Lam could stop.
That thought came to him on a very ordinary morning—no error data, no warnings, no one calling his name on any list.
Just a morning where everything ran smoothly enough for him to pretend he hadn't interfered.
If Lam stopped helping Hai.
If he stopped taking notes.
If he stopped leaving traces outside the system.
Then, over time, the metrics would balance themselves out.
Latency would decrease.
Social trust would increase.
The gazes would return.
No one would blame him.
No one would remember what he did.
The Error Warehouse doesn't hold grudges.
Lam stood before the mirror, adjusting his collar.
In the reflection, he saw a man who was still intact:
a stable job, a not-so-high but secure position, a life of just enough.
A legitimate life.
The phone vibrated.
A message from Hai.
They cut off the power to my apartment this morning.
Without warning.
Lam closed his eyes.
He arrived at Hai's apartment at noon.
Hai was sitting by the window, the light slanting across the room, making it darker than usual because of the power outage.
"What did they say?" Lam asked.
"They said the contract is no longer valid," Hai replied. "But there was no decision to terminate it. It's just… no longer a connection."
"Could you stay somewhere else temporarily?"
Hai shook his head.
"I tried. Hotels wouldn't accept me. Old friends… didn't respond."
He chuckled softly.
"Perhaps I'm about to become a problem without an address."
Lam looked around the room.
Everything was still there: the furniture, the books, the clothes.
Only the recognition was being gradually withdrawn.
"Do you regret asking me for help?" Lam asked.
Hai thought for a moment.
"—No," he said. "At least there's one person who knows how I'm disappearing."
That statement struck Lam more deeply than any indicator.
That afternoon, Lam returned to the Warehouse.
He didn't open Hai's file.
He opened his own file—a privilege every Archivist has, but rarely uses.
The error list appeared concisely and cleanly.
Nothing new.
But in a sub-section, he saw a newly added entry:
Out-of-model interaction behavior: Monitoring
No flagging.
No evaluation.
Just… monitoring.
Lam chuckled softly.
The system had recognized him.
He returned to his desk and sat for a long time in front of the blank screen.
Then he opened an interface he had never used for anyone else before:
Create a supplementary profile — Manual Anchor
An old feature, designed for emergencies before the learning model was fully implemented.
It didn't log errors.
It didn't delete data.
It only did one thing:
anchor an individual to at least one required interaction thread.
Lam entered his name.
Nguyen Van Hai.
The system asked:
Confirm creating a manual anchor?
This action may affect your personal metrics.
Lam paused.
This was where he could turn back.
Just close the interface.
No one would know.
No one would lose anything.
Except Hai.
Except himself, who would have to live with having seen… and chosen to do nothing.
Lam pressed CONFIRM.
No sound.
No notification.
Only a small line of text appeared and then disappeared:
Manual Anchor created.
In that moment, Lam felt it clearly:
he had just left the neutral zone.
That evening, the electricity in Hai's apartment came back on.
Unstable.
Not perfect.
But it did.
Hai called.
"I don't know what you're doing," he said. "But the system just… responded."
Lam leaned back in his chair.
"I'm not sure how long it will last," he said.
"Just long enough for me to stay," Hai replied.
Lam opened his notebook.
The three dots were now connected by a straight line—not beautiful, not smooth.
He wrote the last line on the page:
Existence is not a state.
It is an action repeated—sometimes, by someone else.
Lam closed the notebook.
In the Error Repository, a new adjustment began to take shape.
It no longer revolves solely around Hai.
It no longer revolves solely around Lam.
Instead, it revolves around a question the system has never been trained to answer:
If humans begin to anchor themselves to each other,
how will optimization react?