eTMF · TMF Lead / Doc Mgmt + Inspector
CuRE Compend
Trial-conduct evidence custody — continuous inspection-readiness against the DIA TMF Reference Model.
What it does
Trial-conduct evidence custody — continuous inspection-readiness against the DIA TMF Reference Model. Document-to-data cross-links tie TMF artifacts to study events, so inspection readiness is grounded in actual conduct rather than a generic checklist.
Key capabilities
- DIA TMF Reference Model 3.x classification (zones 01–11)
- Continuous inspection-readiness scoring
- AI-grounded classification + Q&A over the TMF corpus
- Amendment-diff-aware document routing
- AI-assisted CSR / SDRG / ADRG / Module 1 authoring scaffolding
- Persisted Expected Document List (EDL)
- AI TMF completeness / gap detection
- Audit-finding response lifecycle
- Risk-based QC sampling
- TMF-health trending
What sets it apart
- Document-to-data cross-links tie TMF artifacts to study events, so inspection readiness is grounded in actual conduct rather than a generic checklist.
- An Expected Document List gives inspection-readiness a denominator, and AI completeness detection grounds gap-finding on the study's actual config (visits, sites, amendments).
- TMF-health trending, risk-based QC sampling, email capture, safety-letter distribution, CTIS redaction, archive, and blinding partitions are part of the custody surface.
- Site-side eISF embedded inside Capture — no second login (vs. Florence).
- Inspector portal: time-bounded, sponsor-controlled grounded Q&A with citations.
- FDA-meeting-package and IND-assembly document custody extend Compend upstream to the pre-IND / first-in-human journey (Translational bundle, ADR-PLT-100).
Classify a trial document and score its inspection readiness
An eTMF keeps a study continuously inspection-ready against the DIA TMF Reference Model. Type a trial-conduct document title and watch Compend's cue-matcher classify it into a TMF zone with an advisory confidence, then toggle the document's lifecycle facts and watch the six-criterion readiness score recompute. The corpus panel rolls every synthetic document up into a study-level inspection posture with the actionable gap list.
A synthetic IND — ravucizumab (VELT-042). Type a document title; the cue-matcher classifies it live.
These are the operational facts the readiness scorer grounds on — not a guess. Confirm the classification to fill the placement criterion.
TMF classification · advisory (ADR-CPD-003)
Advisory suggestion. No classification is authoritative until a reviewer confirms it (CPD-003). Until then the placement criterion stays unmet.
Inspection readiness · this document
67% · 2 gapsScore = criteria met ÷ 6 (4/6). A document is inspection-ready only when every criterion is met. This is the exact deterministic roll-up Compend runs per document — grounded in the document's own lifecycle, version and classification facts, not a generic checklist.
Study inspection posture · 12 synthetic documents
| Document | Zone | Readiness | Top gap |
|---|---|---|---|
| VELT-042 Clinical Study Protocol v3.0 | 02 · 02.01.01 | 100% | Ready |
| Protocol Amendment 2 — expanded eligibility | 02 · 02.01.02 | 67% | A human-confirmed TMF classification exists |
| Signed Informed Consent Form — Site 104 | 02 · 02.03.01 | 100% | Ready |
| Investigator's Brochure — Edition 4 | 04 · 04.01.01 | 100% | Ready |
| IRB Approval Letter — Central IRB | 03 · 03.02.01 | 100% | Ready |
| Interim Monitoring Visit Report — Site 104, Visit 3 | 08 · 08.02.01 | 50% | A current version with a sealed artifact exists |
| Statistical Analysis Plan v2.1 | 02 · 02.01.09 | 100% | Ready |
| Drug Accountability Log — Site 211 | 05 · 05.01.04 | 50% | A human-confirmed TMF classification exists |
| Vendor Qualification — Central Lab (Quanterix) | 07 · 07.01.01 | 100% | Ready |
| Data Management Plan v1.0 (superseded) | 10 · 10.05.01 | 83% | Document is not archived |
| Site 104 Training Log | 08 · 08.03.04 | 67% | A human-confirmed TMF classification exists |
| Randomization Specification — block, stratified | 09 · 09.01.01 | 83% | A current version with a sealed artifact exists |
Every classification and readiness score is computed in your browser from synthetic document facts — no backend. Click a row to load it into the classifier above. Notice readiness is never about the file existing: a sealed, well-titled document still isn't inspection-ready until its classification is human-confirmed, its placement is valid, it isn't archived, and it hasn't failed QC — the gaps a real inspector would find.
Why this is more than a toy
The classifier and the readiness scorer are a faithful, dependency-free port of Compend's real Phase-1/Phase-2 methodology, from three source files under apps/compend/src/server/documents/: the tmf-taxonomy.ts DIA-TMF-RM-3.x reference vocabulary (zones 01–11, sections, artifacts and their keyword cues), the classify.ts heuristic cue-matcher — scoring each placement by matched-cue count and returning the best with confidence min(0.5 + 0.2·matches, 0.95), or no placement when nothing matches — and the readiness.ts six-criterion inspection-readiness roll-up (score = met ÷ total). Per ADR-CPD-003 the classification is advisory + human-confirm: no placement is authoritative until a reviewer confirms it, so an unconfirmed suggestion leaves the placement criteria unmet — exactly as in-product. Per ADR-PLT-044 readiness is Compend's own operational state (a deterministic criteria checklist, not statistical methodology), so it lives in-app rather than in Calculate. Here it all runs entirely client-side on a synthetic IND — no backend, no real data.
See CuRE Compend in action
Every research ecosystem is unique. Let's discuss how CuRE can be configured for your needs.