I would use all frames where the PCM is in closed loop mode. In that mode the PCM is operating from the VE table AND using fuel trim corrections. So the fuel trim corrections that are being logged in those frames will be true and accurate (well as accurate as the narrow band O2 sensors are functioning - which is not very accurate).
When the PCM is in open loop, the fuel trims logged in the scan tool are usually frozen at their last value in closed loop before entering open loop. i.e. you should see that when you stand on the throttle and accelerate hard. The PCM enters Power Enrichment mode and (obviously open loop) and freezes the LTFT values. No learning takes place, no trims are updated, no O2 feedback is taken into account. The PCM is just running off the VE table with no LTFT correction applied. However, without a wide band O2 sensor you don't get any feedback on the fuel status during that PE mode. So the data is useless for VE tuning.
I would not be concerned at differentiating between learning and not learning. The "Not Learning" condition will coincide with open loop 99% of the time. Some of the other times will be during evap purge and failed O2 sensor conditions.
Even if the PCM is not learning (assuming the O2's have not failed) the PCM will still be using the already learnt LTFT values and so the data logged for LTFT will still be valid.
Bottom line: discard all high throttle transients*, and all open loop data.
* For example, EFILive V7 has a filter that removes all frames where the throttle has changed by more than 10% in 100ms (i.e. since the previous frame).
Regards
Paul