Wednesday afternoon. Bug Café.
Old Zhou glanced at the clock behind the counter. "You're here at one already? No overtime today?"
"Freelancers don't have overtime." Han Luyi carried an Americano to his usual corner seat. "We're always on the clock."
Old Zhou snorted. "I wasn't this self-congratulatory when I was a programmer."
Han Luyi opened his laptop.
CodeSafe's official website was still open in the previous tab. Registrations had climbed to 3,800. Positive reviews had risen from three to seven.
He had read every single one.
He took a sip of coffee, downloaded CodeSafe's public beta, installed it locally, and opened it.
Last time, he had only glanced at the outline. This time, he needed to examine it carefully.
Activate Sight.
Gray streams of code rose from the bottom of the screen and spread before his eyes like a perspective diagram. The frontend framework, backend microservices, data pipelines, model inference layer—CodeSafe's entire tech stack lay exposed at a glance.
First, the good news.
React with TypeScript on the frontend, standardized component separation, smooth interactive animations. The UI looked three tiers better than BugKiller's. They had a dedicated frontend team; that was a gap one person could never catch up to. A Go gateway on the backend, Python for inference, Kafka as the message queue in between. Textbook technical choices. It supported three languages, and detection speed was decent. Its enterprise integration documentation was better than that of many major companies.
Beside the overall code quality, Sight displayed a rating tag:
[Code Quality Rating: B+]
Not bad. A Stanford CS PhD plus Google experience was no joke. This system's engineering standards were top-tier among startups. Zhao Wenyuan had real skills.
Han Luyi continued reading.
Sight drilled into the core detection engine. The code streams grew denser, and the amount of information surged. He focused, and his energy began draining faster.
Then, above the core engine, a line of yellow WARNING slowly appeared:
50,000 lines, 3%.
200,000 lines, 17%.
500,000 lines, 41%.
Three numbers, rising nonlinearly.
CodeSafe's current users were all individual developers and small teams, whose codebases were generally under 50,000 lines. The experience was good, and the positive reviews were genuine. Zhao Wenyuan was not faking anything. His product really was strong at its current user scale.
But enterprise clients often had codebases with hundreds of thousands of lines.
This was not a bug. Bugs could be fixed. This was a fundamental problem with the architectural choice. Feature extraction had no hierarchical pruning; the core engine directly fed everything into a large model for inference. It worked below 50,000 lines because the context window could hold it. Beyond 200,000 lines, the window overflowed, and the model began hallucinating. Detection results went from "accurate" to "wild guesses."
Fix it? Rewrite the core engine.
At least three months.
Zhao Wenyuan probably did not know yet. All the public beta users had small projects, and the data looked beautiful. He definitely thought he was amazing.
Only once he landed a real enterprise client would he discover how thorny this problem was.
Han Luyi deactivated Sight. More than sixty energy points had dropped.
He leaned back in his chair and looked at the line on CodeSafe's homepage: "The most complete AI code review solution."
Su Niannian pushed through the door with a gust of wind from outside. Without even removing her scarf, she walked straight over, sat across from Han Luyi, and took out her laptop. They had agreed to meet at Bug Café.
"I registered for CodeSafe."
"Mm."
"Want to run a comparison test?" She opened a GitHub repository. "This open-source project has a Python backend service, around 80,000 lines. Let's run both through it."
Both computers began running at the same time.
BugKiller's progress bar reached the end first. CodeSafe was nearly a minute slower.
The results came out. Su Niannian placed the two windows side by side.
BugKiller reported 50 issues. CodeSafe reported 36.
Han Luyi activated Sight and scanned it again. Fifty-one bugs.
Using Sight's results as the answer key, he compared them one by one. Of BugKiller's 50, 48 were real and 2 were false positives, with 3 missed. Of CodeSafe's 36, 32 were real and 4 were false positives, with 19 missed.
After spending twenty minutes checking the source code, he gave Su Niannian the false-positive figures.
Su Niannian calculated in her notebook. "False-positive rate: ours is four percent, theirs is eleven."
He did not mention the miss rate. He could not exactly explain to Su Niannian why he knew the true answer was 51. He calculated it silently: six percent versus thirty-seven percent.
On an 80,000-line codebase, CodeSafe's false-positive rate had clearly risen.
But the numbers were not the key issue.
