The CJC reading is added to the temperature determined by the mV reading. Any CJC error contributes directly to reading error. The error is not a percentage, it is a direct 'adder' (O.2).
There are those who can't tolerate that diminished degree of accuracy and resort to external cold junction compensation.
Some very critical calibration applications use a TC immersed in an icebath to ensure cold junction accuracy, which is verified with a certified liquid-in-glass thermometer.
Historically, there were applications where CJC was accomplished by measuring a reference temperature point assumed to be at a stable temperature and then biasing the calculation with the difference between the reference
temperature and the freezing point of water.
I think most of those have gone away, but should that be required, the CJC could be turned off and the reference temperature substituted.
I suspect that there is also a mode to output temp value in mV per the tables, which is what the old mV calibrators did, using copper wire connections.
In the 726 manual, it isn't clear what units the CJC error is in. Deg C?
The 725 manual states:
A thermocouple (TC) input/output terminal and
internal isothermal block with automatic referencejunction
temperature compensation. (page 11 pdf)
Dan