For a ten year old project, it makes sense to have curves more normalized (and each data point not being entirely disjunct). The only small downside would be that we cannot quickly realize a trend that has only happened in the last month.

Generally agreeing on having less data points and clearer data. Now to identify which data sources we have, keep, and should update to the proposal above.

@Dicortazar, what do you think about the proposal to normalize graphs? Maybe we can offer tables with the exact numbers for every month, but then normalize graphs to see the actual trends with more consolidated data.

After a chat today with @Dicortazar, and after he showed the reports they are preparing for OpenStack (example), I'm actually leaning toward quarterly samples for metrics counting people, i.e.

• Number of active users in Gerrit
• Number of users exercising +2/-2 in Gerrit
• Active users in Phabricator

Taking quarterly samples is probably better to count the "long tail" of volunteer contributors. We can assume that almost all professional developers are active every month, but that is not necessarily true for volunteers.

On the other hand, graphs with quarterly metrics also provide less volume of information, less noise. Anyway, something like the population of contributor doesn't change radically from a month to another (there might be exceptions, yes).