Its probably related to the tiff being 16bit. Probably there is a bug in converting that to 8 bit png. Probably could be fixed by adding an explicit conversion step.

Hmm. Some of the 16bit tiff files don't seem affected by this. For example: https://commons.wikimedia.org/wiki/File:Stephenson_grand_army_republic_monument_4a23943a.tif and https://commons.wikimedia.org/wiki/File:Ceremony_dupont_circle_fountain_31125a.tif

Other examples of affected files: https://commons.wikimedia.org/wiki/File:Army_and_navy_club_washington_d.c..tif
https://commons.wikimedia.org/wiki/File:First_custom_house_in_us_yorktown_va.tif

I have not been able to see a pattern yet.

Also appears to be somewhat of a regression. Bug is not apparent on vips-7.38.5. At least locally.

vips im_clip2fmt in.v out.png 2

seems to fix issue.

In T116947#1763578, @Bawolff wrote:

Also appears to be somewhat of a regression. Bug is not apparent on vips-7.38.5. At least locally.

I wasn't testing locally properly. It is reproducible locally.

Change 249676 had a related patch set uploaded (by Brian Wolff):
Make vips reset bit-depth after convolution. (Fix 16 bit tiff)

https://gerrit.wikimedia.org/r/249676

Upstream bug report: https://github.com/jcupitt/libvips/issues/344

Change 249676 merged by jenkins-bot:
Make vips reset bit-depth after convolution. (Fix 16 bit tiff)

https://gerrit.wikimedia.org/r/249676

jcupitt responded upstream, and says another work around is to use integer convolution instead. e.g.

vips conv shrunk.v x.png sharpen_int.mat

with the matrix being:
5 5 131 0
0 0 -1 0 0
0 -5 -11 -5 0
-1 -11 199 -11 -1
0 -5 -11 -5 0
0 0 -1 0 0

Just as a point of interest, the only thing left in https://commons.wikimedia.org/wiki/Category:TIFF_files_affected_by_MediaWiki_restrictions that legitimately should be in there (afaik) is

• https://commons.wikimedia.org/wiki/File:Fairview_Cemetery_Plot_Plan_-_Fairview_Cemetery,_700_Yale_Boulevard_Southeast,_Albuquerque,_Bernalillo_County,_NM_HALS_NM-6_(sheet_1_of_3).tif (Appears to have a corrupt ExifIFD segment)
• https://commons.wikimedia.org/wiki/File:Andromeda_Galaxy_M31_-_Heic1502a_Full_resolution.tiff (Seems to be erroring out on a pool counter fail [Which should indicate either pool full, or pool timeout reached. Although I thought that the timeout was for waiting for locks, not when you got a lock, so I'm not sure why this file specifically would timeout]. Which is odd, not what I would have expected. Apparently used to work as there is a cached thumb for the size used in thumbnail)