Schematic fan diagram illustrating trajectory divergence in reflection experiments.

Central black dot represents Loop 0 origin (all trajectories start from same point).
Six colored curved lines fan outward, representing different LLM providers:
Anthropic (blue), Google (orange), OpenAI (sky blue), xAI (yellow), DeepSeek (green), Moonshot AI (purple).

Trajectories spread apart as reflection loops progress:
- Loops 1-3 labeled "Exploration Phase" (trajectories begin diverging)
- Loops 4-7 labeled "Drift Phase" (trajectories widely separated)

Double-arrow at top shows "Increasing Variance (+2.6x)" indicating 2.6x growth in semantic variance.
Dashed circles mark loop positions (1, 3, 5, 7) showing progressive spread.

Caption notes this is schematic illustration; actual trajectories exist in high-dimensional embedding space.
Batch: full-scale-2025-11-20-v2, N=13112 experiments.