Run the canonical QC pipeline on a Formosan-<CORPUS> dev repo. Sequences clean_xml → orthography_detector (HUMAN INPUT) → standardize → add_phonology → validators, producing a README-style summary at the end. Use when a corpus dev repo needs QC before porting into FormosanBank.
Guided audit of an assistant-built corpus dev repo (../Formosan-<Name>/) before QC + porting. Reads the assistant's preprocessing, maps it onto our current pipeline, runs our validators on its output, and diffs its original tier against the source for dropped characters/punctuation and ungrammatical artifacts — pausing for the maintainer's judgment. Use when starting to review a new dev repo's preprocessing.
Move a QC'd corpus from its Formosan-<CORPUS>/ dev repo into FormosanBank/Corpora/<Name>/ with the standard layout, interactively. Surfaces decisions rather than guessing. Use when a corpus has passed QC and is ready for publication into FormosanBank.
Bootstrap a new Formosan-<CORPUS>/ development repository with the standard layout, Python environment, gitignore, README scaffold, and Claude Code safety rails. Use when starting work on a new corpus that doesn't already have its own dev repo.
Recompute a single corpus's audio-duration stats by pulling its audio from Hugging Face, then deleting the audio. Use when a corpus's audio has changed (or get_corpus_stats --report-stale-audio flags it) and its statistics/audio_durations.csv seconds need refreshing. Never part of CI or a normal merge.
Pull N random <S> elements from a published FormosanBank corpus and format them as a print/email-ready markdown report for a human expert to spot-check (translation quality, naturalness, content accuracy — things no automated validator can catch). Wraps QC/utilities/sample_sentences.py and walks the maintainer through generating + sending the sample. Use when you want a manual expert review of a corpus's contents.