Both sites, or it didn’t happen

The ones still being built

Why these schedules slip

The published revisions are part of the record: first-of-a-kind engineering carries no prior art to borrow from, and the timeline is always the first attempt.

A construction site office with printed programme charts on the wall
Programme wallFirst-of-a-kind hardware has no learning curve to borrow from. The revisions are published as they happen.

The problem with unprecedented machines

Every large scientific instrument listed in this guide was, at the moment of its design review, genuinely novel. Novel means there is no predecessor whose cost overruns and schedule variances can be fed into a risk model. The engineers building ITER at Cadarache cannot check how long it took a previous fusion reactor to weld nine sectors of a vacuum vessel together in a pit — because no previous reactor ever did. The teams assembling the SKA's core infrastructure across the Karoo and the Australian outback have no template for distributing a single aperture synthesis array across two continents. When there is no prior art, the baseline schedule is, structurally, an optimistic first guess dressed as a plan.

First-of-a-kind engineering has no learning curve to borrow from, and the published revisions are part of the record.

This is not a management failure unique to science. It is a known property of projects that sit at the edge of what has ever been manufactured. The aerospace and defence literature calls it the planning fallacy at scale: the same cognitive bias that makes a homeowner underestimate a kitchen renovation compounds when the tolerances are measured in microns and the components are custom-fabricated across a dozen countries.

Aerial view of the ITER construction site with buildings, cranes and surrounding forested hills
CadaracheThe site is laid out around its assembly buildings. The tokamak pit is the smallest part of the footprint.Photo: ITER site 2018 aerial view · Wikimedia Commons

Where the slippage actually lives

The delay rarely arrives as a single dramatic event. It accumulates in the interfaces — the gaps between one nation's contributed sub-system and another's. ITER is a collaboration of thirty-five nations contributing in-kind, meaning components are designed in one country, manufactured in another, tested in a third, and then shipped to Cadarache for final assembly. Each handover is a potential mismatch: a tolerance that was individually acceptable on two separate parts that is collectively unacceptable when they meet. Discovering that mismatch after shipment costs months, not days.

The same logic applies to software. The LHC's trigger system, which must decide in microseconds which collisions to keep, required years of commissioning that no simulation fully anticipated. The LHC trigger and data acquisition systems were revised repeatedly during run-up, because real beam conditions exposed edge cases the design had not modelled. Hardware can be redesigned on paper; redesigning it after it is installed inside a particle accelerator that took a decade to build is a different calculation entirely.

Chronology

  1. JWST original target launch year2007
  2. JWST actual launchDecember 2021
  3. NASA independent review board report identifying integration complexity as primary delay driver2018

JWST is the canonical modern example. Its original launch date was 2007. It launched in December 2021. The NASA independent review board's 2018 report identified integration and test complexity — the physical difficulty of assembling and verifying a system with hundreds of single-point failure modes — as the primary driver of delay, not any single component failure. The sunshield alone required a folding sequence of 178 release mechanisms that had to work in the right order in a vacuum environment no ground test could perfectly replicate.

A crane lowers a large dish-shaped telescope reflector onto its support structure in a desert setting
KarooA dish arrives in panels and is assembled where it stands, which is why the crane is a permanent line in the schedule.Photo: SKA-MPI-2 · Wikimedia Commons

Why revisions are information, not embarrassment

The project schedules that look worst in headlines are often the ones with the most candid published records. A collaboration that revises its baseline and publishes the revision is doing something useful: it is updating the community's understanding of what first-of-a-kind engineering actually costs. The Nuffield Foundation's research on megaproject cost estimation, and work by Bent Flyvbjerg at Oxford's Saïd Business School, both confirm that the projects with the least schedule slippage are those that benchmark against actual reference-class data from comparable prior projects — not the ones that set the tightest initial targets.

For instruments like ITER or the SKA, there is no reference class. The schedule is the reference class. Every revision narrows the uncertainty for the next attempt — whether that next attempt is a follow-on facility, a detector upgrade, or simply commissioning the machine that has just been finished. The record of slippage, published and attributed, is not a failure report. It is the first empirical data point the field has ever had.

Read next