175: The Approach

Wenqing's second analysis arrived April 3.

It was four pages. The first page was the data — session logs from April 1 and 2, tracking the [x] modifier across three daily sessions. The pattern he'd identified:

The modifier was no longer growing at the same rate.

"Linear growth," he wrote, "would suggest the modifier moves at approximately 0.3% per session given the observed rate over the January-February period. Instead, the April 1 session logged 38.4%, the April 2 morning session logged 39.7%, and the April 2 evening session logged 41.2%. The gap between sessions is 1.3% and 1.5% respectively — accelerating. The model I ran in January assumed linear growth because that was what the data supported in January. The current data doesn't support the linear assumption."

He'd included a revised graph. The curve was steepening. Not a clean exponential — he'd noted that too, with the annotation: *does not fit exponential regression; r-squared 0.71 against exponential model versus 0.89 against sigmoid. Treating as sigmoid for revised projections.*

The curve was remembering it had somewhere to be.

"Revised ceiling estimate: April 7 or April 8 rather than April 10. Possibly earlier."

April 7 was four days away. I read the analysis once through, then looked at the graph again. The sigmoid fit made sense given the class design — front-loaded because the class was accruing a new kind of capacity and the early accrual was the highest-marginal-gain phase; decelerating in the middle because the accrual was filling in rather than breaking new ground; then steepening again as it approached the ceiling. Most things with a ceiling worked this way. You didn't arrive at the ceiling gradually. You approached it quickly at the end.

"Additionally," he wrote, "I'm seeing a secondary pattern in the combat log that wasn't present in the January or February data. In formation-context combat, the [x] modifier is behaving differently from its behavior in those earlier months. In January, the modifier applied to MC's own skill output exclusively — Skill A + Skill B + [x] = augmented version of MC's output only. In the current data, there are brief intervals in the log — typically 2–4 seconds in duration — where adjacent guild members' output spikes by a small amount that doesn't correspond to any logged skill activation, consumable use, or environmental modifier. The spike is 1.2% to 2.8% of the affected member's output. Inconsistent but not random — it correlates with the [x] modifier's activation window."

I read that section twice.

Adjacent guild members.

The modifier had been doing something to other people's output. In small intervals, in brief windows, without any logged mechanism to account for it.

I forwarded Wenqing's analysis to TwilightTide.

She replied in nine minutes: *I knew something was reading differently in the healer channel. In the last three sessions, the healing load calculation was occasionally mismatching my own predictions by small amounts. I attributed it to measurement variance in the combat log. It wasn't measurement variance.*

*No.*

*What is it.*

*I don't know yet. Wenqing's hypothesis is that the [x] modifier is beginning to express something that affects the formation's aggregate output, not just mine alone.*

A pause. Then: *I've been watching the combat log for seven weeks. I didn't see it until he showed me what to look for.* Another pause. *He's better at this than I am.*

*He only does this,* I sent. *That's why.*

*Yes. That's the right way to say it.* She was quiet for a moment. The kind of quiet that meant she was thinking rather than searching for words — the distinction was usually visible in the duration. *The first original condition. The one the system is waiting for. If the adjacent-member effect is what you think it is — a formation-wide modifier rather than a personal one — then the first original condition might be when that effect expresses completely. Loud enough that the system can log it as something new.*

*That's my working hypothesis.*

*April 7 or 8.*

*Possibly.*

*I'll watch the log more carefully in tomorrow's session,* she said. *I'll run the standard formation positions and flag every interval that doesn't match my prediction.*

*Yes.*

***

The Tuesday bench. The April maple, which had passed the early-bud stage that I'd seen on the previous Tuesday and was now in the mid-green of a spring that had arrived two weeks faster than the previous year. The kind of green that doesn't have a name because it only exists for about ten days before it becomes something else — too intense to call settled, too full to call new. The tree was in its fastest-moving week and it looked it.

The morning light came in at the April angle — higher than March's, lower than June's, the specific slant that made the bench's near side bright and the far side in shade. The shadow of the maple fell across the path at a length I'd memorized from previous Aprils. The campus had the quality of a morning that had started well and intended to continue.

Wanqing had the problem set. She was reading it this time — the spring seminar's second unit, longer than the first by about forty pages.

"Wenqing moved the estimate," I said.

"I saw the update in the guild records."

"April 7 or 8. Possibly earlier."

She turned a page. "Four days."

"Yes."

She looked at the problem set. "And the secondary pattern. The adjacent member output."

"Wenqing showed it to TwilightTide. She'd been noticing discrepancies in the healer channel's load calculations and had attributed them to variance."

"It wasn't."

"No." I looked at the maple — the mid-green, the specific color of rapid April growth. "The class is starting to do what the design logic pointed toward. The formation context isn't just a modifier to my output. The modifier is beginning to affect the formation itself."

She was quiet for a moment. The bench had its specific Tuesday morning quality — the campus still settling into the day, the light at the angle it held in April before it moved. "Is that what Beigong Yan meant by the first original condition," she said.

"I don't know. He said I'd recognize it when it happened."

"And this."

"This is the approach." I considered it — the sigmoid curve, the accelerating intervals, TwilightTide's flag on the healer discrepancies. "The condition isn't the modification being present in the data. The condition is when it expresses completely — when it does something that can only be done by a Blade Sovereign, not approximated by a Berserker with a [x] modifier. The current pattern is a preliminary form. Something approaching the complete expression."

She looked at me with the full-attention expression. Then she went back to the problem set. "Four days," she said again. Not a question.

"Approximately."

"What will you do differently in sessions between now and then."

"Nothing. The sessions are what the class is accruing from. Running them differently would change the accumulation pattern. Wenqing agrees. We run the same sessions the same way and let the ceiling arrive."

