Overview
Every AI-generated scope item carries a confidence score in the Source column, next to the AI badge. It's the AI telling you how sure it is about that line — and therefore where your review time is best spent.
The score is not a promise about the estimate. It's a signal about the evidence: whether the number came from a dimension on the drawing, or from an assumption the AI had to make.
How to read the scores
High — measured or read directly from the drawing: a dimension, a schedule, a callout. This is the practical ceiling; the AI won't return 100%.
Medium — derived from the drawing, but with a step of interpretation, such as a perimeter taken from a plan or an area matched to another assembly.
Low — shown in red and bold. The AI had too many variables and not enough information to be confident without your input.
Low scores are displayed in red and bold so they stand out without relying on color alone.
Read the item description. When the AI has assumed something, it shows its arithmetic — for example "assumed: 180 LF x 0.5' W x 3' D / 27 = 10 CY" or "Assumed: ~260 LF total perimeter (Main House + ADU) × 10' average height = 2,600 SF". You can check the logic in seconds and correct just the input that's wrong.
Why a score is low
The usual causes, in rough order of frequency:
The sheet isn't in the upload. If the structural set is referenced in the drawing index but wasn't uploaded, framing gets estimated from architectural sections instead of measured. The Assistant tells you when this happens.
The dimension isn't on the drawing. Wall heights, trench depths and slab thicknesses often live in details or schedules that may not be present.
The scope boundary is genuinely ambiguous — whose scope is the demolition, for instance. This is what the focusing questions at the start are for.
Check the Assistant's Show page-to-trade map: any trade showing Primary: None had no dedicated sheet, so expect lower scores across it.
What to do about it
Work in this order — it's the fastest route to a defensible number:
Sort your attention by score, not by dollar value. Scan for the red, bold percentages first.
Read the assumption. If the arithmetic is right and the inputs are right, the number is fine — move on.
If an input is wrong, fix the input. Edit the quantity on the scope table (Edit Scope), or better, measure it.
Measure it on the drawing. Click Edit Takeoff, select the item, and draw the area, length or count. The quantity updates from the drawing and stops being an assumption. See Using the Takeoff Tool.
Then generate the estimate. Every correction you make here is one line, instead of a dozen priced lines later.
The honest version
AI reads your plans and does the repetitive first pass — the part that used to take hours of tracing and typing. It gets you most of the way there, fast. It does not replace your judgment, and it isn't meant to.
The confidence score is how the AI tells you where its judgment was thinnest, so you can put yours there instead. Human plus AI is what makes the estimate right.
Troubleshooting & FAQs
Why not 100%? 90% is effectively the ceiling the AI will return. The remaining certainty comes from you verifying it — that's the design, not a limitation being papered over.
Everything is 60–70%. Usually a sign the upload is missing sheets, or that the drawings are schematic rather than dimensioned. Check Show sheet inventory to see what the AI actually read.
The score didn't change after I fixed the quantity. The score describes the AI's original evidence. Once you've drawn the measurement in the takeoff tool, the quantity comes from the drawing regardless of what the AI originally scored.
Can I turn the scores off? No — they're there to be used. Ignoring the high-confidence lines and concentrating on the low ones is the point.
Quick Reference
Where: the Source column on the project's Scope tab and on Edit Scope.
Red and bold = 60% or below. Review those first.
The description holds the assumption, including the arithmetic the AI used.
Best fix: measure the item in Edit Takeoff so the quantity comes from the drawing.