Trial Intelligence Methodology
The principles behind Trial Intelligence. PrismCDM uses Source Fidelity, Provenance, Explainability, Integrity Protection, Human Governance, and Continuous Versioning to produce inspection-ready clinical intelligence. This page describes the philosophy; the implementation details are shared during the Compliance Review under NDA.
What the engine produces.
Trial Intelligence is the structured, sourced, and versioned understanding of everything a protocol implies for study execution, regulatory success, and submission readiness. It is produced by the PrismCDM engine and consumed by every downstream operational system, from execution through inspection.
The composite Prism Trial Intelligence Index (PTII™) is derived from multiple independently evaluated dimensions that assess protocol complexity, operational characteristics, enrollment considerations, amendment potential, and execution readiness. Individual assessments remain explainable and inspectable within the platform.
The composite is a sentence the executive can act on. The dimensions are the paragraph that explains why.
Every value carries its origin.
Every displayable value is classified by origin and governed by Source Fidelity. Public evidence, customer evidence, recognized reference standards, and platform-generated outputs remain distinguishable throughout their lifecycle.
Source classification follows the value throughout its lifecycle. It is not added during presentation.
Sourced from public records such as ClinicalTrials.gov, sponsor disclosures, and recognized public datasets. Re-derivable by any reader.
Sourced from materials provided by the customer under tenant scope. Never crosses tenant boundaries.
Sourced from recognized reference standards (CDISC, ICH GCP, FDA guidance, MedDRA, WHODrug). Citable to the standard at runtime.
Produced by the PrismCDM engine. Carries the version stack that produced it and the supporting evidence behind it.
Every conclusion answers the same questions.
Trial Intelligence outputs are explainable by design. Every conclusion the platform displays can answer the same set of questions on request: why was this generated, what evidence supports it, which component produced it, what was the assessment framework at generation, can the supporting evidence be inspected, can the output be reproduced, and has it changed since first generation.
The answers are produced by the platform, not by the operator. Explainability is an architectural property, not a workflow.
Stable identifiers across the life of the study.
Findings identify what the protocol means. Recommendations identify what the team should do.
Findings and recommendations carry stable identifiers that survive protocol amendments and platform evolution. A finding identified at protocol import can be referenced by an operational decision six months into execution and an inspection two years after submission. The identifier remains the same; the supporting context evolves around it.
Structured observations about the protocol. Each carries a stable identifier, severity, and the supporting evidence behind it.
Suggested operational actions tied to findings. Each can be accepted, deferred, or declined inside the execution workspace, with the decision captured as an auditable event.
Multiple independent integrity anchors.
Every Brief is sealed with multiple independent integrity anchors that make historical outputs reproducible and tampering detectable. The specific anchors and their relationship to the version stack are shared during the Compliance Review.
Given the same validated inputs, assessment framework, evidence snapshot, and platform configuration, PrismCDM is designed to produce reproducible analytical outputs within the constraints of the configured inference pipeline.
Historical outputs remain inspectable forever.
PrismCDM independently versions the components that influence analytical outputs so historical results remain reproducible and explainable over time. Changes to the engine, the assessment framework, or the evidence snapshot produce new outputs with new provenance, while the prior outputs remain inspectable in their original context.
Historical Briefs are never silently restated when models or assessments evolve. The architectural record of how every historical output was produced is permanent.
How a conclusion gets from a protocol to an inspection.
Every conclusion PrismCDM displays can be traced back to its source through the same canonical chain. Every node is sourced. Every transition is recorded. Every artifact at the end carries the architectural record of every step that produced it.
- Protocol
- Extraction
- Assessment
- Finding
- Recommendation
- Execution
- Evidence
- Submission
- Inspection
AI proposes; operators decide.
Trial Intelligence supports decisions; it does not make them. Regulated activities (protocol amendments, Case Report Form approvals, signed Part 11 records, database locks, query closures, submissions, Medical Monitor decisions, Biostatistics review, Regulatory review) remain with the human role accountable under applicable regulation and the customer's validated processes.
The architectural commitments behind PrismCDM's AI posture are documented at the Responsible AI Architecture page.
A document that cannot structurally contain invented data is a different kind of document than one that merely says it does not contain invented data.
The first is a regulated artifact. The second is a claim.
The implementation details.
The methodology principles on this page are public. The implementation details that realize them (the scoring framework decomposition, the runtime enforcement layers, the specific integrity anchors, the version stack mechanics, the schema model, and the AI inference pipeline configuration) are part of the Compliance Review package shared under NDA.
Enterprise security teams, procurement, regulators, and methodology partners typically receive the package as part of the Compliance Review conversation. Request it through the Contact form with reason “Compliance Review”.