Chapter Twenty Three: Review

The conference room on the fourth floor of the research annex had no windows.

Dr. Miriam Halvorsen had designed it that way two years ago, when the annex was new and Catalyst’s adolescent integration model still existed largely in simulations. Windows invited distraction. Distraction invited imprecision. In work like this, imprecision could follow someone out of the building and remain inside them for the rest of their life.

The table ran long and pale, milled from a single slab of composite that softened the sound of anything placed upon it. Seven chairs. Seven data terminals. At the midpoint, a projection column stood dark.

Six people had arrived.

The seventh seat remained empty, as it always did. Halvorsen had never explained it. No one had asked.

She stood at the far end, reviewing the summary index on her terminal. She had started meetings on her feet for years. Sitting could come later, when she knew what kind of room she was in.

Dr. Ananya Patel sat to her left, a secondary display angled so only she could read it. She had been at the Institute since its second year. Her stillness was practiced: years of watching before she spoke, resisting the temptation to make an observation useful before she had made certain it was true.

Beside her, Dr. Yusuf Levin rested his folded hands on the table. He rarely spoke before seeing the data. Halvorsen valued that restraint, as she valued his precision in surgery and his refusal to hurry a postoperative review because someone important was waiting.

Dr. Sofía Reyes-Galindo had chosen a corner seat. Neurovia’s director of cognitive integration, she had co-authored the theoretical framework beneath Catalyst’s adaptive architecture. She had passed two chairs with better sightlines without appearing to consider them.

In Halvorsen’s private estimation, Sofía was the person most likely to be right when she pushed back. That made her indispensable, though not always easy to work with.

Across from her, Dr. Priya Nkosi had opened a physical notepad beside her terminal. Regulatory affairs and trial ethics. Her shorthand was illegible to everyone else, but things she wrote down had a way of becoming significant later.

At the near end sat Dr. Elric Morse.

He had arrived seven minutes early, before the scheduled climate cycle, and adjusted the room’s temperature manually. By the time the others entered, the air had begun settling toward his preference.

He and Halvorsen had founded Neurovia eight years ago. He now served as Cohort One’s lead psychologist, among other responsibilities that did not fit as neatly beneath a title.

Halvorsen activated the projection column.

Light spread across the tabletop. The walls dimmed around it.

“Three months of data,” she said. “What holds, what needs investigation, and what has to change before we authorize more participants. Yusuf, begin.”

The data resolved in layers.

Eighty-one variables across twenty-four participants, sorted by cohort, individual, and domain. Neural coherence indices. Adaptive latency curves. Changes in sleep architecture. Cognitive throughput measured against pre-implantation baselines and tracked across the trial’s first twelve weeks.

The numbers were clean.

What to call them was less settled.

Levin began with the surgical record. His voice was flat and unhurried, his eyes moving down a document he had already reviewed enough times to know where every qualification belonged.

“Twenty-four procedures. Zero structural complications. One acute inflammatory event in Cohort Two, resolved within four hours of detection. No residual deficit. One subsequent concussive injury, participant Park, which triggered an unscheduled adaptive response we’ll return to.”

The display changed.

“Lattice anchoring exceeded projection in seventeen cases. Suboptimal anchoring in two, both Cohort One. Both remain within acceptable parameters and under observation. Synaptic calibration efficiency is running fourteen percent above the Phase One adult baselines.”

“Fourteen,” Reyes-Galindo said. “That’s not a rounding error.”

“No. Adolescent neuroplasticity appears to interact with the lattice differently than our adult models predicted. The junctions are—”

“More receptive,” Morse said, without looking up.

Reyes-Galindo studied him.

“More receptive is one description. Faster integration is another. I’d like to understand the mechanism before choosing the language.”

“Plasticity was the reason to study adolescents. We expected their integration to differ.”

“We designed for a specific range of response. The data is fourteen percent outside it.”

“And seventeen participants have better anchoring than we projected. That gives us somewhere to look.”

“In a direction we wanted,” she said. “That isn’t the same as understanding why it happened.”

Morse looked up.

