Chapter Twenty Three: Review

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

Dr. Miriam Halvorsen designed it that way two years ago, when the Institute was still a blueprint and Catalyst was still a theory. Windows invited distraction. Distraction invited imprecision. Imprecision in work like this was not an aesthetic failure. It was a moral one.

The table ran long and pale, milled from a single slab of composite material that absorbed sound and reflected light in equal measure. Seven chairs. Seven data terminals, each running a live feed from the Catalyst monitoring infrastructure. A central projection column rose from the table's midpoint, dark.

Six people had arrived. The seventh seat sat empty.

Halvorsen stood at the far end, reviewing the summary index on her terminal without sitting. She started meetings standing. She had done it for years. A room where the person running it had not yet settled communicated something useful: the work was not finished, the posture of everyone present should remain open.

Dr. Ananya Patel sat to her left, posture composed, a secondary display angled at forty-five degrees in a way only she could see. She had been at the Institute since its second year of operation. Her stillness came not from passivity but from years of learning to watch before speaking, and to speak only when the watching had finished.

Dr. Yusuf Levin, lead implantation surgeon, sat beside her with his hands folded on the table. He rarely spoke before the data was presented. Halvorsen respected that about him, as she respected the other things about him: his precision in the OR, his refusal to let anyone rush a postoperative review, the way he carried the moral weight of his work without performance.

Dr. Sofía Reyes-Galindo had taken the corner seat. She had arrived second, after Halvorsen, and had bypassed the chairs with better sightlines without appearing to consider them. She was Neurovia's director of cognitive integration and had co-authored the theoretical framework on which Catalyst's adaptive architecture was built. In Halvorsen's private estimation, she was the only person in the room reliably right when she pushed back, which made her the most important person in it.

Dr. Priya Nkosi, regulatory affairs and trial ethics, sat across from Reyes-Galindo with a physical notepad open beside her terminal. She wrote in a shorthand nobody else had ever decoded, and the things she wrote down had a way of becoming significant later.

At the near end of the table, set apart in the way that some people are always set apart, sat Dr. Elric Morse.

He had arrived seven minutes early. Three minutes ahead of the room's climate system activation schedule, he had overridden it manually. The room had been at his preferred temperature for twenty minutes before anyone else walked in.

He had co-founded Neurovia with Halvorsen eight years ago. He was the lead psychologist for Cohort 1.

Halvorsen activated the projection column. The room's ambient light shifted as the display came up.

“Three months,” she said. “Let's begin.”

* * *

The data resolved in layers.

Catalyst integration metrics came first. Eighty-one variables across all twenty-four active participants, sorted by Cohort, then by individual, then by domain. Neural coherence indices. Adaptive latency curves. Sleep architecture evolution. Cognitive throughput baselines and their rates of change over twelve weeks. The numbers were clean. The trends were not ambiguous.

Levin presented implantation outcomes without preamble. He read from the surgical record in a flat, unhurried voice, his eyes moving down the document at a pace that suggested he had already reviewed it several times and found nothing in it that required emphasis or apology.

“Twenty-four procedures. Zero structural complications. One acute inflammatory event in Cohort 2, resolved within four hours of detection, no residual deficit. One concussive injury in Cohort 2, participant Park, which triggered an unscheduled adaptive response we'll return to.” He advanced the display. “Lattice anchoring exceeded projection in seventeen of twenty-four cases. Sub-optimal anchoring in two. Both Cohort 1, both within acceptable parameters, both being monitored. Synaptic calibration efficiency runs fourteen percent above Phase One adult trial baselines.”

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

“No,” Levin agreed. “Adolescent neuroplasticity appears to interact with the lattice architecture differently than the adult models predicted. The synaptic junction density in this age range is…”

“More receptive,” Morse said, without looking up from his display.

Reyes-Galindo studied him. “More receptive is one word for it. Faster integration is another. I want to understand the mechanism before I endorse the framing.”

“The mechanism is plasticity,” Morse said. “We knew adolescent brains were more malleable. We designed for it. The data is confirming the design.”

“The data is outpacing the design by fourteen percent. Those are different things. Confirmation is when outcomes land where you predicted. This is something else.”

Morse looked up. He looked at her like he looks at a road he has driven too many times.

“Sofía. Everything in science that advances does so by outpacing its own projections. That is not a warning sign. That is the success criterion.”

