Tokens
Context
Utilization
Model & Configuration
Model / Context Window
Block Size (tokens)
Zone Threshold
Instruction Placement
Start → Primacy Boost
Tokenizer
Upload File
PDFCSVXMLTXTJSONMDHTMLPY / JS / TS
🔒 Tokenization is entirely local — data never leaves your browser
Paste or type text0 chars
⚠ Content exceeds model context window — red blocks are truncated. The model never sees overflow tokens.
Total Tokens
0
real BPE (cl100k)
Primacy Zone
0
strongest recall ↑↑
Lost in Middle
0
danger zone ↓
Recency Zone
0
high recall ↑
Overflow
0
truncated ✕
Recall Probability Curve — Asymmetric U-Shape (model-specific)
Based on Liu et al. 2023 "Lost in the Middle"
■ Primacy (stronger than recency) ■ Middle (danger zone — skewed past center) ■ Recency (high recall, weaker than primacy) Dashed = symmetric reference
Memory Architecture Map — Token Blocks
Hover / tap block → decoded token preview + recall %
Ready — paste text or upload a file to visualize token memory architecture
Zone 1 · Primacy (highest recall — slightly brighter)
Zone 2 · Lost in Middle (danger — dip skews past center)
Zone 3 · Recency (high recall — slightly weaker than Primacy)
Zone 4 · Overflow (truncated — model never reads)
Block = 50 tok
Zone Breakdown
Primacy
Middle
Recency
Overflow
Recall probability