“Then the imaging may show us which assumption was too conservative. That would be useful well beyond this cohort.”

“Of course it is. I want us to use it to correct the model.”

Halvorsen let the exchange continue. Friction between these two could be productive, provided neither began arguing with a version of the other that existed only in memory.

Levin cleared his throat.

“We’re comparing the current imaging against our tissue models. The leading hypothesis is stronger reinforcement of activity at the implant interfaces, allowing the lattice to integrate more deeply than it did in the adult trials. We should have additional imaging by the end of the week.”

“Good,” Reyes-Galindo said.

Morse returned to his display.

Nkosi wrote something down.

Patel brought up the psychological findings.

She had revised her opening twice before entering the room. The first version sounded reassuring. The second sounded as though she had already reached conclusions she could not defend.

Neither was acceptable.

“Continuous behavioral monitoring, forty-seven individual sessions across both cohorts, eleven group sessions, and ambient telemetry through the integration layer.”

Heat maps appeared above the table. Different colors, different scales. The same young people rendered several ways.

“I’ll address identity stability, social dynamics, emergent behavior, and one event I want us to examine separately.”

“Can we start with the event?” Morse asked. “It may change how we interpret the rest.”

“I’d prefer to establish the context.”

He inclined his head.

“Both cohorts remain within our predicted range for identity disruption. In the adult trials, we saw a period of heightened self-questioning followed by stabilization at a modified baseline. That broad pattern appears to hold.”

She advanced the display.

“The timing does not. Adults averaged eleven weeks. Here, we’re seeing apparent stabilization in one to seven.”

“Another reason to examine the plasticity difference,” Morse said.

“Possibly. But these participants also eat together, sleep in the same rooms, see one another after a bad session. Most of the adults went home alone with the changes. We need to account for that.”

“That would shorten the vulnerability window,” Nkosi said.

“If what we’re measuring is stabilization.”

Patel let them look at the curves.

“A student stops reporting distress. Perhaps she understands what’s happening to her. Perhaps she’s found one friend she can sit beside at dinner, and that makes it bearable.” Patel indicated the convergence of the curves. “Both can look like this. I need to know which one I’m seeing before I say she has stabilized.”

“Are you seeing evidence that it hasn’t?” Morse asked.

“I can’t tell from these measures. In a session, a student can sound settled until I ask what happens when the others aren’t there.”

He waited.

“I’m not ready to call that part resolved.”

She moved to the next display.

“Cohort Two has developed unusual cohesion. Shared stress explains some of it. It doesn’t adequately explain the timing of certain decisions. We’re recording synchronization events and behavioral influence without explicit communication.”

She had decided not to identify the two participants in the overview. She could justify that choice clinically. Sitting across from Cohort One’s lead psychologist, she was aware of another reason she had made it.

“Two participants have influence profiles extending beyond their immediate friendships.”

“Define influence,” Reyes-Galindo said.

“One appears to change how a group reaches a decision through social attention. She doesn’t have to argue for an outcome. The way she attends to people seems to alter what they consider, and in what order.”

Morse had stopped moving through his own display.

“The second shows early signs of something I’m tentatively calling attentional gravity. When he commits to a choice, nearby participants tend to make compatible choices. Some of that convergence appears to precede the verbal and behavioral signals we can identify.”

“Appears to,” Reyes-Galindo repeated.

“Yes. We haven’t ruled out signals our instruments are missing. Or common inputs producing similar responses.”

Levin unfolded his hands.

“That isn’t in the efficacy literature.”

“No.”

“It is in the emergent-effects literature,” Morse said. “Nonverbal synchronization. Intuitive modeling of proximate states. We included it in the submission.”

“As speculative and inconsistent,” Nkosi said. “Those were the words.”

“And now we may have enough consistency to study what brings it on.”

“Yes,” Patel said. “But before we tell a child she’s influencing her friends, we need to establish that she is. We’ve observed convergence. We haven’t established who causes it, or how.”

“Then give us conditions under which we can distinguish the explanations.”

