Configurable Connections Algorithms

Connections Pro exposes score and ranking controls so you can tune connection results to match your current workflow.

Use this page as a practical setup guide: choose the right candidate pool, pick a score strategy, then add ranking only when you need extra shaping.

Setup flow (recommended)

Choose a results collection key.
- Smart Sources for note-level discovery.
- Smart Blocks for heading/block-level precision.
Select a score algorithm.
- Start with Cosine Similarity as a baseline.
- Move to feedback or metadata weighting after you observe noisy or under-weighted results.
Optionally add ranking/post-process.
- Keep None for speed and deterministic score order.
- Use Re-ranking model or Recency rank when you need final ordering tuned for a specific task.

Quick decision guide

If results are not relevant, adjust score first.
- switch collection (Smart Sources vs Smart Blocks)
- change score algorithm before adding ranking complexity
If results are relevant but out of order, adjust ranking.
- keep score algorithm stable
- add Re-ranking model or Recency rank
If results are relevant but keep surfacing known low-value notes, apply feedback-aware scoring.
- hide recurring noisy notes
- use Similarity Adjusted by Feedback or Similarity Weighted by Feedback

Controls at a glance

Results collection key selects Smart Sources or Smart Blocks as the candidate pool
Score algorithm selects the primary ranking strategy
Post processing allows a secondary rerank stage such as Cohere via Smart Rank when configured

Score algorithms

Algorithm	What it does	When to use
Cosine Similarity	Ranks by embedding similarity between the current note and candidates.	Default when you want stable, feedback-free results.
Similarity Adjusted by Feedback	Penalizes candidates similar to hidden notes.	When you want to reduce recurring noise.
Similarity Weighted by Feedback	Boosts candidates similar to pinned notes and dampens hidden notes.	When you actively pin/hidden to steer results.
Similarity Weighted by Key + Frontmatter	Multiplies similarity based on key fragments and frontmatter matches.	When metadata should emphasize or de-emphasize results.

Practical presets

Default, low-maintenance: Smart Sources + Cosine Similarity + None
Reduce repeated noise: Smart Sources + Similarity Adjusted by Feedback + None
Leverage explicit curation: Smart Sources + Similarity Weighted by Feedback + Re-ranking model
Metadata-driven retrieval: Smart Blocks + Similarity Weighted by Key + Frontmatter + Recency rank

Ranking algorithms

Algorithm	What it does	Notes
None	Keeps the original score order.	Fastest option.
Re-ranking model	Applies Smart Rank to reorder the scored list.	Requires a Smart Rank model in Smart Environment Pro settings.
Recency rank	Orders results by most recently modified items.	Returns score labels with time-ago hints.

Score vs ranking boundary

Score algorithms decide relevance for each candidate.
Ranking algorithms decide final order of already-scored candidates.
Keep metadata weighting (key/frontmatter) in score actions and recency shaping in ranking actions for easier iteration.

Troubleshooting quick checks

Results feel too broad -> switch to Smart Blocks and re-run.
Same low-value notes keep returning -> hide them and use feedback-adjusted scoring.
Valuable tagged/foldered notes are under-ranked -> configure key/frontmatter weights.
Results are relevant but ordering feels off -> add Re-ranking model or Recency rank.

Example weighting config (Similarity Weighted by Key + Frontmatter)

Use JSON in the score algorithm settings:

{
	"key_weights": {
		"Projects/": 1.2,
		"Readwise/": 0.8,
		"#High-value heading": 1.3
	},
	"meta_weights": {
		"status:evergreen": 1.15,
		"type=spec": 1.1,
		"reviewed": 1.05
	}
}

key_weights applies substring matches to key/path/heading signals.
meta_weights supports key, key:value, and key=value matchers.
multiple matches multiply together.

This is best for structural intent. Use recency ranking for freshness intent.

Tip

Try Smart Blocks as the collection for fine grained context when your notes have strong headings. Use Smart Sources for broader note level discovery.