CC430 RTC Compensation

I think you misunderstood the calibration mechanism.

The system cannot change the frequency of its clock source. If it is driven by a 32768Hz crystal, the only thing it could do is dividing it by 32767, 32768 or 32769 to correct any error. Which would be a 30 ppm step.

So the 'calibration' is done differently: every 64 minutes, the system will add 256 or subtract 512 clock pulses per calibration bit.
This way, each second will still be off (and this is why you don't see a change) but some seconds will be off in the opposite direction, so in average, the clock is adjusted.

The 1s output is only for determining the error without any correction, so you can calculate the required settings. It will not reflect any changes of the calibration bits, at least not noticeably, since the effect is only visible for a few pulses inside a 64 minute range. The 256 and 512 Hz frequencies are no subject to the calibration settings at all, as they are directly derived from the input clock.

After all, the RTC is an RTC and not a high-precision high-frequency clock source. It is important that it is not off by a minute every day. it's not important whether it is off by a few milliseconds in one minute and correct again in the next, as long as there are no jumps in the timeline.