Su Niannian opened BugKiller's first detection result. Beneath the bug description was an impact assessment: "Predicted impact after deployment: Approximately 12% trigger probability under high-concurrency scenarios, estimated monthly impact on 12,000 users, repair priority: P0." Below that were three more lines describing the fix path.
She opened CodeSafe's result for the same issue.
"Potential null pointer exception detected. Please check."
One sentence. That was it.
Su Niannian looked over the data from both sides and fell silent for a while.
"Our detection capability crushes theirs." She looked up. "They support more languages than we do, and they started building enterprise integration earlier, but those are all matters of time."
"Right. And their engine will have problems with larger codebases." Han Luyi said. "Once their user base grows, they'll find out on their own."
Su Niannian narrowed her eyes. "How do you know?"
"I guessed from the architecture. Full traversal plus large-model inference. Once the codebase gets big, the context window can't take it. The false-positive rate will explode."
Su Niannian stared at him for two seconds.
She did not press further.
She closed her notebook, flipped to a fresh page, and wrote five words as the title: "Public Beta Plan v0.1."
"Let's go through the competitor analysis first." Su Niannian wrote as she spoke. "CodeSafe has a team of fifteen. I couldn't find their exact funding amount, but judging by Zhao Wenyuan's background, a conservative estimate is fifty times your entire savings."
"Thanks. That's really encouraging."
"We're at a disadvantage in money and manpower. But they're heavy and we're light—that hasn't changed. So we can't fight them on every front. We need to pick one point they can't make up for."
Su Niannian flipped to another new page and tapped the paper with her pen.
"Hit them with impact assessment. Don't touch anything else."
Su Niannian pointed at the comparison on the screen. "CodeSafe tells you there's a bug. BugKiller tells you what impact that bug will have after deployment. That's an entirely different product category."
"What about the launch plan?"
"First, we'll post on the forums, V2EX, and Juejin. We'll have that batch of internal test users help spread the word, and Ma Xiaofei can put out another video. If first-day registrations exceed five hundred, we'll call it a success." Her pen never stopped moving.
"When do we launch?"
Su Niannian opened her calendar and glanced at it. "The first day back after the holiday. We'll finish the launch plan and promotional materials during the break. Everyone will be off for May Day, and on the first workday, the sixth, they'll see it when they browse the forums. If we drag this out, CodeSafe's user base will get established and become harder to steal. Do you have enough time to write the code?"
"Enough."
"Then the sixth."
As she spoke, she started a WeChat call with Ma Xiaofei.
Outside, the sky grew dark. Old Zhou refilled both their cups and turned the lights up a little. Han Luyi wrote code while Su Niannian worked on the plan. They barely spoke in between. Occasionally, Su Niannian would look up and ask, "Did you revise the API documentation?" Or Han Luyi would toss over a piece of logic for her to confirm the product flow.
At eight thirty, Old Zhou began clearing the tables.
"You two, we're closing."
Su Niannian looked up at the time. "That fast?"
Old Zhou smiled and shook his head.
They packed their things and left, walking together toward the subway station while continuing to discuss the launch plan.
At the subway entrance, Su Niannian suddenly stopped. "Luyi."
"Yeah?"
Su Niannian smiled.
"...Never mind. See you tomorrow."
She entered the subway station. Han Luyi turned and headed home.
By the time he reached the door of 502, it was almost nine.
A brown paper bag sat outside the door.
He crouched down and picked it up. Inside was a small cake, the cream spread unevenly, with a smiley face drawn on top in chocolate sauce. A crayon drawing was tucked beneath the bag.
The drawing showed a little person with glasses sitting in front of a computer. Beside him lay an enormous ladybug. Its shell was red, with three big black spots.
He turned it over. Crooked pencil writing covered the back, every stroke pressed down hard:
"Brother Luyi, good luck"
The character for "luck" had been written backward.
Han Luyi stood in the doorway and took a bite. The cream was too sweet, and the cake was a little hard. He took another bite.
He put the rest in the refrigerator.
His phone lit up.
Gu Siyue: "Yu Wei has taken Lin Wanqing's case. The initial assessment says the chances of winning are high."
He replied, "Thank you, Lawyer Gu."
Another message arrived. Su Niannian had sent a Lark document link: "BugKiller Public Beta Launch Plan v0.1."
He opened it, looked through it briefly, and replied, "Got it. Get some rest early."
Han Luyi taped Duoduo's drawing beside his monitor.
Then he opened his editor.
The ladybug stared at him with two wide eyes.
Before you continue