Blackquest 40

Chapter Forty-Five

It's soul-sucking, and I can hardly fathom it, and innumerable bruises and lacerations are making every fine movement of my fingers torture, but here I am—back at the keyboard.

"You've been grinding twenty-six hours on this," I say to Jared and Prisha, who stand together with Paul in my cubicle, "and your code has gotten through the end-to-end Blackquest testing environment how many times?"

The leaders of the two primary modules take a moment counting in their heads. Prisha's typically sleek, well-behaved hair is matted, though otherwise she's come through well enough. The same can't be said for Jared, who smells like brown-wilted cabbage and seems to have a new fringe of neck hair.

"Five," Prisha decides.

Jared whirls on her. "What? We've passed at least two hundreds botlets through the sandbox."

"And the vast majority failed," Prisha says. "The simulator rejects 95% of the botlets' output."

Her glare has more venom than any I've seen her shoot in three months at Codewise Solutions. She's either gained confidence or had her veneer of decorum melted off by the circumstances.

Paul—more practiced at settling squabbles than me—says, "Back up. Why don't you us what you've accomplished so far? Let's focus on that, and not the converse."

This restores civility. Prisha and Jared cool down and describe their progress. The optimization team has unpacked the simulator's matrix of seven variables and calculated a range of possible values for each. They can manipulate the matrix, store its values, compare those values over time. They're able to obtain a handle to the existing algorithm and use it to calculate the output the system would've produced—the output to be lowered.

The injection team, Jared's, has crafted objects ("botlets") that meet the handshake interface agreed to at the start of the project—allowing Prisha's team to catch and use them. The botlets can inject themselves into the host system, evade basic malware countermeasures, and intermittently slip through the shifting sandbox security to the optimization team's waiting arms.

"That's a lot," I say, and mean it. "Now let's talk about that five number. That's five times we've fooled the simulator—beaten the sandbox, taken those seven variables and returned a lower output than before, but within the simulator's accepted bounds. Correct, yes?"

They nod.

"Great. And what's the specific requirement? We need to fool the system for how long?"

Chins fall to chests. Both engineers shuffle furtively until Paul speaks up.

"Perpetuity," he says, the word lumpy from his mouth. "For as long as the—er, the simulation runs."

Our eyes meet, and I know we're both mentally replacing "simulation" with "nuclear power plant." The software needs to swap in allowably lower energy values (who knows what units we're in—Joules, kilowatt-hours) for as long as it remains in place.

What happens if Elite installs our software and it passes a bad value to the plant's core? Does it revert to the old algorithm? Or could a terrible thing happen?

I'm suddenly sympathetic to their insistence on "100%-repeatable success."

We work through lunch, hard and hot, trying to satisfy the simulator by different means. It's a fickle beast. Lopping five percent off the top works occasionally, but not enough to rely on. We change the weightings of the seven variables, testing one at a time, then two at a time, and so forth. The results are inconclusive.

The process is maddeningly slow because Jared's botlets only eke through once every couple minutes.

"We need more from you," I tell him.