At dusk, Han Luyi had some fried rice at Bug Café when his phone buzzed. Su Niannian had sent him a link.
Three Bottom Lines for the AI code Review Field, by Zhao Wenyuan.
Han Luyi opened it.
The three-thousand-word article never mentioned BugKiller by name, yet every paragraph was aimed directly at him.
"The compliance of training data sources is the cornerstone of a product's credibility. When a product's core data source cannot be verified by an independent third party, user trust is built on quicksand."
"Anonymity is a right, but anonymity does not mean exemption from accountability. When an anonymous identity is tied to commercial interests, the community has the right to demand a higher standard of transparency."
"What this field needs is not the narrative of an individual hero who rises to fame overnight, but an engineered system that can withstand scrutiny."
The signature at the bottom read: Zhao Wenyuan, PhD in Computer Science from Stanford, former senior Google engineer, founder of CodeSafe.
The article had been posted simultaneously across multiple tech communities and was quickly republished by numerous technology media outlets.
The comments section was already in chaos.
"To be honest, I'm a little shaken after reading this. NullPointer's detection rate is impressive, but why is it so strong? The data-source statement is one sentence long. Where's the third-party verification?"
"Every word in Dr. Zhao's article hits the mark. Being anonymous while making a product isn't the problem. Being anonymous, opaque, and profitable all at once is."
"I recommended BugKiller to my team during the beta, and now I'm kind of regretting it. What if there really is something wrong with the training data?"
"...So does nobody care that CodeSafe Official liked a sockpuppet post from TechEthicsWatch?"
The last comment had sunk to the bottom of the second page with barely any likes, and the tide was turning the wrong way.
Han Luyi finished the article at Bug Café, his expression grim.
Old Zhou was wiping glasses behind the counter. He leaned over and glanced at him. "Who pissed you off?"
"Nothing. Competitors taking shots at each other."
"Programmers do that too?"
"Coders don't. PowerPoint writers do."
Old Zhou laughed.
A message from Su Niannian came through. "Finished reading?"
Han Luyi started a voice call with Su Niannian. "I just finished it."
Su Niannian skipped the pleasantries, speaking quickly. "He's way too underhanded. On the surface, he's talking about industry standards, but he's really pinning the labels of 'anonymous' and 'opaque data sources' on us. Look at the comments. People are already asking who NullPointer is. He's not just attacking you. He wants to set the rules. Once he does, you'll be judged within his framework forever."
"I know."
"So what do we do?"
"Don't take his bait. He talks standards, I'll talk results. Put the detection rate on the table, and everything else is nonsense."
The other end fell silent for a second. "Okay. Should I sort out the beta data first?"
"Yeah. Detection rate, false-positive rate, impact-assessment coverage. Put it side by side with CodeSafe's public data."
"Give me twenty minutes."
After hanging up, Han Luyi got up and went home.
He sat at his desk and opened a new document.
The title was: Detection Rate Is Everything.
He finished twelve hundred words before eight o'clock.
It was not long and discussed only one thing from beginning to end: what made a good code review tool.
"Users don't care who your founder is, how many PhDs are on your team, or how many pages your white paper spends explaining data sources. Users care about only one thing: can you find the bugs in my code?"
"If you can, you're a good tool. If you can't, everything you say is nonsense."
"These are the data from BugKiller's first batch of beta users."
A table, three rows of numbers: detection hit rate, false-positive rate, and impact-assessment coverage. Every column included the beta sample size and user-verification rate.
"Starting today, these data will be updated in real time on BugKiller's official website. No filtering, no polishing. Any peer, media outlet, or user is welcome to verify them at any time. Any product in the same field is also welcome to publish comparable data and put it side by side with ours."
The final sentence read: "You'll know whether a tool is good after running it once."
Published.
Twenty minutes later, the replies had passed two hundred, with over seven hundred likes.
"That last line killed me."
"Dr. Zhao wrote three thousand words to set the rules. NullPointer didn't even write a thousand, just slapped the data on the table and said: Come on, compare! The difference in caliber is huge."
