When the benchmark fails its own rule

An art audit with a surprise in it

The game has a simple rule for its pixel art: no near-black. In practice that means no pixel darker than a value of 58. A scan checked every installed art file against it, 841 images in all.

The art that had been re-toned for the rule passed cleanly:

The interesting finding was elsewhere. The Sunmeadow buildings and trees are the benchmark, the art every new piece is measured against. They fail the rule too:

So the rule as written and the art it was written to protect disagree. Either it means "no near-black", and the benchmark is wrong, or it means "match the houses and trees", and the threshold is wrong. That is a question for the art direction, not something a scan can settle, and nothing was brightened while it is open. The scan also cannot tell a shadow from an outline, which is part of why a blanket fix would have been a mistake.

Detail that has to match

The same audit measured pixel density. The blossom tree and the tent are drawn at 16 pixels per world unit, while the trees around them sit at about 71 to 82. That makes them roughly five times coarser than their neighbours.

Replacing them is not just a matter of drawing finer art. The editor tool that places these props had 16 pixels per unit written into it, instead of reading each prop's own setting, so a finer replacement would not have been placed at its intended size. Any higher-density piece needs both its density and its pivot point carried through, so that it keeps the same footprint in the world.

The test run

The full set of 714 automated editor tests ran: 712 passed, 0 failed, and 2 were skipped (both are tests that only write a report when asked). Three of the play tests failed on a first batch run and passed when run again, so they are being treated as flaky under batch rather than as broken features.