Verified sequence bundle

The current bundle preserves 245,535 original Ensembl 115 peptide records and 245,535 source occurrences. It contains 230,592 exclude, 13,543 review and 1,400 allow records; 925 records are direct-target candidates before combining all matching occurrences. Record counts are not protein-group counts.

Loader API and examples · Manifest · Coverage · Historical transition receipt · Exact full-sequence comparison.

Selection policy: oncoref.cta-selection.2026-11-08.1. Role policy: oncoref.sequence-roles.2026-11-04.1. Translation policy: oncoref.translation-qualification.2026-11-04.1. Representative policy: oncoref.transcript-selection.2026-11-08.1. Bundle identity: oncoref:sequence-bundle:v1:sha256:c8c060edab62e089fd4651065a857a8eaecb7c68816b016b3ac4c9fd06fa9525. Reference identity: openvax:reference:v1:sha256:ba6f6186a301a3ef9217ea0199ec3b851f9da2eb43558b7ba7e0d6fcd9cd85e9. Native ORFdata identity: dbd1dd752e24ea08f4d787e48fe04e3464ebce248190572782d6e356e010822f.

Coverage is explicit

The peptide inventory has 3,770 source loci without a current decision; these require review. 0 classified loci have no translation in this inventory. The coverage payload retains 0 records outside the classification path and 1,063 ncORF nominations without claiming an empirical ncORF sequence inventory. These lists are different coverage questions and must not be added together as a disjoint denominator.

Every original peptide, including incomplete/NMD/LoF and alternate-locus records, remains present. Coordinate-bearing AA20 segments never bridge ambiguity or stops. Exact raw-sequence grouping retains all records and occurrences. Exclude takes precedence over review, then allow; an allowed CTA occurrence cannot cancel another occurrence's exclusion. Missing records never supply implicit allowance.

Reproduction

Build from the verified native ORFdata snapshot and exact reference manifest:

python scripts/build_cta_sequence_bundle.py SOURCE BUNDLE --release 1.35.0
python scripts/build_cta_sequence_transition.py HISTORICAL BUNDLE

The builder checks source/reference identity, complete qualification-table coverage, original sequence hashes, all payload hashes and counts, and an offline ORFdata round trip. The final release rebuild binds the packaged data version to the published ZIP pin. Historical assets retain their own explicit policy identity; the comparison uses exact full-sequence hashes and never substitutes a partial segment or approximate match for full equivalence.