[rnavarro]

I use defuddle (through JSDOM) to read forum threads, and after upgrading from 0.14.0 to 0.19.5 an SMF topic on eevblog.com came back with 8 of its 19 replies. Every post from reply #9 on was gone. It reproduces on main (556d74e) under both JSDOM and linkedom, and removeContentPatterns: false brings all 19 back.

debug.removals shows the chain: reply #9 is removed as "related post cards", its hr.post_separator as "thin CTA section", and every later post as "trailing non-content".

What matches: in removeByContentPattern, the headingless card loop treats an element as a card grid when at least 70% of its children each hold an img plus a heading or a[href]. One SMF post's div.post_wrapper has three children that each fit: the poster sidebar (status icon and profile link), the post area (message icon and subject permalink), and the footer bar (an icon and a "thanked by" profile link). The first post with a thanker link reads as a three-card grid. hasFollowingProse only counts <p> text, and SMF post bodies are <br>-separated text in divs, so it does not stop the removal, and removeTrailingSiblings then takes the rest of the thread.

The change: a removal target that has a sibling with the same tag and the same child and grandchild tag layout is skipped, since that is one entry in a run of repeated items (posts, comments) and not a trailing block. Classes are ignored because they can be hashed. It uses continue, so a real card grid later in the page is still checked. The tradeoff, also in the code comment: a real related block that sits beside a sibling with the same layout is now kept. That leaves clutter in rather than dropping content.

I looked at two narrower options first. Requiring the cards to share a layout does not separate them here, since postarea and moderatorbar have the same one-level shape. Widening hasFollowingProse to count div text would still drop the last post when that post is the one that matches.

The new fixture content-patterns--forum-thread-post-columns.html is an anonymized SMF-style thread with 11 posts. Without the src change it keeps 3 of 10 replies and the fixture test fails. With it, all 10 are kept. npm test and npm run test:jsdom both pass with no existing expected output changed. On the original eevblog page all 19 replies are kept.