Trigger: AI is used with research data or participant records.
Research AI Evidence Ripple
Make research AI
reproducible, bounded, and reviewable.
Track datasets, models, lineage, consent boundaries, analysis pipelines, RAG systems, and AI agents used in research without turning evidence into bureaucracy.
Reproducibility Evidence
Sensitive context
Why research AI needs evidence infrastructure.
Research AI operates on participant data, proprietary datasets, and models whose outputs may form the basis of published findings. Reproducibility failures, consent boundary violations, and undocumented data provenance undermine scientific integrity and carry legal, ethical, and institutional consequences.
Data categories in scope
People affected
Risk scenarios
What typically goes wrong.
Specific failure modes seen in this sensitive context — without structured evidence.
A research model is trained on participant survey data beyond the original consent scope.
Ethics committee requires evidence of consent boundary. No structured evidence exists. Publication delayed or retracted.
An analysis pipeline produces outputs that cannot be reproduced by a reviewer.
No Model Registry Trace active. No dataset version fingerprint. Environment not captured. Reproducibility claim in publication unsupportable.
A research RAG assistant is built over pre-publication manuscripts and raw interview transcripts.
No document boundary. Pre-publication content accessible via queries. Confidential data at risk. No Retrieval Grounding Evidence.
A research AI agent writes portions of a literature review without disclosure.
No provenance tracking. No disclosure mechanism. Institutional integrity policy potentially violated.
Border control, applied here
Scan the system. Issue the passport. Grant the visa.
What the scan finds in this context, what the passport records, and what conditions the visa attaches.
Currents
the scanWhat AI is running, and what do we not know about it?
Currents opens the file: which models and pipelines touch participant data, whether the processing stays inside the consent that was actually given, which analyses cannot be reproduced today because the dataset version and environment were never captured, and what sits in a retrieval index that should not leave pre-publication.
Ripples
the passportWhat is this system, and what is proven about it?
The Ripple records dataset provenance and version fingerprints, the consent and ethics boundary, run tracking sufficient for an independent reviewer to reproduce the result, the retrieval boundary over unpublished material, and the disclosure position for AI-generated text. Scientific validity is not in it — that is peer review's job.
Droplets
the visaMay it operate HERE, on THIS data, under WHAT conditions?
The Droplet clears the pipeline for one study and one consent scope: these data categories, this transfer agreement, reproducibility evidence live, cross-institutional access only where a DTA is confirmed — with an expiry tied to the ethics review cycle rather than to the publication date.
Scope
What needs a Ripple.
Stakeholder workflow
From trigger to access decision.
Scan
Art. 6 route · Annex III test
Evidence call
Annex IV · Art. 10 bias file
Seal Review
DPO · CISO · Art. 14 owner
Droplet
Conditions bound · expiring
Re-scan
On retrain, drift or model swap
Scan
Art. 6 route · Annex III test
Evidence call
Annex IV · Art. 10 bias file
Seal Review
DPO · CISO · Art. 14 owner
Droplet
Conditions bound · expiring
Re-scan
On retrain, drift or model swap
Principal Investigator
“Research AI will process participant data or produce publishable outputs.”
Require Ripple documenting consent scope, lineage, and reproducibility evidence before deployment.
Ethics Committee / IRB
“AI tools are used with participant data in a research study.”
Review Ripple consent boundary section. Confirm alignment with study protocol.
Data Steward / DPO
“Research data shared cross-institutionally via AI platform.”
Confirm DTA status, data transfer evidence, and privacy section before cross-institutional access.
Access decisions
Context Droplet conditions.
The access decisions that apply in this sensitive context — and the evidence conditions that produce them.
- Consent scope confirmed and documented in Ripple
- Dataset lineage and versioning active
- Cross-institutional DTA confirmed
- Reproducibility evidence: Model Registry Trace active
- Consent scope unclear or broader than original purpose
- Lineage absent — cannot reproduce pipeline outputs
- Cross-institutional transfer without DTA
- AI used to generate or draft publication content
- Research participant data in scope — ethics committee review required
- Clinical data — IRB / ethics approval confirmation needed
- Participant data processed beyond consent scope
- Raw unpublished data accessible via uncontrolled retrieval
- No lineage — publication reproducibility claim unsupportable
Measurement
Evidence families we can structure.
The measurable evidence categories relevant to this context and the evidence signals they produce.
Dataset Lineage & Versioning
Dataset provenance, version history, and transformation lineage from raw data to model inputs.
Model Tracking
Run tracking, hyperparameter capture, and model versioning enabling reproducibility of published analyses.
Consent & Ethics Boundary
Evidence that data processing stays within the scope of participant consent and ethics committee approval.
RAG Document Boundary
Evidence that retrieval scope covers only authorised documents — pre-publication content excluded where required.
Data Export Integrity
Evidence that cross-institutional data transfers occur only under confirmed data transfer agreements.
Privacy
PII detection and de-identification evidence for participant data before analysis or sharing.
Honest scope
What remains not assessable.
AffectLog does not overclaim. These items require external expertise, regulatory process, or long-term study.
Scientific validity of research findings
Scientific validity requires peer review, methodology assessment, and domain expertise — not an AI evidence platform.
Instead: Submit findings for peer review. Engage domain experts. AffectLog evidence supports the infrastructure review — not scientific assessment.
Ethics committee approval
Ethics approval is granted by institutional review boards and ethics committees following dedicated review processes.
Instead: Engage your institutional IRB / ethics committee. AffectLog evidence can support the application — not replace it.
Whether AI-generated content meets journal or institutional disclosure standards
Disclosure standards are set by journals, institutions, and professional bodies — outside evidence platform scope.
Instead: Refer to your institution's AI use policy and target journal guidelines.
Example
Sample Ripple for this context.
ClinicalSynth Analysis Pipeline
Clinical trial data analysis model · Health Research
Access conditions
What we will not overclaim
AffectLog provides technical and operational evidence for research AI. We do not claim scientific validity, ethics approval, or publication standards compliance. We show lineage, consent boundaries, reproducibility evidence, and what review conditions apply.
Common questions
Questions this context raises.
“We already have run tracking internally — we don't need an additional evidence layer.”
Run tracking captures individual experiment runs. AffectLog structures it into a Ripple: consent scope, cross-institutional transfer evidence, RAG document boundaries, and access decisions — making internal tooling accessible to ethics committees, DPOs, and institutional reviewers.
“Research is exploratory — we cannot know in advance what data we'll need.”
Exploratory scopes can be documented as evidence of breadth and intent. A Ripple at research start establishes the consent and data boundaries — updated as scope evolves — rather than reconstructing evidence after publication.
Get started
Make research AI evidence
as rigorous as the research itself.
Map which AI tools are in your research pipeline, establish consent and lineage evidence, and structure reproducibility conditions before your next publication cycle.
AffectLog provides technical and operational evidence. Not scientific validation, ethics approval, or publication standards certification.