Thanks @Asher for testing and finding this out: despite an already converted copy existed, outputting it again converted it again, i.e. it deleted the copy and made it afresh from the untouched source. (there was supposed to be a warning ("Anything you changed inside it is lost") but bueried into a longer warnign text that nobody reads (I wouldn't , so your replacements went with the old copy. The new release 1.9.2  (QUICKFIX) now will make sure: Every change made in a copy is now recorded inside the copy, so the record survives a restart. Output over a changed copy says "the converted copy is your output", lists the changes map by map, and keeps them by default (Esc keeps them too). Only "Convert again, losing them" converts again. Frankly, this developed into an unecessarily complex "slideshow" of confirmation windows and a lot of bloat text. For the time being it works, but I have to look into the logic a bit and try simplifying it.

I see I missed quite some posts from before:   Indeed, equalising height was important (same reason ships would use towers), but the gangplank (epibathra) was used as early as the Siege of Motya in 398 BC.   Indeed, because of the additional weight (another reason I proposed to change this tech to Raw Hides). I think only helepoleis could handle that (the one built at Rhodes was not the only one), or later towers from Imperial Rome (which could anyway spontaneously collapse, like at the Siege of Jerusalem in 70 AD). I also had helepoleis in mind when proposing champion ships and engines, since it would be a powerful siege tower unique to the Macedonians.

No, that would be far too complex - what I have in mind is more a wrapper tool that helps defining the map/game/player characteristics, then hand over to the real Atlas for terrain sculpting, but then help placing entities, particularly walls and palisades (as a start) and visualizes that in 2D and a simplified 3D map. I am trying to use the new tool with Atlas in a kind of "pair-editing" (the tool only looking at and manipulating xml, Atlas basically everything). Atlas is very powerful but appears complex to me. The new tool is an attempt  trying to reduce complexity where possibly without actually taking away Atlas powerful capabilities.   Confirmed: convert_again deletes the copy and regenerates it from the untouched source, wiping any replacements. Please wait for a yet another version 1.9.2 .... (also saw a lot of documentation flaws that needed fixing). (thus demonstrating why short release cycles can be dangerous even with a lot of (automated) testing.)

I added a replacements for things on a map, then I went to the convert tab, and clicked output, and it said there were fails. I then go back to the findings tab, and it undid all the replacements I put in

Like your own version of Atlas?