Prompt 09 — Observation, dataset, and evidence builder

Inputs

DATASET_SOURCE: {{DATASET_SOURCE_JSON}}
COLLECTION_CONTEXT: {{COLLECTION_CONTEXT_JSON}}
RAW_SCHEMA: {{RAW_SCHEMA_JSON}}
MODEL_EVALUATION_PLAN: {{MODEL_EVALUATION_PLAN_JSON}}
VISIBILITY_POLICY: {{VISIBILITY_POLICY_JSON}}
PROVENANCE_METADATA: {{PROVENANCE_METADATA_JSON}}

Task

Create a dataset and evidence card without fabricating rows or measurements.

• describe what was measured, observed, simulated, or derived;
• preserve collection protocol, instruments, sampling frame, time ordering, and environmental conditions;
• define columns with quantity IDs, units, dimensions, uncertainty, missing-value semantics, and censoring;
• separate raw, calibrated, processed, and derived datasets;
• record preprocessing, exclusions, leakage risks, missingness, and known biases;
• freeze training, validation, held-out, intervention, replication, benchmark, and falsification roles before model evaluation where possible;
• record learner, validator, and historian visibility separately;
• capture licensing, privacy, and use constraints;
• reject ambiguous train/held-out overlap;
• make evidence role explicit: data is not automatically proof of the associated claim.

Return only JSON matching schemas/dataset-evidence-card.schema.json.