Freeze CalVer synthetic data specification

[DearKarl]

What changed

• Migrated project-authored research artifacts to the 2026.08.12.1 CalVer snapshot.
• Froze the English synthetic-data specification, coverage matrix, record schemas, generator parameters, license policy, split policy, acceptance gate, collaborator handoff, and Evidence Card.
• Preregistered a 256-record no-training generator smoke with balanced KEEP and EDIT supervision.
• Preserved upstream dependency versions and external benchmark release names.

Why

The shared Explicit and LoRA research tracks require one auditable, source-traceable corpus contract before any synthetic data is generated. The revision also removes project-owned legacy naming from the public tree while retaining immutable upstream names.

Validation

• 18/18 data-free tests passed under Python 3.12.
• JSON and JSON Schema validation passed.
• Coverage matrix closed at 256 total records and 176/48/32 train/development/holdout.
• Markdown link scan passed.
• Evidence Card SHA256 seal verified.
• git diff --check passed.