Chapter 15: Overflow
The bug appeared on a Wednesday at 3 AM, and it was the kind of bug that made Dojun question whether he had ever actually understood computers at all.
Bridge’s task detection algorithm had been working perfectly for two weeks. Feed it emails, calendar entries, and file metadata, and it would cluster them into coherent tasks with 87% accuracy—impressive for a keyword-and-temporal-correlation approach without any machine learning. Hana’s frontend displayed the clusters beautifully. Minjae’s data pipeline fed them reliably. The demo was polished, the pitch deck was drafted, and the Innovation Showcase was nine days away.
Then Dojun added the context-learning feature—the one that let Bridge remember a user’s patterns and improve over time—and everything collapsed.
Not crashed. Not error-messaged. Collapsed. The learning module consumed so much memory that the rest of the application slowed to a crawl. Tasks that previously loaded in milliseconds now took thirty seconds. The UI froze. The email parser timed out. And the task clusters, which had been 87% accurate, dropped to 41% because the learning algorithm’s corrections were overriding correct classifications with incorrect ones.
It was, in programming terms, a catastrophic regression. A feature meant to make Bridge smarter had made it stupid.
Dojun stared at the error logs, scrolling through pages of memory allocation warnings and timeout exceptions, and felt something he hadn’t felt since waking up in 2006.
Helplessness.
Not because the bug was unsolvable—in his previous life, he had debugged infinitely more complex systems. Prometheus Labs’ production infrastructure handled billions of requests daily, and he had personally traced bugs through distributed systems spanning twelve data centers on four continents. A desktop application’s memory management issue should have been trivial.
But the tools of 2006 were working against him. No profiling tools worth the name. No memory visualization. No debugger that could track object lifecycle across Java’s garbage collector. The diagnostic infrastructure he had relied on for forty years simply didn’t exist yet.
It was like performing surgery with a butter knife. He knew exactly what needed to happen. He just couldn’t make it happen with the instruments available.
This, he thought, is what I get for being a time traveler with future solutions and present-day hardware.
He closed the laptop, put his head on the desk, and for the first time since the spring coding contest, seriously considered the possibility that he had overcommitted.
The IEEE paper was due in five days. He had finished the revisions—barely—but Kim Taesik wanted another round of edits on the methodology section. Bridge’s demo needed to be flawless in nine days. The learning feature was broken. And he hadn’t slept more than four hours a night in two weeks.
His phone, facedown on the desk, buzzed with a text. He didn’t look at it. Another buzz. And another. Then a call. He let it ring.
The phone stopped. Then started ringing again.
He picked it up. “Hello?”
“You didn’t answer your texts.” Hana’s voice was sharp with a specific frequency of concern that he recognized—the frequency that preceded anger if the underlying worry wasn’t addressed. “It’s 3:15 AM. Why are you awake?”
“How do you know I was awake? Maybe I was sleeping and you woke me.”
“People who are sleeping don’t answer on the first ring with ‘hello’ in a voice that sounds like they’ve been staring at a screen for nine hours. What happened?”
“The learning module broke Bridge.”
Silence. Then: “Broke as in ‘it needs a fix’ or broke as in ‘we’re screwed’?”
“Broke as in the entire task detection system regresses when the learning feature is active. Memory consumption tripled. Accuracy dropped to 41%. The demo is unusable.”
“The demo that’s supposed to be ready in nine days.”
“That’s the one.”
“Okay.” He heard her take a breath—the deliberate, centering breath of someone switching from panic to problem-solving. “What’s the root cause?”
“The learning algorithm stores too much state. Every time the user corrects a task classification, it saves the correction along with the full context—all associated emails, calendar entries, file references. For a few corrections, it’s fine. For a hundred corrections, the state database overwhelms the memory budget and the classifier starts making decisions based on noise instead of signal.”
“So the fix is to store less state?”
“The fix is to store state more intelligently. Compress the corrections into patterns instead of raw data. But that requires a completely different data structure—something like a Bloom filter or a compressed trie—and implementing that from scratch in Java without proper libraries is—”
“How long?”
“If I build the data structure myself? Three days, minimum. With testing and integration, five.”
“We have nine.”
“I also have an IEEE paper due in five.”
Another silence. Longer this time.
“Drop the learning feature,” Hana said.
“What?”
“Drop it. Cut it from the demo. We present Bridge without context learning. The base task detection is already 87% accurate—that’s more than enough to impress the judges. We don’t need the learning module for the showcase.”
“But the learning module is what makes Bridge special. It’s the difference between a static tool and an adaptive system. Without it, Bridge is just a fancy email organizer.”
