TL;DR
Coordination inherits the whole accuracy chain: registration at 2–4 mm RMS interior, 8–12 mm absolute to control, and an LOD 300 model within ±10–15 mm of the cloud. That stack supports a 0 mm hard clash and roughly 25 mm clearance clash — tighter tolerances just generate false positives. Audit the registration report, verify with 30 model-to-cloud checks at 90% pass, and set LOD by system rather than globally.
Coordination meetings rarely argue about the model. They argue about whether the model is right. When a hanger lands 40 mm off in the field, the question is never "did Navisworks compute the clash correctly" — it is "was the existing condition ever measured properly."
Every clash test rests on an accuracy chain, and most teams only ever see the last link.
The error stack, link by link
| Stage | Typical contribution | Who controls it |
|---|---|---|
| Instrument range noise | sub-millimetre to a few mm | scanner class and range |
| Registration between setups | 2–4 mm RMS interior, 4–6 mm exterior | capture crew and workflow |
| Tie to project control | 8–12 mm absolute | surveyor and control network |
| Modelling from cloud | ±10–15 mm at LOD 300 | modelling team |
| Coordination tolerance | what you set in Navisworks | BIM manager |
The number worth internalising: these do not cancel. A model built to ±10–15 mm of a cloud that is itself ±5mm registered and 8–12 mm from control cannot honestly support a 10 mm hard-clash tolerance against new design geometry. Set the tolerance below the accuracy of the inputs and the report fills with clashes that are not real, the team stops trusting it, and real conflicts get triaged away with the noise.
Clash tolerance is a statement about your measurement chain, not about how careful you want to be.
What accuracy to specify, by decision
Accuracy is not a single project-wide number. It is a per-decision number, the same way LOD is a per-system number — an argument made in full in LOD 200 vs 300 vs 400.
- Feasibility, massing, area takeoffs. LOD 200 at ±20–30 mm is sufficient. Spending on tighter capture here buys nothing.
- Design coordination and general clash. LOD 300 modelled within ±10–15 mm of the cloud. This is the default for renovation work and the level most coordination actually needs.
- Fabrication and tie-ins. LOD 400 for the 5–15% of the building where a fabricated part meets an existing one. A duct connection to an existing riser, a steel beam bolted to an existing column, a pipe tie-in at a live header.
Applying the top tier everywhere is the most common way to overspend. On a 180,000 sq ft Mansfield conversion, specifying LOD by system instead of "LOD 300 everything" cut modelling hours 28%.
Registration quality is the number to audit
Registered accuracy is the input everything downstream inherits, and it is trivially auditable if you ask for the report. What to look for:
Closed loops, not a chain. A registration chain accumulates error at the far end. Loops distribute it. The 250,000 sq ft distribution center covered in our case study used five closed loops and 40 control points to hold ±6 mm across a live facility.
Control density that matches the geometry. Long, narrow, repetitive spaces — corridors, tunnels, warehouse aisles — are where mobile capture drifts. Terrestrial setups at intervals and at every tie-in pin it back down. The 220,000 sq ft plant retrofit paired mobile capture with 38 RTC360 setups placed specifically at tie-in locations.
Campus-scale work on one frame. Twelve buildings across 480,000 sq ft and 18 acres held ±0.02 ft (±6 mm) campus-wide on 47 permanent control points. Buildings registered individually and then assembled will not coordinate against each other.
Residuals reported per setup, not as an average. An average hides the one bad setup that happens to sit under the mechanical room you are coordinating.
Setting a coordination tolerance you can defend
A workable scheme for renovation coordination, given a ±5mm registered cloud and an LOD 300 model:
- Hard clash: geometry interference at 0 mm — flag everything.
- Clearance clash: run at 25 mm for general MEP against existing structure. Below that you are testing your measurement chain, not the design.
- Fabrication zones: tighten only where the model is LOD 400 and the cloud has terrestrial coverage. State the zones by name in the BIM execution plan.
- Soft clash / access: service clearances per trade, unchanged by scan accuracy.
Write the tolerance and its justification into the BEP alongside the accuracy spec. Our scope of work template has language you can lift.
Verifying the model before coordination starts
Do not open a clash meeting with an unverified model. The acceptance test we recommend and run:
- Thirty random dimension checks distributed across floors, model against cloud, with ≥90% required in tolerance.
- A named 5,000–10,000 sq ft sample area reviewed model-to-cloud in full before the remaining area is modelled.
- 95% element completeness in scanned areas, with unscanned areas explicitly listed rather than silently modelled from assumption.
- A live walkthrough of the deliverable within 5 business days of handoff.
Skipping this is expensive in a specific, measurable way. On a 38,400 sq ft hospital wing captured in two overnight shifts at ±6 mm, a verified model removed $147,000 in change-order risk and 23 RFIs before construction — see the hospital wing case study.
Common failure modes in coordination-grade clouds
- Cloud decimated for performance, then modelled from. Model from the full-resolution data, coordinate on the decimated version.
- Mixed coordinate systems between disciplines. Shared coordinates set once, at the start, from the survey control. Everything else is rework.
- Ceilings scanned closed. If the plenum was never opened, MEP above ceiling is assumption. Say so in the deliverable rather than modelling ghosts.
- Model updated, cloud not re-linked. A stale RCP link makes a correct model look wrong.
What it costs to get this right
Scan-and-model for coordination-grade output runs $0.25–$0.60/sq ft all-in. Point cloud capture and registration alone sits in the $0.05–$0.20/sq ft band. Against 15–40 RFIs at $1,200–$2,500 each and change orders at 3–8% of contract value on renovations with poor existing-conditions data, the arithmetic is not close.
For how the deliverables travel between platforms, see point cloud file formats. For capture methodology, see Building 3D Laser Scanning.