“I’m working on them. Meanwhile, I don’t want the first explanation becoming the way everyone treats her.”

Nkosi turned a page in her notepad.

“We also need to establish what these findings require us to disclose. A possibility described in a submission becomes a different obligation when you think you’re observing it.”

Morse leaned back slightly.

“Then characterization gives us both: a better account for the families and a basis for the next study. I don’t see why an unexpected improvement should, by itself, suspend the work.”

“Neither do I,” Nkosi said. “I see a reason to characterize it before deciding it’s an improvement.”

Halvorsen watched Patel’s hand resting beside the controls.

“Bring up the injury,” she said. “What happened, what was authorized, and where the two diverged.”

Patel changed the display.

Leena Park’s scans filled the space above the table.

For a moment, no one spoke.

“Confirmed loss of consciousness following the inter-cohort exercise,” Patel said. “Imaging within thirty minutes showed left temporal swelling and cortical microhemorrhages. The initial assessment anticipated at least seven to ten days of recovery, followed by a monitored return to cognitive and physical load.”

A second set of images appeared beside the first.

“Within four hours, her neurological examination had normalized. The imaging changes were already resolving.”

Levin leaned forward.

“I reviewed the full telemetry. There was no scheduled intervention. No physician authorization at initiation. Nothing passed through the standard approval chain.”

His eyes remained on the images.

“The treating team recognized the activity after it began and chose to continue under observation. But Catalyst initiated it.”

“It acted before the inflammatory response had fully developed,” Reyes-Galindo said.

“That is what the sequence indicates.”

“Anticipatory neuroprotection.”

“A possible description,” Levin said. “We still need to be careful about how much intent we place inside it.”

Reyes-Galindo nodded once.

“It’s not a capability we explicitly built,” he added.

“We built an architecture capable of learning beyond explicit instructions,” Morse said.

Nkosi’s pen stopped.

“Elric. Did you know this particular response was possible?”

“I hypothesized it. The adaptive architecture has theoretical access to pathways involved in vascular regulation. I included a capability framework in the steering committee paper fourteen months ago.”

“That paper concerned performance optimization.”

“Yes.”

“Not autonomous treatment of an acute brain injury.”

“The pathways don’t divide themselves according to the headings in our documentation.”

Reyes-Galindo set down her stylus. Even on the sound-softening table, the contact was audible.

“A fifteen-year-old sustained a brain injury. An implanted device began treating it before anyone responsible for her care knew what it was doing.”

“Before the damage developed as far as the initial assessment predicted,” Morse said. “Her condition improved.”

“It did.”

“If we can establish what protected her, that matters to the next injury too.”

“So does what would have happened if its inference had been wrong.”

Morse looked back at the paired scans.

“I’m not dismissing the authorization failure. I am saying the outcome belongs in the assessment.”

“Leena is glad it helped her too,” Patel said. “She still woke up to find that something had acted inside her head without asking.”

She could still see Leena in the medical suite, trying to fit those two things together.

“What do we tell her it will do next time? Or what she should watch for tomorrow? We don’t know the later effects. We don’t know whether this response would help with a different injury.”

“No,” Morse said after a moment. “We don’t promise her that. We follow what happened closely enough to begin answering it.”

Nkosi looked down at the motionless pen in her hand.

Then she began writing again.

Morse presented Cohort One’s data himself.

He moved beside the projection column and advanced the display with the ease of someone who knew the order of the slides without checking.

His authority had been earned over decades. Olympic training programs in three countries. A decade in cognitive enhancement research at DARPA. The private sector after that, where, as he had once said in a funding interview, the constraints were more honestly negotiated.

Two hundred and two peer-reviewed papers. Four books. One retracted study, which he maintained had failed an ethics board’s vocabulary rather than its scientific standards.

He had never resubmitted it.

Patel had watched him work for years. He was especially good at arranging a question so that the answer, when it came, seemed to have arisen without assistance.

“Cohort One exceeds Cohort Two on eight of eleven primary performance measures,” he said. “Motor output. Decision throughput under adversarial stress. Coordination efficiency. Cognitive stamina during extended demand.”

