Starship Flew Once and Broke Apart. SpaceX Called It a Success.
Four minutes of flight, a vehicle lost on purpose, and the difference between a milestone and an experiment.
On Thursday 20 April 2023, SpaceX flew Starship as a complete vehicle for the first time — the Super Heavy booster and the Starship upper stage stacked together — from Starbase at Boca Chica Beach, in southern Texas near the Mexican border. About four minutes later the rocket was destroyed by its own flight termination system and came down in the Gulf of Mexico. SpaceX did not present that outcome as a failure, and the reasoning is worth following.
CBS News reported that three of the booster's 33 methane-fuelled Raptor engines were not running after liftoff — two from the outer ring of 20 fixed engines and one of the 13 steerable central ones — having either shut down or never ignited. Three more outer engines dropped out over roughly the next 80 seconds. At one minute fifty-five seconds the exhaust plume turned distinctly asymmetrical, and the vehicle then began turning about its long axis. Stage separation never happened, and the automatic destruct command reached both stages a few seconds before the four-minute mark, by which point the rocket had fallen back to around 18 miles. Peak figures were roughly 39 kilometres of altitude, a little over 24 miles, and about 1,400 mph.
SpaceX engineer John Insprucker, narrating the live stream, called it "a rapid unscheduled disassembly, or a RUD, during ascent" — the company's own term for a rocket coming apart. Euronews quoted him adding that "excitement is guaranteed and Starship gave us a rather spectacular end". SpaceX's official statement, quoted by BBC Sky at Night Magazine, said that with a test like this "success comes from what we learn".
Why a company flies something it expects to lose
The flight plan never included bringing the hardware home. CBS reported that the booster was meant to splash down in the Gulf of Mexico and the upper stage in the Pacific; Sky at Night Magazine put the upper stage's target near Hawaii. The vehicle was designed to end in the ocean. What was not disposable was the telemetry.
The reason is scale. The stack stood 394 feet, close to 120 metres, per Euronews, and produced around 16 million pounds of thrust at liftoff — roughly twice NASA's SLS moon rocket, CBS noted — while burning about 40,000 pounds of propellant every second. Thirty-three engines firing in concert, inside a real structure under real aerodynamic load, is not a thing any ground rig reproduces. The only way to learn how that vehicle behaves is to fly it. Even the aborted attempt three days earlier taught something: Sky at Night Magazine reported that the 17 April countdown was stopped by a frozen pressurant valve.
The reaction told you what kind of event everyone understood this to be. NASA administrator Bill Nelson said that "every great achievement throughout history has demanded some level of calculated risk", and that he was looking forward to the next flight test. Elon Musk wrote that the team had "learned a lot for next test launch in a few months". The follow-up was already assumed.
What this means if you are building something
Decide in advance which kind of launch you are running. A milestone is a promise: this works, and people can depend on it. An experiment is a question: we do not know how this behaves under real conditions, and no staging environment will tell us. Both are legitimate engineering. Projects break when the decision is made afterwards, in the meeting where something went wrong.
Write down what would count as success before you press the button. Name the readings you cannot obtain any other way — checkout completion with real payment cards, import time against the client's actual 400,000-row catalogue, how the search index behaves under Monday-morning traffic — and agree in advance that getting those numbers is the win.
Never let a client hear "test" and "finished" about the same event. If the marketing director has told the board the new site goes live on Tuesday, Tuesday cannot also be the day you find out whether the payment gateway holds. Either move the experiment earlier, behind a feature flag or in front of a limited audience, or move the announcement later. What you cannot do is let one date carry both meanings and hope.
Fund the second attempt before you make the first. An experiment whose budget assumes it will work is not an experiment. If the plan has no time or money to act on what the test teaches, the test was a gamble with extra steps.
Instrument everything, or the loss buys you nothing. Four minutes of flight was worth a vehicle because SpaceX was measuring throughout. A deployment that fails and produces no usable record of why is not a calculated risk. It is just a bad day.
Our reading: most client work should ship as milestones, carefully and undramatically. But almost every serious project contains one or two questions that only reality can answer. Name them early, run them as experiments on purpose, and say so before anyone tells the board otherwise.
Sources
