Live demo
Two errors and four warnings, on purpose
MBE3Dstudio opens on a roof-mounted autonomy sensor pod that fails two of its own checks, because a tool that opens on an all-green model teaches you nothing about what it is for.
Needs a modern browser with WebGL. The optional agent calls a server.
The two findings
Each one names a number and a requirement
Nobody wrote these into the demo as text. They are what the analysis pass returns when it runs on the model, and they change the moment you move a part.
-
Lidar field of view violated91.11 % / 97 %
The spinning lidar sees 91.11 % of its declared field against a 97 % minimum. The blocker is named: the cellular / V2X antenna mast, accounting for 8.89 %.
Two parts that are individually fine, mounted where they shadow each other. SYS-014
-
VER-003 failed against ICD-01110.1 % / 5 %
The recorded harness drop verification is marked failed: the compute feed measured 10.1 % at 22 A on the r4 routing, against the 5 % the interface control document allows.
Revision r5 rerouted that feed and the live constraint now passes at 1.23 %. The verification record still describes the run that was replaced, drift the model can show you because the record and the geometry are the same artefact. ICD-011
A third stood here until r6: the pod massed 16.25 kg against the 14 kg the vehicle programme allocated. Pocketing the baseplate between its bracket lands, thinning the shell wall from 2.0 to 1.6 mm and taking the magnesium-chassis compute variant closed it at 13.8 kg, and the same analysis pass that reported it now reports it passing, by 0.2 kg.
Alongside these sit four warnings: a thermally adjacent compute/PDU pair at 265 W across a 20 mm gap, the unallocated function “Maintain the V2X link”, and SYS-040 twice over: no verification activity references it, and no component carries it.
Trace
Coverage you did not have to maintain
100 % of counted requirements are satisfied by a function, 88 % are allocated to a component, 88 % have a verification activity, and only 13 % have actually passed one. That last number is the useful one, and it is the one a hand-maintained matrix is worst at keeping honest.
The matrix is derived from the graph on every render. There is no stored trace document, so there is nothing that can drift.
Analysis
Every rolled-up number, in one place
Mass properties with the full inertia tensor. Power for each of the three modes, with the largest consumers ranked. All eleven harness runs with gauge, length, current, drop and dissipated watts. Sensor coverage with the principal occluder named for each.
Change a mass in the inspector and every one of these updates before you have finished typing. Then undo, and they come back.
BOM
The bill of materials is the same model
Part numbers, suppliers, quantities, unit and extended cost, rolled up from the same assembly tree that produced the mass number. Export it as CSV and it agrees with the review document, because both came out of one pass.
Things worth trying
Five minutes in the studio
- Click a red finding. It selects the parts it names and frames them in the viewport.
- Drag the antenna mast off the lidar. Watch the coverage percentage climb past 97 % and the finding disappear as you move.
- Switch mode to standby. The power roll-up drops from 319.6 W to 43.42 W: the same model, a different question.
- Turn on X-ray and ports. The pod shell goes translucent and every attachment point becomes visible.
- Export the report. You get the Markdown review pack (verdict, constraint table, trace matrix and all) for the state you left the model in.
- Press Ctrl‑Z. Everything you did comes back, including the drags.