Accuracy of RTC (Photon)

I have a Photon application that will occasionally connect to particle cloud. It is very well possible that the system will not connect for several months. In the end it might, and the system will then synchronize of course.
Because of this: is there somwhere a specification of the accuracy of the RTC?
And: is there any reason to believe that the accuracy will be different with- or without a backup battery connected to VBAT?
(Note: this topic is related to- (but not the same as-) my previous request)

The Photon clock typically drifts by 1-2 seconds per day although I have a couple that drift even faster than this.

I am moving to Gen3 processors which do not have a battery backed clock so am adding a DS3231 with battery backup. This comes in two varieties. The larger DS3231SN has a drift rate of 2 ppm and the smaller DS3231MZ has a drift rate of 5 ppm.

@Jan_dM The typical approach with the Photon is to have a Time Sync process run every 24 hours to keep the RTC aligned with Particle cloud time. When you have got a resync happening maybe every several months then another issue will come into view - Particle will consider your device not active - it needs to cloud connect at least once per month.

Regards the accuracy of the Photon RTC it depends upon the temperature (the clock is not temperature compensated like the DS3231). The VBAT being connected will not make any difference to the accuracy of the RTC AFAIK. I suggest you run a test of the RTC time drift over an extended period and see if it is such that you need to initiate a cloud connection to resync.

For the reasons @OziGreybeard has outlined I have used DS3231 based RTC with Gen3 devices - the added advantage of this IC is also you can measure the ambient temperature and the RTC provides 2 alarm clock interrupts.