The build slims down, and the DevLog learns restraint

What changed

A measured before/after of the web export in one session: 17 MB down to 12 MB total; assets 12 MB to 7.3 MB; game art specifically 9.3 MB to 5.1 MB (-45%); JS 5.1 MB to 4.3 MB raw, 1.36 MB to 1.13 MB gzipped. Two separate causes. First, every shipped image outside the map plates and the splatter effects predated the project's own size recipe (longest edge 640px, WebP quality 44) — a new script applies that recipe in place, never regrading or reformatting, never growing a file. It touched enemies, the labyrinth's walls and doors, combat art, portraits, cards and treasure: 9,278 KB down to 5,123 KB. The seven arena backdrops shipped two nights ago went through the same pass, since they live in the same folder as the rest of combat art — same pictures, a fifth the file size. Second, the crash- reporting library was never supposed to run in the browser at all, but the bundler was still pulling its entire ~760 KB browser SDK into the web build regardless, because nothing told it not to. A new web-only stand-in for that library's interface means the browser build never asks for it in the first place; the phone app is unaffected.

What changed

The public site's "does tonight deploy" gate, added two nights ago, was tightened in three ways. The hosting project now rebuilds only when a file the build actually reads changes, instead of on every push. The nightly digest's own publish criteria were narrowed to three concrete triggers — a shipped feature, a UI or comfort change, or a fix to something that blocked play — replacing a vaguer standard the gate's first night had already outgrown. And the landing page can now offer visitors a Play the latest build link straight to the newest Android preview build, recorded by hand after each build and too large (roughly 100 MB) for the site itself to host.

What changed

The enemy, keyword, NPC and equipment lifecycle passes (each on their sixteenth or seventeenth pass) all came back zero-diff — nothing on their surfaces needed creating, updating or retiring this rotation. The card steward's audit also found no change the card surface itself needed, but its research pass turned up something genuine: nothing in the game lets you upgrade a card temporarily, mid-run — the kind of niche Slay the Spire fills with cards like Armaments. This game's only path to a stronger card is permanent, between runs. Building that would mean a new keyword and new engine wiring, past what one steward ships alone, so it was written up as a design candidate instead of built on the spot.