GROQ query to find nested custom blocks with invalid marks array

I can see you're struggling with filtering deeply nested arrays in GROQ. This is a common challenge, and the key is understanding how GROQ handles array traversal in filter conditions versus projections.

The issue you're encountering is that when you write something like:

brodtekst[_type == 'hjelpeartikkel.expandCollapse'].innhold[_type == 'bilde'].marks[]

This creates a flattened array of all marks from all matching bilde objects across all matching brodtekst items. When you then check != null, you're not checking whether specific documents have marks, but rather whether this flattened array contains anything.

Here's a working approach using nested count() functions to check for the existence of marks at the right nesting level:

*[_type == 'hjelpeartikkel' && 
  count(brodtekst[
    _type == 'hjelpeartikkel.expandCollapse' && 
    count(innhold[_type == 'bilde' && defined(marks)]) > 0
  ]) > 0
]{
  _id,
  "problematicBlocks": brodtekst[_type == 'hjelpeartikkel.expandCollapse']{
    innhold[_type == 'bilde' && defined(marks)]{
      _key,
      _type,
      marks
    }
  }
}

This query works by:

1. Using count() at each nesting level to properly evaluate whether matching items exist
2. The inner filter innhold[_type == 'bilde' && defined(marks)] only matches bilde objects that have a marks field
3. The outer count(...) > 0 checks if there's at least one matching expandCollapse item
4. In the projection, the same filter returns only the problematic blocks

If you want to be more specific and only find bilde objects where marks is an array with actual content (not just defined but empty), you can use:

*[_type == 'hjelpeartikkel' && 
  count(brodtekst[
    _type == 'hjelpeartikkel.expandCollapse' && 
    count(innhold[_type == 'bilde' && count(marks) > 0]) > 0
  ]) > 0
]

The count(marks) > 0 ensures you're only matching bilde objects where the marks array actually contains elements.

Why your original attempts didn't work:

• brodtekst[].innhold[].marks[] != null flattens all marks across all documents into a single array, then compares that array to null (which will always be true if any marks exist anywhere)
• While chained array syntax like array[].nestedArray[] works for projections, when you're using it in filter conditions with comparisons, the flattening behavior can produce unexpected results
• Using count() at each nesting level gives you explicit control over what you're checking at each depth

If you're still getting unexpected results, you might also try using length(marks) > 0 as an alternative to count(marks) > 0, though they should behave similarly for arrays. Another option is to add && length(marks) > 0 for extra certainty that it's a non-empty array.