I deleted most of my own digital twin and it still catches the fault. On the Lattice view, and what a monitoring twin actually owes you.
Willy Tai · Smart Cities · AI · Sustainability · August 2026
The Estate Twin on this site has a second view. Press Lattice and the building disappears. What remains is a stepped wireframe with the same outline as the estate, twelve rectangular bands for the floors, a handful of slowly turning shapes for the plant, and particles moving along thin curves where air and chilled water flow. It looks like an instrument, which is what it is. Everything else about the twin, the data, the rules, the fault script, the ledger, is unchanged. The toggle removes polygons and nothing else.
The digital twin conversation usually opens with fidelity. Model every duct, scan every soffit, integrate everything, and the value will arrive at the end. Most building owners never reach the end, because fidelity is bought by the polygon and paid for before any of it earns a cent. The counter-position has been in print since 2018: the Gemini Principles, from the Centre for Digital Built Britain, judge a twin by purpose, trust and function, and ask for only as much fidelity as the decision needs. I have been arguing a blunter version of the same thing for years. A monitoring twin owes you three answers: what is broken, what state everything is in, and roughly where it sits. Fidelity beyond those three answers is cost that never pays back.
The Lattice view is that argument executed in code. Floors become bands, coloured by the worst reading on them. Kind survives as shape: air handlers are octahedra that spin with their airflow, chillers are rings that turn with their load, sensors are small spheres sitting at their rough position on each band. Location survives twice, once in where each node sits, and once in the outline itself, because the wireframe keeps the estate's own massing, a wide podium with a narrower tower, so your eye never loses its bearings. The chiller yard sits on the ground beside the block exactly where it stands in the full view.
Then run the same experiment as Part 1. Inject the bearing fault while the Lattice is up. One octahedron slows and turns red. The three bands it serves turn with it. The flow curves to the chiller go amber and their particles speed up, and the chiller ring brightens as it strains. All three questions are answered in a single glance, and the work order still arrives, still shows the model's confidence, and still waits for a named person. Nothing about the monitoring job needed the building.
I want to be careful about what this does and does not prove. It does not prove that geometry is worthless. The estate view exists because people navigate by recognition, and there are decisions, refurbishment, space planning, fire strategy, that genuinely need more model. What it proves is that the monitoring loop, the part that pages a technician at 2am, runs on relationships and telemetry. In the code this is literal. The lattice never reads the modelled polygons; the only geometry it keeps is the massing outline in the declaration both views are built from, shown here trimmed of its layout coordinates:
const MASSING = {
podium: { from: 1, to: 3, w: 25, d: 15 },
tower: { from: 4, to: 12, w: 17, d: 13 },
ahus: [
{ id: "AHU-R1", serves: [10, 11, 12] },
{ id: "AHU-R2", serves: [4, 5, 6, 7, 8, 9] },
], /* plus layout coordinates */
};
Ten lines of data describe the estate well enough to monitor it, and that says more about the job than about my demo.
At a SimBuild session this year on the PAE Living Building in Portland, one of the most advanced living buildings in the United States, the team described extracting a minimum viable model for operations that was a small fraction of the size of the design model, and found that what operations actually needed was the relationships between assets, systems, spaces and signals rather than the geometry that had cost the most to produce. When the people who own the most detailed models start measuring how little of them operations uses, the fidelity-first default deserves harder questions than it usually gets.
Which raises the question I will take up properly later in this series: if ten lines of data and a browser can carry the monitoring loop, what exactly is being bought when a twin is procured, and what should an owner write into the tender to make sure the expensive parts are the parts that pay? There is a photograph involved in the next part before we get there.