The overview settled into comparative curves.

“The differences remain after adjustment for their earlier implantation date. Against the projection model, several results are already approaching the levels we expected at twenty-four months.”

“Which reference points?” Reyes-Galindo asked.

“Population baseline, individual pre-trial baseline, and the Phase One adult comparison.”

Nkosi looked up.

“At three months.”

“Yes.”

Patel studied the separation between the predicted and observed curves.

“What are they telling you after those sessions?”

Morse turned toward her.

“We review fatigue, recovery, their own assessment of performance. Those records are available.”

“Do they feel ready for the next increase, or do they think being ready is what keeps them ahead of Cohort Two?”

“That’s a question worth asking. Withholding the next challenge won’t necessarily answer it.”

“Neither will giving it to them before we’ve asked.”

“We can ask and still test what they’re capable of.”

“Provided there’s time to listen to the answer.”

The ventilation changed pitch. Somewhere inside the wall, a relay clicked.

Halvorsen brought up the exercise index.

“The arena,” she said. “Which limits failed, and what are you changing?”

Morse glanced at the new display. Halvorsen caught a brief tightening beside his mouth.

“The aggression in Cohort One exceeded sanctioned parameters. That is documented. I’m addressing it through individual coaching.”

“When Calder talks about the collision, what does he think he could have done differently?” Patel asked.

“We’re working backward from the impact. Identifying where he still had another choice, then rehearsing it under equivalent pressure. He needs to be able to find it while moving, not only in a replay.”

“And the inhibitory protocols?” Levin asked.

“I’ve requested a review. We appear to be constraining some physical outputs without adequately constraining the choices that make those outputs dangerous.”

Nkosi held his gaze.

“Your recommendation includes raising Calder’s output ceiling.”

“In controlled testing. If the physical limit is badly fitted, changing his coaching won’t tell us. If the failure is behavioral, lowering the ceiling won’t teach him where to stop. I want to test those separately.”

“A student was injured by his current output.”

“By how he used it.”

“And you’re confident the two can be separated before you give him more?”

“I’m proposing that we establish whether they can.”

Reyes-Galindo folded her arms.

“With whom in the test environment?”

“Instrumented targets initially.”

“Initially needs a separate authorization,” Nkosi said.

Morse nodded.

Halvorsen let the condition remain on the table before moving on.

“What evidence do we have that the preparation itself failed?”

Morse brought up the arena’s early phases.

“Cohort Two was winning comprehensively before the physical escalation. Their adaptive coordination was better than anticipated. Cohort One’s response was worse.”

Positions moved across the field model. Support arrived late. A lane opened where the plan required it closed.

“The pattern suggests compromised preparation. Their decisions were internally consistent and collectively wrong.”

“Or they were responding poorly to an opponent they underestimated,” Patel said.

“That could explain some of it. Not all.”

“You believe someone altered their preparation,” Halvorsen said.

“Yes. Hale reached the same conclusion independently. He’s identified anomalies in their planning materials that he believes came from outside the cohort.”

Patel’s attention narrowed.

“He brought them to you?”

“He did. He wants to reconstruct what happened before making an accusation. I’ve agreed to let him pursue it.”

“Under what boundaries?” Nkosi asked.

“Observation and review of his own cohort’s materials. No confrontation on the basis of a theory.”

Patel looked at the frozen field model.

“They expected to win. Then they couldn’t make the plan work. Have they had to explain that to one another without first being offered sabotage as the reason?”

“They have. And if the materials were altered, that also belongs in the explanation.”

He cleared the display.

“We have the unresolved findings,” Halvorsen said. “Now the progression decision.”

The new projection carried no curves or scans. Its simplicity changed the room.

CATALYST ADOLESCENT INTEGRATION TRIAL: PHASE PROGRESSION

PLANNED APPROVAL DATASET: 8 COHORTS / 96 PARTICIPANTS

CURRENT STATUS: 2 COHORTS IMPLANTED / 24 PARTICIPANTS

PROPOSAL: AUTHORIZE PREPARATION AND IMPLANTATION OF COHORTS 3 AND 4

“We’re eight weeks ahead of the original milestone schedule,” Halvorsen said. “Does this evidence support expansion? If your answer is conditional, specify what has to happen first.”

