A report on why Starliner failed points at the same word twice — overconfidence

Space 4 min 80 sources

NASA's watchdog says the Boeing Starliner crisis that stranded two astronauts came from trusting a design too much and testing it too little — while the week's other space news kept moving.

Key takeaways

NASA's internal watchdog says the Starliner failure that stranded two astronauts for nine months came from over-trusting a design built of proven parts and skipping the test of the whole vehicle together.
Astronauts spent over seven hours replacing a failed wrist joint on the station's Canadarm2 robotic arm, keeping a critical piece of hardware alive to the station's 2030 retirement.
NASA spread nearly $600 million across private Moon landers and two heavy commercial satellites launched — the working machinery of spaceflight kept moving while one capsule got its post-mortem.

The week’s clearest lesson came from a spacecraft that already came home

On June 30, NASA’s own watchdog office — the Office of Inspector General, which audits the agency from the inside — released its report on what went wrong with Boeing’s Starliner capsule [20]. The verdict is blunt: NASA and Boeing were overconfident in a design built from parts that had flown before, so they let Boeing skip testing the whole vehicle as one system [20].

The consequence is already famous. In June 2024, Starliner carried astronauts Butch Wilmore and Suni Williams to the International Space Station on a short test flight [20]. Its thrusters malfunctioned and its service module — the part that steers and powers the capsule — overheated and leaked helium [20]. NASA judged it too risky to bring the crew home in the capsule they arrived in. Wilmore and Williams stayed on the station until March 2025 and came back on a SpaceX Crew Dragon instead [20]. A test flight meant to last about a week turned into a nine-month stay.

The report names three causes: overconfidence in Boeing’s heritage hardware, an unrealistic schedule the program treated as if launch were always six months away — from May 2021 until the actual flight in June 2024 — and limited oversight, with NASA lacking access to simulator data that included loss-of-crew scenarios [20]. NASA did not formally classify the flight as a serious mishap until February 2025, 21 months after it flew [20]. By April 2025 the commercial crew program had lost 21% of its people [20]. The report expects Starliner’s certification to carry human crews to slip to 2027 [20].

The rest of the week kept moving

While one capsule’s post-mortem landed, the working parts of spaceflight carried on.

A robot arm got emergency surgery in orbit. On the space station, joint number five — the wrist — of the 58-foot Canadarm2 robotic arm failed on May 27, drawing too much current and moving wrong [26]. The arm captures arriving cargo ships and moves gear during repairs, so losing it is serious. On June 30, astronauts Jessica Meir and Chris Williams spent 7 hours and 20 minutes outside — the 280th station spacewalk — detaching the arm’s hand and two healthy joints, pulling the dead 200-pound joint, and bolting in a spare from external storage [26][33]. Flight controllers confirmed power and data flowed through the new joint [26]. Canada Day timing was a coincidence; the arm is Canadian-built [23].

NASA spread nearly $600 million across small Moon landers. The agency awarded new task orders under its program that pays private companies to fly science to the lunar surface [7]. It also previewed more Moon-base science awards and signalled fresh opportunities [16], and floated sending a spare nuclear-powered rover — a twin of the ones on Mars — to the Moon under a proposal called PROMISE [8]. The bet is the same one behind cheaper launch: buy rides and hardware from a competitive market instead of building each one from scratch.

Two heavy commercial satellites went up. SpaceX launched a 15,000-pound SiriusXM radio satellite to refresh that constellation [25][13], and United Launch Alliance flew its final Atlas 5 rocket — a workhorse retiring after decades — carrying satellites for Amazon’s Leo broadband network [21]. Rocket Lab, meanwhile, aborted a launch of a Japanese radar satellite at the last second [29], a reminder that a scrubbed countdown is the system working, not failing.

Out past Earth, quieter work

Japan’s Hayabusa2 probe — the craft that returned samples from asteroid Ryugu — is set for a super-close flyby of a small asteroid on July 5, one of the closest encounters attempted by a mission its size [1]. One scientist called it a chance to “discover another beast to put in the zoo of asteroids” [1]. Separately, a new analysis of Ryugu’s gravity field mapped uneven material just under its surface — telling us how these loose rubble-pile asteroids are actually put together [2].

