npm-health: Model Proposal for npm Packages

[dicortazar]

Based on existing literature and state of the art at #2161, this task provides a proposal for the peculiarities of the npm ecosystem. Some of these expectations are repositories that are small in size and with a high bus-factor.

This model will include a set of metrics that should be part of the analysis and the rationale for this:

• Metrics discussion available at npm-health: Define Metrics #2162
• Sources of data discussion available at npm-health: Sources of Data #2163
• Model weights and scoring available at npm-health: Model Weight and Scoring #2164

[canasdiaz]

This task is blocked by #2166 , as we would need to play with data to get metrics and define scores.

[dicortazar]

I've been playing a bit with Census II, Census III and BigQuery. I have now a list of the superset of npm packages as the superset of those three data sources. This includes the package and the URL where development is taking place. We could use this as starting point.

From a more formal perspective, using these three data sources sounds to me like a good initial step:

• BigQuery as this is a list of the dependency weight of a given package for the biggest weight and for the first 1000 of such list.
• Census II and Census II as they are validated by the industry and are official numbers from the Linux Foundation.

Some other numbers, still early and to be confirmed, but:

• The intersection between Census II and II is 244 packages
• The intersection between BigQuery and Census3 is 357 packages
• The superset of BigQuery and Census III is 1036 packages
• The superset of the three is 1150 packages

[canasdiaz]

Note

Quick clarification, the sources are deps.dev (using BigQuery), Census II and III

[canasdiaz]

This is a schema showing the work we are doing here