Overview
Bucket Details - Top 5 from Each
Shows the top 5 hooks from each retrieval bucket with their scores and rankings.
Bucket Composition Strip Plot
Each vertical bar represents one final recommendation, colored by which retrieval buckets found it.
Multi-colored bars indicate hooks found by multiple strategies, suggesting higher confidence. Single-color bars show strategy-specific finds.
Multi-colored bars indicate hooks found by multiple strategies, suggesting higher confidence. Single-color bars show strategy-specific finds.
Score vs Rank Analysis
Two scatter plots showing how normalized retrieval scores translate to rankings.
Left: normalized scores vs raw bucket rank. Right: normalized scores vs final RRF rank. Y-axis inverted so rank 1 appears at top. Look for patterns in how bucket scores predict final rankings.
Left: normalized scores vs raw bucket rank. Right: normalized scores vs final RRF rank. Y-axis inverted so rank 1 appears at top. Look for patterns in how bucket scores predict final rankings.
Content Freshness Analysis
Content age in hours for final recommendations (scatter) and age distributions by bucket (boxplot).
Fresh bucket should cluster near bottom (newer content). Diverse/follow buckets may show wider age ranges. Log scale accommodates both hours-old and months-old content.
Fresh bucket should cluster near bottom (newer content). Diverse/follow buckets may show wider age ranges. Log scale accommodates both hours-old and months-old content.
Performance Timing
Time spent retrieving from each bucket, measured in milliseconds.
Taller bars indicate slower buckets. Since buckets run in parallel, total wall time ≈ slowest bucket time. Look for consistently slow buckets that might need optimization.
Taller bars indicate slower buckets. Since buckets run in parallel, total wall time ≈ slowest bucket time. Look for consistently slow buckets that might need optimization.