Complex problems — whether industrial or legal — are fundamentally systems and information problems. Crivella designs the systems that allow organizations to understand those problems, control them, and ultimately prevail within them. From multi-billion-dollar industrial developments to national defense infrastructure to litigations involving hundreds of thousands of claimants.
Beginning with early securities class actions — including the Aetna and Motorola matters — Crivella applied expertise developed in large-scale industrial systems, manufacturing optimization, and enterprise systems engineering to litigation.
This represented a fundamental shift: litigation was no longer just legal — it became a system to be engineered, optimized, and controlled. Originally developed through work in advanced manufacturing, defense systems, and enterprise transformation, Crivella's methodologies were later extended into the legal domain.
At its core, Crivella's work is based on a simple principle: Complex problems — whether industrial or legal — are fundamentally systems and information problems. Crivella designs the systems that allow organizations to understand those problems, control them, and ultimately prevail within them.
This approach has been applied across multi-billion-dollar industrial developments, national defense infrastructure, cultural and historical institutions, and litigations involving hundreds of thousands of claimants.
Scale of Impact
The goal is not to review documents. The goal is to prove facts.
The most effective plaintiff's counsel teams do not process everything. They never did. They use targeted search and precisely constructed collections to find exactly what they need for the next motion, the next deposition, the next expert report — and they find it in hours, not weeks. This approach produces faster results and better outcomes than exhaustive review, because it is built around the correct goal: proving specific facts from the produced record, not documenting the processing of a corpus.
Unconventional review is not a faster version of conventional review. It is a different theory of what the work is. Legal theory drives collection construction. The context collection grounds the AI and narrows the universe of possible outputs to what is actually in the record — eliminating hallucination structurally, not by hope. AI operating on that bounded collection identifies the relevant documents, passages, and data points, and annotates each with a defensible citation traceable to the produced record.
A small, expert human team reviews a curated, pre-analyzed, citation-annotated set — not three terabytes, not twenty thousand documents, but the documents that matter, identified and sourced by AI working on grounded context. The output is not a review set. It is litigation-ready work product: arguments supported by citations that can go directly into a brief or a deposition outline, defensible under challenge because they derive from what the defendant produced.
The Three Steps of Unconventional Review
01 · Legal theory drives collection construction
What do we need to establish? Each legal theory, each custodian hypothesis, each timeline of interest generates a curated collection — bounded, organized, and purposeful.
02 · AI identifies and annotates within the collection
Working on a grounded, bounded collection — not the full corpus — the AI finds the relevant passages and data points, and annotates each with a citation traceable to the produced record.
03 · Small expert team validates and applies
Attorneys review pre-analyzed, citation-annotated work product — applying judgment to what AI has already identified, sourced, and organized. The brief writes itself.
"Finding the twenty documents that prove the case is more valuable than reviewing twenty thousand that don't. The attorney's job is to know which twenty — and Crivella's job is to find them."
Crivella.ai
Crivella operates at the intersection of systems engineering, artificial intelligence, and complex decision environments, applying these disciplines across four primary domains.
Crivella pioneered what can be described as Litigation Systems Engineering — the application of advanced analytics, artificial intelligence, systems integration, and strategic planning to the execution of complex legal matters.
Representative matters include: Baycol (MDL 1431), Trasylol (MDL 1928), Actos (MDL 2299), DePuy Hip (MDL 2197), Zantac (MDL 2924), Roundup (MDL 2741), and comprehensive product liability clusters spanning mesh and device ecosystems.
Key Innovations
Crivella has designed and implemented fully integrated manufacturing systems, hierarchical control architectures, and real-time process management platforms.
Representative projects include: Timken Company Faircrest Steel Plant (greenfield integrated specialty steel facility), Goodyear Tire modernization programs, Corning Asahi Glass optimization, and Voisey's Bay Nickel Company (~$7 billion development).
Core Capabilities
Crivella bridges strategy, operations, and technology — delivering business transformation plans, value creation strategies, implementation management, and integrated operational systems.
Representative engagements include: US Marine Corps Depot Maintenance Modernization (MARCORLOGBASES), Thomas Steel Strip enterprise modernization, and Inco Alloys value enhancement programs.
