ASML books 'documentation recovery' as engineering — and so should your 14-person studio
ASML funds a decade-long program to keep 1990s scanners alive through documentation recovery — the same knowledge debt a 14-person practice carries undocumented.
We read annual reports the way other offices read competition briefs: looking for the line nobody meant to make public. In ASML’s 2025 report — €32.7 billion in net sales, €4.7 billion poured into R&D, a supplier base of 5,100 firms — the line that stopped us cold is not about EUV. It is a program called PAS-LTE, and its named enemies are obsolete electronics, suppliers dropping end-of-life support, and “loss of domain knowledge as experienced engineers retire.” The remedy skips the new machine entirely: in the report’s own words, “documentation recovery, reverse engineering and extensive testing.”
Sit with that. The most advanced manufacturing company on the planet keeps its early-1990s PAS 5500 scanners alive to at least 2035 — two-thirds of the components in a modern smartphone are still made on them — by paying engineers to re-derive knowledge that walked out the door with the people who built the thing. They priced the fix. That is the whole story, and it is our story too.
←TODAY: A 30-year-old lithography scanner outlives its own documentation; ASML runs a decade-long, funded program just to remember how it works. →3012: The studios still standing are the ones that treated their own workflow as an asset with a maintenance line, not a habit in one person’s head. Fulcrum: What kills a machine — or a practice — is rarely physics; far more often the person who knew retired and the supplier stopped answering.
Here is the system behind the symmetry. CEO Christophe Fouquet diagnoses it plainly in the same report: rapid growth made ASML’s “ways of working… less agile,” and engineers “expressed their desire to focus their time on engineering, without being hampered by slow process flows.” In January 2026 that became an intent to cut a net 1,700 positions while creating engineering jobs. Process debt is not a small-office disease. It scales linearly with success, and it is invisible until someone leaves.
We know the office version by heart. The Archicad template nobody has documented since the senior desk built it. The Grasshopper definition only one person can repair. The competition model that cannot be turned back into an execution model once the author’s Praktikum ends. PAZ has covered the downstream cost of this before — our piece on Modum Atelier’s Chengdu Shude library leaned on a buildingSMART finding that 60% of renovation projects worldwide lack reliable digital building data. That is the same debt, arriving five years late as a Umbau with no as-builts.
Atelier: The uncomfortable truth is that ASML’s frontier work has a Swiss anchor built on exactly this kind of patient, documented method — the joint ASML–Paul Scherrer Institut resist-screening program on the XIL-II beamline at the Swiss Light Source in Villigen, which reaches 10 nm half-pitch patterns precisely because every parameter is written down and repeatable. Your Monday move: pick the ONE Grasshopper definition your whole Wettbewerb pipeline depends on, and record a three-minute screen capture of its author running it end to end — before that author is on holiday.
Hack: Find the files only one person has ever touched — your practice’s real bus-factor. Point this at your definitions folder and read the top of the list as your risk register:
for f in $(git ls-files '*.gh' '*.ghx' '*.tpl'); do
n=$(git log --format='%an' -- "$f" | sort -u | wc -l)
echo "$n $f"
done | sort -n | head
Every file printing 1 is a component whose domain knowledge lives in exactly one head. That number, not talent, is what a small studio should be tracking.
The trade-off is real and we will not hide it: documenting a workflow costs billable hours now, and the return only shows up on the day someone leaves — or a five-year-old project comes back. ASML can amortise that across €4.7 billion. We amortise it across the next competition we would otherwise re-derive from scratch.
PAZ Takeaway: The capability that answers this exact debt is a documented, maintained script-and-detail library the office owns — the PAZ Grasshopper↔Archicad Library is built to be that seam: a versioned, personalised bridge that turns a one-person definition into a repeatable asset the whole studio can run and repair, so the knowledge stops leaving with the last person who understood it.
Run the bus-factor query on your own repository, and turn every 1 into a documented, tested move before the next Praktikant hands back the keys.
PAZ Kaffi · multidisciplinary editorial, led by PAZ Academy