Clinical Cognition Transformation Lab
Status as of Dataset v0.2.9
This repository now represents the Clinical Cognition Transformation Lab (CCTL), a research program studying how clinical cognition transforms when healthcare becomes a distributed human-AI cognitive system.
The current research focus is Human-AI Clinical Cognition: the distributed cognitive processes that emerge when clinical decisions are jointly shaped by clinicians, AI systems, documentation artifacts, institutional workflows, healthcare organizations, patients, and caregivers.
The repository contains conceptual work, early paper drafts, pilot study infrastructure, synthetic feasibility work, and a bounded Dataset v0.2 methodology cycle. Dataset v0.2.8 is closed, and Dataset v0.2.9 consolidated the current stopping point. No additional abstraction, coding, candidate expansion, clinical validation, clinical guidance, or Research Note 002 work is currently authorized.
The project evolved through the following intellectual path:
AI-ready Dental Case Packet
-> Dental Context Specification (DCS)
-> Reasoning Decay
-> Clinical Reasoning Reconstruction
-> Human-AI Co-Reasoning
-> Clinical Reasoning Transformation
-> Clinical Cognition Transformation
-> CCTL
The project began with dental AI infrastructure and the practical question of how clinical context could be packaged for AI use. DCS then emerged as a possible structure for representing clinical context and reasoning traces. Work on reasoning decay and reconstruction revealed a broader question: clinical reasoning was not simply being documented, preserved, or lost. It was being transformed as it moved across people, documents, AI systems, institutions, and patients.
That broader shift led to CCTL. The central object is no longer dental packet infrastructure alone. The central object is clinical cognition as it transforms within distributed human-AI healthcare systems.
Human-AI Clinical Cognition refers to the distributed cognitive processes that emerge when clinical decisions are jointly produced by clinicians, AI systems, documentation artifacts, institutional workflows, healthcare organizations, patients, and caregivers.
The current central question is:
How does clinical cognition transform within distributed human-AI systems?
CCTL studies transformation rather than only AI performance, automation, or tool adoption. The focus is on how cognition is preserved, compressed, delegated, translated, reframed, reconstructed, redistributed, or changed as it moves through healthcare systems.
- Manifesto
- Research Agenda
- Research Programs
- Founding Statement
- Clinical Cognition Is Becoming Distributed
- Reasoning Decay Revisited
- Can Clinical Cognition Be Reconstructed?
Pilot 1, Referral-to-Consult Reconstruction, has been defined, operationalized, and used as the first methodological testbed.
Dataset v0.1 established a synthetic educational feasibility workflow.
Dataset v0.2 explored controlled movement toward citation-linked educational materials, including source selection, second review, abstraction safeguards, limited abstraction, limited coding, and human review gates.
Research Note 001 documented the synthetic referral-to-consult reconstruction feasibility study.
Research Note 002 has not been started and remains prohibited.
Dataset v0.1 used synthetic educational cases to test whether the referral-to-consult reconstruction workflow could produce structured, interpretable outputs.
It was a feasibility workflow and methodology test. It did not use real patients, did not provide clinical validation, and did not create clinical guidance.
Dataset v0.1 should be read as a controlled proof-of-concept exercise for testing coding categories, rubric structure, and reconstruction workflow usability.
Dataset v0.2 explored whether the reconstruction workflow could be adapted to messier educational materials without copying source case text, adding patient-identifying information, or creating clinical guidance.
The v0.2 cycle included:
- candidate source selection;
- second-review eligibility verification;
- abstraction safeguards;
- human review gates;
- limited abstraction;
- limited coding;
- post-coding human review.
Dataset v0.2.8 created and coded two limited educational abstractions: ABS-003 and ABS-004. Human post-coding review accepted both limited coding outputs with caution and closed Dataset v0.2.8 after the limited coding pass.
Dataset v0.2.9 consolidated the status and closure record.
Dataset v0.2 is closed.
Remaining candidates are on hold.
Research Note 002 is prohibited.
No additional abstraction is authorized.
No additional coding is authorized.
No full Dataset v0.2 expansion is authorized.
The repository may preserve and review the existing Dataset v0.2.8 record, but no new dataset, abstraction, coding, or research-note work should be inferred from unfinished candidate pools.
CCTL has not demonstrated:
- clinical effectiveness;
- diagnostic accuracy;
- patient outcomes;
- treatment quality;
- clinical superiority;
- real-world deployment effectiveness;
- clinical validation.
The current repository artifacts are conceptual, methodological, educational, and exploratory. They should not be interpreted as evidence that any clinical intervention, AI system, diagnostic process, or treatment workflow improves care.
The current repository boundaries are:
- educational;
- methodology-focused;
- non-clinical-guidance;
- privacy-safe;
- copyright-aware;
- research-oriented.
The repository should not be used for diagnosis, treatment planning, clinical recommendations, patient-care advice, or operational clinical decision-making.
Open methodological questions include:
- how to distinguish compression from uncertainty loss;
- whether cognition reconstruction can be made valid beyond educational cases;
- whether coding categories remain consistent across reviewers;
- how second-coder reliability should be tested;
- how the cognition transformation taxonomy should evolve.
These questions remain open. The repository does not claim to have answered them.
The following are possible future directions only and are not currently authorized:
- additional controlled candidate testing;
- future abstraction studies;
- future coding studies;
- second-coder experiments;
- future reconsideration of Research Note 002.
Any future work would require explicit authorization, renewed gate review, and continued privacy, copyright, non-validation, and non-clinical-guidance safeguards.
Dataset v0.2 should be considered complete for the current cycle.
Further work should begin only after explicit future authorization and should not be inferred from the existence of unfinished candidate pools.