How to review an AI window takeoff

Write down which pages were analyzed and which were omitted. A clean model from selected pages cannot establish that the whole drawing set has been covered. Check revision scope, additional buildings, floors and alternate work against the complete source.

In the six-sheet teaching case, all six pages contribute useful evidence. For larger sets, review omitted pages separately and reconcile any additional opening information. A page count is a coverage aid, not proof of correctness.

Compare the placed openings with marks in the plan, elevations and schedule. Look for an opening counted twice across views, a mark attached to the wrong wall, mirrored placement and schedule rows with no assigned opening.

Keep unresolved AI proposals and context-only elements separate from placed quantities. An unresolved item should remain visible as work to review; it should not disappear because the main model looks complete.

Check width, height, units, sill and dimension basis against the source. Frame size and rough opening are different meanings. A geometric rectangle does not prove the product family, operation, handing, glazing or performance requirement.

Our synthetic schedule identifies six windows but leaves their operation and subdivisions unspecified. Keep those fields unknown until evidence or an explicit decision resolves them. Editing a model value does not independently verify that value.

Resolve conflicts with an identified source or record the question for clarification. Preserve source references and review status. Compare the exported opening rows and totals with the reviewed workspace before sharing the file.

A preliminary supplier brief supports discussion. Final product selection, performance, installation details and manufacturing dimensions still need project-specific confirmation. Do not turn an AI confidence impression or one successful sample into an accuracy claim.