I have had a look at the emails and added the impression data from them. The sums match your numbers exactly.

please check another timeframe for CTRL and VAR of Test 08

Date Imps Donations Donations/Imps
04.10.2017 72450 429 0.005921
05.10.2017 171020 392 0.002292

On Oct. 5th there are a few irregularities.

1. Between 1h and 8 h and between 14 h and 16 h there seems to be no impressions for both banners (at least Kampagnenmonitor pro Stunden)
2. While looking at the numbers above the number of impressions raised again on Oct. 5th. but there is no implications for the number of donations. That rather looks like a problem in the impression counter.
3. Is the impression number on Oct. 5th correct according to the impression email from Peter?
4. Also starting Oct. 5th. the impressions for CTRL were much lower than VAR. Is that correct according to the emails?

Between 1h and 8 h and between 14 h and 16 h there seems to be no impressions for both banners (at least Kampagnenmonitor pro Stunden)
While looking at the numbers above the number of impressions raised again on Oct. 5th. but there is no implications for the number of donations. That rather looks like a problem in the impression counter.

We have the correct number of mails (96 each) on 4th and 5th of October.
On 4th of October there were 9 mails where the impression data was missing for CTRL and VAR:
2017-10-04 01:45:00
2017-10-04 06:30:00
2017-10-04 17:45:00
2017-10-04 18:00:00
2017-10-04 18:15:00
2017-10-04 18:30:00
2017-10-04 18:45:00
2017-10-04 19:00:00
2017-10-04 23:45:00

0n 5th Oct there were intervals where the emails contained no impression data for the banners:
0:45-9:15 (CET) and 13:30-17:00 (CET).
At 13:15 (CET) only 10 impressions were counted (apparently "ramping up" for the 0 impression data).
At 12:00 (CET) the email is missing the CTRL impressions, the VAR impressions in that email are only 10.

Is the impression number on Oct. 5th correct according to the impression email from Peter?

When recounting the impressions from the mails on the 5th I only get 145110, which is also the value in the database. I'll have to investigate where the 171020 is coming from.
For the 4th it's correct.

Also starting Oct. 5th. the impressions for CTRL were much lower than VAR. Is that correct according to the emails?

CTRL: 72130
VAR: 72980
A difference of -850, which is 1.2%

On the 4th it's reversed:
CTRL: 36690
VAR: 35760
A difference of 930, which is 2.6%

I investigated the mystery of the "VAR is higher than CTRL" and the differing impression counts and found out that we get the impression counts from the 'editor campaign' (5 banners named WMDE_editor_campaign_autumn17_a through WMDE_editor_campaign_autumn17_e ), which were all counted as "var". I've corrected the association in the database and created a ticket for the bug in our impression fetching software: T177629

This does not solve the "impression counts are too low overall" problem, but at least from our side, we can now be confident again, that we're processing the emails we're getting correctly.

@kai.nissen @JeroenDeDauw @Pablo-WMDE : If further investigation for other days is needed during my vacation, I've created a branch for the "analytics" scripts that can reproduce the impression data from the Operations Center. https://github.com/wmde/fundraising-backend/tree/funky-impression-throwaway-scripts
Do not merge.

@tmletzko Do you consider this done or do you need any further data?

jo, done.