CCDD is organised in four levels: raw data, coarsely processed data, finely processed data and supplementary data. The fine level is the one most users will load; it reorganises every match into match-, stage- and clash-level instances, each paired with the three judges' independent judgments.
Per match: one MP4 recording (all audio; video mainly shows the timer and match-control interface) and three judge voting forms as XLSX with every stage score, impression-vote placement, deciding-vote placement and best-debater nominees.
168 complete adjudication instances (148 ordinary, 20 appeal). Repaired ASR transcripts with source-line indices retained; sentence sequences segmented into eight stage types; stage rows aligned one-to-one with the three score sheets.
Match-, stage- and clash-level instances ready for training and evaluation. 148 match instances, 2,698 stage instances, 20,542 clash instances; 504 match-level and 9,163 stage-level supervision signals including appeals.
Concise adjudication summaries and structured reasoning accounts generated from each judge’s commentary with Fable 5 under a 900-character limit, then reviewed by 31 experts for severely unreasonable content (two records were regenerated). Stored separately and explicitly distinguished from the judges’ original commentary, scores and votes.
| Level | Input | Labels | Count |
|---|---|---|---|
| Match | Full transcript organised stage by stage, with format and motion | Hierarchically constructed winner, nine-vote ballot pattern, score normalised to [0, 1] | 148 |
| Stage | One stage plus all preceding context; evaluated side and role identified | The three judge scores for the evaluated side (1–10 each) | 2,698 |
| Clash (entry) | One question–answer or attack–defence unit with the roles of both sides | Inherits the parent stage's score; no separate label | 20,542 |
Stage instances are packaged by type into five files: speech-like stages, questioning side, answering side, free debate, and head-to-head exchange / surprise-attack stages. Clash segmentation yields 6,457 question–answer pairs from questioning and cross-examination, and 4,445 questions, 4,445 answers and 3,328 items from free debate.
Name, type, acting side and a sentence-level utterance list, plus the aligned list of judge scores. Monologue stages carry a score for the speaking side only; questioning and cross-examination carry separate scores for the questioning and answering sides.
Appealed matches carry an identifier marking pre-appeal and post-appeal adjudication. Text volume is about 3.85 million Chinese characters: 2.60 million of transcript and 1.25 million of judge commentary.