Taxonomy reconciliation — embedding clusters × verb/thematic grammar (one grid, two axes)

2026-08-03  ·  taxonomyclassifierspotwhitespace

The verb macro-families (clusters decomposed by primary verb)

Diagnosis: the 34 clusters mix two axes

Some clusters are VERB-defined (Speed Escape, Slap Tower, Training Simulator, Hide and Seek, Ball Deflection); others are THEME-defined (Dinosaur, Anime RPG, Zombie, Prison RP, City RP, Brainrot Runner). Same verb + different themes = different clusters (climb → Tower Obby vs ASMR Tower vs Slap Tower). This is the Plus One Speed Monkey/Slime/Superhero pattern with counts attached: verbs are bounded (~15 macro verbs cover the market), themes are unbounded and carry the novelty.

Consequences

  1. Stage-D classifier scheme: bounded verb list (start from these ~15) + open theme list + one-line fantasy. Cluster names stay as display labels over verb×theme cells.
  2. The whitespace engine IS the grid's empty-cell detector: 0-game combos ([Base Builder]×[Driving RP], [Training Simulator]×[Tycoon]) = empty cells with healthy parents. Sparse midpoints = contested-but-winnable cells.
  3. Competition density per cell is a Decide input: Brainrot Runner n=34 = current meta, late-wave entry risk; Dinosaur n=8 = evergreen fantasy (kid-sleep test passes), sparse.
  4. Trend-native vs evergreen: Brainrot Runner (n=34, largest), Slap Tower (phonk/aura), ASMR Tower, Squishy Trade = trend wrappers on old verbs → fast ramp, short half-life. Dino/Farm/Prison/City = evergreen. Radar should tag every cell with trend half-life class.
  5. Resolves the convergence-table open question "embedding taxonomy vs verb/thematic grammar": not competitors — two axes of one grid. Clusters = what exists; grammar = how to decompose and recombine.