Hellbent

A newer surveillance photo caught Draker smoking outside a facility in Bethesda, Maryland. It was dated November 28, two days ago, the time stamp showing 8:37 P.M. Minutes before Evan had blasted through the door of the pest-control shop, killed everyone inside, and taken the laptop. Van Sciver’s update must have just come in. This second photo had no stamp or coding of any kind.

“The NSA intel put Van Sciver on the trail of drug-treatment places,” Joey said. “From there it was only a matter of time.”

Evan stared at the date on that surveillance photo and knew in his gut that Draker was lost.

“Which means that we’re down to one little Indian,” Joey said.

Evan stared at the last name: David Smith. Moved his fingertip a few inches. The ghost file opened.

A photo of a twelve-year-old boy. A birth certificate. A file painting a familiar story, various foster homes in various poverty-stricken counties. And then it showed a recruitment report from two years ago, listing Tim Draker as David’s handler.

Evan looked for more information, but there wasn’t any to be had. “That’s it?”

“That’s it.”

“Won’t Van Sciver have found him by now?”

“There are 33,637 people named David Smith in the country,” Joey said. “And believe me, with how well Jack’s been stashing people? The kid ain’t using that name anymore.” She jabbed a finger at the screen. “These people are hidden as well as it is possible to hide someone. Everything I know — hell, everything you know about being invisible? We learned from Jack. So I think Van Sciver’s still searching for this kid. I think he’s chasing him down now. And if we don’t find him first, he’s gonna kill the kid like he killed everyone else.”

Evan stood up. Laced his hands at the back of his neck and breathed. “All this…”

Joey completed the thought. “All this is because of you.”

He looked at her.

“Van Sciver’s killing his way to you right now,” she said. “All of us — these five names and however many more Van Sciver doesn’t have yet? We’re all on borrowed time.”

“How do we help that kid?” Evan said.

“We find him.”

“We can’t compete with Van Sciver’s resources. I have access to databases, but he’s at a whole other level.”

“You’re right.” Joey was chewing her thumb again, drifting behind the table, her eyes intense. “When it comes to David Smith we have an absence of data.”

“Right,” Evan said. “How do you look for an absence of data?”

“Deep-learning software,” she said. “Believe me, that’s what Van Sciver’s using.”

She looked over at Evan, saw that he wasn’t following.

“It’s machine learning using advanced mathematics,” she said.

“That doesn’t help.”

She leaned over the table, peering at him from above the laptop screen. “It finds patterns you don’t even know you’re looking for.” She took another turn around the table, passing behind Evan. “Between the name David Smith, potential fake names befitting a thirteen-year-old white kid, facial characteristics, his birth-certificate information, physical developmental changes, purchase patterns for foster kids fitting his analytics, past locations, receipts, meds, and thousands of other factors we’re not aware of but can be extrapolated from on the basis of that thin file”—she jabbed a finger at the screen—“let’s say that there are five billion combinations of data. Being conservative.”

“Conservative.”

“Yes. Without a machine learning system, it would be impossible to correlate all that data, let alone zero in on David Smith under his new name in his new hiding place.”

“Okay,” Evan said. “So what’s the best way for us to do that?”

She paused long enough to flick a smile his way. “Someone who knows where he is—”

“—and a hammer,” Evan said. He stood up. “Seriously, Joey. Can we break into somewhere that has these capabilities and run the data?”

“No. This kind of processing takes time. Days even.”

“What equipment do we need?” he said.

“A pile of hardware,” she said. Mutual exasperation had given the discussion the tenor of an argument. “And like, say, a shit-ton of common graphics-processing unit chips. The mathematics involved in machine learning take advantage of the massive parallelism of the thousands of cores in those things. We’d need giant-ass GPU arrays, computer towers stuffed full of graphics cards, linked together with a high-speed InfiniBand network, running at eighty gigabits and—.” She stopped, looked at him. “More stuff I’d explain to you if I thought you could understand.”

“So how do we do that? Right now?”

“Raid the computer-graphics lab building at Pixar.” She studied his expression. “Joking.”

Frustration mounting, he drifted over and leaned against the couch. The cushions and pillows had been rearranged for her to sleep there, a T-shirt balled up for a pillow.