# BioGNet submission format

Submit exactly one UTF-8 ZIP, up to the configured upload limit. The default limit is 200 MB compressed / 1024 MB uncompressed. Use ordinary relative paths, no encryption or duplicate basenames.

Required files:
- genotypes.csv: sample_id followed by SNP ID columns. ALT hard-call dosage 0, 1, 2; missing NA, blank or dot. SNP columns must occur in markers.tsv.
- markers.tsv: snp_id, chromosome, position, ref, alt. AGPv4, chromosomes 1–10, 1-based positions, biallelic A/C/G/T; exact REF/ALT alignment matters.
- samples.tsv: sample_id, material_type, mother, father. Unique IDs; material_type F1 or inbred. F1 requires directed parent IDs. Select Mixed if the archive includes both types.
- README.md: data provenance, experiment and intended analysis.

Optional for prediction only: phenotypes.csv. Required for evaluation: sample_id, environment, condition (WW/WS) plus the requested trait columns ASI (days), PH (cm), SEW (g/ear). Observed D evaluation/recalculation needs all three traits paired WW/WS per material/environment, using historical fixed transformation parameters; do not refit them to submitted results. Genotype-only D prediction does not require observed phenotypes.

At submission the service checks metadata tables, genotype headers and the first 20 genotype rows. Metadata tables are limited to 200,000 rows. It does not certify the full matrix, paired phenotype completeness or inference compatibility. The team checks those offline, including task-specific frozen 10,000-SNP panel order, alleles, missingness, preprocessing and training-reference GRM. The downloadable 11,388-marker list is the union of historical candidate panels, not an interchangeable inference input.

The ZIP template contains synthetic samples and phenotypes with four example markers. It is a format example only. No inference or email is run automatically. Keep your reference ID and email for status and deletion requests.