ADadaptionlabs
Dataset

Validate that a set of Datasets can be combined

Changed on

Synchronous compatibility check. Returns a single issues array containing both blockers (level: "error") and reconcilable concerns (level: "warning"). compatible: false means at least one error-level issue is present (cardinality, ownership, status, run_id). Warning-level entries — column-type mismatches, DPO sources missing ranked generations — are non-blocking; the merge resolves them automatically (mixed types downgrade to instruction_dataset, type mismatches CAST to a common type, etc.). Nothing is written.

post/api/v1/datasets/combine/validate

Request

Request body

dataset_idsstring[] required

Source Dataset IDs to combine. Must all belong to the caller’s org and be in a combinable state.

Response

Returns {compatible, issues, training_type?, total_estimated_rows?}. Each entry in issues carries a level ("error" or "warning"), a stable code, a preformed user-facing message, and code-specific structured context. Error codes block the merge; warning codes describe reconciliation the merge performs automatically.

compatibleboolean required

Whether the proposed merge can proceed.

training_type'instruction_dataset' | 'preference_pairs'

Resolved training type for a compatible merge. Homogeneous source sets inherit their shared type; mixed instruction + preference sets resolve to instruction_dataset.

total_estimated_rowsnumber

Sum of processed_rows across all sources for a compatible merge.

Changes