Relationship Hierarchy — ls20 frame 0

10 entities · order-1 (shape) · order-2 (pair motifs) · order-3 (motif support)
selected: none (showing all)

Entities

Classes (graph: composite ≥ 0.01)

Connected components of the graph where edges = composite_sim ≥ threshold.
Lower the slider to find more relatedness, raise it for stricter equivalence.

Frame

click a row at left to focus · shift+click another to compare · click a sim-matrix cell to focus that pair · click a motif to highlight all its pairs

Similarity matrices

tiers: <0.1 0.1–0.2 0.2–0.3 0.3–0.4 0.4–0.5 0.5–0.6 0.6–0.7 0.7–0.8 0.8–0.9 ≥0.9

Pair motifs (Jaccard ≥ 0.7 over pair feature bags)

Each motif is a cluster of entity-pairs sharing the same relational features. "Slot consistency" = average composite similarity within slot (high = uniform).

Shared-component frequency

Spread = # of distinct other classes each class is adjacent to. High spread = candidate mechanic-relevant "shared" component.

Insights

loading...

Pair augmentation comparison

A ∩ B A only B only Why scale 0.5× / 2×? Scale invariance lets two shapes that differ only in size match. Set in fuzzy_similarity's scales=(0.5, 1.0, 2.0). Currently uses lossy integer floor-division — a known source of false positives in the matrix.