Service Areas
Crivella designs systems that capture complex knowledge, enable deep analysis, and support institutional decision-making — applied in religious and academic research, national historical institutions, and large-scale archival environments.
Notable implementations include: John Henry Newman Canonization Global Knowledge Systems & AI Research Platform, Gettysburg National Battlefield & Museum strategic technology and interpretive systems.
Applications
A production set is never a clean collection of uniform files. It arrives as a mixture of formats — PDFs, native office documents, email threads, spreadsheets, database exports, presentation files, audio recordings, video, images, and proprietary formats that reflect the full range of how the opposing organization actually conducted its business.
Traditional hosting handles the common formats and struggles with the rest. Files that cannot be processed sit outside the searchable record — effectively hidden in plain sight.
The Crivella platform ingests every format in the production. Audio is transcribed. Video is analyzed. Images are processed. Spreadsheets and databases are normalized. The result is a unified, fully searchable library in which no document is excluded because its format was inconvenient. The library is comprehensive because it has to be — a gap in the library is a gap in the case.
Format coverage
Once ingested, produced materials do not become a database. They become a library. The distinction is not semantic — it reflects a fundamental difference in what the system is designed to do.
A database holds files and returns them when queried by identifier. A library organizes knowledge — by content, by author, by subject, by relationship, by significance — and allows a researcher to explore it intelligently. The Crivella library is organized around the content, the people, the timeline, and the legal issues of the specific matter.
The library is linguistically trained. It understands the internal language of the opposing organization — the terminology, abbreviations, and operational shorthand that employees used in the normal course of business. Search is not keyword matching. It is semantic retrieval: find everything related to this concept, regardless of how it was expressed. The most significant document in the production is never the one with the most obvious keywords.
"The most significant document in a production is rarely the one with the most obvious keywords. It is the one whose significance is only visible when you understand the language the organization actually used — and what it meant when they used it."
The Architecture of Grounded AI
Most AI systems reason from training data — vast statistical models of what is generally true in the world. This produces confident, well-written outputs. It also produces hallucination: the generation of plausible-sounding claims that are not anchored to any specific source and cannot be verified.
Grounded AI is the alternative. Instead of reasoning from general training, it reasons from a specific, curated body of evidence — the actual documents, data, and communications produced in this case. A context collection is the mechanism: a structured set of materials assembled around a specific legal theory, custodian, timeline, or issue, formatted as a retrieval-augmented generation (RAG) input for any large language model.
The difference is consequential. When you ask a general AI what a defendant's safety committee knew about product risks, you get a plausible answer based on how safety committees generally operate. When you ask a grounded AI the same question against a context collection built from that defendant's produced documents, you get an answer traceable to specific documents, specific people, and specific dates. The first answer may be useful research. The second answer is evidence.
This architecture — a managed knowledge repository feeding precisely curated context to AI systems — is what Arthur Ray Crivella patented in 2001 and 2006. The AI industry now calls it retrieval-augmented generation. Crivella has been building, applying, and refining it in world-class litigation for twenty-five years.
The platform is necessary, but the platform alone is not sufficient. Grounded AI is an outcome produced by the complete system: the Rule 26 process that controls what enters the evidence record in the first place; the content identification engine that structures it into precisely bounded collections; the methodology that governs how AI works within those collections; and the human expertise that validates every output before it is used. Each layer enforces the integrity of the one above it. No AI vendor who sells software alone can replicate that system.
Traceable
Every AI output links directly to the documents it was drawn from. Attorneys know exactly what the model saw — and what was outside the collection.
Verifiable
Every claim can be checked against the produced record. No blind trust in the model. No outputs that can't be sourced.
Defensible
Outputs derived from the produced record can survive expert scrutiny, deposition, and cross-examination — because they rest on evidence, not inference.
Issue-Based Collections
All documents bearing on a specific legal claim — organized, sequenced, and structured for AI analysis of what the record shows.
Custodian Collections
Everything a specific individual sent, received, created, or was copied on — enabling AI-assisted analysis of what that person knew and when.
