The drill-down half of the loop: list queries return compact rows, and this expands named GlobalIds into every property set, quantity and material.

Ids that are not returned are reported in not_found, which covers both "no such element" and "hidden by policy" — deliberately without distinguishing them, so the response does not leak the shape of what is filtered out.

Find elements whose property matches a condition, e.g. fire-rated doors on level 3: type=IfcDoor, storey=Level 3, property=FireRating, op=exists.

pset is optional and usually best left out: the same field lands in Pset_DoorCommon on one exporter and a bespoke set on another, so by default every property set on the element is searched. The matched set and value are returned alongside each hit, so the answer says where it came from.

List elements of an IFC type, optionally narrowed by storey, name or PredefinedType. Subtypes are included by default, so IfcBuiltElement returns walls, slabs, doors and the rest. Returns one compact row per element with its GlobalId citation.

Elements that no IfcRelAssociatesMaterial reaches — through a plain material, a material list, a layer set, a profile set or a constituent set, all of which are resolved before the check. Restrict with type to ask the question that actually matters, e.g. only IfcWall and IfcSlab.

The inverse of elements_by_property, and the one most model-audit questions reduce to: which elements of this type have no value for this field? An empty string counts as missing by default, because exporters write one for "not filled in" as often as they omit the property entirely.

Schema, authoring application, units, storeys, and the element counts by type — of the visible model, after policy. Cheap, bounded, and the right first call: it tells an agent which types and property sets exist before it starts guessing names at them.

What values does this field actually take? Run this before elements_by_property rather than guessing: it is how you learn that the model spells fire ratings "EI60", "60 min" and "60" in three different property sets, which no amount of schema documentation will tell you.

Returns each distinct value with a count and up to three example GlobalIds.

Total a named quantity (NetSideArea, GrossVolume, Length, ...) over a selection, grouped by type, storey or PredefinedType. Lengths, areas and volumes are already converted to m, m2 and m3.

Every group reports without_quantity next to count. A total computed over 40 of 200 walls is not a total, and the caller is told so rather than left to assume the model was complete.

Spaces whose floor area falls inside a range, e.g. "rooms below the 8 m2 minimum". Areas are reported in square metres after unit conversion, so the threshold means the same thing on a model authored in millimetres and one authored in feet.

Quantity names are tried in order: the explicit quantity if given, then NetFloorArea, GrossFloorArea, Area. Spaces carrying none of them are reported in notes rather than silently dropped — "no answer" and "no data" are different findings.

The containment hierarchy, which is the map an agent needs before it can ask anything storey-shaped: it turns "level 3" into a GlobalId. Element counts are counts of visible elements, so a policy that scopes the model to two storeys shows zero elsewhere rather than hinting at what it is hiding.

Return the original file text behind a GlobalId: the STEP line that declares it, verbatim. Use it to verify a claim before repeating it, and when a reviewer asks where a number came from.