Kai POV
Friday morning. End of week one.
I read performance reports the way other people drank coffee automatically, at the start of the day, as a way of taking the temperature of everything before the noise began. Six analysts. Five reports came back clean. The sixth was flagged.
Not for errors.
I sat forward.
The flag was a single line from the data team, buried at the bottom of the weekly monitoring log. They sent it to me because anything unusual in the analyst bay came to me first, which was a standing instruction Wen Jie had set up on day one of the firm's existence, and nobody had ever questioned.
The line said: Junior Analyst L. Xia submitted an internal query RE: 2019 fund restructuring document, specifically line-item discrepancy in Schedule C asset valuation. Query routed to the data team for response. Holding pending your direction.
I read it twice.
Then I opened the 2019 fund restructuring document.
It took me four minutes to find what she had found, which told me something, because I had built that document and I knew exactly where everything in it was. Schedule C, line forty-four. A valuation figure that was technically accurate but represented an asset that had been reclassified through a chain of three internal transfers before landing where it currently sat. On the surface: clean. To someone who traced the chain backward through the right source documents: a thread.
She had found the thread.
In four days.
Using read-only access and the standard analyst bay permissions she had been given on Monday morning.
Nobody had found that thread in four years. Not the auditors, because I had structured it to pass the audit. Not the senior analysts, because it required connecting documents across three separate filing years, and nobody had a reason to connect them. Not the investors because they received a different version of the report.
She had no reason to connect them either.
She had connected them anyway.
I picked up my phone and called the head of the data team.
He picked up on the second ring.
I said: "The query from the junior analyst. Do not respond to it. Send her the 2019 restructuring document, the public-facing version, clean, no Schedule C discrepancy. Tell her the original file had a formatting error, and this is the corrected version."
He said: "Yes, sir. Should I"
"That's all," I said.
I hung up.
Then I pulled everything else she had submitted in her first week.
Most new analysts spent week one learning the systems, meeting colleagues, and doing the minimum required work at the maximum allowable pace. It was human. It was expected. I factored it into the onboarding timeline.
Lin Xia had apparently not received that memo.
She had completed all assigned onboarding tasks by Tuesday afternoon. The Wednesday and Thursday work logs showed she had then, without being asked, done four additional things.
First: she had rebuilt the Q2 reconciliation template from scratch, fixing the formula error that had been propagating for eight months, and quietly replaced the broken version in the shared drive with the corrected one. She had not sent an email about it. She had not flagged it. She had just fixed it and said nothing.
Second: she had written a one-page note filed in the general comments folder, where nobody ever looked, identifying the duplicate asset classification and suggesting a correction methodology.
Third: she had drafted a clean version of the distribution timestamp process, reorganized for clarity, and left it in a folder labeled suggested workflow update, discard if not useful. No name on it. No credit claimed.
Fourth: the Schedule C query. Which was the one that mattered.
I sat with all of this for a moment.
Four unsolicited problems were solved, without acknowledgment. One forbidden thread, found and quietly pulled. All of it in four days. All of it done with the careful restraint of someone who understood that you did not announce your findings in a new environment, you documented them and waited.
She was not just good.
She was disciplined.
Those were two completely different things, and the second one was rarer.
I opened a blank document on my screen. I typed one line.
She sees what she is not supposed to see. Handle accordingly.
I stared at it.
Handle accordingly could mean a lot of things. In the original language of my plan, the language of contingencies and risk management, and variables to be neutralized, handled accordingly meant reassigning her. Give the work that kept her busy and pointed away from anything she should not be near. Or let her go with a clean reference and a reason she would accept.
I looked at her work.
I looked at the Schedule C query.
I looked at the corrected reconciliation template that had been sitting broken in the shared drive for eight months while two senior analysts walked past it every single day.
I closed the blank document without saving it.
I sent the assignment through the internal task system at nine-fifteen AM.
Three-company valuation analysis. Harmon Industrial, Pacific Bridge Capital, and Linfield Group. Standard due diligence depth: financial statements, asset review, risk assessment, comparable market analysis. A complete package. The kind of work a confident mid-level analyst could complete in a week if they worked steadily without interruption.
Due date: Thursday.
I watched the task appear on the analyst bay's shared board from my office. Six names on the board. The new task went to one of them.
I watched her screen.
She read the assignment. She was still for about ten seconds, the particular stillness of someone reading carefully, not of someone who was surprised or overwhelmed. Then she opened a new file and started working.
I went back to my own work.
I did not watch her screen again.
I watched it twice more.
Wednesday passed.
I spent it managing three separate threads of the Ren Shao operation: a legal team call in the morning, a position adjustment in the afternoon, and a communication with my offshore structure in the evening. Standard maintenance for the phase I was in. Nothing unexpected. Nothing required more than the baseline level of focus I brought to everything.
At eight-fifty PM, my email showed a task completion notification.
Valuation Analysis Harmon Industrial / Pacific Bridge Capital / Linfield Group. Submitted by: Lin Xia, Junior Analyst.
I checked the timestamp.
Wednesday. Eight forty-seven PM.
I looked at the due date on the original assignment.
Thursday.
She had submitted it a day early.
I told myself this meant nothing. Early submissions happened. People worked fast. It was data, not significance.
I opened the document.
The first section was Harmon Industrial. I read it in four minutes. It was technically correct, well-structured, and appropriately cautious in its conclusions. Good work. Better than I would expect from someone a week into their first professional position.
The second section was Pacific Bridge Capital. I read it in six minutes. Also correct. She had found a leverage issue in the balance sheet that the public filings obscured slightly and noted it clearly without overstating its significance. That was not a beginner's read. That required understanding context, not just numbers.
The third section was Linfield Group.
I read it twice.
She had identified the primary risk factor, a contingent liability buried in a footnote on page forty-one of the annual report, and built her entire risk assessment around it. Not just noted it. Built around it. The way you built analysis when you understood that the footnote was the story and the rest of the document was just the frame.
I had known about the Linfield footnote. I had included Linfield in the assignment specifically because of it. As a test.
She had found it. She had understood what it meant. She had centered her analysis on it without anyone telling her it was important.
I sat back.
I thought about this for a moment.
I thought about the Schedule C query that I had quietly suppressed this morning. About the reconciliation template she had fixed and not claimed credit for. About the way she had sat down on Monday morning and read the organizational chart on the wall with the focused attention of someone making a map.
She had been making a map.
She was still making it.
And she was better at reading terrain than anyone I had hired in four years. Possibly better than anyone I had met.
This was, by any reasonable measure, a problem.
It was also, and I noted this with the same flat honesty I tried to bring to all assessments, including unwelcome ones, the most interesting professional problem I had encountered in a very long time. Everything else on my desk was a problem I had already solved. I was executing a plan I had already written. The moves were already made. I was just watching them land in sequence.
This was not a problem I had already solved.
This was a problem that kept changing shape every time I looked at it.
I sent the analysis to my review folder and wrote two words at the top before closing it.
Capable. Dangerous.
I did not add a third word. I did not need to.
The third word was already there in the space between the other two, in the part of my thinking that I kept separate from the plan, in the fact that it was nine PM on a Wednesday and I was sitting in my office reading a report twice when I had already understood it the first time.
The third word was: interesting.
I did not write it down.
I closed the file.
I went back to work.
The work did not hold my attention the way it usually did, and I sat with that fact for exactly thirty seconds before I decided it was irrelevant and moved on.
It was the least convincing decision I had made in four years.