“That's a formulation, not an argument. I want the mechanism.”

Halvorsen did not intervene. She rarely did when these two were running. The friction produced something she had found more useful than either of them alone.

Levin cleared his throat. “We're running comparative histological models. Best current hypothesis is enhanced long-term potentiation density at the synaptic junctions. The lattice is integrating more deeply than it does in post-adolescent tissue. Confirmatory imaging data should be available by end of week.”

“Thank you,” Reyes-Galindo said.

Morse returned to his display.

* * *

Patel presented the psychological findings.

She chose her words carefully; recognizing the decisions made impacted people who had no idea this meeting was happening.

“Three months of continuous behavioral monitoring, supplemented by forty-seven individual sessions across both Cohorts, eleven group sessions, and ongoing ambient telemetry through the Catalyst integration layer.” She advanced through a series of heat maps, each one a different way of looking at the same underlying question. “I'll address four areas. Identity stability. Social dynamics. Emergent behavior indicators. And one anomaly I want to flag separately.”

“Start with the anomaly,” Morse said.

“I prefer to contextualize first.” She did not look at him.

Morse inclined his head by a degree.

“Identity stability. Both Cohorts are within predicted parameters for Catalyst-induced identity perturbation. The disruption model we built from adult trial data holds. Participants experience a period of heightened self-questioning, followed by re-stabilization at a modified baseline. What's different here is the speed. Adult trials averaged eleven weeks. Current Cohorts complete the cycle in one to seven.”

“Plasticity again,” Morse said.

“Possibly. Or social acceleration. These participants are doing this together, which adult trial subjects were not. Peer dynamics may be compressing the timeline. Participants working through perturbation in relation to each other rather than in isolation.”

“That's encouraging,” Nkosi said, looking up from her notepad. “Faster stabilization means shorter vulnerability windows.”

“Or it means we're not seeing the full perturbation. Some of it may be resolving socially rather than psychologically, which can present as stability while masking incomplete integration.” Patel held the pause. “I'm not saying that's what's happening. I'm saying I cannot currently distinguish between the two, and I think that matters.”

“You're speculating,” Morse said.

“I'm flagging. That's what I'm here for.”

She moved to social dynamics.

“Cohort 2 has developed unusual cohesion. Not the kind that comes from shared stress, though that's present, but something more structural. Decision synchronization events. Behavioral influence without explicit communication.” She paused, noting the convention that required her not to name individual participants in a room that included their academic competitors' lead psychologist. “Two participants in Cohort 2 show influence profiles that extend beyond their immediate social circle.”

“Define influence profiles,” Reyes-Galindo said.

“One participant appears to affect group decision convergence when she communicates socially, even when the communication is not directive. Her presence in a conversation seems to shift how the group arrives at conclusions. Not by arguing toward them, but by the quality of her attention. A second participant shows early indicators of what I'm calling, tentatively, attentional gravity. When he commits to a choice, proximate decision-makers converge toward compatible choices without being asked or informed. The convergence appears to precede any verbal or behavioral signal he could plausibly be sending.”

Silence settled over the table.

“That's not in the efficacy literature,” Levin said in the careful tone of a man identifying something he would prefer to be wrong about.

“No. It isn't.”

“It's in the emergent effects literature,” Morse said. “It's precisely what the adult trials hinted at. Non-verbal synchronization, intuitive modeling of proximate states. We noted it in the FDA submission.”

“We noted it as speculative and inconsistent,” Nkosi said. “Those were the exact words. Speculative and inconsistent.”

“And now we have two cases in three months in adolescent subjects. Which is no longer speculative.”

“It's also not understood,” Patel said. “And I want to be careful about that distinction. Documented and understood are not the same thing. We have observed something. We do not know what it is, what its mechanism is, what its trajectory is, or what it will look like at twelve months or twenty-four. Calling it no longer speculative gives us more purchase than the data actually provides.”

“Understanding follows observation. We observe, we document, we adjust the framework. That is the process.”

“It is also the ethical process,” Nkosi said, “that requires us to slow down when we encounter effects we cannot characterize. Those two things are not in conflict unless someone decides they are.”

Morse looked at her. His expression did not change. “Priya. If we slow down every time an outcome exceeds projection, we will never have outcomes worth slowing down for.”

