By knowing Columbia's planned trajectory, the relative position of the planet and the constant rate of the video frames, they worked out the timing, speed and direction of the flight to within a fraction of a second.

The second video was from Springville, Calif., near Sequoia National Forest. In it, they spotted the star Deneb and the compass star Polaris. Other stars — Alpha Cepheus, Vega and Beta Cassiopeia — could be plotted. This allowed them to confirm the timing of the flashes and the position of the shuttle.

By enhancing the stars digitally and then working with the constellations they revealed, the two men determined that Columbia's actual flight path had been barely a mile off its predicted reentry. Navigation experts at the Jet Propulsion Laboratory quickly confirmed their work.

The skunk team could then estimate the mass and size of the debris the spacecraft was shedding.

By knowing the properties of the exotic materials used in shuttle construction — recorded in three 22-year-old notebooks archived in the Caltech library — they calculated how quickly each piece was left behind by the spacecraft.

A theory took form.

The brightest flash — an event that NASA several weeks later would label "Flash #1" — was most likely a vapor of burning aluminum droplets, spewing perhaps from the spacecraft's melting framework.

"You see a bright flash and then a haze that is left behind," Beasley said. "The haze slows to essentially zero almost instantaneously, so it is impossible that it is any sort of solid material."

Perhaps something inside the shuttle had exploded. Melting metal scattered "like dumping a bucket of tinfoil chaff out of an airplane," Beasley said.

But what could have blown off so early in reentry without destroying the entire spacecraft?

Working with little sleep, the skunk team focused on a smaller but more substantial flash in the videos.

NASA would later label it "Debris Point #6."

From its speed, Dimotakis deduced that this chunk of Columbia might be up to 4 feet long.

Anxious to find it, the team arranged to charter a small plane, with Caltech donating the money to pay for it.

By this time, it was Saturday afternoon, two weeks after the accident. They would start their own air search Sunday morning, as soon as Dimotakis could refine his calculations for the most likely landing spot.

It was midnight by the time he and Beasley had finished cross-checking their results.

They canceled the flight.

They had underestimated how far something flung at 4 miles a second from an altitude of 200,000 feet would fly before hitting the ground.

The debris, they figured, had traveled 240 miles. It could have landed anywhere in an area 30 miles by 10 miles centered on the Gunlock Reservoir in southeastern Utah.

They sent their analysis to NASA and to the independent investigating board. Civil Air Patrol pilots were soon searching the area. Nothing turned up.