Timeline Collections
Documents from a specific period of interest — reconstructing what the organization knew, decided, and communicated during the events at issue.
Deposition Preparation Collections
The complete documentary record relevant to a specific witness — structured for AI-assisted preparation and real-time examination support.
The Crivella platform does not lock you into a single AI system or a proprietary model that the platform vendor controls. Context collections are designed to work with any current large language model — Claude, GPT-4, Gemini, or any model released after this page was written.
This architecture reflects a deliberate design choice: the value is in the library and the collections, not in the model. Models improve rapidly. By treating AI models as interchangeable consumers of the same structured context, the platform ensures that every improvement in model capability is immediately available to the legal team — without re-processing documents, re-building libraries, or re-negotiating vendor relationships.
The investment compounds. The library built from the first production in a matter becomes more powerful with every subsequent production, with every new AI model, and with every new legal theory that the context collections are organized around. A platform that gets more capable over the life of the matter — without additional processing cost — is a fundamentally different kind of litigation infrastructure.
AI model compatibility
The value is in the library. Models are consumers of the context it generates. As models improve, the same library produces better answers — automatically.
The Crivella digital library is not a production repository. It is the central knowledge environment for the entire matter — holding everything the case generates, not just what the other side was compelled to produce.
Every analysis, memorandum, deposition outline, expert communication, draft motion, and research note produced by the legal team lives in the same digital library as the produced documents it was built from. The separation between what was received and what was created is maintained — privilege and work product protection are preserved by design — but both are accessible within the same intelligent search and retrieval environment.
This integration is what makes AI most powerful in litigation. A context collection assembled for deposition preparation can draw simultaneously on the custodian's produced documents and the attorney's prior analysis of them. An AI query about what the record shows on a specific issue returns not just the underlying documents but the work product built on top of them — giving the legal team a compounding knowledge base that grows more capable with every analysis produced.
Expert work product — technical reports, analytical methodologies, supporting calculations, and correspondence with testifying experts — is managed with the same rigor. The library becomes the evidentiary foundation for expert testimony: organized, searchable, and available for AI-assisted preparation of both direct examination and cross.
Legal Memoranda & Analysis
Issue analyses, case summaries, and legal research — indexed alongside the produced documents that support them.
Deposition Outlines & Preparation Materials
Examination outlines, witness profiles, and exhibit sets — accessible through the same library environment as the produced documents they reference.
Expert Reports & Technical Work Product
Testifying and consulting expert materials — organized to support direct examination and withstand cross-examination on methodology.
Pleadings, Motions & Court Record
Filings, orders, and court record documents integrated into the library — so that AI context for any query reflects the full procedural posture of the case.
The platform maintains a complete history of every action taken and every change made over the life of the matter. Every document reviewed. Every annotation added. Every context collection created or modified. Every AI query run and every response generated. Every exhibit flagged, every analysis produced, every version of every work product document. All of it — attributed to the user who performed it, timestamped to the moment it occurred, and preserved in a form that cannot be retroactively altered.
This history is not an administrative byproduct. It is substantively valuable in at least three ways.
First, it is the documentation of methodology. When an expert witness is asked how they reached their conclusions, the platform history is the verifiable record of every step taken: what was reviewed, in what order, what was found, and what analysis followed. That record is defensible because it was created contemporaneously — not reconstructed after the fact.
Second, it is institutional continuity. Litigation matters run for years. Attorneys rotate. Institutional knowledge walks out the door with every personnel change — unless it has been captured. The platform history ensures that every new member of the legal team can see exactly what was done, by whom, and why, from the first day of the engagement forward.
Third, it is a defense against sanctions and spoliation allegations. A complete, immutable record of how the produced documents were handled — and when — is among the most valuable assets a legal team can have when its conduct is scrutinized. The platform provides that record automatically, as a feature of the system, without additional effort by anyone.
"When the opposing party challenges your methodology, or the court asks what was done with the produced record, the answer should not be a reconstruction. It should be a printout."
The history record includes
The platform manages both sides of the production equation. The same AI capabilities applied to defendant productions are applied with equal rigor to what plaintiff's counsel creates and produces on behalf of their own clients — and to the aggregate case inventory that drives every settlement decision in the matter.
