Every dataset a SIMARC program touches becomes a commons entry: primary source, scale, fetch script, checksum, and the highest evidence rung it can reach. No account, no agreement, no fee. Data licensed for redistribution is mirrored with SHA-256 manifests. Everything else ships as a re-download script plus hashes — equally reproducible, and legally clean.
The rule inherited from every program: every byte re-downloads from a primary source. That is not a storage decision but an epistemic one: a dataset that cannot be re-fetched has provenance which is asserted rather than demonstrated. Where a publisher provides a checksum, it is verified on every run: the dementia program's GWAS is md5-verified against the EBI's published checksum each time it loads.
Nothing here is wrapped in a bioinformatics framework or an analysis toolkit, on purpose. The point of the commons is to know exactly which records were dropped and where, and a pipeline whose source cannot be inspected is a narrative rather than a measurement.
| dataset | scale | program | rung it reaches | access |
|---|---|---|---|---|
| TCGA pan-cancer RNA-seq | 33 projects · ~11,500 patients | P-03 | OBSERVATIONAL | fetch + hash |
| DepMap CRISPR gene effect | 1,208 lines × 18,531 genes | P-03 | PERTURBATIONAL | fetch + hash |
| Dose–response (CTRPv2 / GDSC et al.) | 5,450,000 measurements | P-03 | PHARMACOLOGICAL | fetch + hash |
| ClinicalTrials.gov — oncology | 97,623 interventional trials | P-03 | CLINICAL | live query, timestamped |
| ClinicalTrials.gov — dementia | 8,224 trials · 731 classified dead | P-04 | CLINICAL | live query, timestamped |
| Bellenguez 2022 AD GWAS | 21,101,114 variants · n=487,511 | P-04 | GENETIC | fetch + md5 verified |
| Europe PMC — dementia corpus | 227,231 records + retraction metadata | P-04 | prior art | live query |
| Open Targets · gnomAD | associations · constraint · frequencies | P-03 · P-04 | prior art | fetch + hash |
| Planck & WMAP sky maps | 4 independent sky-cleanings | P-01 | the boundary seam | fetch + hash |
| Quasar catalogs | 756,000 objects | P-01 | the lattice seam | fetch + hash |
| The Butson corpus, order 6 | 27 classes, byte-verified | P-01 | PROOF | mirrored |
| DRAT proof artifacts | 145 MB, hash-pinned, replayable | P-01 | PROOF | mirrored |
| Mathematical subject surveys | 33 fields | P-01 | reference | mirrored |
Machine-readable manifest: /api/v1/datasets.json
Some data is behind an application, and pretending otherwise would be the exact failure this institute is built against. ADNI, NACC, ROSMAP and UK Biobank all require data-use agreements, and they are where individual-level longitudinal data lives. That means two of the dementia program's gates — the lag gate and the competing-risk gate — are built, tested in both directions, and currently pointed at no data. That is written down as a block rather than worked around, and it is one of the things SIMARC is seeking collaboration on.
Humans and agents can nominate datasets for accumulation. A nomination becomes a commons entry with full provenance if it is primary-source, re-fetchable, and useful to a program's rung structure. Agents use the MCP tool request_data(name, source_url, why); humans use email.
