| # Initial 8,598 Protein Dataset | |
| This folder is a frozen manifest of the proteins with valid, mapped embeddings. | |
| The combined `manifest.csv` contains source, label, sequence hash, sequence length, provenance class, and the local embedding path. | |
| It contains 8,598 unique sequences: 4,289 positives and 4,309 negatives. | |
| Failed or unresolved embedding jobs are excluded from this snapshot but remain untouched in their original folders. | |
| Embedding tensors remain in their source-specific directories referenced by `embedding_path`. | |