In personal injury MDL and mass tort litigation, each plaintiff is required to submit a Plaintiff Fact Sheet — a detailed, court-ordered questionnaire, mandated by the MDL transferee court through a Case Management Order, describing their medical conditions, medical history, exposure facts, and the circumstances that substantiate their individual claims. The PFS is the foundational document of every case in the matter. It is also one of the most exploited targets in the defense playbook.
Incomplete fact sheets, inconsistencies between stated conditions and underlying medical records, and failure to supplement prior submissions as new information develops are among the most common grounds on which defense counsel moves against individual claims. Because the PFS is a court-ordered obligation — not merely a discovery response — the consequences of non-compliance are severe. Courts in MDL proceedings have dismissed individual cases with prejudice for persistent PFS deficiency. A case lost to a deficient fact sheet is not a case that lost at trial. It is a case that was surrendered before it began.
In a large MDL with thousands of plaintiffs spread across dozens of participating law firms, maintaining quality and consistency across the full inventory without systematic support is not possible. But the obligation runs in both directions. A clean, complete, and internally consistent PFS inventory is not merely a defensive posture — it is a settlement leverage tool. In global settlement negotiations, per-case values are directly influenced by the documentation quality of the inventory being presented. The well-documented claim commands more. The deficient claim gives ground it should never have yielded.
The Crivella platform applies AI quality assurance to every Plaintiff Fact Sheet in the inventory. Completeness is verified against the required fields of the specific matter's fact sheet form. Stated conditions are checked for consistency with produced medical records. Prior submissions are tracked and flagged when supplementation is required. Cases with deficiencies are identified and escalated to the responsible firm before defense counsel has the opportunity to exploit them — and before a preventable loss becomes a permanent one.
Completeness Verification
Every required field checked against the matter-specific form requirements. Missing or inadequate responses flagged before production.
Medical Records Cross-Verification
Stated conditions and medical history checked for consistency against produced medical records. Gaps and inconsistencies identified before defense can exploit them.
Amendment Tracking
Prior submissions tracked against current medical records and case developments. Required updates identified and escalated to the responsible firm.
Cross-Firm Consistency
Quality standards applied uniformly across all participating firms — so that the weakest fact sheet in the inventory reflects the standard of the strongest.
In a large MDL, the individual case is rarely the unit of analysis that drives the most consequential decisions in the matter. Global settlement is negotiated at the population level. The question is not what any single plaintiff can recover — it is what the full inventory of claims is worth, in aggregate, given the distribution of conditions, the quality of medical documentation, the mix of strong and weak cases, and how that inventory compares to prior settlements in similar litigations.
The Crivella platform builds and maintains context collections covering the complete case inventory — separately for each participating law firm and in aggregate across the entire matter. These collections are AI-accessible: attorneys and their advisors can query the full inventory in natural language, ask questions about the distribution of claims and conditions, model settlement scenarios against real case characteristics, and identify patterns in the claimant population that inform both valuation and strategy.
The statistical description of the case inventory is updated continuously as new fact sheets are submitted, medical records arrive, and case characteristics evolve. What was true of the inventory six months ago may not be true today — and the settlement strategy built on an outdated picture of the claims population is a strategy built on the wrong foundation.
As the matter matures, the case inventory becomes the primary instrument for settlement planning. Tiers of cases can be evaluated against realistic recovery ranges. The overall population can be analyzed for concentrations of strength and weakness. The effect of proposed global settlement structures on individual firm inventories can be modeled before any agreement is reached. The platform makes all of this possible — not as a separate analytical exercise, but as a continuous capability of the same system managing the cases.
"Global settlement is built from individual case characteristics. The attorney who understands the shape of their own inventory — with precision, in real time — negotiates from a fundamentally different position than the one who does not."
What the inventory AI can answer
The most powerful litigation AI is not the kind that answers questions when asked. It is the kind that watches the evidence record continuously, understands what changes mean in context, and tells you what you need to know before you know to ask.
The standard model of AI-assisted legal work asks the attorney to bring documents to the AI — paste text into a chat window, upload a file, describe what is relevant. This approach has fundamental limits: it is manual, it is incomplete, and it exposes confidential materials to environments not designed to hold them.