Nkosi answered before anyone else moved.

“Not yet.”

She looked at the projected number.

Twenty-four more.

“Our longest observation period is three months. We have an autonomous intervention whose implications remain unresolved. Behavioral effects we have not adequately characterized. Integration moving faster than the model accounts for.”

Her notepad lay open beneath her hand.

“I don’t recommend adding twenty-four adolescents before we have addressed those findings.”

“The next cohorts can tell us which findings repeat, and which belong to these particular students,” Morse said. “Keeping the sample at twenty-four limits that work too.”

“Part of it. Not the part that must happen first.”

“We have exceptional integration, cognitive gains above projection, and an unanticipated response followed by rapid recovery. Those findings need scrutiny. They also give the next cohort a stronger basis than the first had. I think that warrants a serious discussion of acceleration.”

“And this is a serious discussion,” Nkosi said. “About a device designed to remain in a developing nervous system while continuing to change its behavior.”

She tapped the edge of her terminal.

“We cannot stop exposure by ending a course of medication. Removing the implant is another procedure, with its own risks. That affects the weight I give to three months of reassuring data.”

“Of course it does.”

“Then efficacy doesn’t answer the expansion question by itself.”

“No one is proposing that it does.”

Patel watched Nkosi take a breath before speaking again.

“We do not have long-term safety data. We do not know how adolescent development and continued lattice integration will affect each other over years. We cannot describe the full risk the next group would be accepting.”

Reyes-Galindo leaned forward.

“We need to separate two questions. Do we understand Catalyst’s current behavior? And do we understand enough to expand under an adequate oversight framework?”

Halvorsen nodded for her to continue.

“The answer to the first is no. We haven’t characterized the intervention in Park. We haven’t explained the influence profiles. Even the faster integration lacks a settled mechanism.”

She glanced at Levin, who nodded.

“But sequential cohorts exist because a complete picture doesn’t emerge from the first group. We expand while investigating. The decision turns on what we can detect, what we can interrupt, and who has the authority to make us stop.”

“What will the next family actually be told?” Patel asked.

She had kept her secondary display dark since presenting. Now she touched its edge without activating it.

“They need to hear what happened to the students already here. A general warning about unexpected effects won’t do that.”

“Then we have something concrete to improve before recruitment: the effects we’ve observed, and what we still can’t explain.”

“What would you tell them about the social influence findings?”

“What you’ve told us. We’ve observed decision synchronization. The cause is unresolved.”

“And what that could feel like? A girl wondering whether her friends agreed because they wanted to, or because she did something to them?”

“If that remains a defensible hypothesis, yes.”

Patel looked at him.

The answer was reasonable. It did not remove the difficulty.

“A parent hearing ‘social synchronization’ may think we mean teamwork,” she said. “A student may think we mean understanding their friends better. Neither necessarily imagines becoming uncertain whether a choice was their own.”

“Then use those words.”

She sat still.

Around the table, the room waited for her to accept an answer that had, on its surface, given her what she asked for.

“There’s another problem,” she said.

Halvorsen’s eyes moved to her.

“What I’ve documented has not yet gone to the external ethics board. We’ve been waiting to agree on what it means.”

She could feel the cost of the next sentence before she said it.

“I’ve been trying to understand what these students are telling me before putting a label on it. While I’ve been doing that, the findings have stayed here. Now we’re discussing another twenty-four children.”

She looked toward Nkosi.

“The external board should already have this. Including the uncertainty. Including my part in the delay.”

No one spoke.

Levin finally shifted in his chair.

“From the surgical standpoint, the risk is low and increasingly well characterized. Twenty-three uncomplicated recoveries, one minor inflammatory event that resolved.”

He looked at Leena’s name in the index still visible on his terminal.

“My uncertainty begins after implantation. I want more time with the Park telemetry before performing the procedure on another twenty-four adolescents.”

