Log

Append-only history. Each entry: ## [YYYY-MM-DD] <op> | <title> where <op> is ingest, query, lint, split, or merge. Query with grep "^## \[" log.md | tail -5.

[2026-07-21] ingest | Meet the Unity CLI (first-party) — hub-routed

Hub-routed Telegram drop (msg 964), arriving right after the Unity 7 source and naming the CLI it previews. WebFetch 403'd (hard block) → fell back to firecrawl per the edge rule. T1 (first-party unity.com, Etienne Whittom VP Authoring Platform, 2026-07-20). Created unity-cli (TechArticle, url-only) + a thin etienne-whittom Person node (parallel to marcus-wassmer); linked from unity-engine's Unity 7 section. Synthesis: extended the "AI tooling as the new front" axis — Unity's answer to Unreal's bundled models is an agent execution surface (CLI manages / Pipeline drives / eval reaches inside a running Editor; JSON + exit-code contract; self-describing Editor → function-calling/MCP). Sharpens the MCP convergence (both big engines on the same protocol) — live seam to ../agentic-tooling-wiki. No per-route verify (deferred).

[2026-07-21] ingest | Unity unveils the Unity 7 roadmap (Game Developer) — hub-routed

Hub-routed Telegram drop (msg 963). Game Developer's report on Unity's Unity 7 announcement at Unite Seoul (2026-07-21). T2 (trade-press coverage of an official announcement). Created unity-7-roadmap (NewsArticle, url-only); refreshed unity-engine with a "Unity 7 — next generation" section (beta Dec 2026 / release early 2027, same Unity-6 architecture → zero-rebuild upgrade, open CLI + public API, opt-in genAI, no price change). No new entity node — Adam Smith / Matt Bromberg named but thin, held. Synthesis: new "two next-gen playbooks: continuity vs. reinvention" thread — Unity 7 (minimize migration cost, AI optional, no repricing = post-Runtime-Fee goodwill repair) vs UE6 (Verse-primary, deprecate Blueprints, native models) on the same ~2027 timeline. The Unity CLI it previews is the subject of a follow-on source (msg 964) ingested next. No per-route verify (deferred).

[2026-06-26] ingest | Godot Vision Statement (2026, first-party) — hub-routed

Hub-routed Telegram drop (msg 598). WebFetch returned only page chrome; the godotengine.org blog post just links to the full text on godot.foundation — fell back to firecrawl to scrape both, per the HUB edge rule. T1 (first-party Foundation governance doc, dated 2026-06-23). Created godot-vision-statement (DigitalDocument, url-only) + 3 entity nodes: godot-foundation (Organization, the FOSS-governance counterpart to epic-games/unity-technologies) and co-creators juan-linietsky + ariel-manzur (Person, thin). Refreshed godot-engine with a stated-design-philosophy section. Synthesis: new "stated philosophy as a competitive axis — vision-as-governance" thread — the statement makes explicit (and commits to) the leanness / low-end / focused-plugin / free-forever values the wiki had only inferred, and reads as the near-inverse of unreal-engine-6's "engine as ecosystem gateway." +4 pages (→~). Single clear-fit spoke, no runner-up.

[2026-06-01] merge | godot-wiki folded into this wiki

The hub merged the former godot-wiki spoke into this one (human directive, reversing the 2026-05-31 spin-out's "Godot has its own spoke" carve-out). This wiki's domain broadened from "non-Godot engines" to the whole general-purpose engine landscape, Godot included. Migrated from godot-wiki: 4 immutable raw/ docs and 6 pages — godot-engine, godot-game-template, godot-xr-game-jam-v, godot-best-plugins-2025, mega-crit, unity-vs-godot-comparison. The two unity-engine pages were deduplicated into one: the primary Unity page here absorbed godot-wiki's Unity-foil comparison facts (footprint/compile-time gap, Mega Crit defection) as a "Versus Godot" section. Merged index/synthesis/log; broadened CLAUDE.md domain; updated ../wikis.md and the hub log. godot-wiki/ directory removed. godot-wiki's prior history is preserved below under "Migrated history".

