skryba corpus

Verify a curated corpus against each sample's declared whitespace-word count and SHA-256 digest, then select a deterministic seeded subset. This command is offline.

Invocation

skryba corpus <PATH> --seed <SEED> [--samples <SAMPLES>]

PATH is required and must name a JSON corpus with a non-empty samples array. --seed is required: it orders the sample, and Skryba assumes none. --samples defaults to all samples.

skryba corpus corpus.json --seed <N>

Output and state

Stdout is JSON containing the expanded corpus path, verified_samples, and selected sample entries with id and whitespace_words. Selection order is deterministic for the seed. No file is written, no cache is changed, and no network credential is used.

Refusals

• A missing --seed is refused by the parser with exit 2.
• Unreadable input: reading curated corpus <PATH>.
• Invalid JSON or schema: parsing curated corpus <PATH>.
• No samples: curated corpus <PATH> contains no samples.
• Empty sample: curated sample <ID> carries no text.
• Word-count mismatch: curated sample <ID> declares <DECLARED> words and carries <ACTUAL>.
• Digest mismatch: curated sample <ID> declares sha256 <DECLARED> and carries <ACTUAL>.