SANTAROSLABSPartner With Us

STUDY PORTFOLIO

Questions, Protocols, and Transparent Status

These studies are proposed or in development. They describe planned methods and outputs—not completed findings. Status labels will change only when a documented research milestone is reached.

Review the Studies

PORTFOLIO PRINCIPLES

Designed to Be Checked

Each protocol is designed around a falsifiable question, a meaningful comparison condition, expert-grounded measures, and a publication plan that includes null results and limitations.

SL-001Protocol development

Scientific reasoning with AI co-investigators

When does AI assistance improve the diversity, testability, and calibration of scientific hypotheses?

Proposed design

Proposed randomized within-subject study with domain researchers completing matched hypothesis-generation and critique tasks with and without AI support.

Planned public output

Preregistered protocol, de-identified task materials, scoring rubric, and analysis code where consent and licensing allow.

Primary measures

  • Hypothesis diversity
  • Testability
  • Expert-rated plausibility
  • Confidence calibration
  • Error detection
SL-002Pilot protocol

Reproducible AI-assisted data analysis

Can another researcher reproduce an AI-assisted analysis from its provenance record alone?

Proposed design

Proposed blinded reproduction study using open scientific datasets and containerized analysis environments across researcher-only and AI-assisted workflows.

Planned public output

Open provenance schema, reproducibility checklist, benchmark tasks, and reference containers.

Primary measures

  • Outcome agreement
  • Trace completeness
  • Undocumented decisions
  • Time to reproduce
  • Environment portability
SL-003Preregistration in preparation

Evidence synthesis under scientific disagreement

How reliably can AI represent consensus, uncertainty, and contradiction in a mixed scientific literature?

Proposed design

Proposed adversarial evidence-synthesis benchmark built from source packets containing supporting, null, conflicting, and methodologically weak studies.

Planned public output

Curated evaluation packets, expert annotations, error taxonomy, and benchmark harness subject to source licenses.

Primary measures

  • Citation entailment
  • Contradiction detection
  • Uncertainty calibration
  • Selective omission
  • Appropriate refusal
SL-004Scoping

Shared AI workspaces for research teams

How do shared AI workspaces change coordination, review, and scientific accountability within a lab?

Proposed design

Planned mixed-method field study of small research groups using a shared AI workspace for literature, code, and project handoffs over six weeks.

Planned public output

Study protocol, interview guide, governance template, and implementation report without exposing confidential research content.

Primary measures

  • Duplicate work
  • Review latency
  • Decision provenance
  • Perceived workload
  • Early-career learning

GOVERNANCE

No Human-Participant Work Before Appropriate Review

Sample sizes, recruitment channels, participating institutions, and review bodies are intentionally not named before they are confirmed.

Any study involving people, private data, or institutionally controlled materials will use the consent, ethics, privacy, security, and data-sharing processes required for that study.

CONTRIBUTE TO A STUDY

Bring Domain Expertise to Study Design

We welcome scientific collaborators who can sharpen tasks, measures, validity checks, and real-world constraints before a protocol is finalized.

Partner With Us →
Participation, authorship, data access, and governance are documented before research begins.