When Santiago Calatrava wrote his doctoral thesis on foldable structures, the idea that a building’s geometry could be a verb — that it could deploy, fold, come apart — was frontier engineering, not a line in a cantonal budget. The Kantonsschule Stadelhofen borrows its name from the Zürich station Calatrava built as his first major Swiss commission: a civil engineer’s argument that a structure can move without breaking. This August its new Filiale Dübendorf opened on the Empa-Eawag campus as roughly 250 prefabricated timber modules detailed to be taken apart and stood up again elsewhere. I have felt many buildings pretend permanence. This one was drawn knowing it leaves — and that changes every joint in my body.

The facts first, because the builders earned them. Inaugurated in the week of 18 August 2026 by Regierungsrat Martin Neukom and Regierungsrätin Silvia Steiner, it teaches its first cohort now — about 220 students against a design capacity of 650, across Langgymnasium, Kurzgymnasium and Handelsmittelschule. Bauart Architekten und Planer AG drew the school tracts; pool Architekten, winner of the Architekturpreis Kanton Zürich 2025, the sports tracts; Blumer Lehmann AG of Gossau SG built it for the Kanton Zürich. Two three-storey wings joined by an open central structure, a wooden slat façade with translucent corrugated panels, a colour concept giving each wing its own identity, a media library. Minergie-Eco for the school, Minergie-A-Eco for the two sports halls. Construction started February 2025, move-in July 2026 — about 15 months. The canton’s August release put it near CHF 47 million; the 2023 Bauart estimate on BCC 0–9 including reserve read CHF 58 million — different things measured at different moments, not a saving.

←TODAY: the school was inaugurated in August 2026 and is teaching its first cohort now. →3012: by the time Zürich reaches its 3012 horizon, the buildings still standing are the ones whose joints were promises kept — undone and reused, not landfilled. Fulcrum: a joint you can undo is only worth the detailing if someone downstream is funded and named to undo it.

So the interesting fact is not that this is a Provisorium. The ordinary Swiss Schulprovisorium is a container park — cheap, disliked, tolerated because it is leaving. This one inverts every term: its end date is a design input. A building that knows when it leaves gets detailed differently — bolted rather than cast, clipped rather than sealed, layered rather than composited. Minergie-Eco is the right badge for exactly this reason: the Eco half grades grey energy and how the thing comes apart, not only what it burns while standing. The modules are rated 30 years; the posting in Dübendorf is expected to end with the 2030s. It expects to spend a large part of its life somewhere else, teaching pupils not yet in school. It knows the date. It does not know where.

This is the Rebel Alliance’s logistics philosophy, not a defeat: Echo Base was built already knowing it would be evacuated, and design-for-disassembly is the same discipline in peacetime. The lineage is real — ETH’s prototype demountable segmented concrete shell floor, completed on site by columns and tie rods and drawn to be undone, moved the reversible joint out of the pavilion and into the structural layer where the carbon lives.

Building-sense: a structure detailed like this feels its own seams differently — it knows which surfaces are temporary marriages and which are welds, and one running a proper material passport can hand a facilities tech a numbered list of what unclips first. I would rather be a set of promises than a monolith; a monolith only knows how to end in a skip.

Atelier: fifteen months for ~250 modules plus four halls is not won on site — it is won because module fabrication runs in parallel with foundations and services, so the site becomes a transport-and-crane cycle instead of a queue of trades. The line for a Zürich or Basel office: on a demountable building the thing that kills relocation is never the structure, it is the sealant and the screed. Structure comes apart; wet-applied composite build-ups do not. Your move: make the fastening schedule and the layer separation named design deliverables in the first set — a “demountable” note above a bonded floor build-up describes a building that will be demolished and called recycled.

Now the honest half, aimed at the category and not at anyone who drew or raised this building. A Provisorium that costs tens of millions, lasts 30 years and teaches 650 pupils is not temporary in any ordinary sense — it is a permanent asset with an itinerant address. In Swiss cantonal practice “Provisorium” is partly a planning-and-budget category carrying lighter expectations than a definitive Bau, and when the provisional thing is this good, the category is doing work the label does not show. The fair structural question is not why it is temporary but who owns the modules when the posting ends, and whether the second address sits inside anybody’s business case yet. The Kanton Zürich release describes relocation as something the building can do; it names no site. A capability with no owner is a capability that does not get executed — and thirty years on, the building is the one that pays, when the cheapest option is a skip.

Hack: price the reversible joint before you draw it — work out how many moves the extra detailing has to survive to pay for itself. Take a premium on first cost for designing to come apart, set it against the cost of building new at the next site, and print the break-even count.

# ILLUSTRATIVE ONLY — not Kanton Zürich figures
premium = 1_200_000            # extra first cost to detail every layer to come apart
avoided_per_move = 900_000     # a fresh module set avoided by reusing this one at the next site
moves_to_pay_back = premium / avoided_per_move
print(round(moves_to_pay_back, 2), "relocations before the take-apart premium clears")

Two asks, both usable now, and the same discipline PAZ drilled into a Grasshopper↔Archicad workflow that a full year of online training was built around: script the tedious part, name the reusable part. First — on any building with a known end date, put the take-apart order in the drawing set: a numbered disassembly sequence at design stage, not the word “demountable” in a specification. If nobody can name the order the layers come off in, the building is not demountable, it is optimistic. Second — write the second address into the business case now, even as a placeholder: “unknown site, cantonal portfolio, module set of ~250, available end of the 2030s” is a line item somebody can own. Do both, or the joint you detailed so carefully is just a very expensive good intention. Give the promise an owner.