VUBU did not recover from the Mann image.
The company suspended public image generation four days after the story broke. Initially the suspension was described as temporary. Existing customers retained access to their archives, but no new images were produced.
A week later, subscriptions stopped being charged.
A month after that, VUBU announced that its consumer service would not return.
The statement admitted no fault. There was still no evidence of manipulation, intrusion or data corruption. No employee had been identified as responsible for the Mann image. No outside actor had gained access to the account.
The problem was more fundamental.
The system had apparently done what it was designed to do.
That was now the worst possible explanation.
Responsibility for understanding it passed gradually from the company to institutions whose names sounded less vulnerable to panic.
The National Computational Assurance Directorate began a formal audit of VUBU’s image-generation architecture. A specialist police cybercrime unit worked alongside it. Company engineers remained available under supervision, although their role increasingly resembled that of witnesses being asked to explain a machine they had helped construct.
The audit continued for two months.
It found nothing.
Or rather, it found enormous amounts of information and almost nothing that answered the question.
The production pipeline behaved consistently.
Account authentication was intact.
No administrator had inserted an image.
No undocumented external service had supplied one.
The archived inference package associated with the Mann account generated the same result when rerun.
Delete the output.
Run it again.
The poster returned.
Every time.
The investigators tested older images. They tested random accounts. They rebuilt environments from archived containers and restored models from historical checkpoints.
VUBU continued doing what VUBU had always claimed to do.
This became, in official language, the difficulty.
The deeper the audit went, the less the system resembled a fraud and the more it resembled a machine whose central principle had somehow escaped the people operating it.
Most of the architecture could eventually be understood.
Some of it was ugly. Some of it was brilliant. Some had accumulated in layers across years of development and survived chiefly because nobody had been brave enough to rewrite it.
And then there was Anton Vey’s code.
The auditors encountered his name repeatedly.
Not in comments.
Anton had disliked comments.
His name appeared instead in repository history, old architecture documents, internal tickets and modules that newer developers had learned not to modify unless absolutely necessary.
At the centre of several of those modules was a dependency referred to only as BU.
The designation appeared everywhere once they knew to look for it.
BU_INDEX.
BU_OFFSET.
BU_COORD.
BU_SAMPLE.
BU_RETURN.
The data itself was inaccessible.
Anton had encrypted it.
Not merely secured it behind VUBU’s ordinary access controls, but encrypted it separately using a scheme whose key had never been committed to the repository.
The dataset appeared to be enormous.
Even that was difficult to establish.
Some references treated BU as stored information. Others behaved as though it were an interface rather than a conventional database. Certain functions appeared to request records. Others supplied what looked more like coordinates.
The documentation amounted to three surviving notes.
The first read:
DO NOT MIRROR BU.
The second:
LOCAL RECONSTRUCTION IS NOT REQUIRED.
The third was longer.
BU DATASET MUST REMAIN ENCRYPTED OUTSIDE ACTIVE INFERENCE ENVIRONMENTS.
PUBLIC SAFETY REQUIREMENT.
DO NOT REMOVE.
No explanation followed.
The sentence appeared to have been written by Anton himself.
For several days, the auditors treated public safety as an old security concern: personal data, surveillance exposure, perhaps some abandoned government contract.
Then they found that Anton had used the phrase elsewhere.
Always in connection with BU.
Never in connection with customers.
By January, the audit had produced twelve thousand pages of findings and no explanation for Davey Mann.
The police investigation remained open.
VUBU remained offline.
And in a secured government facility, running on hardware disconnected from the public internet, Anton Vey’s code continued asking an encrypted dataset called BU for generated images of tomorrow.
There was one possibility the audit report did not mention.
Shah remembered Eleanor Price raising it after one of the original interviews, when the recorder had been switched off and the conversation had acquired the looser, faintly indiscreet quality of people standing up to leave.
The developers had wondered about BU themselves.
Not recently.
Years ago.
Anton had never expanded the initials in any document they could find, and when asked directly had tended to answer some adjacent question instead. But he had said enough around it that, eventually, a consensus formed among the senior engineers.
They never used it publicly.
They barely used it privately.
It sounded ridiculous, Eleanor had said. Worse than ridiculous. It made a respectable predictive technology sound like something devised in the back room of a university philosophy department.
And once VUBU worked, the question ceased to matter.
Eleanor thought Anton had understood the necessity of that misunderstanding from the beginning.
Nobody was going to fund a machine whose business plan depended upon a particular theory of time being true.
It had to be predictive technology.
It had to be AI.
Those were ideas investors recognised, with markets attached to them and language already waiting.
Anton had been pragmatic enough not to argue.
But, Eleanor suspected, he had also been incapable of leaving the truth entirely alone.
If he could not say what the machine was doing, he could at least put it in the name and wait for somebody to notice.
The company called it predictive modelling.
The engineers called it inference.
The customers called it tomorrow.
Only Anton had continued calling it VUBU.
Shah had asked Eleanor what the initials meant.
She had looked briefly uncomfortable.
“We think the first part is literal.”
“View?”
“Yes.”
“And BU?”
Eleanor had hesitated.
Then she told him.
View the Block Universe.