Discrepancies with Appnexus (>10%) and Sonobi (>20%)

[headerbidding]

I have >20% revenue discrepancies with Sonobi and >10% revenue discrepancies with Appnexus (brealtime and oftmedia aliases).

AOL, Conversant, Sovrn and Sekindo show no major discrepancies (less than 2.3%).

I am currently running version 0.18.0 and the discrepancies persist.

Sonobi and bRealtime were unable to help me.

For now I have applied a bidCpmAdjustment of 22.5% for Sonobi and 11% for bRealtime.
Despite this adjustment, Brealtime still shows a 14% impression discrepancy and Sonobi shows a 5% impression discrepancy and still a 10% revenue discrepancy. (I will need to increase the Sonobi bidCpmAdjustment to 33%)

This test page shows my implementation:
http://www.flixxy.com/prebid.htm
And here is a sample live page:
http://www.flixxy.com/the-incredible-flyboard-air.htm

[protonate]

Changes made with #989 and #990 address some issues that may have caused an auction to end early or a "bids back" callback to not fire. Prebid 0.19.0 is released now and can be tested to see if discrepancies are reduced.

[headerbidding]

Great! Thank you for such a quick response!
I will test it as soon as 0.19.0 is available on the "easy download" page ( http://prebid.org/download.html )

[headerbidding]

The discrepancies with Appnexus and Sonobi persist, even with Version 0-19-0, and a bidCpmAdjustment of 0.9 ! Example:
BRT (Appnexus) Report: 2,249 $5.06 $2.25
DFP Report: 2622 $5.20 $1.98

AOL, Conversant, Sovrn and Sekindo show no major discrepancies (less than 2.3%).

[dmitriyshashkin]

@headerbidding do you have discrepancy with Sonobi only in terms of revenue or impression do not match too? We have major discrepancy with them, but strangely enough impressions differ only by 2% while revenue is off almost by 30%.

[headerbidding]

Sonobi shows a 5% impression discrepancy and a revenue discrepancy of 20-30%

[headerbidding]

Changing the timeout from 500ms to 1200ms fixed the discrepancy with Sonobi (both revenue and impressions) on a test page.
Why would that be?
Is there any correlation with Issue #1046 which happens "semi-frequently on publisher's pages who use very tight timeouts of 500ms or less."