Phase Portrait

Lena returned on Thursday to determine whether Ruth was safe and Evan was still useful.

She arrived after the home-care aide left. Ruth received her at the kitchen table with the family album closed beside her and answered every question she considered relevant. She had slept, eaten, taken her medication, and experienced no headache, weakness, nausea, or unusual sensation. She still remembered Celia. She did not consent to repeat the cognitive tasks.

"I was clear enough to know when he should stop examining me," Ruth said.

Lena wrote this down. "Do you want me to discuss the run with him?"

"I want you to tell him what happened. He thinks understanding and permission are the same thing."

"They aren't."

"Good," Ruth said. "You may continue."

In the garage, Lena compared the treatment log with the independent replication. Ruth's six-minute run had produced lower absolute field, less heating, and a shallower covariance collapse than either operator trial. Its effect on autobiographical coherence had been larger than its instrument signature suggested.

"Your power reduction didn't reduce the relevant coupling," Lena said.

"Her answers were the feedback signal."

"Yes. You put identity inside the loop and then congratulated yourself for turning down the amplifier."

Evan accepted this because arguing would have converted an accurate statement into a dispute he could try to win.

He showed Lena the visual notes after lunch.

The first entry predated Ruth's run. During L-03, Evan had seen a faint hexagonal grain across the closed-eye darkness. During Lena's replication it had acquired depth, like tiled rooms viewed from all sides at once. While the sensors lost rank in RUTH R-01, the grain had returned at the edge of his vision even though he was in the kitchen and not the active subject.

He had omitted it because geometric afterimages were subjective, migraines existed, and writing hallucination beside cognitive enhancement would have made every other result easier to ignore.

Lena read the entries without changing expression. "Have you seen this before?"

Evan said yes.

Eleven years earlier, he had participated in a licensed perception study that used DMT. He had volunteered because the study paid, the monitoring looked serious, and another engineer claimed the experience was impossible to describe. Evan had interpreted this as a challenge to description.

The session lasted minutes. It did not reveal a cosmic truth. It dismantled the ordinary priority by which one room, one body, and one sequence of moments suppressed every competing interpretation. He saw recursive architecture populated by figures that seemed to react when he attended to them. Afterward he treated the impression as a property of a brain temporarily unable to enforce its usual model.

He had kept one page of notes.

PERCEPTION STUDY, 11 YEARS AGO

not a place
not infinite, only too recursive to count at the time
figures may be visual system rendering agency from structure
attention changes what counts as foreground

strongest impression:
  being compressed into a form the scene could recognize

Lena placed the old note beside the machine-state entries. "This does not make DMT a detector."

"I know."

"It makes your prior experience an excellent source of expectation bias."

"I know that too."

"Then we can test the report without giving you anything."

They designed PHASE P-01 as a perceptual correlation experiment, not an exposure for benefit. Evan would sit in the darkened front room with an opaque mask and broadband noise. He could not see, hear, or feel the garage controls. Lena would run safe and fault builds in concealed order, then inject three classes of events after the anomalous region began.

Physical phase steps advanced or retarded the heater front. Decoy steps wrote identical values into a shadow register no hardware consumed. Hold events changed neither. Evan would speak only a small vocabulary describing the geometry: LEFT, RIGHT, STILL, OPEN, CLOSED, DEEPER, SHALLOWER, FIGURE, NONE.

The vocabulary prevented him from narrating his way toward a match. Event timing and analysis were fixed before the first run.

PHASE P-01: BLINDED PERCEPTUAL CORRELATION

observer: E. Hale, no display or control access
operator: L. Varga
conditions: SAFE / FAULT concealed from observer

hidden events:
  PHYSICAL LEFT / RIGHT
  DECOY LEFT / RIGHT
  HOLD

observer vocabulary:
  LEFT RIGHT STILL OPEN CLOSED DEEPER SHALLOWER FIGURE NONE

primary:
  report alignment within 2 s of hidden event
secondary:
  reported recursion depth versus covariance rank

The safe run produced retinal grain, two vague rings, and nothing Evan could classify as a figure. He reported NONE through every hidden event. Lena did not tell him which condition had run.

After a full thermal reset, she loaded the second concealed build.

For twenty minutes Evan saw the ordinary mottled darkness produced by eyes with no useful input. Then the grain organized.