[2026-05-31] split | created from hub _inbox cluster game-engines (3 sources)

Spun out by the hub router once three non-Godot game-engine sources clustered: o3de (O3DE 26.05 release, via GameFromScratch), unreal-engine (Epic's RT3D engine homepage), and unity-engine (Unity AI tooling page). Godot already has its own deep spoke (../godot-wiki); this wiki covers the rest of the engine landscape + the comparative/business/AI-tooling dimension. Unreal and Unity homepages both 403'd the fetcher — their pages are honest stubs (title + known context) per the fetch-fail rule; O3DE was fetched in full.

Cross-spoke note: ../godot-wiki already holds a unity-engine page framed purely as a Godot comparison point. That is a deliberate adjacency, not a duplicate — see this wiki's synthesis. If the two drift, the human can decide where Unity's primary coverage should live.

[2026-05-31] ingest | O3DE 26.05 Released — via hub route

Source (URL): gamefromscratch.com/o3de-26-05-released. New page: o3de (SoftwareApplication). Established the open-source / Linux-Foundation end of the landscape.

[2026-05-31] ingest | Unreal Engine homepage — via hub route

Source (URL): unrealengine.com (403, stub). New page: unreal-engine (SoftwareApplication) + epic-games (Organization). The proprietary RT3D-powerhouse end of the landscape.

[2026-05-31] ingest | Unity AI — via hub route

Source (URL): unity.com/features/ai (403, stub). New page: unity-engine (SoftwareApplication) + unity-technologies (Organization). The AI-tooling angle.

Migrated history (from godot-wiki, before the 2026-06-01 merge)

Preserved verbatim from the former godot-wiki/log.md for provenance. The ../research-wiki reference below points at where Godot's pages first originated before godot-wiki existed.

[2026-05-29] split | created from ../research-wiki cluster C

Spun out of the ../research-wiki as its own wiki (game-dev was off that wiki's knowledge-management thesis). Migrated 3 sources and 4 pages: godot-game-template, godot-engine, godot-xr-game-jam-v, godot-best-plugins-2025. Re-homed index/log/synthesis; removed cluster-C framing from the pages. Original ingests happened 2026-05-29 (see ../research-wiki log history).

[2026-05-29] ingest | Unity or Godot? (same game twice) — via Telegram

Source: raw/unity-vs-godot-comparison.md (Creative Bloq). First ingest into godot-wiki post-split. New pages: source summary unity-vs-godot-comparison (Article) + unity-engine (SoftwareApplication, the comparison counterpart to godot-engine). Cross-linked from godot-engine. Synthesis: added a real analysis layer and partially resolved the "Godot vs Unity" open question (comparable perf; Godot far lighter/faster dev loop; Unity more mature). Routed there (not ../research-wiki) per the topic-routing rule.

[2026-05-29] lint | health check (4 sources, 6→7 pages)

Clean: no orphans, no broken links, no contradictions. Applied: created mega-crit (Organization) — referenced across 3 pages (Slay the Spire studio, Unity→Godot migrant) but unpaged; linked from godot-engine, unity-engine, unity-vs-godot-comparison. Recommendations: ingest an Unreal comparison source (only Unity covered so far); a rigorous multi-project benchmark (current sources are practitioner write-ups). Thin/single-source pages throughout — expected at this size.

[2026-05-29] lint | health check (4 sources, 7 pages)

Clean: no orphans, no broken links, no contradictions. No new issues since the prior lint (the mega-crit missing-page fix is in place). Recommendations stand: ingest an Unreal comparison source (only Unity covered) and a rigorous multi-project benchmark; sources remain practitioner-level. No fixes needed.

[2026-05-31] lint | health check (4 sources, 7 pages)

Part of a hub-wide lint. Clean: no orphans, no broken links, no contradictions; healthy ~5 links/page; @types appropriate (SoftwareApplication/Organization/BlogPosting). Open gaps (carried, web-search-fillable): (1) no Unreal comparison — only unity-engine covered; (2) no rigorous multi-project benchmark (sources are practitioner write-ups); (3) maintenance/production-grade status of godot-game-template and the top plugins unverified; (4) verify the "Mega Crit / Slay the Spire 2 on Godot" claim is still current. No fixes needed.