“A fancy email organizer that works is infinitely better than an adaptive system that doesn’t.” Her voice was firm now—the voice of a designer who understood that shipping a working product beat shipping a perfect concept every time. “We’re not building the final version, Dojun. We’re building a demo. A proof of concept. The judges need to see that Bridge can detect tasks and organize information. They don’t need to see it learn—they need to believe it could learn. That’s what the pitch is for.”
“The pitch.”
“I describe the learning module as a planned feature. ‘Bridge v2.0 will incorporate adaptive learning based on user corrections.’ Future tense. Roadmap. Vision. The judges don’t need to see it working—they need to see that we’ve thought about it.” She paused. “Dojun. You’re trying to build Prometheus Labs in three months on a Compaq Presario. That’s not ambition. That’s self-destruction.”
He flinched. Not at the word self-destruction—at the word Prometheus. She had no way of knowing the significance. She had chosen the metaphor innocently, a Greek myth reference, the titan who gave fire to humanity and was punished for it.
But the resonance was shattering.
“You’re right,” he said quietly. “I’ll cut the learning module.”
“Say that again. I want to record it.”
“You’re right. The demo doesn’t need it. We ship what works.”
“Good. Now go to sleep.”
“It’s 3:30 AM.”
“Exactly. You need at least five hours before the IEEE revision. Set an alarm for 8:30. And Dojun?”
“Yeah?”
“Eat breakfast. Real breakfast. Not ramyeon.”
“I’m out of rice.”
“Then buy rice. You won a national championship and a research publication. You can afford rice.” A pause, softer. “We’ll fix the learning module after the showcase. Together. It’s not gone—it’s postponed. Okay?”
“Okay.”
“Sleep well.”
“You too. And Hana?”
“Hmm?”
“Thanks for calling at 3 AM.”
“That’s what partners do. Good night.”
She hung up. Dojun set his alarm, closed his laptop, and lay down on his bed without changing clothes. The ceiling stain stared down at him—the lopsided heart, unchanged since March, a constant in a life full of variables.
She was right. About all of it. The demo didn’t need the learning module. The pitch could carry the vision. And he was, in fact, trying to build Prometheus Labs on a Compaq Presario, driven by the same manic perfectionism that had consumed him the first time around.
The product is the means. The people are the end.
He had written that in his journal. He believed it. But when the code called to him at 3 AM, when the optimization was almost within reach, when the elegant solution dangled just beyond the limits of 2006 hardware—he forgot. Every time, he forgot.
This is the pattern, he thought. This is how I lost Hana the first time. Not in one big explosion, but in a thousand small choices to stay at the screen instead of picking up the phone.
She had called at 3 AM. She had noticed his silence. She had pulled him back from the edge of the same spiral that had destroyed them before.
And she didn’t even know she was saving him from repeating history.
Saturday. Namdaemun Market.
His mother took one look at him and put down her cutting knife.
“Sit,” she said. It was not a request.
He sat on the cracked plastic stool. She placed kongnamul-guk in front of him—she always had it ready, as if she could sense his arrival through some maternal radar that operated on a frequency invisible to science.
“When was the last time you slept a full night?” she asked.
“Define ‘full night.'”
“Seven hours. Minimum.”
“…I don’t remember.”
“Aigoo.” She sat down across from him on an overturned crate, her knees almost touching his. The market hummed around them—vendors calling, customers haggling, the eternal background music of commerce. But in their tiny space behind the banchan counter, it was just mother and son. “Tell me.”
“Tell you what?”
“Everything. You look like your father did before he left. Running from something or running toward something—either way, running until you collapse.” She reached across and took his hand. Her fingers were rough and warm, scented with garlic and sesame. “Talk to your mother.”
So he talked. Not about the time travel—never about that—but about everything else. The IEEE paper deadline. The Bridge demo. The Innovation Showcase. The ZDNet article. The pressure of being watched, evaluated, questioned. The fear that he was building something he couldn’t sustain. The terror that the people who believed in him would eventually discover that his success was built on knowledge he couldn’t explain.
He didn’t use those words. He said “I’m overwhelmed” and “I took on too much” and “I’m scared I’ll let people down.” But his mother heard what he meant, the way she always did.
“Do you remember when I first opened this stall?” she asked when he finished.
“Vaguely. I was four or five.”
“You were three. I had no money, no experience, and no idea what I was doing. I bought vegetables wholesale, cooked all night, and set up a table on the street outside the market because I couldn’t afford a stall inside. My first day, I sold twelve containers of kimchi and three of japchae. I made seventeen thousand won.”
“That’s not much.”
“It was nothing. It was less than nothing after the cost of ingredients. I went home that night and cried because I didn’t know how I was going to feed you.” She squeezed his hand. “But do you know what I did the next morning?”
“You went back.”
