Barbados historical handwriting: v055 review checkpoints

This repository supplies trained model artifacts for the automatic v055 competition review solution. The code package is delivered separately to the organizers. It contains no competition images, label tables, OCR predictions, or submission CSVs.

The reviewed public score was 0.927525446 (weighted WER 1.306287832; weighted CER 1.985048359). The private score has not been supplied. Postcompetition manual transcript corrections are not used.

manifest.json maps every historical checkpoint file and fitted selector to a SHA256-addressed file under weights/. Identical files share one stored object. Restore each target_path beneath the configured trained-model directory and link it into the runtime workspace. Pin a completed commit when downloading; do not use an in-progress upload as a complete dependency set.

Public pretrained initialization models remain available from their upstream repositories; this repository does not duplicate those bases. Model families include TrOCR full-line/word/phrase OCR, Kraken CTC, Qwen OCR LoRA variants, and CPU selectors. The selected checkpoint inventory plus historical initializers/support selectors is retained conservatively and is not claimed to be the smallest possible model set.

Upstream model licenses and the competition's data-use terms continue to apply. This card does not grant new rights to the competition dataset or imply that the complete raw-image pipeline has passed a new clean-environment run. For the precise training, inference, runtime and validation scope, consult the separate organizer code package.