The people running the Mars Climate Orbiter expected to celebrate on September 23, 1999.[1] The spacecraft had been flying for nine and a half months.[2] That day it was to fire its main engine for about 16 minutes, slow down, and settle into a long loop around Mars that took 14 hours to complete.[3]

Part of the way through the burn, the spacecraft was due to pass behind the planet as seen from Earth, and its radio signal would cut out. Controllers in Pasadena, California and Denver, Colorado expected it back at about 2:26 in the morning, Pacific time. It never came back.[2]

A week later NASA announced what had gone wrong. Two teams had used different units of measurement, and the mix-up sat in a computer file that fed the navigators wrong numbers for months.[4][2]

Two teams and one file

The orbiter was built and flown by two groups in two states. Lockheed Martin, in Denver, built the spacecraft and ran it day to day. NASA's Jet Propulsion Laboratory in Pasadena was in charge of the mission, and its navigators worked out where the spacecraft was and where it needed to go.[2]

It had left Florida on December 11, 1998. Its job was to become the first weather satellite around another planet, and to relay messages for a second spacecraft, the Mars Polar Lander, due to land that December.[2]

On the way to Mars the orbiter needed small corrections of its own. Spinning wheels inside it kept it pointed the right way, and over time they built up more spin than they could handle. Every so often, small thrusters fired to get rid of it. Each firing also nudged the whole spacecraft slightly off its path, so the navigators needed to know exactly how hard each push had been.[2]

After each firing, the spacecraft sent its data to the ground, and a computer program turned it into a file. The navigators used that file to track the spacecraft. A written rule said the numbers in the file had to be in newton-seconds, the metric way of measuring a push. The program wrote them in pound-seconds instead. It was the Lockheed Martin team that used English units.[1] A pound of force is about 4.45 newtons, so the navigators' software treated every push as about 4.45 times weaker than it was.[2]

The popular version of this story, that NASA forgot to convert its measurements, is only partly right. The spacecraft's own software used metric units and was correct. The wrong numbers came from a file on the ground, which broke a rule that existed on paper. The investigation board's report found the rule was not properly used when the program was written and tested.[2]

A small error, repeated

One wrong push would not have mattered. There were far too many. The orbiter fired its small thrusters 10 to 14 times more often than the navigators expected, because its single solar panel was lopsided and sunlight pushed on it unevenly. The engineers had planned a daily flip of the spacecraft to cancel that out, nicknamed barbecue mode. Later studies decided it was not needed, and it was dropped. Nobody told the navigators.[2]

Each extra firing added a small error, and over nine months the errors piled up. By the time the spacecraft reached Mars, it was about 170 kilometers (105 miles) lower than planned.[2]

The navigators did sense that something was off. For the first four months the files were so full of format errors that they could not be used, so the team relied on emails from the contractor to learn when a firing happened. When the files were fixed in April, it became clear within a week that their data looked wrong. Through the spring and summer the navigators saw gaps between what their calculations predicted and what the radio signals showed. Those concerns were passed along only informally. As the spacecraft got closer, they ran three different ways of working out its position, and one of them kept putting it nearer to Mars than the others. The differences were never resolved.[2]

The team was small and stretched. For the final approach, navigation was handled by the team leader and one navigator, while the same organization was also running two other Mars missions. When the numbers disagreed, people mostly sent emails instead of filing the formal problem reports the project had for the purpose.[2]

The last week

On September 15, eight days before arrival, the team made its last planned course correction. It was meant to keep the spacecraft 226 kilometers above Mars at its closest. Over the following week, the navigators' calculations showed that figure kept shrinking. About an hour before the engine burn, better tracking data put it as low as 110. The lowest altitude the spacecraft could survive was 80.[2]

So on the numbers the team had, the orbiter was still above the line. Calculations done afterward, with the right numbers, put its closest point at 57 kilometers.[2]

There was one last chance. A spare course correction had been kept as a backup. The idea came up in conversation shortly before the burn, but it was never carried out. The team had not tested the procedure or agreed on when to use it, and the burn schedule stored on the spacecraft left no time to send new commands. The investigators also noted a widespread feeling that orbiting Mars was routine.[2]

Four minutes behind Mars

The engine started at 09:00 universal time, or about 2 a.m. in California. About four minutes later, the spacecraft went behind Mars. Its signal dropped out 49 seconds earlier than predicted. It was supposed to return after 21 minutes. It did not.[2]

Nobody knows exactly what happened next. The board said the orbiter was either destroyed in the atmosphere or skipped back out into space and ended up in orbit around the sun.[2] A laboratory spokesman said its propulsion system overheated and was disabled as it dipped deep into the atmosphere.[1]

On September 27 the navigators sat down with the spacecraft's engineers to talk about their puzzling numbers. On September 29, they found the cause.[2]

What NASA said

On September 30, NASA made it public. Edward Weiler, who ran space science at NASA, said, "People sometimes make errors." He said the real failure was in NASA's engineering process and the checks that should have caught the error. Edward Stone, director of the Jet Propulsion Laboratory, said, "Our inability to recognize and correct this simple error has had major implications."[4]

Tom Gavin, a senior manager at the laboratory, said no one was pointing fingers at Lockheed Martin. "A single error like this should not have caused the loss of Climate Orbiter," he said. He also said a NASA document written while the Cassini mission to Saturn was being built had made metric the standard for all measurements. The orbiter had cost $125 million.[1]

The board's first report came out on November 10. It named one root cause, the wrong units in the ground file, and eight contributing causes. Among them were too little communication between the teams, too little training for the navigators, and no full test of the ground software from end to end. Another was the first-ever handover of a Mars-bound spacecraft from the group that built it to a new team running several missions.[5] The board said mistakes happen on spacecraft projects and that other checks usually catch them before they matter. This time, none did.[2]

The next spacecraft

The Mars Polar Lander was already on its way, due to land near the planet's south pole on December 3. The board's first report was written mainly to help it, and its advice included checking that every part of the lander's design and operation used the same units.[2]

The lander was lost too. Its last signal was sent before it entered the atmosphere, as planned, and nothing came afterward. In March 2000 an independent review team said the most likely cause was a different mistake. When the lander's three legs snapped open during the descent, the shock probably set off a false signal from a sensor meant to detect touchdown. The flight computer was not programmed to tell a false signal from a real one, so it believed the lander had landed and shut off the descent engines while it was still about 40 meters, or 130 feet, up. It would have hit the surface at about 50 miles an hour.[6]

Engineers will never know for sure, because no data came back. A software patch could have fixed the problem before the lander reached Mars, if anyone had realized it was there. NASA sources said it was not fully uncovered until tests on a similar lander built for a 2001 mission, after the crash.[7]

The same day, NASA said it would restructure its Mars program, put off a planned 2001 landing mission, and hold money in reserve for surprises, something the cost-capped team behind the two 1998 spacecraft had never had. Weiler added that, of roughly $18 billion in space hardware NASA had launched since 1992, it had lost about $500 million, counting the two Mars failures. By his math that was a 97 percent success rate.[6]