Nkosi set her pen down. “I don't think that sentence means what you intend it to.”

Halvorsen spoke for the first time since opening the meeting. “The anomaly, Ananya.”

* * *

Patel pulled up the neurotrauma data.

“Participant Park sustained a Grade Three concussive injury during the inter-Cohort exercise. Loss of consciousness confirmed. Imaging obtained within thirty minutes showed left temporal swelling and cortical micro-hemorrhages consistent with a recovery timeline of seven to ten days minimum, with monitored return to cognitive load protocols.” She advanced the slide. “She was neurologically clear in under four hours.”

The room went quiet. Not the quiet of disagreement or processing. The quiet of arriving at the same conclusion and waiting to see who would acknowledge it first.

Levin leaned forward. “I reviewed the telemetry myself, in full, before this meeting. There was no scheduled intervention. No physician authorization. No flagging through the standard adaptive response approval chain.” He looked at the table. “Catalyst initiated a targeted neurological response on its own inference architecture. Without being asked.”

“It performed anticipatory neuroprotection,” Reyes-Galindo said slowly. “Before the inflammatory cascade had fully expressed itself.”

“That's not a capability we built,” Levin said.

“We built the architecture that learned it,” Morse said.

Nkosi's pen stopped moving.

“Elric.” She did not raise her voice. She rarely raised her voice. “Please tell me you did not know this was in the capability set.”

Morse met her gaze. “I hypothesized it was possible. The adaptive inference architecture has theoretical access to vascular modulation pathways. I presented a capability framework paper to the steering committee fourteen months ago.”

“That paper was about performance optimization. Not autonomous medical intervention.”

“The line between the two is much narrower than we acknowledge in regulatory filings.”

Reyes-Galindo set her stylus down. The small sound it made was the loudest thing in the room. “Elric. This is a fifteen-year-old child. Catalyst performed an unsanctioned neurological procedure on a child, without authorization, without consultation, without anyone in this room knowing it was going to happen.”

“Catalyst,” Morse said, “prevented a young person from spending ten days in pain and cognitive impairment.” He paused, “I am genuinely struggling to identify the harm.”

“The harm,” Patel said, “is that none of us, including Catalyst, knew what the downstream effects of that intervention would be. You cannot measure harm that has not expressed itself yet. The fact that it has not expressed yet is not evidence that it will not.”

Morse was quiet for a moment.

“Then we continue to monitor. Which we are doing. Which we were always going to do.” He looked around the table with the steadiness Patel had been tracking for the last forty minutes. “This is what a trial is. We proceed. We observe. We learn from what we see. That is not recklessness. It is method.”

* * *

Cohort 1 data came next. Morse presented it.

He moved to the projection column without thinking about it. He advanced slides without apparent effort, his voice and presence carrying earned authority. Patel had observed Morse for years, watching his career arc resolve into a pattern. Olympic training programs in three countries. A decade at DARPA's cognitive enhancement division. The private sector after that, where, as he had once phrased it in a funding interview, the constraints were more honestly negotiated. Two hundred and two peer-reviewed papers. Four books. One retracted study, retracted not because it was wrong but because the ethics board reviewing it had not yet developed adequate language for what it was describing. He had never resubmitted it. He had never needed to.

In every room Patel had watched him work, the pattern held. He did not push for outcomes. He created the conditions under which outcomes occurred and then presented them as discoveries.

“Cohort 1 outperforms Cohort 2 on eight of eleven primary performance metrics at the three-month mark.” He let the overview display settle before advancing. “Motor output. Decision throughput under adversarial stress. Team coordination efficiency. Cognitive stamina across extended high-demand tasks.” He advanced again. “These results are not marginal. They represent a statistically significant divergence from baseline projections across all three reference points.”

“Which baseline,” Reyes-Galindo asked.

“All three. Population baseline, pre-trial individual baseline, and the Phase One adult comparison baseline.” He let that sit for a moment. “Cohort 1 is performing at levels we projected for participants twenty-four months into the integration protocol.”

“At three months,” Nkosi said.

“At three months.”

“That should concern you,” Patel said.

Morse looked at her. “Why.”

“Because acceleration that far ahead of projection means either your model was wrong, or something is proceeding faster than we have the capacity to understand and respond to. Both are reasons to pause, not reasons to present as a success story.”