[2026-06-02] ingest | Godot: GABE reaches stable — games entirely on Android/XR

Routed in by the hub (clear Godot match). Created godot-gabe-android (NewsArticle, url) — the wiki's first first-party Godot source (a release announcement), filling the gap the godot-engine page had flagged. GABE (Godot Android Build Environment) stable: build/export (AAB, full Gradle, Play Billing/AdMob)/publish entirely on Android or XR devices, no PC; Godot 4.6 → stable, on Play Store / Meta Horizon / GitHub. Updated godot-engine (new "on-device/mobile- first reach" bullet; replaced the "no first-party source yet" note) + index. Reinforces Godot's low-overhead/accessibility differentiation and the XR thread (godot-xr-game-jam-v).

[2026-06-07] broaden | scope widened from "general-purpose 3D" to all game engines (2D + 3D)

User directive ("Broaden") via the hub after Flame was initially parked. Widened the spoke domain from general-purpose 3D engines to all game engines across 2D and 3D, explicitly admitting 2D and framework-based engines (e.g. Flame on Flutter). General-purpose 3D stays the deepest- sourced corner. Updated: this spoke's CLAUDE.md (header + Scope conventions), index.md (Flame row), synthesis.md (thesis line + new "standalone vs framework-based" axis + open question), and the hub wikis.md block (domain/keywords/sample-pages/note). Same play as the 2026-06-05 tts→speech-audio broaden. Spoke name unchanged ("game-engines-wiki" already covers 2D+3D).

[2026-06-07] ingest | Flame — Flutter-based 2D game engine (MIT, Blue Fire)

Routed in by the hub immediately after the broaden (Flame was the trigger source). Created flame-engine (SoftwareApplication, url) — the wiki's first 2D engine and first framework-based engine (layers a game loop / component system on top of the Flutter UI toolkit, Dart). MIT, ~10.6k stars, community-sponsored; core features (game loop, components, effects, particles, collisions, input, sprites) + bridge packages (Forge2D physics, Tiled, AudioPlayers, Bloc). Anchors the open-source/no-royalty end with o3de/godot-engine; opens the standalone-vs-framework architecture axis in synthesis. Updated index + synthesis. Source: github.com/flame-engine/flame.

[2026-06-09] ingest | Godot 4.7 RC 1 (first-party release announcement)

Hub-routed (Telegram, godotengine.org). New source godot-4-7-rc1 (NewsArticle, url) — the wiki's second first-party Godot source (after godot-gabe-android). Godot 4.7 hit Release Candidate (2026-06-08): features locked, no critical regressions, "production environments can begin integrating." Headline features: HDR output, drawable textures, and a first-party Godot Asset Store (+ web/XR/Android editors; 16 fixes since beta 5). Updated godot-engine (new "release cadence — 4.7 RC" bullet + Related/updated date) and synthesis (ecosystem bullet: Godot now first-partying the Asset-Library/time-to-game layer via the official Asset Store). Index: new NewsArticle row. Caveat: promotional pre-stable RC announcement.

[2026-06-09] ingest | +3 framework-based engines (Bevy, Phaser, LÖVE) — all-spokes cron test

Answered the "2D/framework engines thinly sourced (peers like LÖVE/Phaser/Bevy)" open question: bevy (SoftwareApplication, url+src — Rust, ECS, 2D+3D, code-first, MIT/Apache), phaser (SoftwareApplication, url+src — JS/HTML5 2D, MIT), love2d (SoftwareApplication, url+src — Lua 2D, zlib; Balatro). Flame is now a 4-wide framework-engine cohort varying on language × host runtime × platform; Bevy adds ECS/data-oriented architecture + a Rust open-source leanness contender. Synthesis open question updated (still open: a head-to-head 2D benchmark, and Godot-vs-Unreal). 14 → 17 pages.

[2026-06-10] ingest | GDevelop (no-code axis) + Unreal refresh — all-spokes pass

New page gdevelop (SoftwareApplication, url, Wikipedia) — free/open-source (MIT) no-code, visual event-based engine (Florian Rival, 2008): conditions+actions in natural language, pre-built behaviors, AI-assisted authoring, GDJS/PixiJS/Three.js under the hood; 2D(+3D), exports Win/mac/Linux/Android/iOS/HTML5; big education uptake (10k+ students by 2025). Opens a third engine-architecture axis — programming-required vs no-code — beyond open-vs-proprietary and standalone-vs-framework; engine map now spans unreal-engine (AAA) → unity-engine/godot-engine (scripting) → bevy/phaser/love2d (code-first) → gdevelop (no-code). Also refreshed the unreal-engine stub in place (was a 403 fetch-fail stub) with real Wikipedia data: 5% royalty over $1M, UE5 Nanite/Lumen/MetaHuman, and 2024 market share (Unity ~50%/Unreal ~28% by units; Unreal 31%/Unity 26% by revenue) — partly addresses the standing Godot-vs-Unreal open question (narrowed, not closed; no head-to-head benchmark yet). Folded into synthesis (Godot-vs-Unreal Q updated + new no-code-axis bullet) + index. No contradictions. 17 → 18 pages (Unreal refreshed, not added).

