TL;DR
Vet accuracy before signing: demand four numbers in millimetres (±5mm registered, 2–4 mm RMS interior, 4–6 mm exterior, 8–12 mm to control), ask whether registration used closed loops and how many control points, request a real registration report with per-setup residuals, run a 5,000–10,000 sq ft sample area first, and put a 30-check, 90%-pass acceptance test in the contract.
Accuracy is the one thing in a reality capture proposal that cannot be assessed by looking at the deliverable photos. Every point cloud renders beautifully. The differences that matter are in numbers most proposals never state.
What follows is a vetting process you can run in a single call and a sample review, before committing.
Step 1: Get four numbers in writing
A defensible accuracy commitment has four parts, all in millimetres:
| What | Reasonable target |
|---|---|
| Registered cloud accuracy | ±5mm |
| Interior registration residuals | 2–4 mm RMS |
| Exterior registration residuals | 4–6 mm |
| Absolute accuracy to project or state plane control | 8–12 mm |
If a modelled deliverable is in scope, add one more: model within ±10–15 mm of the cloud at LOD 300.
Adjectives — "survey grade", "high precision", "millimetre accurate" — are not commitments. A provider who will not put numbers in the proposal will not be held to them later.
Step 2: Ask how registration was done, not just how accurate it is
Three questions, and what the answers reveal:
"Closed loops or a registration chain?" Chains accumulate error at the far end. Loops distribute it. A 250,000 sq ft distribution center held ±6 mm on five closed loops and 40 control points.
"How many control points, and are they permanent?" Density matters most in long repetitive spaces. A twelve-building, 480,000 sq ft campus across 18 acres used 47 permanent control points to hold ±0.02 ft (±6 mm) on a single frame.
"Mobile only, or mobile plus terrestrial at tie-ins?" Mobile capture is fast and drifts; terrestrial setups pin it. On a 220,000 sq ft plant, 38 RTC360 setups were placed specifically at tie-in locations inside a 5-day field program.
Step 3: Request a registration report from a past project
Not a rendering — the report. What to look for:
- Residuals listed per setup, not as a project average. An average hides the one bad setup, which is statistically likely to be the crowded mechanical room you care most about.
- Overlap percentages between setups.
- The named coordinate system and control source.
- Date of capture and the instrument used.
A provider who has these reports on hand produces them in a day. A provider who does not, does not generate them.
Step 4: Run a sample-area test before the full award
The cheapest insurance in this business: scope a named 5,000–10,000 sq ft sample area, have it captured and modelled first, and review it model-to-cloud before the rest proceeds. Problems that would have propagated across 200,000 sq ft get caught on 5,000.
Step 5: Put an acceptance test in the contract
The test protects both sides, and it needs to be written before mobilisation, not negotiated after delivery:
- 30 random dimension checks, model against cloud, distributed across floors, ≥90% in tolerance
- 95% element completeness within scanned areas
- Unscanned areas enumerated in writing rather than modelled from assumption
- A live walkthrough of the deliverable within 5 business days of handoff
- A stated remedy if the test fails
Copy-ready language is in our scope of work template and the accuracy structure in how to write a scan deliverable spec.
Step 6: Make the quotes comparable before comparing them
Capture and registration alone runs $0.05–$0.20/sq ft. Scan-and-model runs $0.25–$0.60/sq ft. A two-times price spread between proposals almost always means one of these differs:
- LOD specified globally versus by system
- MEP included above ceiling or excluded
- Native cloud delivered or withheld
- Acceptance test included or absent
- Registration to control or internal only
Normalise those five and the quotes usually converge.
Red flags
- Accuracy stated only as a scanner spec sheet number. Instrument accuracy is not project accuracy — registration and control dominate.
- No registration report available from any past project.
- Resistance to an acceptance test.
- No willingness to name field days or delivery in business days.
- A model quoted without the underlying cloud as a deliverable. Without the cloud, nothing can be verified later.
What good looks like in practice
Delivered work with numbers attached: a 38,400 sq ft hospital wing captured in two overnight shifts at ±6 mm with a federated Revit model in 8 business days, removing $147,000 in change-order risk and 23 RFIs. A 96,400 sq ft warehouse conversion captured in 2.5 field days, LOD 300 model in 18 business days, $85,000 of contingency released. Those are the kinds of specifics that should be available on request from anyone bidding your project.
Vetting a provider for an Ohio or nationwide project? Call 614-210-3679 or start with Building 3D Laser Scanning.