THE UNWRITTEN MACHINE / RESEARCH ATLAS / V0.5

What survived
the control?

An inspectable record of bounded particle simulations. Each candidate is replayed beside a paired neutral-rule world with identical starting positions. A visual difference is an observation about this model, not a discovery of life or intelligence.

—COMMITTED TRIALS
—PASS PAIRED CONTROL
—MANUAL CAMPAIGN CYCLES
—REPLAYED IDENTICALLY

Loading committed observations…

SELECT A TRIAL

—NOVELTY DISTANCE
—INITIALIZATION ROBUSTNESS
—MEAN PAIRED CONTRAST
—CONTRAST SPREAD

INTERACTION RULES

NEUTRAL MATRIX / SAME START

Select a candidate to inspect its endpoint.

HOW TO READ THIS

The same seed initializes both views. The control replaces all species attraction values with zero while keeping the short-range repulsion, drag, radius and starting positions. The distance compares five endpoint features for two paired initializations.

“Pass” requires both paired distances to reach 0.06. This threshold was chosen as a descriptive filter, with no statistical calibration. Replay means identical output from the same deterministic code; it does not establish independent replication. Novelty uses 12 fixed reference seeds plus previously committed trials under the same protocol. Later cycles have a larger comparison set, so their scores are not directly comparable as evidence of progress.

All campaigns were run manually in a bounded local process. This public page reads a static JSON archive; it does not run scheduled jobs or paid inference.