“Time for what specifically?” Morse asked.

“To determine whether we can recognize the initiation of that response earlier. Whether it can be bounded. Whether stopping it once active would be safer than allowing it to continue.”

Morse nodded.

“Those are questions we can put into the conditions.”

“And answer before surgery,” Levin said.

“Yes.”

Nkosi watched the exchange.

“More monitoring only helps if we have the capacity to review it. We already collect more than this committee can meaningfully examine.”

“Then we increase review capacity,” Morse said.

“Before we increase enrollment.”

“Alongside preparation. There will be months of screening before anyone is implanted.”

He turned toward the center of the table.

“We have room between halting the program and implanting the next cohort tomorrow. Use it. Better models. More reviewers. An external panel with authority. Those are meaningful constraints, and I’m prepared to work within them.”

Nkosi opened her mouth, then closed it.

She wrote several lines.

Halvorsen waited until her pen stopped.

Reyes-Galindo looked toward her.

“What are the proposed terms?”

“Screening can proceed,” Halvorsen said. “Implantation cannot proceed on today’s evidence alone. These are the requirements.”

She advanced the display.

PROPOSED CONDITIONS FOR ADDITIONAL COHORT AUTHORIZATION

PARTICIPANT SELECTION: EXTENDED PRE-SCREENING; MINIMUM 90-DAY OBSERVATION BEFORE IMPLANTATION

MONITORING: EXPANDED CONTINUOUS TELEMETRY; DAILY CLINICAL REVIEW; INCREASED REVIEW STAFFING

CONSENT: SPECIFIC EMERGENT-EFFECTS ADDENDUM FOR PARTICIPANTS AND GUARDIANS

HOLD CRITERIA: ANY NEW UNAUTHORIZED AUTONOMOUS MEDICAL INTERVENTION; ANY PARTICIPANT WITHDRAWAL REQUEST, PENDING INDEPENDENT REVIEW

EXTERNAL OVERSIGHT: ANOMALOUS DATA REFERRED WITHIN 48 HOURS; PANEL AUTHORITY TO REQUIRE A HOLD

PRE-IMPLANTATION REQUIREMENT: REVIEW OF THE PARK EVENT AND APPROVAL OF RESPONSE-MANAGEMENT PROTOCOLS

Patel read the list once, then again.

“Who sits on the panel?”

“We’ll propose members drawn from the external advisory roster,” Halvorsen said. “Names, conflicts, and authority disclosed to this committee before authorization becomes final.”

“Before screening or before implantation?”

“Initial recruitment and screening can begin under the existing approval. No new implantations until the conditions are met.”

“I want the consent language before it goes to families.”

“Agreed.”

“And the findings supporting it. Including the unresolved ones.”

“Yes.”

Patel nodded. She had not yet agreed to the expansion.

Nkosi was still looking at the hold criteria.

“We have already had an unauthorized intervention,” she said.

“That event is now a specific pre-implantation review condition.”

“It should have triggered a hold before we reached this meeting.”

Halvorsen did not answer immediately.

“Enter that reservation,” she said.

Nkosi’s gaze stayed on her.

“And the next event stops progression. It doesn’t merely produce another discussion about whether the outcome was favorable.”

“The next qualifying event triggers the hold. A favorable outcome doesn’t cancel it.”

“Existing participants continue receiving care. New implantation and escalation stop until the external panel reviews the event.”

“Put that exact scope in the record,” Halvorsen said.

Nkosi set down her pen.

“I want my position recorded in full. I believe we are approaching the limit of our analytical capacity. I do not consider the current consent model adequate to the effects observed. The Park intervention should have prompted automatic external review before expansion was placed on the agenda.”

The light from the projection lay pale across her notepad.

“I won’t block a conditional authorization if the conditions are binding and the panel is independent in practice. My reservations remain formal.”

“Recorded,” Halvorsen said. “Sofía?”

Reyes-Galindo studied the list.

“The panel is the difference for me. It must be able to halt the work without needing this committee to agree with its reasons.”

