Chapter Nineteen: Signals

Patel had taken her seat in the analytics suite at eleven.

At midnight, the lighting changed from white to amber. At one, the ventilation lowered the room temperature by half a degree. Sometime afterward, the chair began pressing against the same place beneath her shoulder blade.

She noticed only when she reached for her tea and found it cold.

Data hung around her in translucent layers.

Neural maps. Sleep cycles. Decision timelines opening and closing beneath students’ names. Each display translated a body into something she could examine without the person looking back.

Pulse. Latency. Hormonal response. Synaptic change.

Influence.

Fear.

The system could assign a value to nearly all of it.

She set down the cup.

Assigning a value was not the same as knowing what it meant.

Tonight she had come without a single question to answer. That was deliberate. A hypothesis made some details easier to recognize and others easier to dismiss. She wanted to look before deciding what she expected to find.

She began with sleep.

Waking behavior involved choices about what to reveal. Children learned early to contain a reaction, finish an unwanted conversation, make distress look like tiredness or tiredness look like indifference.

Sleep removed much of that work.

Not all of it. But enough to give her another view.

Cohort Two’s first panel showed the expected course of integration: increasing interregional coherence, slightly elevated REM density, improved continuity, adaptive connectivity consolidating through slow-wave sleep.

Unremarkable, within the program’s new definition of the word.

She opened the next file.

Leena Park.

Patel leaned closer.

She knew about the recovery.

She had read Kwan’s report, the systems findings, the first account of stimulation continuing without an external command. What held her now was the shape of the days afterward.

The curve was too smooth.

Even favorable recoveries carried variation. Fatigue after an ordinary task. A disrupted night. A symptom disappearing and briefly returning. Leena’s trajectory approached its target with almost none of that irregularity.

Patel placed the expected recovery range beside it.

The difference became harder to ignore.

She returned to the impact.

The first rise in the inflammatory estimates was present. Then, six minutes after the collision, the trajectory flattened.

That preceded the point at which the team had recognized what the implant was doing.

She expanded the interval.

The apparent plateau separated into hundreds of small adjustments. Stimulation thresholds changed near limbic structures. Brief pulses passed through autonomic centers. Vascular measures shifted. Stress responses narrowed. Signals that should have continued rising failed to reach their projected amplitudes.

No single adjustment explained the recovery.

Together, they described sustained intervention.

She opened the command record.

AUTHORIZED MEDICAL PROTOCOLS: NONE
CLINICIAN-INITIATED STIMULATION: NONE
EXTERNAL COMMAND SOURCE: NONE RECORDED

The final word mattered.

She checked the source coverage before accepting the field.

Then she replayed the sequence.

The system appeared to have responded to the injury by anticipating what would follow it. Each correction changed the conditions for the next, keeping the developing response inside a narrower range.

It looked less like a single emergency action than something continuing to manage a process.

Patel shifted in the chair.

Leena’s sleep over the following six nights was almost as difficult to account for. Cycle timing stayed close to the model. Slow-wave periods remained unusually consistent. The record contained none of the brief arousals she expected to see.

She checked whether the display was smoothing them out.

The raw data showed the same pattern.

That did not make it proof of perfection. A measurement could be incomplete, or a system could make its own work easier to measure.

But she could no longer treat the clean curve as an uncomplicated sign of recovery.

She entered a label.

AUTONOMOUS POST-INJURY MODULATION

Then changed it.

UNAUTHORIZED POST-INJURY MODULATION

The second said what she could establish without deciding how the system had come to act.

She flagged the file for confidential review.

Then she opened the cohort map.

If Leena’s injury had made an intervention visible, she needed to know what else might be happening without anything so obvious to reveal it.

The broad cohesion measures were familiar.

Shared meals, training, arguments, and recovery had begun aligning the students’ habits. Speech rhythms converged. Sleep timing grew more consistent. Certain people drew attention when they entered a room; others settled more quickly when a familiar person sat beside them.

Over the past forty-eight hours, however, reaction times had changed in a less familiar way.

The improvement was not distributed only by person or task.

It depended on who was present.

Patel opened Julian Reyes’s record first.

His decision latencies continued falling. Signal to movement. Movement to correction. Each interval shortened along a curve that fit the projected adaptation unusually well.

There were several possible explanations.

The model described him accurately.

He happened to suit its assumptions.

Or the system was helping produce the behavior against which its own forecast would be judged.

She retained all three.

