For three years the digital twin has been a demo. Beautiful renderings, a live data overlay, a conference-stage "look, the building breathes." What it rarely was: a control loop that actually decides something and is held to a measured number afterward. That line — from visualization to functional twin that closes the loop — just got crossed at three different altitudes in a single week. The interesting part is that the proof points are not one vendor's slide. They are three independent operators, three building types, three M&V regimes.
Three altitudes, three proof points
| Altitude | Operator / system | Reported result | What "functional" means here |
|---|---|---|---|
| Data center (mission-critical cooling) | Phaidra RL control | ~80% reduction in thermal-spike "Max-Q" excursions | Closed-loop setpoint control, not dashboards — the model acts on the cooling plant |
| Corporate HQ (mixed-use office) | Delta, Taipei flagship (Omniverse twin) | ~20% building energy savings | Twin drives HVAC + lighting scheduling against occupancy reality |
| Commercial portfolio (distributed) | PassiveLogic / Trane footprint | ~14,000 buildings under autonomous-control deployment | Functional control replicated across a fleet, not a flagship one-off |
Stack those and a pattern appears that no single case makes alone: the functional twin works at the tightest control loop (a data hall where a thermal excursion is a six-figure event), at the messiest control loop (a human-occupied office where comfort and schedule fight), and at fleet scale (where the hard problem is replication, not the pilot). Demo-to-deployment isn't crossing once. It's crossing at every altitude at the same time.
The M&V caveat that separates signal from marketing
A practitioner reads those three numbers with one reflex: measured how? The honest answer is that they sit on different rigor tiers, and lumping them together is how vendors mislead.
- Phaidra's ~80% is an excursion-count reduction — an operational telemetry metric, clean to verify because the events are discrete and logged.
- Delta's ~20% is a whole-building energy claim and should be read as an IPMVP Option C (whole-facility) result: defensible only with a calibrated baseline and weather normalization. Treat it as benchmark-grade, not contract-guaranteed, until the baseline model is published.
- The 14,000-building figure is a deployment count, not a savings claim — a scale signal, not a performance number. Don't let it borrow credibility from the other two.
That distinction is the whole game. The functional twin is real; the discipline of which number means what is what separates an operator who captures the savings from one who buys a story.
Why this is a composer problem, not an OEM problem
Here is the trap. Each of those three results comes from a different stack — an RL controls specialist, a hardware OEM's twin, a controls platform. An owner with a mixed portfolio cannot standardize on one of them without orphaning the others. The instinct to "pick the winner and build our own twin on top" is the OEM mistake: you become a productizer of one substrate and lose the rest of your buildings.
AISB's position is the opposite — the composer pattern. The durable asset isn't the simulator or the controller; it's the per-client calibration layer (our κ=20 hierarchical-Bayesian calibration matrix) that attaches to whichever substrate a building already runs and normalizes the M&V across them. The Fusion Model thesis is exactly this: don't reinvent the thermal model that Phaidra, Delta, or PassiveLogic already proved — compose above them and own the measurement-and-verification discipline that makes a fleet comparable. We argue the full case in the composer-vs-productizer split, and the architecture-evidence read is in the DSX roster analysis.
What to do this quarter
If you operate buildings, the functional-twin crossing changes one thing in your 2026 plan: the question is no longer "will autonomous control work?" — three altitudes just answered yes. The question is "can I measure it the same way across a heterogeneous portfolio?" Start by classifying each building's existing controls substrate and tagging the M&V option you'd actually defend (A/B/C/D). That inventory is the prerequisite to composing, and most owners don't have it yet.
Want the inventory done for your portfolio? Ask the agent — it's free — to map your control substrates to IPMVP options and flag which buildings are ready to compose. For the evidence base behind these claims, see our proof page.
Methodology note: third-party results (Phaidra, Delta, PassiveLogic/Trane) are reported from public sources and reflect each operator's own measurement basis; AISB has not independently audited them. Savings percentages are not transferable guarantees — actual results depend on baseline, climate, and building type, and should be verified under the appropriate IPMVP option.