Context Length
0
tokens in context window
Memory Entropy
0.000
mean attention distribution H
Collapse Score
0.000
⟁ Hidden State Drift — D(t) = 1 − cos(h_t · h_{t−1})
≋ Attention Entropy per Layer
⊞ Layer-wise Hidden State ‖h_t‖₂
⟳ Memory Decay — Early Token Retention
◈ Token Retention Heatmap — Layer × Token Position
LOW RETENTION
HIGH RETENTION
▸ System Log
[SYS] MEMOSCOPE initialising…
[NET] connecting to inference stream…