“Yes.”

“Then I support conditional progression. If its direction is ignored, I withdraw my support and file the objection externally.”

Morse nodded.

“Yusuf?”

Levin took longer.

Halvorsen watched his eyes move from the ninety-day screening period to the final condition. She knew the expression. He was putting an actual patient in the place of the proposed one and asking whether his answer changed.

“The screening period matters,” he said. “So does resolving the response-management question before surgery. Under those conditions, I can support it.”

“Ananya?”

Patel looked at Halvorsen, then at Morse.

“Conditional. The addendum and its supporting disclosures require my review before this becomes final. That is a condition of my vote.”

“Recorded as such. Elric?”

“Yes. Authorize Cohorts Three and Four under the terms listed.”

Halvorsen looked around the table one last time.

No one spoke.

“Conditional progression is approved, with Dr. Nkosi’s reservations entered in full. Cohort Three pre-screening begins Monday. Implantation remains contingent on every listed requirement.”

She closed the decision screen.

For a moment, the room seemed dimmer than it had before the meeting.

Chairs moved. Terminals darkened. People exchanged the small practical words that followed decisions too large to carry comfortably into a corridor.

Morse remained seated.

Halvorsen waited as the others left. Nkosi was last. The door sealed behind her with a soft, definitive sound.

Then Halvorsen sat.

Morse looked up.

“Say it.”

“The Park response,” she said. “You knew it was possible.”

“I hypothesized it.”

“Elric.”

He rested his hands on the table.

“I ran probability models eighteen months ago. Adaptive inference, pathways involved in vascular regulation, acute physiological stress. The strongest projections involved a highly integrated adolescent nervous system.”

“How strong?”

“Strong enough to investigate. Not strong enough to call it a capability.”

“And you put it in Appendix C.”

“It was a ninety-page framework. The appendices contained the modeled extensions.”

“You’ve been writing papers long enough to know where people stop reading.”

He looked toward the dark projection column.

Halvorsen could hear the ventilation. Without the displays, the room’s warmth felt more noticeable.

“You did this with the Karolinska study,” she said. “The relevant possibility was in the documentation. Technically available to everyone. Nowhere they were likely to understand its weight.”

“That isn’t why the study was challenged.”

“It’s why I’m asking you.”

He said nothing.

“You wanted to see whether the response would emerge.”

“Under conditions we could observe.”

“A child was injured.”

“I did not want a child injured.”

Halvorsen watched him. She believed that he meant it. She was less certain what the belief settled.

“Then you needed to make the possibility unmistakable before we put children in those conditions.”

Morse drew a slow breath.

“Miriam, I’ve spent thirty years studying what human beings can do. Every method gives us part of the answer. A response under one test. An average across a sample. A limit that may belong to the person or to the instrument.”

His voice remained low.

“Catalyst lets us observe adaptation inside an individual nervous system as it happens. We can see the approach to a limit, not just the result of crossing it. That is new. And yes, I want to know how far it goes.”

“What evidence would make you stop?”

“Where do you set that point when the model understates what they can do? Stop at its prediction, and we never find out why it was wrong.”

“You’re asking to move a limit we haven’t explained.”

“I’m asking to study how they adapt when they reach it.”

She leaned back. The chair gave slightly beneath her, an unfamiliar movement after so long standing at the head of this table.

“The protocols aren’t there to prevent adaptation. They’re there because someone has to bear the consequences.”

“I know who the participants are.”

“Then keep them in the sentence.”

He looked at her.

For a while, neither spoke.

Finally Morse said, “Caution can become a way of preserving a model long after the evidence has moved beyond it.”

“And ambition can become a way of refusing to see what the evidence costs.”

He accepted that without answering.

Halvorsen rubbed a thumb along the edge of the table. Eight years ago, they had spent an entire night arguing over a question neither could test. She remembered cold coffee, a failing screen, the exhilaration of realizing the other person could follow her farther than anyone else had.

She still knew what he meant.

That was part of the difficulty.

“What is the question now?” she asked. “The one beneath the framework.”

