This is from the MiXCR documentation:

Under the hood analyze amplicon is equivalent to the execution of the following MiXCR actions options:
> mixcr align -s hs -p kAligner2 input_R1.fastq input_R2.fastq alignments.vdjca
... Building alignments
> mixcr assemble alignments.vdjca clones.clns
... Assembling clones
> mixcr exportClones --chains IGH clones.clns clones.txt
... Exporting clones to tab-delimited file

So when using analyze amplicon, you are using mixcr exportClones under the hood and not mixcr analyze, hence the difference in output.

I believe what you are experiencing results from a bug, please see this issue. If this is the case, all these files should be the same according to the bug report. Fortunately the developers have already addressed the bug in version 3.0.3.

This is from the MiXCR documentation:

Under the hood analyze amplicon is equivalent to the execution of the following 
MiXCR actions options:
> mixcr align -s hs -p kAligner2 input_R1.fastq input_R2.fastq alignments.vdjca
... Building alignments
> mixcr assemble alignments.vdjca clones.clns
... Assembling clones
> mixcr exportClones --chains IGH clones.clns clones.txt
... Exporting clones to tab-delimited file

So when using analyze amplicon, you are using mixcr exportClones under the hood and not exportAlignments, hence the difference in output.

I believe what you are experiencing results from a bug, please see this issue. If this is the case, all these files should be the same according to the bug report. Fortunately the developers have already addressed the bug in version 3.0.3.