I can't reproduce, can you upload a screenshot?

Hmm, no idea what that is. It seems that there's some (invisible) content inside the reference at that position, and the editor faithfully highlights it. It could be something added to the markup of references by a gadget, maybe?

It is actually <span class="mw-reflink-text">[14]</span> which is display:none in default Cite style sheet.
It is followed by an ::after pseudoelement.

ext.cite.parsoid.styles.css:

/* Without a group */
.mw-ref a::after {
	content: '[' counter( mw-Ref, decimal ) ']';
}

/* Hide the default rendering in browsers that support content
 * generation by selecting for :after pseudo-element. */
/* stylelint-disable-next-line selector-type-no-unknown */
x-thereisnoelementwiththisname::after,
span.mw-reflink-text {
	display: none;
}

However, ve.ce.MWReferenceNode.less overrides it with the following, which has been fixed with 4ac067c0faf6 by @thiemowmde (T334595). The patch date matches with my regression status conjecture.

.ve-ce-mwReferenceNode span.mw-reflink-text {
	color:transparent;
	position:absolute;
	display:inline
}

However, that does not explain why the absolute position for 16th ref in sample page is wrong with Firefox.