“Both are reasons to collect more data. Which we are doing.”

“There is a difference between collecting data and being carried along by it faster than you can process what you're seeing.”

Morse smiled briefly. The smile of a man who had heard this before. Who had heard it in grant review meetings and ethics board hearings and late-night arguments with people he had respected and then quietly stopped calling. Who had, each time, waited for the room to finish, and then continued.

“The inter-Cohort exercise data,” Halvorsen said. “The injury event, and the rematch outcome.”

Something in Morse's expression compressed. Not visibly. Not in any way someone who did not know him would catch. Halvorsen caught it.

“The first exercise outcome reflects the intensity differential we have been cultivating in Cohort 1. Their aggression escalated beyond sanctioned parameters during the exercise. That is documented, and it is being addressed through individual coaching protocols.”

“Being addressed how,” Patel asked.

“By me. Directly.”

“And the escalation itself,” Levin asked. “Participant Calder's behavior specifically.”

“Rafe Calder is a high-expression competitive profile. Under adversarial stress conditions, he pushed into territory that the inhibitory protocols should have constrained more effectively. I've flagged a recommendation to review the inhibitory threshold parameters for high-expression physical profiles. The current settings may be under-constraining in that domain.”

Nkosi put her pen down with a deliberateness that was its own sentence. “You're recommending we loosen the inhibitory protocols because a participant injured another participant. That is not the direction of adjustment I would have anticipated.”

“I'm recommending we calibrate more precisely. Not all restraint is appropriate restraint. If Calder's output ceiling is being artificially depressed by a parameter that does not fit his profile, we are not measuring what Catalyst can do with him. We are measuring what it can do after we have clipped the part of him that is most scientifically interesting.”

The room was quiet.

“The most scientifically interesting part,” Reyes-Galindo repeated. She said it without inflection.

Morse continued as if she had not spoken.

“The data leading up to the injury is also interesting. Cohort 2 was winning, Comprehensively.” He advanced to comparative overlays. “Their tactics reflected a level of coordinated adaptive strategy that was…unexpected. The nature of the defeat suggests Cohort 1 was operating from compromised intelligence or compromised confidence. Their decision cohesion collapsed in a way that does not fit the internal dynamics I have been observing. Someone disrupted their preparation.”

Patel's attention sharpened.

“You believe someone interfered with Cohort 1's preparation,” Halvorsen said.

“I do. I should add that the conclusion is not mine alone. Hale has come to me independently with the same assessment. He has been observing his cohort's behavior since the loss and has identified anomalies in their preparation materials that he believes were introduced from outside the cohort. He is investigating. I have agreed to let him pursue it on his own terms.”

“It could also be competitive pressure producing exactly that effect,” Patel said. “Cohort 2 performing better than expected is, by itself, sufficient to destabilize a group that believed it was dominant.”

“It could be. I intend to find out which.”

* * *

“Which brings us,” Halvorsen said, “to the final item.”

She advanced the projection column. The display was simple, formatted for a decision rather than an analysis.

CATALYST ADOLESCENT INTEGRATION TRIAL — PHASE PROGRESSION

FDA MILESTONE: 8 COHORTS / 96 PARTICIPANTS REQUIRED FOR FULL APPROVAL

CURRENT STATUS: 2 COHORTS IMPLANTED (24 PARTICIPANTS)

PROPOSAL: AUTHORIZE IMPLANTATION OF 2 ADDITIONAL COHORTS

“We are eight weeks ahead of the original milestone schedule. The question before this committee is whether the data from the first two Cohorts is sufficient to authorize expansion to Cohorts 3 and 4.”

Nkosi spoke first.

“My answer is no. Not yet.”

“We have twenty-four participants implanted for three months. We have one unsanctioned autonomous intervention by the device itself, the full implications of which we do not yet understand. We have emergent behavioral effects that do not appear in the efficacy literature and were not described in the FDA submission. And we have performance results outpacing our projections by a margin we cannot yet explain. Those are not the conditions under which I recommend implanting twenty-four more adolescents.”

“Those are precisely the conditions that make expansion valuable,” Morse said.

“The data we have is exceptional,” Morse continued. “Integration rates above projection. Cognitive outcomes above projection. And yes, one adaptive response we did not anticipate, which resolved a medical emergency in four hours rather than ten days. If this were a drug trial and we had these efficacy numbers, we would not be debating expansion. We would be accelerating it.”

