cupel 1.2.0

Context window management pipeline for LLM applications
[test]
name = "RecencyScorer ranks items by timestamp"
stage = "scoring"
scorer = "recency"

# 3 items with distinct timestamps.
# countWithTimestamp = 3, so denominator = 2.
# oldest (Jan) has rank 0 → 0/2 = 0.0
# middle (Jun) has rank 1 → 1/2 = 0.5
# newest (Dec) has rank 2 → 2/2 = 1.0

[[items]]
content = "oldest"
tokens = 100
timestamp = 2024-01-01T00:00:00Z

[[items]]
content = "middle"
tokens = 100
timestamp = 2024-06-01T00:00:00Z

[[items]]
content = "newest"
tokens = 100
timestamp = 2024-12-01T00:00:00Z

[[expected]]
content = "oldest"
score_approx = 0.0

[[expected]]
content = "middle"
score_approx = 0.5

[[expected]]
content = "newest"
score_approx = 1.0

[tolerance]
score_epsilon = 1e-9