Early indications suggest a train that derailed Monday morning near Dupont, Washington, killing several people, was traveling about 50 miles over the speed limit on the stretch of track where it crashed. If that proves correct, it will be the latest example of delayed implementation of a safety failsafe in the United States leading to a fatal train wreck.

The Amtrak Cascades train was traveling from Seattle to Portland, traveling a brand-new, faster route, when it jumped the rails south of Tacoma, sending cars careening off the tracks and off an overpass over Interstate 5, where it struck several vehicles. There’s still no confirmed death toll from the wreck. But the National Transportation Safety Board said late Monday night that the train was traveling 80 miles per hour, close to the 79 mph limit on the run, citing the train’s data recorder. The train was supposed to slow to 30 miles per hour as it curved over the highway.

Amtrak argued that any such judgments are too hasty. “It’s being investigated by the NTSB and anything beyond that, until their findings are out, is pure speculation,” Gary Banks Olson, assistant superintendent of operations, told the Seattle Times.

But while it is true that the NTSB’s full investigation will probably take months to complete, the cause of a wreck is often quickly clear from the data recorder, which is similar to those used in airplanes. Train-safety experts who examined photos and footage of the crash said that the arrangement of the cars—including the direction of one locomotive that left the tracks, and jackknifed cars on the overpass—suggested the train had been moving too fast headed into the curve. (The train had locomotives at both ends, a common configuration.)

Amtrak said Monday that Positive Train Control, a technological failsafe intended to avoid collisions, excessive speed, and other human errors, was installed on the stretch of track where the train derailed but was not in effect and was still in testing.

Mechanical and equipment failures are more common causes of train wrecks than human error, but human error often plays a role in the most spectacular crashes. In May 2015, Amtrak 188 derailed in Philadelphia, killing eight people. The NTSB concluded that the engineer had been distracted by radio traffic and failed to slow down headed into a curve. In 2016, a New Jersey Transit train going too fast crashed at Hoboken Terminal. In 2008, a Metrolink train collided with a freight train in Los Angeles, killing 25, an accident NTSB said was caused by its engineer texting and missing signals.

Each of these cases would likely have been preventable with PTC, which acts somewhat like an autopilot on a train. If an engineer fails to slow down to speed limits, or blows through a signal, the system will override the manual controls and slow the train down.

The 2008 crash led Congress to mandate that PTC be in place on all passenger, commuter, and hazardous-materials trains by the end of 2015. Railroads warned that they would be forced to halt service on January 1, 2016, if an extension was not granted. Toward the end of 2015, with railroads complaining they could not roll the system out fast enough, Congress extended its deadline to 2018. Amtrak, which is perpetually strapped for cash, has struggled to pay for the installation of the system.

Yet horrifying and fatal crashes continue to pile up in the absence of PTC. In the curve where Amtrak 188 crashed in Philadelphia in 2015, PTC was in place on the southbound side of the track but not the northbound side. NJ Transit had received a waiver excusing it from installing PTC at the Hoboken stretch where the 2016 crash occurred. Now comes the Washington state case, where PTC was installed but not yet in place.

“If a PTC system has been active, this train might not have derailed,” then-NTSB Chair Christopher Hart said when the board completed its investigation of Amtrak 188. “Close to 200 passengers would not have been injured and eight other passengers would still be alive today.”

Even if PTC would have prevented Monday’s derailment, that leaves many questions about why the train was overspeed. For example, was the train moving too fast because of an equipment problem, or because the engineer didn’t try to slow down? If the latter, what happened? Was the engineer distracted, whether by a phone or something else? Could the engineer have blacked out or had a health problem?

Richard Beall, a longtime train-wreck investigator, told me railroads ought to have more than one person in the cab of a train, especially in the absence of PTC. “If this guy had a brain fart and didn’t realize where he was and, oh crap, I’ve got a 20 or 30 mph curve coming up, that’s why you have another pair of eyes up there to help you with that stuff,” Beall said.

Because Monday’s voyage was the first run with passengers on the new line, the safety preparations and training ahead of the launch will come in for scrutiny, with questions about whether the engineer’s unfamiliarity with the route might have contributed to speeding through the curve.

The Washington State Department of Transportation said that the tracks would be inspected in January 2017, with dry runs using the actual trains starting in February. But it’s not clear how many test runs the train made, what the staffing was, or when that testing concluded, which might shed some light on how familiar the crew was with their route.

“What troubles me here is this is a startup operation,” Beall said. “Usually if you have a startup operation you’re running the equipment daily for months with no one on it to get the crews familiarized with all the lines.”

There were 78 passengers and five crewmembers on board, according to Amtrak. Around 77 people were hospitalized after the crash, including some people injured when the train fell off the overpass onto busy I-5. But officials have not yet released a solid number for those killed, citing confusion at the scene, multiple responding agencies, and the difficulty of inspecting all parts of a dangerous scene.