- Dataset card: CC BY-NC 4.0 licence (as the benchmark datasets), layout, file formats, species and sources.
- `reference.gff3.gz` -> `reference_longest_isoform.gff3.gz` (19 folders, server-side copy).
- New `reference_all_isoforms.gff3.gz` for the 17 test species: the same genes with every isoform of the source release, plus the reference transcript where that release lacks its exact CDS. Each gene's isoforms equal the paper's any-isoform accepted set (`build_any_isoform.isoform_sets`) on 17/17 species.
- Medicago and Mimulus `genome.fa.gz` (+ `.fai`, `.gzi`) now hold the whole assembly (42 / 23 records, was 14 / 14 contigs >= 131 kb, which dropped 97 / 41 reference genes). The 14 records already hosted are byte-identical.
- Drops `plants/reference_curves.csv`, which nothing reads since the curves cell left notebook 03a.
Merge together with the ntv3_revised Space PR that switches the notebooks to the new reference name.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Remove flat root copies superseded by the <clade>/<split>_species/ layout
genome.fa.gz: NCBI GCF_002263795.3_ARS-UCD2.0_genomic.fna.gz (md5 b856ff696adbb5bebd858cd28c96878a), headers cut to the seqid, sequence and NCBI soft-masking unchanged, bgzipped + indexed; 1,958 records, 2,770,686,120 bp incl. chrY. reference.gff3.gz / pseudogenes.bed: export_reference() on the benchmark's annotation.filtered.gff3.gz (NCBI RefSeq GCF_002263795.3-RS_2024_12), NC_/NW_ seqids, so the released GFF3 scores with no renaming. Replaces ARS-UCD1.2 as the benchmark cow. Not listed in metadata.tsv, which is the plant tutorial's split table.
Add the 03b demo predictions: released NTv3_annot_plants on Medicago NC_053042.1:11.0-12.6 Mb
