Game Development · Case Dissection. Browse Why Does My Godot Web Build Run in the Editor but Stop in the Browser?, Why Did My Balatro Save Load With the Wrong Stats After Corruption?, Why Does Your Game Keep Getting Bigger Until You Cannot Finish It?, with requirements, routes and walkthrough steps.
A threaded Godot 4 web export needs SharedArrayBuffer, available only on a cross-origin isolated page in a secure context. Without Cross-Origin-Opener-Policy: same-origin and Cross-Origin-Embedder-Policy: require-corp, or the Progressive Web App service worker workaround, the documentation says the project will not run. Single-threaded export, the default since Godot 4.3, avoids the headers but cannot use threads.
An independent tester flipped every bit of his own three Balatro save files (version 1.0.1o-FULL). Because each .jkr file is a Lua table loaded with loadstring and raw DEFLATE carries no checksum, 12,694 of 61,672 corrupted files (20.6%) loaded with no error and different data. This article reports that third-party test only, not an official LocalThunk finding, and turns it into save practices for the reader's own project.
Evolve Games canceled The Swarm after roughly six months: three to four months of small, well-meant design changes made a richer game that needed about twice the team and twice the time. Team fragmentation then made the math impossible. This case turns that postmortem into a Monday-morning scope check for your own project.
Hello Games launched No Man's Sky in August 2016 without the multiplayer and base building many players expected, and trust collapsed. Free updates rebuilt it in stages: Foundation in November 2016, Pathfinder and Atlas Rises in 2017, NEXT in July 2018, Beyond in August 2019, and Frontiers in September 2021. Steam reviews moved from Overwhelmingly Negative to Mixed, then Mostly Positive in September 2021, then Very Positive in November 2024.
Raph Koster, designer of Ultima Online's resource system, corrects the famous story: the ecology AI was disabled for radial-search and pathfinding cost, and the closed resource bank collapsed when players hoarded goods like wool shirts, leaving no wool to spawn more sheep.
A common documented Unity workflow problem, not a single named shipped-game incident: two developers edit the same scene, Git conflicts, and plain text merging cannot understand scene structure. Unity 6000.6 docs describe UnityYAMLMerge as the semantic merge tool for scene, prefab, and meta files, and the Game Developer Unity and Git guide walks through the same conflict and the git mergetool flow.
At launch in December 2020, Cyberpunk 2077 characters froze in T-pose. Hotfix 1.05 fixed two named quests only, and patch 1.23 in June 2021 was still fixing T-pose spots. In 2024 engine lead Charles Tremblay said the team chose a visible T-pose over a hard crash when memory ran out.
John Van Scoyoc October 2026 devlogs list three AI-written Unity bugs: invented APIs, fake null on destroyed objects, and renamed serialized fields that lose tuned values. Two are backed by Unity 6 docs. This checklist article is developer observation, not a single project postmortem.
BLAZOWACHO built err Friday for GMTK 2025 in about 12 to 15 hours, entirely in the Godot editor, and tested the HTML export only at the end. In the web build the death loop failed and enemy AI froze. The postmortem gives no root cause, so this article reports the symptoms, keeps the cause open, and adds Godot 4.4 web-export checks.
Factorio multiplayer runs on deterministic lockstep: every player simulates the whole game locally and only actions cross the network, so any tiny difference is a desync. Friday Facts 340 traces two deep cases, a Serpent nil placeholder that changed table order and a unit-group speed value that was recalculated on load, plus the older pattern of hidden state, hash order and invariants.
Facepunch Devblog 80 admits a save bug on the original server release of a Thursday patch: long-running servers that updated saw a lot of buildings fall down, while subsequent restarts were fine. The postmortem gives no mechanism, no affected count, and no restore claim, so this article states what is known, keeps the separate sound-pooling problem separate, and turns the incident into patch-day process lessons.
An independent experimenter ran Brotato on a weak ARM handheld, saw 5 to 10 FPS from around wave 20, failed first with a custom physics server, then rebuilt the test harness and found three small causes: unseparated collision layers, repeated per-frame work, and unbatched blood particles. The FPS gains are the author's own harness results, not official Brotato benchmarks.
