Skip to content

Observer Ladder Clustering Summary

Source artifact: checkpoint_sequence_summary.json

This analysis uses only the saved wormhole ladder checkpoint summary. Features were standardized before clustering.

Checkpoint Feature Table

Checkpoint Manual regime Discovered cluster OPL mean OPL max portal_hit_density throat_event_density crossings_per_pixel segments_per_crossing AverageSegmentsPerRay
mouth near-side cluster_2 9.9599 15.8071 0.1465 0.0969 0.6495 153.2590 99.5403
mouth_to_throat_approach near-side cluster_2 9.7287 15.5162 0.1633 0.1048 0.6987 139.6225 97.5569
throat throat cluster_2 9.5078 15.1926 0.1750 0.1139 0.7479 128.1728 95.8566
post_throat_backstep_01 bridge cluster_1 7.5908 12.2611 0.0964 0.0555 0.2098 366.0292 76.7983
post_throat_exit_approach far-side cluster_0 8.1171 14.8980 0.1798 0.2111 1.6544 50.3105 83.2336
exit_lookback far-side cluster_0 8.4337 16.3070 0.2557 0.2198 1.4200 60.9614 86.5624

Clustering Comparison

Method k ARI vs manual labels Silhouette
agglomerative 2 0.1429 0.4320
agglomerative 3 0.5946 0.5547
agglomerative 4 0.2857 0.4020
agglomerative 5 -0.0976 0.1067
kmeans 2 0.1429 0.4320
kmeans 3 0.5946 0.5547
kmeans 4 0.2857 0.4020
kmeans 5 -0.0976 0.1067

Interpretation

  • The best alignment with the current manual regime labels was agglomerative at k=3, with ARI=0.5946 and silhouette=0.5547.
  • The bridge checkpoint post_throat_backstep_01 falls in cluster_1 with cluster size 1.
  • Bridge outlier verdict: yes. It is isolated as a singleton cluster under the best-performing automatic clustering pass.
  • The automatic clustering still groups the near-side mouth approach and throat progression together more strongly than the manual taxonomy separates throat from the rest of the near-side leg.

Figures: