It shows its work: how the assistant earns a number

A takeoff number is only useful if you can defend it. That’s the principle behind the chat assistant at chat.construction.ai — before it gives you a quantity, it earns it. Here’s what that looks like in practice.

It starts with scale, because that’s where the money mistakes live. The assistant reads the title block and the scale bar, then asks you to confirm before a single scaled measurement gets reported. Printed dimensions are reported as written; anything scaled off the sheet is labelled an estimate. And if the title block and the printed dimensions disagree — the classic half-size print trap — it tells you, instead of quietly doubling your takeoff.

Counting starts at the legend, not the plan. Filled, open and half-filled symbols are tracked as separate types, because your legend says they’re different fixtures. Grid bubbles, keynote circles, section callouts and elevation markers are excluded, not tallied. Every count cites the legend entry it’s based on, and the breakdown comes zone by zone, so you can spot-check one area before you trust the sheet.

Schedules get reconciled, not assumed. If the fixture schedule says 33 and the plan shows 31, you hear about the gap — along with its best guess at which enlarged plan is hiding the other two. Callouts and match lines are followed across the sheets you selected, and mismatches get flagged instead of averaged away.

Assumptions sit out in the open. Formulas are shown with every input tagged by source — printed, noted, or scaled. Before you rely on any quantity, it asks you to validate one real measurement. When a scan won’t resolve, you get “I can’t read that” instead of a confident guess. And it never certifies code compliance — that’s your engineer’s stamp, not ours.

None of this slows you down — it makes the ten minutes you spend with the assistant worth something. Budgetary figures only, verify before you bid, and validate against the site. Try it on your next set at chat.construction.ai.