“I went back. And the next day, and the next. For thirty years, Dojun-ah. I went back every morning, not because I had a plan or a strategy or a guarantee. I went back because the alternative was worse.” She let go of his hand and pointed at the stall around them—the containers of banchan, the hand-painted sign, the cracked plastic stool. “This didn’t happen because I was smart or talented or lucky. It happened because I showed up. Every day, no matter what. That’s the only thing that works in the long run.”
“But what if showing up isn’t enough?”
“Then you show up differently. You change what you’re selling, or where you’re selling it, or how you make it. But you never stop showing up.” She stood and went back to her cutting board, picking up the knife with the practiced grace of three decades. “You’re taking on too much, Dojun-ah. I can see it. But the answer isn’t to do less—it’s to do the right things first. What’s the most important thing right now? Not the most urgent. The most important.”
He thought about it. The IEEE paper was urgent—deadline in five days. The Bridge demo was urgent—showcase in nine days. But the most important thing?
“The people,” he said. “Hana. Minjae. Professor Kim. You.”
“Then take care of them. The papers and the demos and the competitions—those are your banchan. They’re what you sell. But the people are your stall. Without the stall, the banchan is just food rotting on a table.”
“When did you become a philosopher?”
“I’ve always been a philosopher. You were just too busy staring at computers to notice.” She chopped radish with rhythmic precision. “Now eat your soup. And after you eat, go home and sleep. The computers will still be there tomorrow.”
“The deadline—”
“Will the paper be better if you write it exhausted? Will the demo work better if you code it half-asleep? Will the judges be more impressed by a young man with dark circles and shaking hands?”
“No.”
“Then sleep. Sleep is not laziness, Dojun-ah. Sleep is maintenance. Even computers need to restart.”
He ate the soup. It was, as always, the best thing he’d ever tasted. When he finished, his mother packed him a bag—rice, kimchi, japchae, kkakdugi, and a container of sesame oil—enough food for a week.
“This is too much, Mom.”
“It’s not enough. You’re too thin. But it’s what I have.” She pressed the bag into his arms. “Bring your partner next time.”
“Hana?”
“Who else? I want to meet the person who calls you at 3 AM to tell you to sleep. That’s a good person, Dojun-ah. Don’t lose her.”
“How did you know about the 3 AM call?”
“You mentioned it. Just now. While you were talking.” She shrugged. “A mother listens.”
He hugged her. She was small and warm and smelled of the market—garlic, sesame, the clean sharp scent of fresh vegetables. He held on longer than usual, and she let him.
“Go home,” she said into his shoulder. “Sleep. And call me tomorrow.”
“I will.”
“And eat the japchae first—it doesn’t keep as long as the kimchi.”
“Yes, Mom.”
“I love you, Dojun-ah.”
“I love you too.”
He slept for eleven hours. It was the longest continuous sleep he’d had since February, and when he woke on Sunday morning, the world felt different. Not fixed—the IEEE deadline was still five days away, the Bridge demo still needed nine days of work, and the learning module was still broken. But the problems felt smaller, or he felt bigger. Either way, the math worked.
He opened his laptop and looked at Bridge’s codebase with fresh eyes. The learning module was still there—disabled, commented out, waiting for a future that wasn’t nine days away but maybe nine months. Hana was right. The demo didn’t need it. The judges needed to see a product that worked, a team that believed, and a vision that was worth investing in.
He had all three. He just needed to stop trying to build the fourth thing that nobody had asked for.
He opened the IEEE paper and read Kim Taesik’s latest annotations. The methodology section needed tighter statistical language. The conclusion was too speculative. The references were missing two recent publications. All fixable. All doable in five days, with sleep.
He opened Hana’s sprint schedule—the color-coded one she had made him sign. Sprint one was on track. Sprint two started tomorrow. Sprint three was the pitch, and Hana had already drafted fifteen versions of the opening line, each more compelling than the last.
The schedule had a note in red at the bottom, in Hana’s handwriting:
Park Dojun will remain functional. Signed and sealed. Violators will be fed only cafeteria bibimbap for the rest of the semester.
He smiled. Then he opened his email and started working.
Not at 3 AM. Not in a fugue of sleep-deprived desperation. At 9 AM on a Sunday morning, after eleven hours of sleep and a breakfast of his mother’s rice and japchae, with a clear head and a full stomach and the knowledge that the people who mattered most were exactly where they should be.
His mother at the market. Hana in Busan, sketching wireframes. Minjae in the computer lab, testing data pipelines. Kim Taesik in his office, editing papers. Seokho in Daejeon, solving problems that nobody had asked him to solve.
All of them, in their own ways, showing up.
And Dojun, at his desk, writing code that was good enough—not perfect, not revolutionary, just good enough to ship—felt the particular peace of a man who was finally learning, after sixty-three years and two lifetimes, the difference between building something great and building something that mattered.