Question
Frame an answerable scientific question and state which observations could change the conclusion.
Nonprofit Computational Research Lab
Building Verifiable Methods for AI-Assisted Scientific Research
We study how AI changes the learning and teaching of scientific reasoning, evidence synthesis, code, and data analysis. Every case file makes its outcomes, assumptions, human review, and current status explicit.
Current public research record
- Portfolio
- 4 proposed studies No recruitment or data collection
- Current stage
- Scoping and protocol drafting Internal study identifiers only
- Completed outputs
- None claimed Planned work is labeled separately
- Last reviewed
- 28 August 2026 Status changes require a documented milestone
MISSION
Make AI-Assisted Research Easier to Verify
Santaros Labs is a nonprofit lab focused on how scientists and learners build research judgment with AI tools. We study instruction, feedback, independent transfer, evidence quality, and reproducibility. The goal is not faster output alone. It is durable learning and research that others can inspect.
RESEARCH LIFECYCLE
From Question to Verification
Protocol
Specify the comparison, candidate outcomes, exclusions, analysis plan, and review requirements.
Execution
Record sources, data transformations, code, model details, and material human decisions.
Verification
Use independent checks, sensitivity analysis, reproduction, or replication where the design permits.
MEASUREMENT DOMAINS
What We Measure
Reasoning quality
Are hypotheses testable, assumptions visible, and confidence aligned with correctness?
Candidate measures: testability, calibration, expert error ratingsReproducibility
Can an independent researcher reconstruct the workflow and reproduce the reported result?
Candidate measures: trace completeness, environment recovery, outcome agreementEvidence support
Does each scientific claim remain faithful to its cited source and the uncertainty in the literature?
Candidate measures: citation support, contradiction detection, omission rateLearning transfer
Can learners apply a scientific practice independently after AI-supported instruction ends?
Candidate measures: delayed transfer, error detection, appropriate escalationEVIDENCE STATUS
Research Status Stays Visible
Study concepts, protocols, registrations, data collection, analyses, and completed outputs are distinct states. Each public record should show what exists, what does not yet exist, and what changed.
Preregistration-ready protocols
Reproducible computational workflows
Mixed-method evaluation
Transparent limitations
Cross-disciplinary collaboration
COLLABORATION AND SUPPORT
Support Method Development and Independent Verification
Santaros Labs welcomes grants, compute credits, open-data partnerships, methodological review, domain expertise, educators, learning scientists, and replication partners for AI-assisted research training.
Support enables:
- Protocol development
- Research staff
- Secure compute
- Open-source tooling
- Replication studies
RESEARCH COLLABORATION
Bring a Scientific Workflow Worth Testing
Share the research question, current workflow, data constraints, and the decision that better evidence would support. We will begin with scope and method, not a predetermined claim.
No study begins until scientific accountability, data terms, and required oversight are documented.