Data provenance
Dataset name or accession, cohort scope, assay, access conditions, inclusion criteria, and known batch or site effects.
Technical notes / evidence trail
Technical records for current research.
When published, VHL notes will document data provenance, representations, model configurations, negative results, evaluation, limitations, and corrections for current research. Each note will supplement, rather than replace, the corresponding manuscript.
Current availability
No standalone technical notes are public yet.
The research records available now are the two private manuscripts listed below. For signed essays about the research direction, visit the Blog.
Current evidence register
The current research record is focused on RNA.
These manuscripts are under patent filing and peer review. Researchers can request a private copy; full files are not posted on the website.
What a technical note contains
Enough information to inspect the result.
Split definition
The unit held out for testing—sample, donor, site, or study—and why that split matches the stated generalization claim.
Method record
Model version, configuration, preprocessing, comparison methods, and code or reproducibility status.
Limits and corrections
Negative results, uncertainty, subgroup failures, leakage risks, changes to interpretation, and dated corrections.
Coverage
What the current RNA evidence trail must answer.
Generated pseudo-cells versus measured single cells
Which properties of cell composition, expression, pathways, and immune state can be evaluated—and which claims remain out of scope.
What survives RNA compression
How a learned RNA token is tested across assay resolutions, cohorts, biological tasks, and language-model outputs.
Study-held-out and donor-held-out tests
Why random sample splits can overstate RNA-model performance and how cross-study transfer is reported.