transform-pipeline-verification

What problem does it solve?

This Skill helps prevent wasted cost and bad training data by showing you how to run and inspect Lightning Rod transform pipeline outputs at intermediate and full stages before scaling up.

Core Features & Use Cases

• Iterative pipeline verification: Run a QuestionPipeline with the minimum stages you need (including seeds-only) and inspect the resulting dataset immediately.
• Quality and distribution spot-checking: Validate dataset fields such as is_valid, label distribution, and sample-level fields like question_text, label, reasoning, and invalid_reason.
• Server-side dataset linting: Run the dataset linter to catch structural issues (e.g., duplicates, missing required fields, label inconsistencies) before splitting or training.
• Use case: A notebook workflow where you generate 10–50 samples, confirm validity and label quality, then use estimate_cost and rerun at larger max_questions once the pipeline looks healthy.

Quick Start

Generate a small seeds-or-full transform run, download the produced dataset rows, spot-check validity and label fields, and then lint the dataset before you split or train.