipped. A model that lives in someone's head protects no one; a model with named stages becomes a shared standard.

What follows defines each stage, explains what it protects against, and notes when it matters most. Apply all five for anything load-bearing; calibrate down only for low-stakes orientation work.

R — Retrieval: Confirm the Foundation

The model starts before you run a single query, with the foundation the platform stands on.

What this stage covers

Confirm the platform retrieves answers from a real, current, jurisdiction-appropriate legal corpus rather than generating from a model's memory. This is a property of the tool, established once during vetting and rechecked periodically.

What it protects against

Ungrounded generation—the root cause of fabricated citations. A platform that does not retrieve first cannot be trusted regardless of how the later stages go. The overview of what to understand about these platforms treats this as the central question.

I — Inspection: Read the Output Critically

With a grounded tool in hand, the per-query work begins.

What this stage covers

Treat each answer as a hypothesis. Read it critically, note every claim and citation, and identify what you would need to confirm before relying on it. Frame the output as a lead, never a conclusion.

What it protects against

Passive acceptance—the mindset that turns convenient output into uncritical reliance. Inspection is where you decide what to verify. The best practices expand on this stance.

G — Good-law Check: Verify Authority Still Holds

Existence and accuracy are necessary but not sufficient.

What this stage covers

For every authority you intend to rely on, confirm it is still good law: not reversed, overruled, or materially limited. Use the platform's treatment indicators and trace forward for high-stakes authority.

What it protects against

Building arguments on foundations a later court already removed. This stage is easy to skip once you have found a case you like, which is exactly why it is its own named step. The common-mistakes catalog documents how often it gets dropped.

O — Outside Verification: Open the Source Yourself

The stage that defines the whole model, because it is the one most often skipped under pressure.

What this stage covers

Open and read each cited authority in the underlying database. Confirm it exists and confirm it says what the tool claims. This is non-negotiable for anything that leaves your desk.

What it protects against

Both fabricated citations and mischaracterized real cases—the two failures that produce sanctions and embarrassment. The step-by-step query process sequences this verification explicitly.

When it applies

Always, for load-bearing authority. Orientation-only research for personal understanding can defer this, but nothing reaches a filing without it.

R — Review: Close the Loop

The final stage extends beyond a single query to the ongoing health of the practice.

What this stage covers

At the matter level, document what was verified so diligence is provable. At the practice level, periodically reassess the platform's corpus currency and vendor data practices, and track error and time-saving metrics.

What it protects against

Unprovable diligence and quiet drift. Without review, a tool that was safe at adoption can become unsafe as conditions change, and good work becomes hard to defend if challenged.

Applying the Model in Practice

The stages are not equally heavy on every task, and the model is meant to flex.

Calibrating to stakes

A quick orientation query touches Retrieval and Inspection lightly and may defer the rest. A brief headed to court runs the full sequence with rigor. Match effort to consequence.

Making it a shared standard

The model's real value appears when a whole team uses the same named stages. Training, auditing, and catching skipped steps all become possible once everyone speaks the same five-part language. The beginner's introduction is a gentle on-ramp to the verification stages.

Where Teams Tend to Break the Model

A model only helps if people follow it, and the predictable ways it breaks are worth naming so you can watch for them.

Collapsing Inspection and Outside verification

Under deadline, people read the answer and treat that as verification, skipping the step of actually opening the source. Keeping these as two distinct, named stages makes the collapse visible: if you only did Inspection, you are not done.

Treating Retrieval as a one-time concern

A platform that was grounded and current at adoption can drift as its corpus ages or the vendor changes practices. The Review stage exists partly to re-touch Retrieval periodically, so the foundation does not silently erode.

Skipping the Good-law check after a satisfying find

The moment you find a case that says what you want, the search feels finished. The named Good-law stage interrupts that satisfaction with a required question: is this authority still standing?

Embedding the Model Into Daily Work

Naming stages is only half the job; the model has to live in the workflow, not in a document nobody reopens.

Make the stages visible

Some teams put the five stages on a research template or a checklist attached to each matter, so the sequence is in front of the lawyer rather than in memory. Visibility is what keeps the Outside verification stage from quietly disappearing on a busy day.

Use the model for training and audit

Because every stage has a name and a purpose, you can train new lawyers against it and audit completed work by asking which stages were run. The common-mistakes catalog maps neatly onto the stages a team skipped, which makes the model a useful diagnostic when something goes wrong.