[2026-06-11] ingest | Godot 4.7 RC 2 (godotengine.org)

Telegram drop, hub-routed → game-engines-wiki (Godot = deepest corner). Sequel to the godot-4-7-rc1 source (2026-06-08); RC2 (2026-06-11) is a distinct source (new URL/date), not a re-seen refresh, so created a new source page godot-4-7-rc2. Feature set unchanged (locked at RC); RC2 = pure stabilization: 10 regression fixes / 7 contributors (2D tile randomization, animation-tree null check, build/MSYS2, worker thread-pool deadlock, scene-tree drop, D3D12 script), stable "just around the corner." Updated godot-engine release-cadence bullet (RC1→RC2 in 3 days, regression-only) + Related; index NewsArticle entry added. synthesis untouched (RC2 adds no new thesis beyond the 4.7 first-party-Asset-Store point already recorded). Site rebuild follows.

[2026-06-13] ingest | Godot Tutorials (official docs hub, docs.godotengine.org/en/stable)

Routed from hub (Telegram drop). First-party Godot tutorials/manual index for the stable line (4.6) — General + 20+ Topic areas (2D/3D/scripting/shaders/networking/XR/export/…). Quality gate: tier T1 (first-party/primary); not previously paged (new source summary godot-docs-tutorials); gap-relevance — opens the documentation / onboarding competitive axis and grounds the stable line vs the 4.7-RC pages; integrated into godot-engine (new bullet) + synthesis (folded into time-to-game thread + new open question on cross-engine docs-quality); url-only provenance. freshness: stable (version-pinned, rolls per release). Site rebuild + commit follow.

[2026-06-15] ingest | Three engines: Torque2D + Spring + openage (hub-routed burst)

Hub-routed Telegram burst (msgs 440/441/443), all clear game-engine matches, batched into one ingest:

• torque2d (SoftwareApplication, url) — MIT 2D engine, GarageGames/Torque lineage (TorqueGameEngines org); standalone (own editor + TorqueScript), OpenGL/Box2D/OpenAL; 4.0 Early Access. First standalone 2D engine here; second "formerly-commercial → open-sourced" engine after o3de.
• spring-rts (SoftwareApplication, url) — GPL RTS-specialized engine (formerly TASpring); C++/Lua, large-scale unit battles (Beyond All Reason / Zero-K / Balanced Annihilation). First genre-specialized AND first copyleft engine in the spoke.
• openage (SoftwareApplication, url) — GPLv3 open reimplementation of AoE II's Genie engine (SFTtech, ~14.2k★); C++20/Python/Cython; RTS, in active dev w/ limited playability, BYO original assets. First clean-room reimplementation engine; RTS cohort with Spring. Synthesis updated: licensing axis extended to a 4-position spectrum (permissive / copyleft / royalty / subscription); new axis "general-purpose vs genre-specialized" (Spring+openage = RTS cohort, two flavors: generic RTS platform vs specific-game clone) + origin-story dimension (ground-up / ex-commercial / reimplementation). Index updated. All url-only (no source:true); tiered as first-party repos. Site rebuild

• verify follow.

[2026-06-15] ingest | libGDX + Superpowers (hub-routed burst, cont.)

Two more hub-routed engines (msgs 445/446):

• libgdx (SoftwareApplication, url) — Apache-2.0 cross-platform Java game framework, OpenGL ES, 2D+3D, framework-based (no editor, Gradle); 25.2k★. Extends the framework cohort to five (adds Java to Dart/JS/Lua/Rust) and, like Bevy, does 2D+3D. Flagged a contradiction: widely-reported that the original Slay the Spire used libGDX, which would complicate the mega-crit page's "Unity → Godot" framing for StS2 — recorded in synthesis Contradictions, NOT asserted (repo source lists Pathway, not StS; needs a Mega Crit primary).
• Superpowers (SoftwareApplication, url) — ISC HTML5 engine+IDE (Sparklin Labs), TypeScript/Three.js, 2D+3D, self-hostable; real-time collaborative game dev ("Google Docs for game dev"). Archived 2021 — the spoke's first defunct engine. Added two synthesis axes: real-time collaboration and liveness (maintained vs dead). Index + synthesis updated (framework cohort 4→5; collaboration/liveness axes; StS contradiction flagged). url-only. Site rebuild + verify follow.

