Guide: Setting up a 'golden dataset' of correct code to steer suggestions.

Hi everyone! I'm Liam92, and I've been lurking here for a bit while trying out Windsurf. I'm really impressed by the in-editor AI, but I've hit a common snag: sometimes its suggestions drift from our internal patterns, especially for SQL transformations and data model definitions. I'm eager to use it more but cautious about having to constantly correct it.

I come from a SQL and data modeling background, and we have a pretty established set of patterns—like how we name CTEs, handle incremental loads, or structure our dbt models. I've seen folks here mention the concept of a "golden dataset" to steer suggestions. Could we maybe build a community guide on setting that up effectively?

My thinking is, if I could point Windsurf at a directory of our best, approved SQL files and Python data pipelines, it would learn from that correct context. But I'm not entirely sure about the practical steps. Like, is it as simple as linking to a GitHub repo full of exemplar code? Or do you need to feed it specific, curated examples in a certain format? I'm worried about it picking up on any stray, non-standard comments or patterns in a broader codebase.

Also, has anyone done this specifically for data engineering concerns? I'd love to hear about:
* What file types and structures you included (e.g., `.sql`, `.py`, `.yml` for dbt).
* How you kept the "golden" dataset clean and maintainable.
* Whether you saw a noticeable improvement in suggestion accuracy for your niche patterns.

I'm hoping we can pool our experiences. A good guide would help newcomers like me commit to using the tool more deeply, knowing we can anchor it to our own standards 😊. Thanks in advance for any tips or stories you can share!