"Run the sessions the same way and let it arrive."

"Yes."

She turned another page. The campus was the April quality — warmer than March, the afternoon light staying past six, the exam period still three weeks away so the campus still had some of its normal daytime movement. People on the paths between buildings, the particular unhurried pace of a week that hadn't yet started closing. A good week of the semester, before the pressure of the final phase.

"The workshop," she said.

"What about it."

"Father called Monday. The April sessions are full — he added a third Saturday slot in response to demand. Twelve students in three sessions. Xiaoyu is running a waitlist for May." She looked at the problem set. "The model's May projection was nine students in two sessions. He's running ahead of it."

"Your model said the constraint on growth was his available time and energy."

"Yes. He's testing the upper bound of that constraint." She turned a page. "I'm updating the model this week with the April data. The May projection will adjust upward. I'll also check whether the energy profile is tracking within Doctor Yan's projected range or above it."

She was updating it quarterly. She had been tracking this since May 2016. The April 3 workshop data would go into the updated model before the week was out. There was something particular about the way she handled the model — not a duty, exactly, more like an ongoing attention, the kind that doesn't announce itself because it's simply always running. She kept the model current the way I kept the planning layer current. It was how she worked.

"Your spring seminar ends when," I said.

"June 10."

"And then."

"Then I'm in Hangzhou for the summer seminars. Starting June 20 — the same pattern as last year."

The same bench on Tuesday and Thursday. The same thermos in the evening sessions. The third summer in a row, probably, or close to it. The bench would have its summer quality again — the maple in full leaf, the light differently angled, the campus in its summer rhythms. I'd been at this bench through three seasons of it now and I could hold all three versions in memory at the same time. That was one of the things that accumulated without anyone planning it. The bench became a place where different versions of the same time coexisted — the April bench had March in it, had last April in it, would have next April somewhere underneath when it arrived.

"Good," I said.

She looked at me for a moment. Then she looked back at the problem set without saying anything else. That was accurate too — the specific look that was its own kind of acknowledgment, the thing that didn't need the rest of the sentence.

***

The April 5 evening session logged a [x] modifier average of 43.1%.

Wenqing updated the record at eleven PM: *The acceleration is continuing. The secondary pattern — adjacent member output spikes — is appearing more frequently: 7 intervals in tonight's session versus 3 in the April 2 session. Duration is extending slightly: up to 6 seconds in one interval, compared to the previous maximum of 4. Still small in absolute terms (1.2–3.4% of adjacent output) but the frequency increase is statistically significant against any random-variance model.*

*Revised estimate: April 6. Tomorrow evening's session.*

He'd moved it by a day. The sigmoid was steeper than his model had predicted. I noted the adjustment and went back to the session log — not to find anything new in it, but because I'd been reviewing session logs for twenty-two months and the habit was the habit.

I sent TwilightTide: *Wenqing updated the estimate. Tomorrow.*

She replied: *I saw. I'll be in the session.* Then, after a pause: *The three AM window. I'll run my standard protocol and leave the formation position open for whatever tomorrow's session shows. I'll be watching the healer log in real time.*

*Yes.*

*Are you ready for it,* she sent.

I thought about Beigong Yan walking to the inner area. The Sword Sovereign walking east. The twenty-one days of sequences that were the transition itself, not instruction. *I'll recognize it when it happens.*

*The class is,* I sent. *I'm running the sessions the same way I've been running them.*

A pause.

*That's the right answer,* she said. *See you in the tunnel relay.*

***

Zhu Yuhan sent a brief message at midnight April 5.

*The Iron Hills session was good tonight.*

She'd been running the three AM window consistently since December. Not every morning — the frequency varied with her schedule — but the habit of checking in afterward had become consistent. The brief message at the end of a good session, the one-line report that wasn't reporting anything the session log didn't already have. It was a different kind of communication from Wenqing's analysis. Less information per word, more acknowledgment per word. Both necessary.

*Yes,* I sent.

*The [x] modifier. Tomorrow.*

*Wenqing's estimate.*

*What does it feel like,* she sent, *from the inside. Waiting for something you've been building toward for a long time.*

I considered the question. Zhu Yuhan asked questions the other council members didn't ask — not because they weren't interested, but because she was interested in a different layer of the thing. The experience layer rather than the data layer. I'd noticed this about her for a long time and had never thought to name it until now.

*Normal,* I sent. *The sessions are the same sessions. The class is doing what it does. The estimate is Wenqing's model, not a guarantee.*

*But you believe it.*

*Yes.*

*Why.*

*Because the model has been accurate since January. And because the secondary pattern — the adjacent member effect — is what the class's design logic said the class would eventually do. The evidence is consistent.* I looked at the dorm ceiling, the specific texture of it in the late-night quiet. *It doesn't feel like waiting. The sessions are the work. Whatever happens tomorrow is what the work produces. I'm not waiting for it. I'm doing it.*

She was quiet for a moment. Long enough that I thought she might have closed the thread.

*Okay,* she said. *I'll be at the sixteen-meter position.*

*I know.*

*I know you know,* she said. *I just like to say it.*

I looked at the ceiling for a while.

The [x] modifier was at 43.1% and Wenqing thought tomorrow was the day. TwilightTide would be in the tunnel relay. Zhu Yuhan would be at sixteen meters. Wanqing was updating the model this week. The April maple was in mid-green and the campus was in its last good week before the exam pressure and Father's workshop was running twelve students in three sessions.

Everything was ongoing. Everything was the next version of itself. The sessions tomorrow would be the sessions tomorrow — no different in form from the sessions last week, no different in form from the sessions next month. The only difference would be in what they produced. And that difference you discovered by running them, not by waiting for them.

I set an alarm for 2:45 AM and went to sleep.