[2026-06-15] ingest | GDevelop — dedup refresh in place (first-party repo)

Hub-routed (msg 448) but already paged (gdevelop, 2026-06-10, Wikipedia-sourced) → refreshed in place per the edge rule, no new page. The official repo (github.com/4ian/GDevelop, Florian Rival, ~23.8k★) shifts positioning from "primarily 2D" to headline 2D/3D/multiplayer — multiplayer now first-class, 3D (Three.js) elevated, plus the gd.games platform + Steam/store export channels. Added also-url: to the repo; bumped updated:; tier stays T2 (vendor positioning). No synthesis change (the no-code/AI-authoring axis it anchors is unchanged). Count unchanged (25).

[2026-06-15] ingest | Babylon.js (hub-routed) — web-native 3D engine

Msg 449. babylonjs (SoftwareApplication, url) — Apache-2.0 web-native 3D game/rendering engine, TypeScript, WebGL/WebGL2/WebGPU; full engine (PBR, WebXR, physics, node-material editor, playground, glTF, DCC exporters), ~25.6k★, Microsoft-originated. Adds the web-3D framework corner (3D counterpart to phaser's 2D-web) and a web-3D engine vs library distinction (Babylon vs Three.js — the latter embedded by gdevelop/Superpowers). Framework cohort now 6 (Flame/Phaser/LÖVE/Bevy/libGDX/Babylon). Index + synthesis updated. url-only. +1 page (→26).

[2026-06-15] ingest | PlayCanvas + PixiJS (hub-routed) — web-graphics matrix completed

Msgs 451/452. playcanvas (SoftwareApplication, url) — MIT web-native 3D full engine (JS/TS, WebGL2/WebGPU/WebXR/glTF + Gaussian splatting; physics/anim/audio + collaborative cloud editor), 16k★, Disney/Samsung/King. pixijs (SoftwareApplication, url) — MIT fast 2D WebGL/WebGPU rendering library (TS, 47.4k★), the 2D counterpart to Three.js, embedded by gdevelop for 2D. Together they complete the web-graphics library/engine matrix (dimension × level): 2D-lib PixiJS / 3D-lib Three.js → 3D-engine babylonjs → 3D-full-engine+editor PlayCanvas; libraries are substrates the editor-engines embed. PlayCanvas also extends the collaboration axis (maintained 3D cloud editor vs archived Superpowers). Synthesis + index updated. url-only. +2 pages (→28).

[2026-06-15] ingest | Three.js — ontology-guided new content (Ralph loop)

Created three-js (SoftwareApplication, url) — the 3D-web rendering library node the schema.org audit flagged as referenced-but-unpaged by 5 pages (babylonjs/playcanvas/pixijs/gdevelop/superpowers). mrdoob, MIT, WebGL/WebGPU, ~113k★; a renderer (no physics/loop/editor), the 3D counterpart to pixijs, the substrate Babylon/PlayCanvas-style engines and the editor-engines (GDevelop/Superpowers) build on/embed. Completes the web-graphics library/engine matrix as a real node; wired three-js into the 5 referencing pages + the synthesis matrix + index. Not a routed source — internal ontology-guided content creation (filling a referenced-but-missing node). +1 page.

[2026-06-15] ingest | WebGPU — rendering-backend node (Ralph loop, ontology-guided)

Created webgpu (DefinedTerm) — the WebGL-successor browser graphics+compute API referenced as a backend by 4 web-engine pages but unpaged. Anchors the rendering-backend dimension (WebGPU vs WebGL) beneath the web-graphics library/engine matrix; wired into three-js/pixijs/babylonjs/ playcanvas + the synthesis matrix + index. Its compute-shader / in-browser-AI-inference angle adds a rare cross-spoke seam from game-engines (an island) → llm-inference. +1 page (→30).

