Change growth threshold algorithm do use percentage points increase

[diox]

Also contains a refactor to make the code easier to test, as well as admin & reviewer tools tweaks to make the consequences of changing that threshold more obvious.

If the average hotness in a tier is 0.1, and the growth threshold is set to 50, we used to flag add-ons in that tier with hotness over 0.15 (percentage increase) now we'll flag over 0.51 (percentage points increase).

Fixes mozilla/addons#15178

Screenshots

New admin features:

Testing

You can add addons with various hotness and average_daily_users values to your local environment to test and trigger flag_high_hotness_according_to_review_tier task (if the add-on is flagged, it should get a NeedsHumanReview instance with HOTNESS_THRESHOLD reason attached to the latest signed version in each channel that hasn't already been explicitly reviewed by a human), or trust the unit tests that do that.

[KevinMind]

Looks good in principle.. I have a slight doubt/question about the algo change.

[diox]

With the old algorithm this would have been:

        return self.average_growth * (
            1 + self.growth_threshold_before_flagging / 100
        )

[KevinMind]

@diox is it possble to get more specific criteria for a given addon, which values should be set to what in order to trigger or not trigger the flag?

[diox]

It's difficult since it depends on the average hotness of your non-disabled extensions, and what the threshold for the given tier (which themselves depend on average_daily_users). That's why I suggested to trust the unit test.

On a local env, assuming a completely empty database (or at least no add-ons with a average_daily_users value that would matter for the tier we're creating), you can try doing the following:

• Submit 3 unlisted extensions, run auto_approve to make sure they are signed
• Update their average_daily_users value to 1500 each via shell
• Update their hotness value to 0.4, 0.5, 0.6 respectively
• Create a UsageTier in the admin with lower_adu_threshold set to 1000, upper_adu_threshold set to 5000, and growth_threshold_before_flagging to 9

With that setup, the average hotness amongst add-ons in that specific tier that was created should be 0.5, since the growth treshold is 9, that should flag only the one with 0.6 hotness (0.5 + 9/100). So triggering the task (flag_high_hotness_according_to_review_tier() in a shell - optionally flag_high_hotness_according_to_review_tier.delay() if you want to use celery) should result in that add-on to be added to the manual review queue, and its review page should indicate that it has been flagged for growth.

[KevinMind]

Looks good to me 🚢