Bugzilla

> We should be able to add a filter to reject null sample_id and have the baseline_clients_first_seen logic work correctly.

I don't think this is strictly true, since there seem to be clients with 3 values in the sample id. There must be slightly different calculations for the sample ids that are causing the difference.

> It would definitely be interesting to track this down. But it looks like this still points to a relatively short period of time that was affected by multiple sample_id values, in 2019.

This looks to be mostly true for firefox tv, although the band seems to be a bit wider going from August til December of 2019 (https://sql.telemetry.mozilla.org/queries/79640/source#197816). Looking at fenix, we see duplicates as recently as March of this year. It's increasingly rare from starting September of 2019, although occasionally present (https://sql.telemetry.mozilla.org/queries/79637/source#197814)

> We should be able to add a filter to reject null sample_id and have the baseline_clients_first_seen logic work correctly.

I don't think this is strictly true, since there seem to be clients with 3 values in the sample id. There must be slightly different calculations for the sample ids that are causing the difference.

> It would definitely be interesting to track this down. But it looks like this still points to a relatively short period of time that was affected by multiple sample_id values, in 2019.

This looks to be mostly true for firefox tv, although the band seems to be a bit wider going from August til December of 2019 (https://sql.telemetry.mozilla.org/queries/79640/source#197816). Looking at fenix, we see duplicates as recently as March of this year. It's increasingly rare from starting September of 2019, although occasionally present (https://sql.telemetry.mozilla.org/queries/79637/source#197814)

edit: my work with the redash plots are a bit messy and I forgot to save my changes. It looks like the query for firefox tv shows duplicates fairly recently too.