Docs · Accuracy
Reading the evidence graph
The evidence graph is the backbone of every audit-ready bid. Each takeoff row points back to a sheet and a spec paragraph; each spec paragraph knows what takeoff lines cite it. Reverse-link queries answer 'who's relying on this section?' in one hop.
Anatomy
Four kinds of nodes: plansheets (DRAWING documents), spec sections (SPECIFICATION documents), manual pages (manufacturer datasheets uploaded for complicated equipment), and takeoff items. Edges are typed and directional — a takeoff item cites a spec section, and we record both sides of that link so reverse queries answer in one hop.
The forward query — what backs this takeoff line?
- Open the takeoff edit table; click any row's evidence indicator
- Sidebar opens to the spec paragraph + sheet snippet that justify the row
- Click the source-page link to jump to the exact bbox on the source PDF
- Multiple supporting evidence rows show as a stacked list
- Confidence per evidence (not just per row) so you see which support is strong
The reverse query — what cites this spec paragraph?
- Open any spec paragraph in the spec viewer
- Sidebar shows every takeoff line citing it, with their costs
- Useful when an addendum changes a paragraph — you see who's affected before re-runs
- Useful for clarifying RFIs — you see what depends on the answer
- Reverse queries answer in O(log n) via a composite index on (target_kind, target_id)
Related
Read these next.
Accuracy
Review queue
Confidence gates, bulk-confirm thresholds, edit-then-confirm workflow, audit trail.
Accuracy
Symbol library + active learning
Promote detections to ground truth, train the recognizer per-org, manage trade-specific definitions.
Workflow
Plan-to-bid playbook
End-to-end walkthrough of how a 1,000-sheet plan set becomes a bid packet — every step on the record.