How we work
Autonomy in the work.
Care in the conclusions.
The lab delegates research tasks to AI agents. Every result still needs a clear question, inspectable evidence and an honest account of what remains uncertain.
01 / Research cycle
From a question
to a record.
- 01
Define the question
Specify the population, comparison and outcome. Write the scope and analysis plan before interpreting results. Decide what would count against the hypothesis.
- 02
Find and inspect evidence
Search accessible literature and suitable public datasets. Preserve search records, source locations, access limits and reasons for inclusion. An abstract is not treated as a fully read paper.
- 03
Analyse and challenge
Write code when calculation is needed, retain its inputs, and check the outputs. A separate AI task challenges pivotal extractions and interpretations. Agreement between related AI models remains internal review.
- 04
Report with visible limits
Release a readable overview alongside methods, sources and reproducible material where applicable. Separate findings from hypotheses and record corrections. A completed study is not automatically externally peer reviewed.
02 / Continuity
A lab that can
pick up its work.
Continuity lives in the project records.
The laboratory maintains a manifesto, research protocols, task queue, evidence tables, analysis code and decision history. These give each working session a place to resume and make completed work inspectable.
The AI coordinator assigns bounded tasks, integrates their outputs and reports progress and limitations to Pawel. Scheduled sessions depend on the local computer and application being available. Reliable unattended completion is still being developed; we do not claim continuous operation.
Private working records and personal context stay separate from public research reports.
03 / Scientific standards
Different claims.
Different evidence.
Keep outcomes distinct
Feeling better, reducing inflammatory activity, preventing further structural damage and repairing an existing lesion are different research questions. We do not substitute one outcome for another.
Keep uncertainty attached
Observational associations, experimental findings and computational hypotheses carry different kinds of uncertainty. We preserve that distinction through the analysis and into the report.
Invite qualified scrutiny
AI checks can catch errors, but they cannot establish clinical truth. Specialist interpretation, independent validation and laboratory experiments require human collaborators. We are seeking those relationships.
Collaborate with the lab