Scoring methodology
A score you can act on
Popularity is easy to measure and easy to game. Usefulness is harder. The practical score is our attempt to measure what a working developer actually cares about before installing a plugin.
Practical score
ruflo
What goes into the number
Five dimensions, each weighted by how much it affects day-to-day usefulness.
The five dimensions
Each score is computed from public GitHub signals and the project's own documentation.
Maintenance
Is the project alive — recent commits, releases, and issue activity.
- How
- Commit activity over the last 90 days, release recency, and issue health using a Wilson score for small samples.
- Why
- DeepSeek Harness moves fast. An unmaintained plugin breaks quickly, so this carries the highest weight.
Practicality
Can you actually use it — docs, examples, and installation notes.
- How
- README completeness: install and usage sections, code examples, and feature documentation. Skills are also checked for a complete SKILL.md.
- Why
- A powerful plugin with no instructions is not useful. Documentation is the first signal of practicality.
Ecosystem pull
Is the community adopting it — stars and participation.
- How
- Log-normalized stars against the 99th percentile of the index, plus fork participation estimated with a Wilson score.
- Why
- Stars are not quality, but a project nobody uses carries more risk. Popularity is a rough proxy for trust.
Ease of setup
How quickly can you get from discovery to a working install.
- How
- Clear install commands parsed from the README, and whether extra tokens or API keys are required.
- Why
- One-command installation is the core promise of the marketplace. Heavy setup lowers the score.
Signal quality
Is the project's metadata complete and trustworthy.
- How
- Coverage of description, license, topics, homepage, and README.
- Why
- A missing license or description is a basic trust gap. Completeness is the floor, not the ceiling.
Fusion: weighted geometric mean
The final score is not a simple average. A weighted geometric mean means one weak dimension pulls the whole number down — exactly as it should.
A plugin with 95 popularity but 20 maintenance would average 57 arithmetically, but only 44 geometrically. A popular project that no longer works is not a good recommendation.
Confidence adjustment
New or incomplete entries are down-weighted with a Bayesian prior. Thin data cannot produce a suspiciously perfect score.
How to read a score
Use these bands as a starting point, then read the detail page.
Strong across every dimension: active, documented, adopted.
Solid overall, with one or two dimensions trailing.
Works, but check the weak spots before installing.
New, incomplete, or stale. Read the repository first.
Questions
Why not just sort by stars?
Stars are cumulative history. They do not reflect documentation quality, setup friction, or recent maintenance. The five dimensions separate those signals, and the geometric mean prevents one metric from hiding a weakness.
Does a low score mean a plugin is unusable?
No. A low score usually means incomplete metadata, slower maintenance, or heavier setup. It is a first-pass filter; read the detail page, the README, and judge against your own needs.
How often are scores updated?
Scores refresh with the daily index. New entries are down-weighted by a confidence prior until their data is complete enough to score fairly.
See the score in context
The methodology is more useful when you apply it. Browse the highest-scoring entries and compare.