Checking all 118 window marks — mark-existence based consistency review

The limits of sample-based review

Window review has traditionally been sample-based, because manually matching hundreds of window marks on the drawings against the schedule, one by one, was physically impossible. The problem is that a single mismatch leads to ordering errors, rework, and delayed handover.

Mark-existence based full cross-checking

The doAZ window verification console detects 118 window marks on the unit floor plans and cross-checks every one of them against the schedule (the registered list), mark by mark. An example of the verdicts from an actual verification screen:

• 118 detected on the plans · 9 distinct marks · 4/9 consistent on both sides
• 2 missing from the plans — marks listed in the schedule but absent from the drawings
• 3 absent from the schedule — marks drawn on the plans but not registered in the schedule

The principle matters here: consistency is judged on mark existence. Counts are treated as reference only, because the plans (the sum across all units) and the schedule (the registered list) aggregate at different levels, which makes a 1:1 comparison inappropriate — the marks missing from the plans or absent from the schedule are what actually needs review. Forcing a comparison between counts that aggregate differently, and then declaring a "match", is exactly the kind of plausible wrong answer we refuse to produce.

The evidence sits right there on screen

Every verdict is placed at its coordinates on the drawing. Click a mark box and that location is highlighted, opening alongside the corresponding row in the schedule. The reviewer does not take the AI's verdict on trust — they look at the evidence and confirm it.

This structure is what doAZ means by Zero Wrong-Link: the principle of conservative matching that never creates a link that is wrong.