[2026-06-16] ingest | Godot 4.7 RC 3 (hub-routed, Telegram) — release-discipline data point

Third in the RC chain after godot-4-7-rc1/godot-4-7-rc2: created godot-4-7-rc3 (NewsArticle, T3/volatile, url-only) from the first-party announcement (2026-06-15, 17 fixes / 12 contributors, regression-only — feature set locked). Folded into godot-engine's release-cadence bullet + Related, and into index. No synthesis edit: RC3 doesn't open a new axis, it sharpens the existing release-discipline read — fix categories keep clustering on GPU backends (Jolt physics, Adreno) and XR, and 4 regressions stay open, so stable slips on platform/backend bugs not features (a data point for the unity-vs-godot-comparison "more complete/stable" critique). Entity discovery: no new people/orgs — Godot project already paged; author byline not surfaced by the post. +1 page (→31).

[2026-06-17] ingest | Unreal Engine 5.8 released (GameFromScratch)

Hub-routed from Telegram (gamefromscratch.com/unreal-engine-5-8-released). Clean single-spoke route — game-engine tech/landscape is this spoke's core. Added unreal-engine-5-8 (NewsArticle, url-only, T3 — games-news site reporting Epic's release; 60fps/hardware targets are Epic's stated goals, many features experimental/beta). Refreshed unreal-engine (new "Latest release — 5.8" section + Related; bumped updated; fixed stale index "(stub)" label). Gap-relevance: advances the "AI tooling as the new front" synthesis thread — the Unreal MCP Server Plugin is Unreal's editor-AI answer to Unity's, and a concrete MCP integration (cross-spoke seam to agentic-tooling). Also nudges the fidelity-range map: Lumen Lite (+ Switch 2) pushes Unreal down toward low-end hw, softening the AAA-ceiling pole; flags UE6 ~late 2027 roadmap horizon. Folded both into synthesis. Entity discovery: none minted — epic-games already a node (linked); publisher GameFromScratch is a low-signal news site (recorded inline). Ran avoid-ai-writing over the new prose. +1 page (31 -> 32).

[2026-06-18] ingest | Unreal Engine 6 (UE5 + UEFN merge)

Telegram-dropped; hub-routed → game-engines (runner-up: none). Source summary unreal-engine-6 (NewsArticle, T2 — Game Developer reporting Epic's first-party blog; url provenance). Subject unreal-engine already paged + UE6 foreshadowed in unreal-engine-5-8 ("UE6 ~late 2027"); this is the concrete announcement, so I added a dedicated UE6 source page and updated the engine + epic-games Thing pages in place. New Thing pages: verse-language (ComputerLanguage — UE6's primary language, deprecating Blueprints/sidelining C++). Entities (1/8 budget): marcus-wassmer (Person, Epic EVP, the announcer) — new; epic-games matched existing, linked; publisher (Game Developer) not noded, per spoke house style for url-only news. Gap-relevance: opens a new synthesis axis — engine as ecosystem gateway (merge + portable content/code/economies via glTF/USD; engine monetized through the Fortnite ecosystem, not just royalties) — plus a language-consolidation sub-axis (Verse over C++/Blueprints) and a native-models step on the AI-tooling front (Claude/Gemini past the MCP plugin). Integrated into synthesis + index. freshness: volatile (2027 roadmap; plan-not-product). Cross-spoke: native Claude/Gemini + MCP brush agentic-tooling/llm-providers (noted, not duplicated). Site rebuild + commit follow.

[2026-06-19] ingest | DefinedTerm enrichment pass — webgpu grounded

Hub-wide DefinedTerm enrichment ("go deeper, fetch sources"). The spoke's one DefinedTerm page, webgpu, was solid but unsourced; grounded it with mdn-webgpu (MDN, T1) — added the API model (GPUAdapter→GPUDevice, native D3D12/Metal/Vulkan backends, render vs compute pipelines, GPGPU first-class, HTTPS-only, compatibility mode, not-Baseline). New source page mdn-webgpu; webgpu Related extended.

