Last Thursday, 17 September 2026, nine talks ran at the Berner Fachhochschule in Burgdorf, and if you read the programme as a sentence it says something the abstracts do not. One talk on a bridge that fell. One on propping a motorway viaduct while a station underpass is dug beneath it. One on keeping two 110-year-old railway bridges for another fifty years. One on a closed Italian viaduct reopened by sensors and a digital twin. Two on new bridges. The eighth Burgdorfer Brückenbautag, run by BFH’s Institut für Siedlungsentwicklung und Infrastruktur with the SIA Fachgruppe für Brückenbau und Hochbau under Prof. Dr. Dirk Proske, spent most of its day on structures that already exist. BFH’s own framing for the event is blunt: significantly more bridges are being maintained in Switzerland than newly built, and the stock «erreicht häufig seine planmässige Lebensdauer».

More than 4,300 bridges stand on my road cousin’s national network, about 6,000 on the SBB one. They are arriving at the end of their design lives more or less together, because they were built more or less together. Switzerland has decided to keep them and recalculate them rather than replace them — the sustainable choice, and the cheaper one, and the one that rests entirely on a single instrument.

An inspection is a question, not a look

An inspection is a question put to a structure with a particular instrument, and every instrument has a probability of detection for each damage mode. Near certainty for a spalled soffit. Near zero for a cracked wire inside a grouted duct. A clean report from a low-detection instrument is not evidence of a sound bridge. It is evidence that the bridge is not broken in the ways that instrument can see.

←TODAY: The Brückenbautag ran last Thursday in Burgdorf; the Aare railway bridges at Interlaken are under rehabilitation right now; the Carola forensic report is public and complete.
→3012: The asset that survives the century is not the strongest one — it is the one whose failure mode someone named early enough to buy the right microphone.
Fulcrum: Interval is a procurement variable. Detection probability is a physics one. Only the second decides whether the report means anything.

Dresden, told straight

At 02:58 on 11 September 2024, part of Zug C of the Carolabrücke in Dresden — built 1967–1971 in three parallel prestressed-concrete decks — dropped into the Elbe. No traffic was on it. Nobody was hurt. The bridge had been regularly inspected and technically monitored, which is the whole case and the reason Dresden is a lesson rather than a scandal.

Prof. Steffen Marx and the Institut für Massivbau at TU Dresden ran the forensic work — interim report December 2024, final report 27 May 2025 — and published what they found, which is exemplary conduct and should be named as such. According to Marx, the cause was hydrogen-induced Spannungsrisskorrosion in Hennigsdorfer Spannstahl, cracks initiated while the bridge was being built, in tendons that sat ungrouted in their ducts too long: «Diese Korrosion wurde bereits beim Bau initiiert und vollzog sich über viele Jahre im Inneren der Brücke unsichtbar von aussen.» — invisible from outside, for five decades. At the fracture, TU Dresden found over 68 % of the prestressing tendons severely damaged.

The only instrument that hears that mechanism developing is Schallemissionsmessung: microphones listening for the crack of a wire breaking inside concrete. As Michael Heeg reported for ingenieur.de on 11 September 2026, «Nicht kaputt genug, um aufzufallen» — not broken enough to be noticed. The variable that matters is not how often you inspect. It is whether the method can detect the most dangerous damage at all.

Burgdorf did not file this under foreign news: Gregor Schacht of MKP GmbH | Marx Krontal Partner gave the 10:20 talk, Ausblick nach Carola für unsere Spannbetonbrücken — for our prestressed-concrete bridges. ASTRA’s own condition report says the plain thing about age: many motorway sections date from the 1960s, 70s and 80s. Same decades, same structural generation. I will not print a count of Swiss bridges carrying SRK-prone steel, because no such count exists in front of me.

The Swiss answer is a number, not a verdict

ASTRA’s Netzzustandsbericht 2024, published 23 October 2025, reports 92 % of Kunstbauten good or acceptable across about 2,254.5 km, replacement value about CHF 146 billion, CHF 1.7 billion spent on maintenance in 2024. SBB’s Netzzustandsbericht 2025, published 17 April 2026, holds the grade at 2.8 of 5 — but reports a Nachholbedarf of CHF 9.5 billion, up 12 % in a single year, over half of it in the track rather than the bridges, with 186 km renewed against the 230 km needed to stop the backlog growing. Stabilisation is the stated goal for 2040.