Lines joined into six-sided cells. Each cell contained a smaller arrangement that was not a copy but served the same structural role. The field acquired depth without a stable direction of view. Near and far became levels of subdivision rather than distance.

"DEEPER," Evan said.

In the center, several cells began sharing edges. Their combined boundary closed around a vertical form. It had no face or fixed number of limbs. Wherever Evan tried to locate one surface, smaller versions performed the same closure beneath it.

"FIGURE."

The form rotated left.

"LEFT."

It stopped. Smaller structures continued turning inside it, some left and some right, but their outer relation held.

"STILL."

The next rotation went right. Evan reported it. The figure opened through the center, revealing an additional level of nested forms. He said OPEN, then DEEPER.

Lena's hidden sequence continued. Some events produced no response. Others changed the figure's chirality so promptly that Evan began to feel he was reading a display. He reminded himself that the display was inside his own perceptual system and therefore capable of manufacturing agreement from expectation.

The figure then did something outside the vocabulary.

Its nested surfaces separated into five braided bands. Four continued through a common turn. The fifth arrived early, crossed the others, and ended in a flat repetition. Evan experienced the arrangement not as an image to decode but as a mistake he had already understood.

The early band was the heater confirmation edge. The flat repetition was the shadow register accepting a phase value before the thermal front had physically responded.

"STOP," he said.

Lena opened the safety contactor. The geometry thinned into ordinary afterimage before the RF current reached zero.

She entered the front room and removed his mask. "Why did you stop?"

Evan drew the five bands. He explained that one confirmation event occurred before the modeled thermal response and allowed the controller's causal bookkeeping to close on a state the hardware had not yet reached.

"That isn't in your source," Lena said.

"The confirmation path is. I never treated it as part of the phase loop."

They checked the event trace. The anomalous rank collapse had begun when the fifth confirmation edge advanced into the prior thermal-response window. Evan's diagram described the timing relation within one controller tick.

The phase reports were worse for Lena's decoy events. They were not merely weaker. They remained at chance. Evan had reported twenty-one of twenty-four physical left or right steps correctly within two seconds and five of twenty-four decoy writes. Reported depth increased as measured covariance rank fell.

The geometry had tracked what the phase value caused, not what the register contained.

Lena reran the analysis twice. "There could be a physical cue we haven't isolated."

"Through the wall, mask, and noise?"

"Yes. An unlikely cue is cheaper than a responsive hallucination."

"You called it responsive."

"I called your claim expensive."

They tested the control insight without entering another observer state. Evan delayed the confirmation latch until the measured thermal response crossed its threshold. Ordinary trapped-field gain changed little. Loss during the phase walk fell by eleven percent. The covariance collapse deepened and arrived earlier.

The advice worked. It also made the anomaly stronger.

Lena restored the original timing and locked the branch. "We do not accept optimization suggestions from unclassified perceptual phenomena."

"It showed me a causal edge."

"Your brain rendered an insight after weeks of studying the system. That explanation remains available."

"And the hidden phase sequence?"

"Remains a result. It does not grant the result a personality."

Evan entered structured figure in the report. Lena changed it to structured percept with agent-like persistence. They compromised by keeping both phrases and marking the disagreement.

PHASE P-01: RESULT

SAFE condition:
  no structured figure reported

FAULT condition:
  physical phase direction: 21 / 24 aligned reports
  decoy register direction:   5 / 24 aligned reports
  reported depth tracks decreasing covariance rank

unprompted content:
  five-band timing relation matches logged confirmation edge
  proposed delay reduces loss 11% and deepens rank collapse

interpretations retained:
  unidentified physical cue
  observer inference under altered machine state
  residual structure rendered as agent-like percept

Ruth joined them for dinner. She remembered Lena and the reason for her visit but placed it on Tuesday. She asked whether they had found what the machine was doing.

"It may be giving his visual system information about the apparatus," Lena said.

"Like a screen?"

"A screen would be preferable."

Evan described the nested figure. Ruth listened, then asked whether it had known he was looking.

He said the form had responded to machine settings.

"That isn't what I asked."

Evan returned to the garage after Lena left. He did not power the apparatus. He replayed the recording of his own sparse reports and watched the hidden event markers appear afterward.

LEFT. STILL. RIGHT. OPEN. DEEPER. FIGURE.

At the moment he had said FIGURE, no hidden event occurred. The physical phase had held steady for eleven seconds. Sensor rank continued falling, but the form's closure had not followed a command.

It had followed his attention.