examples-guide

What problem does it solve?

It helps you choose the correct dataset-building and training pattern (RL-style forecasting vs content-learning SFT vs tabular mapping) so you don’t waste cycles on the wrong approach.

Core Features & Use Cases

• Decision tree for dataset/training selection: guides whether to use forward-looking GRPO, content-learning SFT, or tabular sample mapping based on your inputs and labels.
• Answer-type framing guidance: explains when to use binary, multiple choice, numeric, or free response to improve training signal quality and labeling reliability.
• Practical forecasting guardrails: emphasizes temporal splitting, avoiding leakage, linting before splits, and ensuring prediction dates precede outcomes.

Quick Start

Ask: “Given my data type and goal (teach domain facts, predict future outcomes, or process a table), which Lightning Rod training pattern and answer type should I use, and what are the key steps and pitfalls to avoid?”