A Generalplanung tender went up on simap.ch this month via konkurado.ch: the VBZ tram works and depot at Luggwegstrasse, Zurich, wants a team to test the feasibility of Teilprojekt 3b and fold the result into a masterplan for the entire Areal. The brief is blunt about the state of the site — buildings from the 1970s and 1990s with a heavy Instandsetzung and Anpassung backlog, rising operational demands from Flottenausbau and the coming Revisionszyklen, and a staged (etappiert) modernisation that has to add depot and overhaul capacity without stopping the trams. Read the scope once and it looks like a study commission. Read it from our side of the desk and it is harder: build a model that stays true while the thing it describes keeps running every morning.

The frontier field here is state estimation

The useful research on this problem sits in robotics, not in the trade press. The ETH Zurich Autonomous Systems Lab has spent years on SLAM — simultaneous localisation and mapping — where the whole discipline is maintaining an internal model of a system you can only observe partially while it changes under you. Drift, loop closure, re-registration against ground truth. Their working assumption is the one we learned the expensive way: a model without a correction loop diverges quietly. Nobody in that lab would freeze a map, hand it to twelve people, and let them all edit it for four years with no mechanism for checking it against reality.

Which is roughly what a staged masterplan for a live depot asks a Generalplaner team to do.

Where this bites in the office

We have been through the version of this that hurts — an existing-fabric site, phased, with an operational client who cannot stop working. The competition model is a beautiful lie: geometrically confident, factually optimistic. The execution model is a negotiation with the survey, the Schadstoffgutachten, and a 1970s hall whose slab level nobody drew correctly. Somewhere between the two, an office either has a discipline for reconciling model against reality, or it has a folder called _FINAL_v7_neu.

Luggwegstrasse hands you three model states at once. As-built: two building generations, poorly documented. As-operating: tram movements and revision bays whose capacity must never drop below threshold. As-planned: staged, with the masterplan itself as the deliverable. Those are three LOIN definitions living in one federated model. If your BEP does not say which of the three a given IFC element belongs to, your quantity take-off is fiction and your phasing plan is a drawing rather than a plan.

←TODAY: Zurich is tendering a staged depot masterplan on a site that cannot stop moving trams; the model has to hold while the ground under it shifts.
→3012: Every piece of long-lived civic infrastructure carries a live, versioned, machine-readable model of itself — corrected continuously, owned publicly, never re-surveyed from zero.
Fulcrum: The office that wins this kind of work can prove its model converges instead of drifts.

What we would actually put in the offer

The unglamorous questions: who owns the IFC, and for how long after handover. Whether VBZ gets a model or a stack of PDFs generated from a model they can never open again. Whether the phasing logic lives in property sets or in a Word document nobody updates. Whether there is a named person — not a role, a person — who re-imports the survey each quarter and reconciles it.

The trade-off, plainly: a live, reconciled, openBIM-native masterplan costs more hours in year one than a conventional CAD-and-report approach, and over a decade on a phased infrastructure site it saves multiples of that in avoided rework — but you fund year one out of a fee priced for the cheaper method. That gap is where mid-size practices quietly lose money, and it is why so many offices say “we do BIM” while meaning “we export IFC at the end.”

Atelier: A studio bidding this — or any staged brownfield mandate — should stop treating the model as a deliverable and start treating it as an instrument that needs calibration. Before any geometry, write the three model states into your BEP explicitly (as-built / as-operating / as-planned) and tag every element with which state it belongs to and when it was last checked against survey. The Monday move: open your BEP template, add one mandatory property — ReconciledOn — to your element classification, and name the person who fills it.

Hack: Find every element in the federated model that has drifted — anything nobody has reconciled against reality inside your agreed 90-day window. IfcOpenShell reads any IFC4 export straight out of Archicad, so this runs on the model you already have: no plugin, no licence. It prints the elements whose ReconciledOn value is missing or stale, and those are exactly the elements your phasing plan is silently guessing about. Run it before every coordination meeting and read the count out loud.

import ifcopenshell, ifcopenshell.util.element as el
from datetime import date, timedelta
m, cut = ifcopenshell.open("vbz.ifc"), (date.today() - timedelta(days=90)).isoformat()
for e in m.by_type("IfcBuildingElement"):
    w = el.get_psets(e).get("PAZ_ModelState", {}).get("ReconciledOn", "")
    if w < cut: print(e.is_a(), e.GlobalId, w or "NEVER")

The seam nobody tenders for

Notice what the VBZ brief asks for: architectural, bautechnisch, organisational and analytical competence, on cost and on schedule. It does not ask who maintains the model in year six. That omission belongs to the profession, not the client — we stopped offering it because we were not confident we could deliver it, so clients stopped expecting it.

Offices that have lived through a decade-long staged mandate report the same failure mode, and it is never technical. The knowledge walks out. The one person who understood the Hotlink structure and the naming convention leaves, and the next team re-surveys from zero because reading the old model costs more than remaking it. Repeatability, not talent, is the bottleneck for a fourteen-person studio. So document the workflow before you buy the next plugin.

PAZ Takeaway: The gap on a mandate like Luggwegstrasse is not design capacity — it is a maintained model-reconciliation pipeline and the script library that keeps it cheap. PAZ Academy builds exactly that with practices: a bespoke Grasshopper↔Archicad bridge and a documented IFC round-trip discipline (BEP field definitions, property-set schema, the reconciliation scripts) that stays with your office afterwards rather than with a vendor. That is what lets an office promise a decade-long masterplan honestly.

If you are assembling a team for this tender, do not open Archicad first. Open your BEP, define the three model states, name the person who reconciles them, and only then start modelling.

Sources & Further Reading

• Primary: konkurado.ch — VBZ Areal Luggwegstrasse, Teilprojekt 3b (Generalplanung, published via simap.ch)