Phase 22 · alignment stress and mechanism controls
Is it the structure, or just the angle?
AT-0175 → AT-0182. A motif can look selective because the identity readout happens to point somewhere the actuator barely reaches. Randomise the orientations, score the worst case rather than the best, and most of that advantage evaporates.
Method
Each system is stable, sparse and driven by one actuator. For every trial the actuator, the identity readout direction, the address readout direction and the initial state are redrawn at random. Selectivity is the ratio of address response to identity response. The number that counts is the tenth percentile, not the maximum.
Internal toy benchmark, seed 175: eight-variable stable sparse linear systems, randomised actuator plus randomised identity and address readout directions across twelve trials per motif. Most motifs lose selectivity under random alignment; a small minority retain median selectivity in the region of 2.5 to 3.1x, with the strongest observed motif near a median of 2.55 and a tenth percentile near 1.59 under matched controllability measures. Status: CANDIDATE STRUCTURAL SELECTIVITY IN TOY MODEL.
Controls
Running alignment stress test…