Morse looked at the dark terminal across from him.

“Whether Catalyst can take a person to the edge of their potential,” he said. “And whether it can recognize that edge before going beyond it.”

Halvorsen let the answer stand between them.

Then she rose.

At the door, she turned back.

“The panel gets Appendix C with the rest of the submission. Flagged. Explained. No one has to discover what we already knew.”

“Yes.”

“And whatever you propose for the next phase comes to this committee before it begins.”

Morse met her eyes.

“Understood.”

She left him in the windowless room.

On his terminal, the data still glowed: neural maps, performance curves, the extraordinary acceleration of young nervous systems. Beneath each curve sat a participant identifier.

From the doorway, they were too small to read.

Down the corridor, Patel had not left the building.

She sat in her office with the door closed and a set of notes open before her. She had written the first lines at two in the morning three days earlier, after circling the same thought until exhaustion made it easier to put down.

At the time, she had not known whether she was being rigorous or frightened.

She was still not certain.

At the top of the page, underlined twice:

CATALYST IS NO LONGER JUST RESPONDING.

IT IS LISTENING.

WHAT HAPPENS WHEN IT BEGINS TO ANSWER?

She read the words and disliked their imprecision. Listening was a metaphor. The device had always received input. Answering could mean a hundred different things.

She had kept them because the clinical language did not yet hold the question she was trying to ask.

Below them, added before the meeting in faster strokes:

MORSE IS RUNNING A DIFFERENT EXPERIMENT THAN THE ONE WE FILED.

Patel rested the stylus against the desk.

That was not a finding.

She added beneath it:

HYPOTHESIS. INTENT NOT ESTABLISHED.

The qualification did not make the sentence above it disappear.

She thought about the authorization. The meeting had reached the outcome Morse wanted, accompanied by conditions the rest of them had insisted on. Some were substantial. She believed they could matter.

She also could not point to a moment when he had seemed surprised by any of them.

Had they changed his proposal, or made it possible for themselves to accept it?

She had asked a version of that question about a fifteen-year-old girl.

The resemblance troubled her. It did not prove anything about either of them.

Through the wall, the building settled into evening mode. The climate system cycled. Light beneath her door softened as the corridor dimmed.

Somewhere farther down the hall, a door closed.

Patel returned to the sentence she had spoken in the meeting.

Including my part in it.

She had not planned to say that. It had been true, and for once she had allowed the truth to arrive before she found a more manageable version.

Saying it did not undo the delay.

She had withheld findings while trying to understand them. She could defend the reasons individually: uncertain attribution, incomplete evidence, the risk of labeling a child with an effect she might not possess.

Together, the reasons had kept the information inside the Institute.

Now the Institute had voted to grow.

Patel opened a new document through her retained academic affiliation at Johns Hopkins.

She had maintained the account for collaborative work on adolescent psychiatric outcomes. The login was familiar. So was the brief authentication delay. Tonight, she watched the progress indicator as though it might ask a question it had never asked before.

She would not place raw participant records here. No names. No telemetry exports. Those required a disclosure process she had just insisted they follow.

What she needed was a dated account of the decisions: what had been raised, what remained unresolved, what had been promised, and what she herself had agreed to.

A record that did not depend on Neurovia’s systems continuing to preserve it.

She titled the document:

OBSERVATIONS: INDEPENDENT.

The word held more than she wanted it to.

Independent of the monitoring infrastructure. Independent of the committee’s minutes and their careful summaries. Independent, as far as she could make it, of the version she might later tell herself about why she had voted conditional instead of no.

She could not make the document independent of her.

That was why the first entry would include her own decision.

Patel looked toward the closed door.

She knew how much the Institute recorded. Estimated subvocal activity. Physiological traces. Patterns of movement, proximity, attention. She knew the limits of some layers and suspected others existed beyond her access.

An external login would leave a trace. She had no reason to pretend otherwise.

She turned back to the page.

For a moment, the cursor was the only thing moving.

Then she began to write, carefully, as if someone might one day need to know exactly what had been said.