VCMI random maps feel noticeably different from HotA’s and H3 SoD rmg. Tldr: The scenery is repetitive (identical obstacles are bunched together making the map look ugly), there are large empty spots on the map with nothing to pick up, the start is often blocked by a guard and there is less space to move around because vcmi places more guards, vcmi places a lot of trash artifacts like scrolls or level 1 shrines of thought objects which feel unrewarding to pick up or visit.
So I did what any sane person would. I made tooling for statistical map analysis, generated large batches of my favorite map template in both h3 SoD/Hota and vcmi and made a report. Here is my analysis and findings in more detail.
VCMI RMG vs HotA RMG: main differences found by map comparison
Basis: template Clash of Dragons G, 252×252, 1 level, 2 players, no water. 30 VCMI maps (monsters strong, HotA mod active) compared with 20 HotA 1.8.0 maps (monsters 3). Values are medians per map. Every finding below is statistically strong (Cliff’s |d| ≥ 0.83, q < 0.0001) unless marked otherwise. Full report: VCMI random map generator compared with HotA: problems found
| # | Problem | VCMI | HotA | Likely code | Impact |
|---|---|---|---|---|---|
| P1 | Identical obstacles cluster together | Same-obstacle pairs within 8 tiles: 1.28× random | 0.32× random (HotA actively keeps them apart) | ObstaclePlacer, ObstacleSetHandler |
Visual monotony |
| P2 | Large empty areas, treasure in heaps | 4.1% of passable tiles are >4 tiles from any object; largest empty radius 8.6 | 0.6%; radius 6.3 | TreasurePlacer (pile assembly), ObjectManager weights |
Dead stretches of map |
| P3 | Few big obstacles instead of many small ones | 7,350 obstacles, median 4 tiles each | 12,720 obstacles, median 2 tiles (same blocked area) | ObstaclePlacer |
Blockier look, makes P1 more visible |
| P4 | More guards, weaker top end, more closed start | 1,272 monsters; p90 guard 157k; 1,770 tiles reachable from start without fighting; 8 guards between starts | 1,108 monsters; p90 205k; 2,510 tiles; 5 guards | TreasurePlacer (one guard per pile), ConnectionsPlacer |
Slower, more boxed-in start |
| P5 | Different object mix | 1,243 artifacts (324 scrolls); 533 gold piles; 175 dwellings; ~85 decorative animals | 690 artifacts (118 scrolls); 1,154 gold piles; 307 dwellings; none | TreasurePlacer value weighting, HotA mod object config. Partly caused by template conversion |
Treasure value +24%, mostly from artifacts; some HotA objects never appear |
| P6 | More roads, fewer rivers (low priority) | 1,630 road tiles, 21 rivers | 1,084 road tiles, 38 rivers | RoadPlacer, RiverPlacer |
Cosmetic, slightly faster routes |
| V1 | Same seed does not give the same map | 3 of 3 reruns differed in 2–3 zones (single-threaded) | n/a | Probably pointer-ordered containers (std::set<shared_ptr<…>>, maps keyed by shared_ptr<Zone>) |
A map can’t be regenerated from its seed |
Where VCMI does well or the comparison doesn’t apply: VCMI is fairer between the two players than HotA. There is usually a similar amount of free unguarded space between two players and similar guard strength and closer total object value.