TWO BEDS / ONE SHARED RESERVOIR
Care for what
you cannot see.
Every journey takes time. Water arrives later.
A reading becomes a memory when Lumen leaves.
Lumen is finding its bearings.DRAG TO ORBIT · SCROLL TO ZOOM
A control sends Lumen to that mechanism. Weather advances on every travel and operation tick. Pausing stops physics; an intervention pauses automatic decisions.
01 / THE COST OF NOT BEING THERE
What stayed in memory?
Reed bed · B
Solid: actual moisture in the recorded evidence. Dashed: Lumen’s estimate. Shaded: the bed’s viable range. Evidence plots can show more than the learner sees.
02 / CLAIMS HAVE TO EARN THEIR PLACE
Does memory change the outcome?
The new nursery tests local readings, learned delays and a finite water supply. The maze comparison tests transfer on layouts excluded from every training and selection run.
Two-bed care
Five controls · four training seasons · eight evaluation seasons per control. 2,880 actions per control; 14,400 total.
Transfer to unfamiliar mazes
Evolution with transfer, without transfer, and random challenges. Three training seeds; exactly 12,000 training/search actions per arm and seed.
Studies run in the Mac service and continue while you switch pages. Completed work survives service restarts; interrupted work is charged conservatively. Evaluation seeds become exposed after this first run.
A place to test a small claim.
The body moves along an authored map. The controller receives local moisture readings, nearby equipment states and current weather. It remembers observations and fits simple response models with different delay hypotheses. These are explicit design choices, not evidence of general intelligence.