“This is not a drug trial. A drug is metabolized and cleared. This device is permanent, adaptive, and learning. Those are categorically different risk profiles, and treating the evidence standards as equivalent is not a methodology, it is a rhetorical move.”

“They are different. Which is why the bar for efficacy data should be higher. And it is. Every metric we have clears it.”

“Every metric we have measures performance. We have no long-term safety data. We have no data on what happens when an adolescent nervous system continues to develop alongside an integrating implant over years rather than months. We cannot characterize the risk we would be asking the next twenty-four children to accept because we do not yet know what that risk is.”

Reyes-Galindo leaned forward.

“I want to separate two questions. Whether the data is sufficient to understand Catalyst's current behavior, and whether it is sufficient to authorize expansion. Those are not the same question, and I think we have been treating them as though they are.”

“Go on,” Halvorsen said.

“On the first question, no. We do not fully understand the adaptive intervention. We do not fully understand the influence profiles in Cohort 2. The rate of integration in adolescent tissue is not mechanistically characterized, and I am not comfortable pretending otherwise. But we rarely fully understand a system before expanding a trial. The whole point of sequential Cohort design is to accumulate data across a broader population while we continue to investigate what we are seeing. The question is not whether the picture is complete. The picture is never complete at this stage. The question is whether the oversight framework is adequate for what we do not know yet.”

“The consent model assumes participants are accepting a characterized risk,” Patel said. “If we cannot characterize the emergent effects to the ethics board in honest terms, we cannot meaningfully obtain consent for them. What I would have to put in a consent document to be genuinely honest would include things we have not disclosed to anyone outside this room.”

“We disclose that emergent effects have been observed,” Morse said. “We disclose that their nature is not yet fully understood. That is honest consent.”

“It is technically honest. I am not certain it is meaningfully honest. There is a difference between telling a family we have observed some unexpected effects and telling them what those effects actually are. One of those families has a daughter who can now influence group decisions in ways she does not consciously control. I do not know how to consent someone to that. I do not know what language makes that legible to a parent.”

Morse looked at her steadily. “You cannot consent someone to every possible outcome of a trial. If that were the standard, no trial would proceed past Phase One.”

“No. But you can consent them to what you know. And what I know, what I have documented, is currently in a file that has not been shared with the ethics board, because we have not agreed on what it means yet.” Patel paused. The next sentence cost her something. “That is the thing that concerns me most in this room today. Including my own role in it.”

The room went quiet. Even Morse.

Levin cleared his throat. “From an implantation standpoint, the procedure is safe. Twenty-three clean procedures, one minor inflammatory event, all resolved, surgical risk profile well-characterized and low. My concern is the post-implantation picture, which is less settled than I would want it to be before I perform twenty-four more procedures on adolescents. I would want more time with the Park data before I am comfortable saying that we understand what the device will do in an acute crisis.”

“Yusuf,” Morse said with a gentleness that was genuine and not entirely without calculation, “in thirty years of medical device trials, how often has the post-implantation picture been fully settled before Cohort expansion.”

Levin considered it honestly. “Not often.”

“Never. Expansion is how the picture settles. Sequential Cohort design exists precisely because we cannot get the data we need from two Cohorts alone. We do not sit on what we have indefinitely waiting for a certainty that can only come from the next dataset.”

“We also do not accelerate past the point where we can responsibly analyze what we are collecting,” Nkosi said.

“We are not past that point.”

“I think we may be approaching it.”

“I think,” Morse said making an important distinction, “that you are conflating approaching the edge of our current models with approaching an ethical limit. Those are different things. One requires better models. The other requires a stop. I am arguing for better models. I am not arguing for a stop. There is a third option between proceeding blindly and halting everything, and it is proceeding carefully with enhanced monitoring and real accountability. That is what I am proposing. That is all I am proposing.”

Nkosi opened her mouth. Closed it. Wrote something in her notepad. Several lines. Halvorsen noted the duration.

Reyes-Galindo looked at Halvorsen. “What are the terms of the proposed authorization.”

Halvorsen advanced the display.

