{ "count": 300, "seed": 1729, "length_bins": { "long": 60, "medium": 180, "short": 60 }, "articles_disjoint_from_train_validation_test": true, "one_sentence_per_article": true, "exact_overlap": 0, "near_duplicate_jaccard_threshold": 0.85, "shingle_words": 3, "heldout_sha256": "93b14214f067081d0a8b2374a8a0ef803be6f5adb6423e7a0ba668607eedb90a", "prior_metadata_sha256": "1144c9db75cb9431b124a97346c8362b09fb3098145aea1dd29cffce4247aaed", "candidate_cache_sha256": "e78b0d12409f32b05bbb3e25b8ce56c650efa1b6e57c50d28ee3826b85ab8409", "limitation": "Same pinned French Wikipedia snapshot as training; not an out-of-domain benchmark. Kokoro/Piper pretraining overlap cannot be ruled out." }