The Crivella model inverts this. Rather than bringing documents to Claude, Claude comes to the documents — querying Crivella's context engine directly during a session. An attorney working in Claude asks a question about what the record shows on a specific issue. Claude queries the relevant context collection in Crivella's system. The response is grounded in the actual produced documents, organized and curated by Crivella's platform, without those documents ever leaving Crivella's controlled, privilege-protected environment.
Every AI interaction is traceable to a specific collection. Every collection is bounded by Crivella's access controls. Client materials do not train AI models. The confidentiality of the entire record — source documents, metadata, case strategy — is maintained throughout every AI interaction, under the same governance that governs the platform itself.
How It Works
Attorney queries Claude
A question about what the record shows — a custodian's knowledge, a timeline, a pattern across productions.
Claude queries Crivella's context engine
The relevant context collection is retrieved from Crivella's platform — curated, current, privilege-protected.
Grounded response
Claude's answer is traceable to specific documents in the produced record — not the model's general knowledge.
Work product captured
Claude's analysis is written back into Crivella's system — indexed, linked to its sources, available for future sessions.
Every analysis Claude generates — a chronology, an issue summary, a deposition outline, a pattern report — is stored back into Crivella's platform as work product. It is indexed alongside the source documents that grounded it. It is retrievable under the same search and access controls as the rest of the library. And it becomes available as context for every future session.
The analysis produced on a key custodian last month becomes part of the context collection used to prepare that custodian's deposition this month. The issue summary generated after Production 3 becomes context for the AI query run after Production 7. Each session builds on every session before it.
Over the life of a complex matter, Crivella becomes not just a document repository but an accumulated record of every analytical judgment — human and AI — applied to the case. The system compounds. A matter managed on Crivella from Rule 26 through trial arrives at trial with years of structured intelligence already organized and available, not buried in email threads and partner-only folders.
What gets captured
A large complex litigation generates hundreds of thousands of documents across dozens of productions spanning years. The evidence record changes continuously. No team — regardless of size — can maintain comprehensive, current awareness of a data environment at this scale. Important developments get missed. Connections go undrawn. Cases are weaker than they could be.
Crivella's agentic infrastructure addresses this directly. Rather than waiting to be asked, these agents watch the evidence record continuously, understand what changes mean in the context of the specific matter, and surface what matters — automatically.
Production Monitoring Agents
When a new production arrives, it is automatically compared against existing context collections, pending theories, and custodian profiles. Documents with high relevance to active issues are flagged and surfaced to counsel — without waiting for manual review to surface them days or weeks later.
Alert Agents
Configured to notify specific attorneys when new evidence is relevant to their matters — a document contradicting prior testimony, a communication involving a custodian claimed to have no relevant involvement, a gap in production metadata that reveals selective withholding.
Reporting Agents
Scheduled analysis without manual intervention: weekly summaries of new relevant documents by theory; privilege log statistics updated with each production; PFS completion rates and deficiency trends across the plaintiff inventory; custodian document volume changes that may signal selective production.
Cross-Matter Intelligence Agents
In complex litigation involving multiple firms or consolidated matters, agents that maintain awareness across the entire case inventory — surfacing patterns, flagging anomalies, and identifying developments in one matter that are material to others in the portfolio.
Retrieval-augmented generation — the architecture by which AI models are grounded in specific document collections rather than trained data — is the dominant paradigm in enterprise AI as of this writing. It is the focus of billions of dollars in investment, hundreds of startups, and the active development roadmaps of every major AI vendor. Every major eDiscovery platform is now racing to add it. Venture capital is flooding the space.
Arthur Ray Crivella invented this architecture in two steps, separated by five years. The first patent — US 6,883,008, filed 31 July 2001 — describes the multimedia knowledge-management substrate: a repository in which content is identified, organized, and made available for intelligent retrieval. The second — US 7,779,007, issued 17 August 2010 and active through approximately mid-2028 — protects the engineered-relevance bounded-collection mechanism: the property that distinguishes precise, intentional retrieval from the similarity-score-based retrieval most "RAG" systems implement today. Together they predate transformers, GPT, and the current AI wave by two decades. When law firms are evaluating AI vendors and trying to distinguish genuine capability from recent opportunism, those dates settle the question.
