Tiers of memory: the OS metaphor

Tier lab: a support agent runs three sessions and twenty turns, and its twelve facts want 97 tokens of need. The core block holds eight lines of 48 tokens — 384, against 576 for all twelve — so four facts sit outside it. Evicting by need costs 9 tokens of page faults (9.3%), by recency 24 (24.7%), by age 63 (64.9%): the rule that decides what stays is the memory. Then price the tier in the window: 384 tokens of core is 384 tokens of conversation the queue cannot hold, and a second block leaves the queue 410 tokens when the 70% warning fires.

This Memory Engineering lesson is part of a paid plan. Sign in with a plan that includes this course to read it.