"The screenshot of CodeSafe Official liking that TechEthicsWatch sockpuppet post got dug up again. What an ugly look."
"TechEthicsWatch had zero posts and zero replies, registered yesterday and disappeared today. Do you still need hard proof that it was a sockpuppet?"
Zhao Wenyuan pinned a reply on his blog: "This article does not target any specific product."
The top reply beneath it had twenty-three hundred likes.
"Dr. Zhao, your article was published thirty-six hours after BugKiller's beta post. Do you generally only start caring about industry standards after a competing product gets popular?"
Zhao Wenyuan did not reply again.
Friday afternoon, Ma Xiaofei dropped a short-video link into the 404 Group.
"One move straight to the peak!"
The title was: <I Tried the Hottest AI code Review Tool on the Internet, BugKiller, and the Result Was...>
Han Luyi opened it. It was Ma Xiaofei's usual style: fast-paced, packed with reaction memes, exaggerated yet compelling. He used one of his own small open-source projects for the demonstration. After BugKiller finished running, it found three bugs he had not known about, each with an impact assessment and a suggested fix.
"This thing scanned my code once and dug up every landmine I planted three months ago."
"The second bug was even crazier. It said there was a chance of deadlock under high concurrency. I was thinking, this crappy project of mine barely has any concurrency. But when I looked closely, this method was referenced by another project, and that project had concurrent scenarios. How did it know? How did it know??"
"The third one... forget it, I'm not showing it. A misspelled environment variable sitting there for three months. The production environment not blowing up was pure luck. Embarrassing."
His final summary was: "There are two kinds of tools. One makes you think, 'Not bad, I guess.' The other makes you immediately go dig through your own codebase after seeing it. BugKiller is the second kind. It's in free beta, not an ad. But NullPointer is welcome to pay me."
Zhang Haoran said in the group, "How many lines does that project even have? Three bugs is a pretty high hit rate."
Ma Xiaofei: "Shut up."
Han Luyi: "Didn't you say it wasn't an ad?"
Ma Xiaofei: "ICBC 6222 XXXX XXXX XXXX, recipient Ma Xiaofei. Thanks, boss."
In forty-eight hours, the video reached one hundred and twenty thousand views. Ma Xiaofei gained nearly ten thousand followers, charging straight toward seventy thousand. BugKiller broke out of the beta-testing circle on tech forums and entered the view of the broader tech audience.
After seeing the numbers, Su Niannian sent one message: "Ma Xiaofei's pretty good. This video worked better than two weeks of promotional copy from me."
Han Luyi screenshot it and sent it to Ma Xiaofei. Ma Xiaofei replied instantly: "Goddess Su??? You two talk every day now???"
"She helps review the product."
"Review. The. Product. Got it, got it."
Monday.
Su Niannian sent three screenshots without a single word.
The first was a tech-media headline: <CodeSafe Officially Enters Public Beta: The "Most Complete AI code Review Solution" Built by a Stanford PhD Team>.
The second was a popular tech-forum post: <CodeSafe's First Week: Enterprise Integration Is Seriously Strong>. The poster tested Python, JavaScript, and Java, concluding, "It surpasses every code detection tool currently on the market. Its integration documentation is better written than the product itself, and it works right out of the box."
The third was a screenshot of Zhao Wenyuan's social-media post. No one knew who had leaked it.
There were only four words.
"Let the product speak."
The BugKiller beta group was discussing it too. "CodeSafe has entered public beta. When are we going public?"
Su Niannian said nothing, and Han Luyi did not reply either.
On the forum homepage, NullPointer's post was still pinned as the hottest post, with seven thousand likes. Zhao Wenyuan had not posted anything for three days.
Han Luyi opened CodeSafe's official website. On the registration page, the first week's numbers were still rising.
He activated Sight.
Gray streams of code floated up from beneath the page. Front end, back end, data pipelines—CodeSafe's technical architecture spread across the screen like an X-ray diagram.
He watched for several seconds, then turned off Sight and closed the page.
His expression did not change. He opened his editor and continued writing v0.4.
Before you continue