Re 1: This will depend on the case. In case the inline filters are replacing a subquery (e.g. in case `FOR doc IN collection RETURN { name: doc.name, filteredSubAttributes: (FOR sub IN doc.attr FILTER sub.whatever == 30 RETURN sub.name) }`), they may be faster, because the overhead of a full-blown subquery can be avoided. In other cases, they may be on par with other constructs, and some cases they may be sub-par. We don’t yet have good heuristics yet, but I am pretty sure these operations can be optimized if required.
Re 2: [***********] will flatten up to 10 levels without errors, regardless of how deeply nested the value is. It will go depth-first into the value up to 10 levels deep and pull each member of an array up one level. It won’t recurse further than 10 levels, so nested structures beyond level 10 would remain. Only arrays and sub-arrays will be flattened, so if an array has less than 10 levels of nesting, then this won’t be a problem, and all nested values will be pulled up to the top level.
Re 3: yes, you’re right. The article is adjusted now. Thanks!