Eviction and Scoring

Motivazione

Scoring prevents unbounded growth while keeping durable, useful observations.

Teoria

Hook mode:

score=0.3t+0.2r+0.5a score = 0.3t + 0.2r + 0.5a

Passive mode:

score=0.4t+0.3r+0.3d score = 0.4t + 0.3r + 0.3d

t is type weight, r recency, a access, and d diffusion.

#keep

Observations containing #keep are protected from pruning.

ADR: deterministic scoring

CI merge should be reproducible so teams can explain why observations survived or were evicted.