PROPOSED CONDITIONS FOR ADDITIONAL COHORT AUTHORIZATION:

Participant selection: extended pre-screening, 90-day minimum observation period

Monitoring: continuous telemetry, elevated from current protocol; review cadence increased to daily

Ethics disclosure: emergent effects addendum to consent documentation

Halt criteria: any adverse event requiring autonomous Catalyst intervention without physician authorization; any participant requesting withdrawal

Independent review: all anomalous data flagged to external neurological oversight panel within 48 hours of identification

“An independent panel,” Patel said. “Who sits on it.”

“We would convene it from the FDA's external advisory roster. Names disclosed to this board before authorization is recorded as final.”

Patel looked at the conditions the way she looked at things she was still deciding whether to trust. “The emergent effects addendum. I want to see the language before it goes to families.”

“Of course.”

Patel nodded. She did not say yes. She did not say no.

Nkosi looked at the halt criteria for a long moment.

“Any adverse event requiring autonomous Catalyst intervention without physician authorization,” she read aloud. “We have already had one of those.”

“Which resolved favorably,” Morse said.

“Which resolved favorably this time. The halt criterion covers future instances. I want to make sure we are all clear that this condition is binding. Not a target. Not a guideline. If it happens again, we stop.”

“If it happens again, we stop,” Halvorsen said. “That is the condition. It is binding.”

Nkosi wrote something. Set her pen down with the finality of a woman closing a door she intended to open again.

“I want it in the record that I have significant reservations. I believe we are moving faster than our analytical capacity. I believe the consent model for emergent effects is not yet adequate for what we have observed. And I believe that the Park intervention should have triggered an automatic review hold before any expansion discussion took place. That it did not is itself something the record should reflect. I will not block authorization if every condition listed here is binding and not advisory, and if the independent panel has genuine authority and genuine independence. But my reservations are formal, and I want them documented.”

“Noted. Formally. Reyes-Galindo.”

Sofía was quiet.

“The independent panel is the difference for me. If the oversight is real, not nominal, actually external, actually empowered to require a halt, then the monitoring conditions are sufficient for me to support expansion.” She looked at Morse directly. “The moment that panel flags something and this committee does not act on it, I withdraw my authorization retroactively and formally.”

“Understood,” Morse said.

“Levin.”

Levin looked at the conditions for longer than seemed strictly necessary. Halvorsen understood what he was doing. He was running the scenario one final time to be sure he had considered everything he could consider.

“Ninety-day pre-screening is meaningful. That is more than we had for Cohorts 1 and 2. The surgical risk is acceptable. I can support expansion under these conditions.”

“Patel.”

Patel looked at Halvorsen. Then at Morse. Then back at Halvorsen.

“I need the addendum language before this is recorded as final. I am not voting no. I am voting conditional, and I mean that as a genuine condition, not a formality.”

“Conditional is recorded as conditional. Morse.”

Morse had not moved. He had not, to Patel's careful observation over the last two hours, changed a single thing about his posture or expression during the entire meeting. He had listened with complete attention and responded with complete precision. He had never once looked like he doubted the outcome.

“Yes. Authorize Cohorts 3 and 4.”

Halvorsen looked around the table one last time. The habit dated to her early career running clinical trials, before any of this, when the stakes had seemed large and by comparison were quite small.

“Authorization for Cohorts 3 and 4 is approved, conditional on the listed terms, subject to Patel's addendum review, and with Dr. Nkosi's reservations formally entered in the record. Implementation begins when participants return from break. Cohort 3 pre-screening initiates Monday.”

The projection column dimmed.

People gathered their displays, stood, exchanged small words. The room emptied slowly.

Morse remained seated.

Halvorsen noticed. She dismissed the others with a glance and waited as the door sealed behind Nkosi with a soft definitive sound. She sat down.

In their eight years of working together, this was an unusual position for her to take. She almost always remained standing until a room cleared. Morse recognized what it meant.

“Say it,” he said.

* * *

“The Park incident,” Halvorsen said. “You knew it was in the capability set.”

Morse looked at the dark projection column. “I hypothesized it.”

“Elric.”

“I ran probability models eighteen months ago. Adaptive inference operating in vascular-adjacent pathways, under acute physiological stress, in a highly integrated adolescent nervous system. The math suggested it was possible. Not certain. Possible.” He looked at her. “I did not suppress it. I submitted a capabilities framework paper to the steering committee.”