No current competitor in the plaintiff-side MDL litigation support market can claim this combination: the foundational IP, the twenty-five years of production deployment, the 80+ matter track record, and the integrated service model covering Rule 26 advisory, document platform, case inventory management, and litigation intelligence. They can build toward it. We are already here.
This is not a company that adopted AI when it became fashionable. It is a company that invented the underlying architecture before the field existed — and has spent a quarter century refining it in the most demanding document environments in American litigation.
US Patent 7,779,007 B2 · Active
Issued: 17 Aug 2010 · Term: Active through ~mid-2028
Identifying Content of Interest. Protects the engineered-relevance bounded-collection mechanism — the property that makes a recommendation reproducible against a specific assignment boundary, rather than against a generic similarity score. The load-bearing exclusion right behind every Context Collection the platform produces.
US Patent 6,883,008 B2 · Architectural Priority
Filed: 31 July 2001 · Issued: 19 Apr 2005
Multimedia knowledge-management substrate. Term has matured into the public record; what remains is the established 2001 priority on the architecture, defensive prior art preventing competitors from claiming the same architectural ground in subsequent filings, and permanent inventor credit for Arthur R. Crivella and Wayne West on the foundational work.
130M+ Pages in Active Production
The platform runs today against more than 130 million pages of production volume across two active multi-district litigation engagements — including the Roundup Products Liability Litigation and the pelvic-mesh products MDLs. Both remain active and operate under federal protective orders.
Not AI-Assisted Work. Not Human-Supervised AI.
The platform was designed from its 2001 origin for two co-equal users: human professionals and machine intelligence. Both operate on the same knowledge foundation, through the same intake pipeline, under the same review discipline. Neither is secondary to the other. The three modes are detailed in the section below.
The Grounded Intelligence Console was designed from its 2001 origin for the integration of natural and artificial intelligence under one discipline — not for human work with AI bolted on the side. Three modes of inquiry operate on the same Knowledge Object graph, the same provenance discipline, the same reliability grading, and the same review gate. The capital-project Knowledge Kiosk operates in hybrid mode by default; the other two modes remain available.
Mode 01 · Natural Intelligence
A practitioner conducts direct investigation, curation, and authoring through the platform's advanced search and Context Collection management. Every result, excerpt, and finding is captured as a Knowledge Element with full attribution to the human authority record.
Mode 02 · Artificial Intelligence
A programmatic agent operating under a user's authority and responsibility — autonomous inquiry and authoring, with every contribution catalogued as a first-class Knowledge Element carrying authority record, run identifier, grounding-confidence value, and human-review routing. Nothing commits without human review.
Mode 03 · Hybrid · Default
The practitioner supplies judgment and context; the agent supplies retrieval, synthesis, and structured output. Both contributors commit through the same intake pipeline; the dialogue itself is catalogued as a sequence of Knowledge Elements, with either contributor's turn independently auditable years later.
The symmetry is not architectural housekeeping. It is the property that makes three things simultaneously achievable that most platforms force a trade-off between:
The platform is the infrastructure that connects Rule 26 advisory to litigation document intelligence. Together the three services constitute a complete strategy — from the first meet and confer through trial or settlement.
Stage One
Rule 26 Technical Advisory →
Ensure the right documents enter the case. Control the scope, the custodians, the format, and the metadata from the start.
Stage Two · You Are Here
Grounded Intelligence Console
Transform produced documents into a comprehensive AI-ready context engine. Every format. Every custodian. Every production.
Stage Three
Litigation Document Intelligence →
Deploy the platform at litigation speed — early motions, witness surprise, and momentum that compels early resolution.
Continuous
Privilege Intelligence
Running alongside every stage — inadvertent disclosure detection, privilege log statistical analysis, and the Privilege Log Assessment Report.
If you have an active matter with a production in hand — or a production coming — we can have the library operational within days of receipt. Tell us where you are and we will tell you what is possible.