Output structure that's friendly for LLMs

[D3MZ]

The way the documentation is setup, it makes it difficult for an LLM to read, RAG, or train on.

Problem 1

I'm personally looking for Standard Deviation function. But all of your aggregate functions are listed as links, not on the page directly. Without prior knowledge of the function name, I'll have to scan through all the shorthands and click on the pages that I think are relevant.

Problem 1.5

I believe this structure is why your own chat LLM is "lazy" to answer questions like "list all the aggregate functions"

Based on the knowledge sources provided, I cannot provide an exhaustive list of all aggregate functions in ClickHouse... [more verbose explanation here]

Solution

Single document is OK:
Have a TOC with the same logically grouping as you have now.
Have all the functions to be formatted as

1. Name
2. Description.
3. Examples with output
   Ideally the examples should be less than 1000 tokens each, so RAG will pull the entire code as context.

[gingerwizard]

#2888 will fix