A backlog outrunning renewal, and it is worth being precise about whose fault it is: nobody’s. The design lives arrive together because the boom built them together. What deserves the sharp end is interval thinking — the habit of writing inspection regimes as frequencies and then treating a completed schedule as an answer.

Proske and colleagues at BFH built the alternative: a risk assessment of roughly 13,000 bridges for SBB and ASTRA, collapse probabilities computed from operator data, each loss weighed against the whole system. SBB’s bridge ranking is risk-based, not condition-based — about 30 parameters, twelve adjusting failure probability to actual condition, fifteen describing what the loss would do to the network. Proske et al. note the reason cleanly: closing one bridge may barely register on the network, while the loss of another degrades travel nationwide. Their 2022 paper describes SBB as an international pioneer here, and on the evidence that is fair.

The legal floor under all of it is SIA 269 (2011), Grundlagen der Erhaltung von Tragwerken, and its vocabulary is the engineering: Aktualisierung — updating the structural model with what you actually find on site; Erfüllungsgrad — how far an existing structure meets today’s requirement, as a number rather than a pass/fail stamp; Verhältnismässigkeit — proportionality of intervention. That is what lets a recalculated 1968 girder stay in service on its measured capacity instead of its original drawings. Prof. Dr. Robert Wagner of BFH names the cost of doing it honestly: «Je älter die Brücke ist, desto schlechter ist in der Regel die Dokumentation.» Proske calls the work «ein gewisses detektivisches Gespür», and on the choice in front of the country: «Es besteht die Möglichkeit, die Brücken abzureissen und neu zu bauen oder sie länger zu nutzen, was natürlich nachhaltiger ist.»

Atelier: Three of Thursday’s talks are the same job in three registers. Keep it: Basler & Hofmann AG on the Instandsetzung Aarebrücken Interlaken — BLS’s two heritage-protected Aare railway bridges, about 110 years old, CHF 40 million, January 2026 to May 2027, bought another fifty years. Prop it: Emch+Berger AG Bern on the Unterfangung des Weyermannshausviadukts N12, underpinning a live motorway viaduct while the Personenunterführung Europaplatz Nord is built beneath it. Listen to it: CAEmate’s Hannes Grosse on the Parchi viaduct on Milan’s Tangenziale Est, closed for degraded Gerber saddles and reopened by Milano Serravalle under distributed fibre-optic sensing and a physics-informed digital twin — monitoring as the instrument that hears what the inspection cannot, applied before the fall rather than as autopsy after it. Every one is Bauen im Bestand, and the rule holds identically in a 1970s Zürich school: the existing structure’s true capacity is a measurement, never a drawing. Monday move: take the next existing-building job on your desk and add one line to the survey scope before it goes out — which failure mode can this method not see? — and require the surveyor to answer it in writing.

Hack: Put a number on why detection probability beats frequency, so the argument survives contact with a client who wants more visits instead of a better instrument. Bayes handles it: a clean inspection is evidence only in proportion to how likely that method was to catch the damage at all. The POD values below are illustrative — nobody’s measured figures — and yours should come from the literature for your damage mode and typology.

def after_clean(prior, pod, n=1):
    for _ in range(n):
        prior = prior * (1 - pod) / (prior * (1 - pod) + (1 - prior))
    return prior
print(after_clean(0.05, 0.10, 5))  # five visual passes -> 0.031
print(after_clean(0.05, 0.90, 1))  # one acoustic pass  -> 0.005

Five clean visual inspections at a 10 % detection probability move a 5 % belief in hidden damage to about 3 %. One inspection at 90 % moves it to half a percent. The frequency did almost nothing; the instrument did everything. That asymmetry is the entire post-Carola argument, and it fits on one slide.

What I would ask for, if anyone asked me

I carry loads across structures older than the norms that govern them, and what my generation ended up regretting was never the bridge that fell — falling bridges get inspected, funded and rebuilt. It was the damage mode nobody wrote down, sitting inside an asset everyone believed was surveyed. A report that cannot return a certain answer is not a reassurance; it is a gap wearing a signature.

So, two asks. Write the damage mode into the inspection brief: for every existing structure, name the failure mechanism invisible to a visual inspection at that typology and age, and specify the instrument that can see it — method first, interval second. And put the documentation hunt in the first deliverable of every Bauen-im-Bestand job, billed as design work, because that is what it is and that is where SIA 269’s Aktualisierung actually happens. Do not wait for a norm revision to make either mandatory. Write both into the next scope you send out.