Tamura’s job was to watch the universe.

More precisely, his job was to operate and maintain the Integrated Space Observation Data System — he wasn’t personally watching anything. Satellites watched. Sensors recorded. Servers processed. Tamura checked the logs every day to make sure the whole machine kept running without a hitch.

The system tracked 940 million celestial objects per day. Galaxies billions of light-years away, their redshifts, luminosities, shapes, and positions all logged and updated in real time. The system’s name was ARGUS.

“Tamura, about this month’s performance report —”

His supervisor, Section Chief Kanda, cracked the office door and stuck his head in. Tamura didn’t turn around.

“I need a little more time.”

“It’s already the 25th.”

“I’m aware.”

A silence. The door clicked shut. Tamura let out a long breath.

The problem was simple: for the past three weeks, the monthly performance report would not complete.

The process would run. Ninety-two hours would tick by. Then, in hour ninety-three, the system would crash. The log always said the same thing: Authentication error: facial recognition failed for report approver.

ARGUS was a state-of-the-art system with biometric security built into its core. At the end of each month, finalizing the report on the server required the System Administrator to stand in front of a camera and pass facial recognition. The System Administrator was Tamura.

He’d tried everything. Head-on. Chin up. Glasses off. Glasses on. Under fluorescent light. In natural light. Morning. Night. Same error every time.

Authentication failed. Match rate against registered profile: 17%.

Seventeen percent.

Tamura dug into ARGUS’s codebase. The facial recognition module used computer vision technology developed as a byproduct of astronomical observation — an AI trained to detect luminosity changes in galaxies down to 0.01% precision, repurposed for human face identification.

High accuracy by design, the developer notes said. False positive rate well below the 0.1% industry standard. Subject identification at cosmic-observation precision.

Tamura opened the system administrator database and pulled up his own registration photo. It had been taken three years ago, on his first day at the company. He’d been heavier back then. Seven kilograms heavier, to be precise.

ARGUS never missed a 0.01% shift in a galaxy’s brightness. Naturally, it hadn’t missed the seven-kilogram shift in Tamura’s face either.

He submitted a re-registration request. HR replied by email: Updating the facial data for a System Administrator can only be performed once the existing administrator has passed facial recognition.

Tamura stared at the screen for a while.

Outside his window, ARGUS hummed along perfectly. The dashboard showed 940 million objects updating in real time. Galaxies ten billion light-years away, tracked moment by moment.

The door opened again.

“Tamura, the report —”

“Kanda-san.” Tamura finally turned. “You recognize my face, right?”

”…Sorry?”

“Do you think I look different from three years ago?”

Kanda considered this for a moment. ”…You were heavier, weren’t you. Back then.”

“Thank you.”

Tamura stood in front of the ARGUS camera and tried one last time.

Authentication failed. Match rate against registered profile: 17%.

An eye that could see to the edge of the universe had spent three years failing to see the face of the man sitting right beside it.