And astronomers reported a potentially habitable planet, GJ 3378b, orbiting a faint red dwarf star just 25 light-years away in the constellation Camelopardalis [40] — close, in cosmic terms, and the kind of nearby target future telescopes will want to study.

02 · Lesson · why it matters

Every part had flown before. That was exactly the problem.

When each piece has a track record, the whole can go untested — because "proven" quietly becomes "no one needs to check the seams."

A short flight that lasted nine months

Two astronauts flew to the space station for a week and could not come home in the ship they arrived in. Its thrusters faltered. The part that steers and powers the capsule overheated and leaked. So they stayed nine months and rode home on someone else’s spacecraft.

The briefing gives you the facts. The report behind them gives you a sentence worth carrying: they were overconfident in a design built from parts that had flown before, so they skipped testing the whole vehicle as one system.

Read that twice. The problem was not a part nobody trusted. The problem was parts everybody trusted.

Confidence pools at the joints

Picture how trust actually gets built. This thruster has flown. That valve is heritage. This computer is a version of one that worked. Each piece arrives with a record, and each record is real.

But a spacecraft is not a pile of parts. It is parts wired to each other, sharing heat, sharing power, sharing a schedule, firing in sequences none of them ran alone. The reliability lives in the connections as much as the components — and the connections have no track record, because they are new every time you assemble them differently.

Here is the quiet move. When every part is trusted, the confidence adds up into a feeling that the whole is trusted too. It isn’t. Component reliability and system reliability are different things. But the second one costs money and time to prove, and the first one is already sitting there for free, so the mind reaches for the free one.

The seam belongs to no one

There is a second reason the whole goes unchecked, and it is about people, not physics.

Every part has an owner. A team built the thruster; they know it cold. A team owns the flight computer. Ask any of them “does your piece work?” and you get a confident, honest yes. But the seam between two parts — where one team’s assumptions meet another’s — sits in the gap between two org charts. It is nobody’s job in particular.

Integrated testing is the expensive, awkward act of making the seam somebody’s job. You run the whole thing together and watch what the pieces do to each other. Skip it, and you haven’t removed the risk. You’ve just moved it to the one place no one was looking, and agreed not to look there.

Why the schedule made it worse

The report names a second cause alongside overconfidence: a schedule that acted, for three years, as if launch were always six months away.

These two are not separate failures. They feed each other. A tight schedule makes the expensive, seam-checking test the obvious thing to cut, because it is slow and it might find something — and finding something means more delay. Trust in the parts gives you the permission to cut it. “They’re proven. We’ll be fine.” The clock supplies the motive; the confidence supplies the excuse. Together they point straight at the one test that would have caught the problem.

You are already inside this

It is easy to file this under “aerospace” and move on. Don’t. The shape is everywhere you can’t see it, including in the systems you depend on.

The bank whose payment software is fine, and whose fraud-check software is fine, and which loses your money in the handoff between them. The hospital where every specialist is excellent and the patient falls through the space between two departments. The team where each person did their part and the project still failed, because the part nobody owned was how the parts fit together. You have been the person who said “my piece works” while the seam quietly rotted next to it. Everyone has.

And notice the arrangement underneath. “Proven heritage” is not a neutral fact — it is a reason, and reasons get used. It let a schedule stay unrealistic and a test get skipped, and it did so while sounding like caution. The word that gave everyone confidence was the same word that let everyone stop checking.

What the whole looks like from here

The uncomfortable part is that no single seat could see it. The thruster team saw a good thruster. The schedulers saw a plan that had to work. The managers saw a heritage design and a wall of green checkmarks. Each was looking honestly at their piece. The failure lived precisely where no one was assigned to look — and everyone, standing where they stood, had a fair reason not to look there.

So the next time you hear “it’s fine, all the parts are proven,” you don’t have to disagree. Just ask the humbler question the whole invites: proven apart, or proven together? Because the parts are almost never the thing that fails. The space between them is.

03 · Lab · your turn

The Seam Check

Rehearse choosing whether to test proven parts together, and feel the failure land in the untested space between them.

04 · Hope · carry this

The most reassuring thing about the Starliner report is that it exists at all — a system honest enough to audit its own overconfidence is a system that can still learn. Two astronauts came home safe because someone chose caution over the schedule, and that choice is always available to us.