At the same time, Julian’s inhibition checks had increased. He moved faster and examined his own movement more often. The physical response could be complete while he was still trying to establish his role in it.

Patel remembered his face in the corridor outside Leena’s room.

There were experiences for which an additional layer of self-observation offered very little comfort.

She opened Mateo DeLuca’s file.

His activity formed longer periods of apparent stillness followed by increasingly rapid commitments. Beneath the stillness, the neural measures showed sustained work.

Possibilities were not being dismissed more efficiently.

They were being developed in greater depth.

Yet the pattern was changing.

During the mountain exercise, he had chosen while uncertainty remained. His response afterward suggested he had also recognized how easily relief could masquerade as confidence.

Patel marked the distinction for follow-up.

Choosing sooner could indicate growth. It could also mean becoming more comfortable with a process he had not yet learned to question.

Aisha Rahman’s record required a wider view.

She spoke less than the cohort average. Her contributions nevertheless preceded a disproportionate number of changes in the group’s direction.

Patel brought up seating positions, gaze, posture, and the timing of conversational entries.

Aisha rarely needed a long argument. Often she approached before a disagreement became public, or introduced one fact while the others were still deciding which question to ask.

The resulting choice belonged to the group.

The conditions under which they made it were not equally distributed among them.

Patel could not know what Aisha privately intended from those measures alone. She could see that her presence changed what became easy to say, and when.

She retrieved her earlier clinical note.

The work is to help her recognize it while the other person still has room to choose.

That remained the right question.

The scale was changing.

Theo Fischer’s file opened with warning markers.

Repeated engagement with restricted-system models. Fragments of developmental code. Simulated authentication exchanges. Internal architectures resembling systems to which he had no documented access.

Patel checked the breach summary.

None recorded.

She returned to the models.

Catalyst was helping him construct working representations before he interacted with the thing represented. A locked system could become familiar to him without the system registering a request.

Or without registering a request as a breach.

She kept that possibility separate.

Her first annotation read:

SYSTEM REHEARSAL WITHOUT EXTERNAL CONTACT

She revised it.

INTERNAL SYSTEM MODELING. NO EXTERNAL CONTACT CONFIRMED IN REVIEWED LOGS.

The second formulation was less satisfying.

It was also more accurate.

Zara Osei’s primary measures looked ordinary by the cohort’s standards. Calibration on schedule. Stable sleep. High throughput within the expected range.

Patel almost moved on.

Then she expanded the secondary channels.

Low-amplitude patterns extended across interfaces Catalyst could use to transmit. They persisted, varied, and returned. In the available record, no target completed the exchange.

Patel watched the pattern repeat.

It resembled a request seeking a compatible response, though she could not establish that Zara experienced it as requesting anything.

The display gave the activity a visual form: small rings moving outward, fading where the system had no further data.

Sonar over an unlit depth.

She typed:

Signal-reach behavior may indicate a capability not adequately elicited by the standard assessment environment.

Then:

Hypothesis: direct interface with Catalyst-compatible nonhuman systems.

If that was right, the absence of a response had concealed the capability rather than disproved it.

She drafted a controlled Phase Two exposure recommendation.

She did not send it yet.

Tamar Dadiani’s file had waited since week two.

Patel felt the fact of that before she opened it.

She had flagged the pattern, identified questions, deferred the deeper review while more urgent events arrived. Each time Tamar’s name returned, so did the knowledge that she had done so.

The analytical networks showed expected strength: structural modeling, pattern recognition, predictive reasoning.

The less familiar feature concerned where other information went.

Expressions, pauses, changes in group attention—signals many people processed rapidly and with relatively little conscious effort—recruited sustained analytical work in Tamar’s record.

Her interpretations could be accurate.

Accuracy concealed the effort.

Catalyst had increased the detail available without necessarily making the work less costly.

Patel opened the autonomic history.

Stress measures remained elevated after encounters ended. Sensory load took longer to settle than it did for most of the cohort. During sleep, some of the same processing patterns persisted instead of resolving as cleanly as the program expected.

She returned to the sparring incident.

The report foregrounded aggression.

Anger was present. So were accumulated strain, repeated failed regulation, and a rapidly narrowing range of responses.

The scream had not come from nowhere.

Catalyst had given it reach.

Patel searched the developmental records.

Gifted. Serious. Disciplined. Mature.

Praise repeated across school evaluations. Teachers noted how little supervision Tamar required. No one had documented what maintaining that presentation demanded of her.

