§ Evaluation Dossier · Evidence, Not Claims

Evaluated across 1,035 regulated documents.

At DIA 2026, Clinplex presented a multi-framework regulatory and quality gap-detection methodology evaluated across SOPs, CAPAs, deviations, audit records, CMC submissions, validation packages and preclinical documents. 96% gap-detection accuracy against SME review.

DIA 2026 · Abstract 115882 · Track 8: Regulatory AI · Poster P312
View the DIA abstract → Request the presented poster
96%

Gap-detection accuracy

Measured against SME-panel ground truth.

3.0×

More gaps surfaced

12.3 regulatory gaps per SOP vs 4.1 under single-framework review.

73%

Less assessment time

Reduction in cross-framework regulatory assessment time.

98%

Critical-tier sensitivity

Agreement with SME review on Critical findings.

§ 1 · The Dataset

How we tested it.

A 1,035-document regulated corpus drawn from biologics, small-molecule and CDMO operating environments.

SOPsCAPAsDeviationsAudit recordsCMC submissionsIQ / OQ / PQ validation packagesPreclinical study reports
1,035 documents→Clinplex evaluation→Independent SME review→Consensus adjudication
§ 2 · Ground Truth

Human review was part of the benchmark.

7
SME reviewers
Quality & Regulatory subject-matter experts formed the review panel.
312
Stratified review subset
Documents independently reviewed by the SME panel.
Blinded
Randomized review
Reviewers assessed documents without seeing the AI's findings first.
Consensus
Adjudication
Disputed findings were resolved by panel consensus, not a single reviewer.
κ = 0.87
Inter-rater agreement
Cohen's κ for AI-vs-SME agreement.
95% CI
Confidence intervals
Reported for both accuracy and κ.
§ 3 · The Comparison

Multi-framework analysis found substantially more.

PerformanceSingle-framework reviewClinplex multi-framework
Average gaps per SOP4.112.3
Critical-tier sensitivity0.710.98
Specificity0.880.94
Cross-document linkageNot evaluatedSOP → CAPA mapped
13 additional cross-regulatory gaps were identified that were not captured under single-framework assessment.
§ 4 · Why the Difference

Why 3× more findings?

A conventional review asks
"Does this SOP meet 21 CFR 211?"
Clinplex asks
"Does this SOP satisfy the combined obligations created by 21 CFR 211, ICH Q7/Q9/Q10, EU GMP and WHO requirements — and the related records that implement those obligations?"
Evaluated: FDA 21 CFR 210/211 · ICH Q7 / Q9 / Q10 · EU GMP Annexes · WHO TRS · 21 CFR Part 11 — ICH Q9-aligned risk classification
§ 5 · Reproducibility

We tested consistency,
not one successful output.

Repeated analyses
Outputs were evaluated for stability across repeated runs.
Environment checks
Evaluation included controls for operating-environment variance.
Test / retest stability
Consistency was evaluated over time rather than from a single point-in-time run.

Detailed reproducibility protocol and evaluation artifacts are available under NDA.

§ 6 · Limitations

Study limitations.

The reported evaluation covered selected organizations, regulatory classes and document types. The results should not be interpreted as universal performance across every organization, document type or regulatory framework. Broader cross-organizational evaluation and longitudinal characterization remain ongoing work.

§ 7 · Product Evidence

See what the result actually looks like.

Every Clinplex finding carries the full evidentiary chain — from the source record to the proposed fix.

01Source evidenceThe record and context supporting the finding
02Regulatory basisThe governing requirement, cited
03Impact and rationaleWhy the issue matters and what related evidence is affected
04Proposed remediationA review-ready path to address the finding
Request a redacted sample analysis →
§ 8 · The Record

Review the underlying work.

DIA 2026 Global Annual Meeting · Abstract 115882 · Poster P312

AI-Enabled Multi-Framework Regulatory and Quality Gap Detection for Pharmaceutical Operations Across the Product Lifecycle

Now run it on
your own records.

One closed deviation, its CAPA and the governing SOP. Ranked gaps with citations and a drafted remediation. Free.

Run an Inspection Stress Test → For regulators →