Let it evolve

The five stages are a durable backbone, but the specifics—how you document Review, which treatment indicators you trust—should adapt as your tools and practice change. A model you never revisit becomes folklore; one you periodically refine stays useful.

Applying RIGOR to a Single Matter

To make the stages concrete, consider how they run end to end on one realistic research task.

From question to filing

You start with a grounded platform you vetted during the Retrieval stage. You frame a precise query and read the answer critically—Inspection—noting each claim and citation as a hypothesis to test. For every authority the answer offers, you run the Good-law check against treatment indicators, discarding anything reversed or overruled. Then you perform Outside verification, opening each surviving source and confirming both that it exists and that it says what the tool claimed. Finally, at the Review stage, you log what you verified and, periodically, reassess the platform itself.

Why the order matters

The sequence is not arbitrary. Inspection identifies what to check, the Good-law check filters out dead authority before you invest effort reading it, and Outside verification is the irreducible step that catches the failures with real consequences. Running them out of order wastes effort or, worse, lets an unverified citation slip toward a filing.

Comparing RIGOR to Working Without a Model

It is worth being explicit about what the structure buys you, since the stages can look like overhead at first glance.

What ad hoc research misses

Without a model, the same lawyer doing the same task will verify thoroughly on a good day and skip steps on a bad one, with no way to tell the difference afterward. The failures cluster precisely in the steps that feel skippable—Good-law and Outside verification—because nothing forces them. The common-mistakes catalog is essentially a list of stages people drop when no model holds them in place.

What the model adds

By naming each stage, RIGOR converts a private, variable habit into a shared, auditable standard. You can train against it, review work against it, and diagnose a failure by asking which stage was skipped. That repeatability, more than any single stage, is the point.

Frequently Asked Questions

What does RIGOR stand for?

Retrieval, Inspection, Good-law check, Outside verification, and Review. Each stage maps to a specific failure mode that AI legal research tools tend to produce.

Which stage is the most important?

Outside verification—opening and reading each cited source yourself. It catches both fabricated citations and mischaracterized real cases, the two failures with the most serious consequences.

Do I have to run all five stages every time?

No. Calibrate to stakes. Orientation-only research may touch the early stages lightly, but anything load-bearing or headed to a filing should run the full sequence.

How is this different from a simple checklist?

A checklist lists items; the model assigns each recurring risk to a named stage, which makes the process trainable, auditable, and easy to notice when a step is skipped.

What does the Review stage actually involve?

Two things: documenting per-matter verification so diligence is provable, and periodically reassessing the platform's corpus and vendor practices while tracking error and time metrics.

Key Takeaways

  • The RIGOR model assigns each recurring AI-research risk to a named stage: Retrieval, Inspection, Good-law check, Outside verification, Review.
  • Retrieval confirms the tool grounds answers in a real corpus—the foundation everything else depends on.
  • Inspection and Good-law checks guard against passive acceptance and reliance on overruled authority.
  • Outside verification is the non-negotiable core: open and confirm every cited source.
  • Review closes the loop with provable diligence and ongoing reassessment, and the whole model calibrates to stakes.

Search Articles

Categories

OperationsSalesDeliveryGovernance

Popular Tags

prompt engineeringai fundamentalsai toolsthe difference between AIMLagency operationsagency growthenterprise sales

Share Article

A

Agency Script Editorial

Editorial Team

The Agency Script editorial team delivers operational insights on AI delivery, certification, and governance for modern agency operators.

Related Articles

General

Rolling Out AI Hallucinations Across a Team

Most teams discover AI hallucinations the hard way — a confident-sounding wrong answer makes it into a client deliverable, a legal brief, or a published report. The damage isn't just to the output; it

A
Agency Script Editorial
June 1, 2026·11 min read
General

A Model Behind an API Is Only Potential

Large language models don't do much on their own. A model sitting behind an API is potential, not capability. What converts that potential into something useful—something that drafts, classifies, summ

A
Agency Script Editorial
June 1, 2026·11 min read
General

Case Study: Large Language Models in Practice

Most teams that fail with large language models don't fail because the technology doesn't work. They fail because they treat deployment as a one-time event rather than a discipline — pick a model, wri

A
Agency Script Editorial
June 1, 2026·11 min read

Ready to certify your AI capability?

Join the professionals building governed, repeatable AI delivery systems.

Explore Certification