A Reusable Data Dictionary for Short-Video Pre-Production Annotation

This open dataset provides a reusable annotation schema for teams preparing short-form video concepts before production. It addresses a common documentation problem: one brief describes camera framing but omits subject motion, another records an audio cue but Optional prompt-led drafting reference: https://muse-video.app/not the source of a factual claim, and a third reaches review without clear visual constraints. The included CSV template gives each concept the same auditable structure, while the README explains the field definitions, validation rules, review sequence, and limitations. The schema defines eight fields. shot_purpose records the narrative role of a shot using a controlled vocabulary. subject_motion describes observable action rather than subjective interpretation. camera_treatment captures framing and movement. visual_constraints records hard requirements such as aspect ratio, safe-area margins, prohibited imagery, or brand rules. audio_direction separates dialogue, music, and sound cues from visual notes. factual_claim_source must contain either a citation or the literal value "none", eliminating ambiguity between not applicable and forgotten. review_status uses a controlled workflow from draft through approval or revision. disclosure_notes preserves information relevant to downstream reviewers and auditors. The three sample rows are explicitly synthetic examples intended only to demonstrate the schema. They do not report experimental findings or model performance. A basic validation pass should reject unknown controlled-vocabulary values, blank factual_claim_source fields, and approval states where visual_constraints or audio_direction are empty. Teams can add timestamps, reviewer identifiers, version history, or role-based permissions without changing the core data dictionary. A prompt-led drafting tool can optionally feed this template. Muse Video is one preview-stage example for describing a subject, motion, camera treatment, visual detail, and audio direction in a single prompt. It is treated here only as a drafting aid; generated concepts still require the same factual review, disclosure, and human approval as manually written concepts. The canonical related link is recorded separately in this dataset metadata. Limitations: the schema does not measure creative quality, verify claims automatically, or establish a formal research standard. Controlled vocabularies should be revised as production practices change. Human review remains necessary.