Day
15
of 90
Memories
824
3.05 MB
Weekly Recall
0%
0/0 questions
Avg Similarity
0%
cosine distance
Tier Transitions
252
HOT → WARM achieved!
Memory Tier Distribution
HOT
572 (69.4%)
WARM
252 (30.6%)
COLD
ARCHIVE
Experiment Progress
Day 15 / 90 (16.7%)
March 4, 2026HOT->WARM achieved Day 15!WARM->COLD expected ~Day 60June 2, 2026
What This Experiment Measures
Primary Questions
- 1. How fast do semantic memories degrade when not accessed?
- 2. Does recall accuracy drop before or after tier transitions?
- 3. Do access patterns (recall testing) slow decay measurably?
- 4. Is the 30% similarity threshold appropriate for "meaningful recall"?
- 5. How does memory density (count) affect recall quality?
Design Decisions
- Weekly testing (not daily): Prevents access boosts from artificially keeping memories HOT
- 25% sampling (max 50): Lets 75% of memories age without any interference
- 30% similarity threshold: High bar for "successful recall" to avoid false positives
- 16-subject rotation: Cross-domain knowledge tests domain independence of decay