We have well documented paper evidence that we have in common a Henry and Gwenllian Williams 2nd great grandparents for me and 3greatgrandparents for him.
So you are 3rd cousins, once removed.
The mean shared cM for that relationship according to DNA Painter is 48 cM, with a range between 0 and 192 cM over 4514 submissions. 1,686 of those submissions were between 26 and 50 cM, and 2,826 submissions were at 50 cM or less.
At a match of 50 cM unweighted, you are pretty much on the mean value for shared cM at that relationship level. That in itself is not proof of your relationship, but it corroborates your documentary evidence very well.
It is notable that from an unweighted match across 3 segments, Ancestry's Timber algorithm has effectively stripped out over half of the unweighted match length.
The unweighted CiM tool at DNA-sci.com allows comparison between weighted match lengths (after Ancestry's Timber calculation) and the unweighted length or comparable figures from other testing sites that don't attempt to strip IBD segments. The stripped match length of 23 cM and unweighted length both suggest that matches in the region of 3C to 3C1R grouping are the highest probability.
For Ancestry's stripped length of 23 cM in particular, the tool suggests the highest probability (30.7%) for a 3C1R, Half-3C, Half-2C2R or 2C3R match, so I would be entirely happy with that as corroboration of your documented 3C1R relationship based on the paper research.