TELLSEED TRAINING-DATA CAMPAIGN BRIEF Template, not a completed study. Replace every placeholder before collection. Guide: https://tellseed.com/guides/collect-ai-training-data 1. Purpose Decision or model behaviour to improve: Intended use of the resulting dataset: Uses this dataset is not suitable for: 2. Source material Source owner and permission to process: Input format and stable external item key: Which items are synthetic and how they are labelled: Personal or confidential information to remove: Languages, domains and important edge cases: 3. Task What the contributor sees: Expected response format and allowed values: Definition and boundary example for every label: Rule for missing context, ambiguity and skips: Whether AI assistance is allowed and how it is declared: 4. Contributors Who can make a valid judgement and why: Recruitment method and expected coverage gaps: Information participants receive about data use: Permissions, retention and withdrawal process: Name/email collection, if necessary, and its separate purpose: 5. Pilot and collection Total item pool: Items shown per round: Ordering: sequential, shuffled or coverage-weighted: Target independent judgements per item: How independence and coverage will be checked: Published campaign version: Pilot findings and instruction changes: 6. Review Acceptance, correction and rejection criteria: Required review count: Who resolves ambiguous cases: How collection judgements differ from later review decisions: 7. Export and evaluation Raw and reviewed counts expected: Split grouping: scenario lineage, contributor or none: Split ratios and seed: Additional duplicate, source-document or subject overlap checks: Separate held-out evaluation set: Provider-specific conversion and validation: Manifest, file hashes and preparation script version: Tellseed hosts collection and review workflows. You arrange the audience, source permissions, dataset governance and model training. A synthetic pilot is not evidence of model performance or representative human preferences.