Per-mode averages across 100 random seeds (top-10 per mode)
For every seed, each engine picks 10 recommendations. We count unique artists, content types, scripture books (distinct books cited), and themes within each top-10, then average across all seeds. Bolded cells mark the best mode per row.
| Metric (mean over 100 seeds) |
Engagement |
Discovery |
Depth |
Unique artists in top 10
Higher = more distinct voices per recommendation set
Unique scripture books in top 10
Higher = wider biblical coverage per recommendation set
Unique content types in top 10
Sermon / podcast / music / scripture — cross-format breadth
Mean cosine similarity to seed
Higher = more similar to seed. Discovery & engagement can both stay close; depth trades similarity for biblical depth.
How much do the three modes agree?
Per-seed, we take each pair's top-10 and count shared items. Averaged across 100 seeds.
—
of an average recommendation set is shared by all three modes
Max possible overlap per pair is 10/10 (identical lists). "All three" measures items appearing in every mode's top-10.