“You submitted a capabilities framework that was ninety pages long and placed that projection in appendix C. That is not disclosure. That is deniability with extra steps. You did the same thing with the Karolinska study eleven years ago. You knew exactly what you were doing then. You knew exactly what you were doing now.”

Morse was quiet.

“You wanted to see if it would happen,” she said.

Still quiet.

“Not in the sense that I hoped someone would be injured.”

“That is not the reassurance you believe it is.” She said it without heat. Factually.

He looked at her then intensely. That was the hardest thing about Elric Morse to be on the receiving end of. You felt it as a pressure that was not comfortable even when you trusted him. Especially when you trusted him.

“Miriam. I have spent thirty years studying what human beings are actually capable of. Not what they are believe about themselves. Not what the protocols permit us to measure. What they can do.” “Catalyst is the first tool in the history of this work that can answer that question in full. Not approximately. Not statistically. In real time. In individual nervous systems. With precision that no prior methodology has ever approached. If we constrain it to avoid surprises, we do not learn what it can do. We learn what we were already comfortable with. And I did not spend thirty years building toward this to learn what I was already comfortable with.”

Halvorsen sat with that for a moment.

“The ethics protocols exist because the people in those trials are children.”

“The ethics protocols exist because the people who wrote them were afraid. Fear produces caution. Caution produces protocols. Protocols produce exactly the kind of constrained, predictable outcomes that will never tell us what we need to know. About what these brains can become. About what Catalyst can become inside them.”

“And what do we need to know.”

Morse was quiet for a long moment, looking at the dark terminal across the table.

“Whether Catalyst can take a human being all the way to the edge of their potential. And whether it knows, on its own, where that edge is.”

The room was very still.

Halvorsen stood. She walked to the door and stopped before opening it.

“Whatever you intend to pursue in the second phase, Elric…” She did not finish the sentence.

The door opened. Closed.

Morse sat alone in the windowless room. The data was still faintly visible on his terminal: neural maps, performance curves, the extraordinary acceleration of young nervous systems that had no idea what they were in the process of becoming and no language yet for what it would feel like when they found out.

* * *

Down the corridor, Patel had not left the building.

She sat in her office with the door closed and her display open to a set of notes she had written at two in the morning three days ago, when everything had gone quiet and the thought she had been circling for weeks finally stopped moving. She had not shared the notes with anyone. She had not been certain, when she wrote them, whether she was being rigorous or paranoid. She was still not entirely certain now.

She read them again.

At the top of the page, underlined twice:

Catalyst is no longer just responding.

It is listening.

What happens when it begins to answer.

Below that, added this morning before the meeting, in different handwriting. Faster. Less deliberate.

Morse is running a different experiment than the one we filed.

She read both lines for a long time.

Through the wall, the building settled into its quiet evening mode. The climate systems cycling. The corridor lights shifting to their lower overnight register. Somewhere down the hall, a door closed. Then nothing.

She thought about what she had said in the meeting. The thing she had not been planning to say. She had said it because it was true and she had not been able to leave it unsaid. But saying it did not change what she had been doing. She had been withholding from the institution that paid her to disclose. For what she believed were good reasons.

She opened a new document. Not on Neurovia's servers.

She had retained an academic affiliation at Johns Hopkins through her tenure at the Institute. The affiliation gave her access to a research server she could legitimately claim was for ongoing collaborative work on adolescent psychiatric outcomes. The claim was thin but defensible. If anyone at Neurovia checked, the access logs would show a clinical researcher using a legitimate university server. They would not show, unless someone went looking with intent, what she was actually putting on it.

She titled the document:

OBSERVATIONS — INDEPENDENT.

She sat with the title for a moment. The word independent meant several things at once. Independent of Neurovia's monitoring infrastructure. Independent of the ethics board's review chain. Independent of any institutional process that could compel disclosure or quietly disappear the documentation. Independent also of her own role at the Institute, which she had chosen and which had paid her well and which she was now beginning to operate against, in small careful ways, on a server that was not theirs.

She knew the move was a risk. The Institute's monitoring was extensive. Sub-vocal mapping in the corridors. Pheromone trace differentials. Layers she had access to and layers she suspected existed and could not confirm. If Morse decided to look for what she was doing, he would find traces.

She began to write. Carefully. As if it might matter later.