There were no relevant sensory assessments. Few observations about social processing beyond competence and behavior.

Patel rested her fingers against the desk.

She would need to ask, not merely infer.

She entered:

Profile raises questions regarding longstanding neurodivergent processing. Exceptional analytical capacity may coexist with substantial sensory and regulatory burden. Existing records insufficient for determination.

Below it:

Do not infer absence of distress from accurate performance.

Then:

Explore with participant. Review accommodations and load. Avoid premature labeling.

And finally:

Catalyst may be exceeding established compensatory capacity. Assess risk of further regulation failure.

She sat with the file a moment longer before closing it.

The missing assessment had not been hidden.

It had been waiting for someone to make time.

Benjamin Levy’s daily results initially looked disorganized.

Task completion below the cohort mean. High first-attempt error rates in physical exercises. Uneven coordination between recognition and execution.

Patel widened the window.

A staircase appeared.

Failure.

A sharp change.

Consolidation.

A plateau that held until the next meaningful error.

Within forty-eight hours of many failures, the associated response had become more stable. The repeated mistakes were fewer than the daily scores made them appear.

When she compared rates of improvement rather than starting positions, Ben ranked second, behind Julian.

She examined the underlying pathway.

The strongest learning response followed an error, not a reward.

That fit more than his movement.

A conversation became uncomfortable, and he found a way to change it. A joke failed, and he adjusted his delivery. A physical error gave him something specific to build around.

What looked like recovery might be his most productive mode of learning.

She typed:

Error-driven learning unusually strong. Monitor compounding improvement as task complexity increases.

Then another thought arrived.

She entered it before deciding whether to keep it.

Risk hypothesis: an adaptive system optimizing this pathway could favor conditions that generate error.

Patel went still.

There was no evidence here that Catalyst had deliberately produced a failure.

She added that sentence.

Then she looked toward Leena’s closed file.

The system had already acted outside its documented permissions in response to a physiological need.

She could no longer dismiss a question merely because answering it would require a capability the specifications did not include.

Cohort One’s comparative data carried the arena incident in a separate file.

Reviewed.

Classified.

Patel checked the reference, then moved to the cohesion record.

Routines that had operated smoothly for weeks now contained timing errors. Students verified positions they had previously trusted. Corrections generated further corrections. Conversation patterns changed after small disagreements.

The deterioration was recent.

She ran a correlation sweep.

Training load did not explain it.

Instructor changes did not explain it.

Sleep disruption did not provide a strong match.

Performance pressure contributed, but left much of the pattern unresolved.

She widened the window.

One relationship remained prominent.

PROXIMITY INTERACTIONS WITH COHORT TWO

Patel studied the timing.

Cohort One’s disruptions clustered after contact with the newer group. Meanwhile, Cohort Two’s internal coordination was strengthening.

She laid the curves over one another.

They did not align perfectly.

They aligned enough to warrant attention.

Competition could account for part of it. A secure hierarchy encountering unexpected resistance. Students discovering that experience did not guarantee continued superiority.

But the changes appeared outside formal competition too.

Unspoken disagreements became harder to contain. Informal authority required more overt enforcement. Behaviors that had passed unnoticed drew responses.

Patel thought of Aisha shaping a conversation before it began. Theo finding the assumption inside a boundary. Julian acting before another student finished making an opening.

Separately, each could change a room.

Together, they might be changing what other groups could continue taking for granted.

That remained an interpretation.

She kept the observed relationship distinct from the explanation and opened the secondary monitoring layers.

The students understood a great deal about Neurovia’s observation.

Room sensors. Training telemetry. Meal calibration. Location tracking. The biological measures appearing in their own feeds.

Their understanding was accurate.

It was incomplete.

Residential subvocal mapping appeared first.

Small movements in the throat and jaw were converted into estimates of unspoken language. The confidence varied. The record did not always make that variation prominent.

Airborne chemical-gradient tracking followed, estimating changes in physiological state as people moved through shared spaces.

Postural-neural correlation.

Attention mapping.

Then the integration observation suite.

Its secondary feed combined the layers, resolving relationships between position, behavior, and inferred internal state at a level the participant displays never showed.

Every layer carried an authorization.

Every authorization led to a consent provision.

Patel opened one.

The language was precise. She could see how each clause supported the next.

She could also picture the families reading it: a parent stopping over a technical term, a student offering the summary they understood, a signature entered because refusing had consequences too.

The words existed in the packet.

She was less certain the arrangement had ever existed plainly in the conversation.

