That's not a source. That's yet another citation.
Here's the source:
The testing in which those cells were evaluated included charges to 4.4V, which brings any conclusions into question.
Claiming a "memory effect" is misleading based on the general public's understanding. The "memory" effect as re-defined by that paper is minimal and has negligible implication to this thread or any other application related to the focus of this forum. It doesn't affect capacity. Nor does it affect SoC computations in any meaningful way.
Here's a more accurate presentation of the data:
"Short cycled" (not fully charging between partial or full discharges) LFP cells will experience slight alterations in voltage response which can slightly change prior SoC to voltage correlations. Test data indicate this phenomenon is localized in the charge/discharge curves and occurs almost immediately. There is no evidence this condition accelerates or changes over time. "New" cell voltage response can be obtained by a single charge to full. Once a cell has been short cycled and a SoC to voltage response range is established, there is no data to suggest that this behavior will change over time unless the cells are charged to full thus resetting the original relationship.
Note that I didn't use "memory effect" anywhere in there because there is no memory effect in these cells based on any definition outside of that research paper.
Please do not continue to spread misinformation.