Upload README.md

README.md

@@ -36,7 +36,7 @@ readouts in a structured set of tables. The final dataset contains:
 - **1,120 oligo records** (1,000 generated non-control oligos + 120 controls)
 - **5,600 assay instances** across multiple time points and replicates
 - **16,800 replicate-level toxicity readouts**
-- **128 tables** including 8 core modeling tables (oligo metadata, modifications, position-level modifications, biophysical, dose, assays, readouts, controls) plus per-source
+- **127 tables** comprising 8 core modeling tables (oligo metadata, aggregate chemistry, position-level modifications, biophysical, dose, assays, readouts, controls) plus per-source
   evidence modules and supporting metadata files
 
 The dataset is intended for use in developing, training, benchmarking,
@@ -188,6 +188,19 @@ The full calibration specification is recorded in
   high) is a deliberate benchmark stratification chosen to produce a
   non-trivial classification problem with both common and rare strata;
   it is not a claim about wet-lab oligonucleotide toxicity prevalence.
+- Of the 1,120 rows in `oligos.csv`, 1,089 represent oligos with
+  explicit sequences (`sequence_length > 0`) and 31 represent context
+  records without an administered oligo: vehicle/mock controls,
+  clinical risk-score contexts, extracellular-vesicle delivery
+  contexts, small-molecule rescue contexts, and non-administered
+  injury-model contexts. These context records carry an
+  `oligo_modality` value that explicitly indicates they are not
+  administered oligos (e.g., `vehicle_mock`,
+  `context_not_administered_oligo`,
+  `extracellular_vesicle_context_not_oligo`,
+  `clinical-risk-score-context`). Users restricting analysis to
+  sequence-bearing oligos should filter on `sequence_length > 0` or on
+  the modality field.
 - Original source PDFs are not redistributed; sources are identified
   through title-level metadata only.