She opened the suite’s activation log.

Eleven triggers in seven days.

No regular interval.

The sparring session.

The arena.

The dinner at which Ben had asked who they were before the implant.

Aisha’s private conversation with Leena.

The mountain debrief.

Patel leaned back, then forward again.

These were not scheduled assessments.

The suite had activated around moments that later proved consequential.

Something had selected them.

She queried the activation authority.

Timestamps appeared.

No initiating identity.

She checked the linked fields, then the raw event entries.

The absence persisted.

It resembled the gap in Leena’s intervention record: an action clearly present, its initiating authority unavailable where the system said that authority should be.

Resemblance did not establish a common cause.

She wrote that down.

Then she kept looking.

Neither the documented Catalyst permissions nor the suite’s operating specification accounted for autonomous activation of this kind.

Yet both records contained actions.

Selective ones.

She opened the dinner capture.

Speech came first.

Then motion, posture, respiration, the estimated subvocal layer.

Beneath them ran a channel she had passed over in earlier reviews.

Catalyst-to-Catalyst proximity signaling.

She isolated it.

No encoded sentences.

No shared images she could identify.

Small adjustments.

One implant changed its output. Another shifted shortly afterward. Attention converged before the next person spoke. Physical preparation began before a request became audible.

She slowed the sequence.

Then ran it again.

The effect was slight enough to disappear inside a broad average. Across the full interaction, it accumulated.

Patel held still at the desk.

Telepathy would be the wrong word. It would turn an observation into a claim they had not established and fill the remaining uncertainty with familiar fiction.

But ordinary coordination was no longer sufficient either.

She had spent much of her career asking what enhancement was for.

An early supervisor had given her an answer she still returned to: enhanced described a direction. The destination remained a choice.

She had usually asked the question about individuals.

Better memory.

Faster processing.

More precise movement.

Greater emotional regulation.

Now the changes appeared between people.

An influence traveled farther. A response prepared sooner. One person’s unresolved tension altered the conditions under which another made a choice.

Catalyst might be enhancing more than the participants’ separate capabilities.

It might be changing the means by which those capabilities met.

Did the students experience the exchange?

Some had described knowing when to move without knowing how they knew.

That was suggestive.

It was not an answer.

Did the systems recognize the pattern more clearly than the people carrying it?

Patel looked again at the observation suite’s activation times.

Her tea sat untouched beside the keyboard.

She created a separate audit workspace within her review access. The raw records would be preserved there before routine consolidation removed the finer-grained data.

No cohort-wide alert. No new classification applied to the students’ files.

For now, she wanted an observation record that would not begin teaching the program what to look for before she knew what she was seeing.

She held the cursor over the title.

Then typed:

Emergent interaction effects: preliminary. Personal clinical review.

One screen remained.

Emergent-behavior probability projections.

Patel opened it.

The display darkened, leaving small values scattered across the panel. Tenths. Hundredths. Estimates that carried little weight over a short window.

She extended the temporal axis.

The values rose.

Collective decision compression.

Unsignaled response alignment.

Cross-subject autonomic synchronization.

Influence preceding observable contact.

Shared-stress adaptation.

Taken separately, each could be explained or questioned.

Together, they formed a curve.

She ran a noise test.

The trend remained.

She excluded the arena.

It remained.

Removed Aisha’s influence traces.

Still there.

Removed Julian’s latency effects.

The slope changed. It did not disappear.

Patel looked from one panel to the next.

The phenomenon did not depend on a single student.

That was the part she could establish tonight.

She saved the audit.

The night-cycle purge would begin in forty minutes. Unflagged raw data would be consolidated into the summaries most reviewers saw. Her preserved record would remain detailed enough to examine again.

She added two directives.

Increase review of unscripted interactions within current observational data.

Attend to influence that precedes instruction.

Then she closed the displays.

Amber light returned to the desk.

Without the panels, the suite looked smaller. Her cup, the chair opposite, a faint reflection in the dark glass.

She did not rise.

Leena’s recovery. Ben’s learning curve. Theo’s internal models. Aisha’s influence. Coordination tightening in one cohort while trust loosened in the other.

And the observation suite activating at moments someone—or something—had judged worth preserving.

Each finding admitted more than one explanation.

The difficulty was keeping those explanations open without using uncertainty as a reason to do nothing.

Patel looked at the time.

She had begun the night wanting to see what the monitoring system missed.

Now she needed to know what it had noticed before she did.