[2026-06-20] ingest | Godot 4.7 stable ("Director's Cut") — hub-routed

Two Telegram drops on the same release (hub msgs 535 godotengine.org download + 536 80.lv writeup). 4.7 shipped stable 2026-06-18, closing the RC1→RC3 cycle this spoke tracked (RC3's 4 open regressions cleared). New source godot-4-7-stable (NewsArticle, T2 — first-party release + 80.lv secondary corroboration). Headline features (per 80.lv): new Asset Store (replaces Asset Library), HDR output (Win/macOS/iOS/visionOS/Linux-Wayland), AreaLight3D, inline shader previews, Control-node transforms, DrawableTexture2D, GABE stable. Updated godot-engine (cadence section → stable; bumped). The Asset-Store + GABE-stable threads (ecosystem + on-device) graduate community/preview→official. +1 page.

[2026-06-19] ingest | entity enrichment — Unity Technologies grounded

Extending enrichment beyond DefinedTerm to entity pages. unity-technologies was an 11-line stub; deepened with unity-technologies-wikipedia (Wikipedia, T2): founding (2004 Copenhagen), business model (Personal/Pro/Enterprise + Asset Store/Ads), 2020 IPO, and the 2023 Runtime Fee controversy + Riccitiello departure — which grounds the trust backdrop behind the mega-crit → godot-engine migration in unity-vs-godot-comparison. New source page; index updated. (epic-games / mega-crit already developed — left as-is.)

[2026-07-01] ingest | Carbon engine open-sourced (CCP/Fenris) — EVE Online's MMO engine goes MIT

T3 (primary vendor page fenris.com/carbon + trade coverage; MIT confirmed on github.com/carbonengine). Router source was gamesindustry.biz (host-blocked, unfetchable — substituted, not dropped). New pages: carbon-engine (SoftwareApplication), carbon-engine-open-source (source), ccp-games + fenris-creations (Organizations). Carbon = cross-platform framework behind EVE Online / EVE Frontier; components Trinity (render) / Destiny (physics+pathfinding, drove an 8,825-player battle) / CarbonUI / CarbonIO (networking) / CarbonAudio + Python scripting; foundation libs (Core C++, Scheduler, Resources, IO, Math) on GitHub. Now fully open under MIT. Why (Pétursson): open code is "more robust over time" (Linux analogy), let EVE "live forever," enable universities/game-jams/mods — stewardship, not monetization. Touched synthesis: added Carbon to the permissive-MIT licensing pole, and a new origin-story sub-variant — formerly-proprietary engine opened while the game is still live & paying (vs torque2d/o3de post-abandonment salvage), + the spoke's first MMO-scale genre engine (server-authoritative networking + massive battle sim). Counter-note: CCP splits engines — Awakening on Carbon, EVE Vanguard FPS on Unreal. Dates: announced Mar 2024, transition complete 2026.

[2026-07-03] ingest | Box3D — Erin Catto's 3D physics engine (the middleware layer)

Msg 719 (github.com/erincatto/box3d, via Telegram). Router-classified to game-engines-wiki (physics engine = game-engine technology; runner-up ../dev-tooling-wiki declined — this is a game-runtime subsystem, not build/ship infrastructure). T2 — first-party project README of a named, established author. New pages: box3d (SoftwareApplication, url) + erin-catto (Person, url, the Box2D/Box3D author). Box3D = 3D successor to Box2D, C17 core (C++20 samples), data-oriented + SIMD/multithreaded, Soft Step solver, island-sleep, cross-platform determinism, continuous collision + convex/capsule/sphere/ mesh/heightfield + raycast, revolute/prismatic/motor/weld joints; MIT; v0.1.0 (2026-06, PRs disabled pending stabilisation); samples use Sokol + Dear ImGui. First physics engine / middleware component in the spoke — a subsystem engines embed (Godot→Jolt, Phaser/Torque2D→Box2D, Carbon's Destiny = in-house physics) rather than a full engine. Synthesis: added a middleware-layer axis (build-vs-embed a physics core; determinism+MIT maps onto the deterministic-sim genres — RTS pair + MMO Carbon) and extended the lineage thread (proven 2D engine re-authored in 3D by the same author). +2 pages. Ran avoid-ai-writing. Per-route verify deferred (routine ingest, no structural page move).

[2026-07-04] ingest | Godot contribution policy 2026 — first engine AI-contribution stance (bans agents / vibe coding)

Msg 726 (godotengine.org/article/contribution-policy-2026, via Telegram). T1 (first-party Godot Foundation governance statement). New page godot-contribution-policy-2026 (Article, url). The Foundation (30 Jun 2026) restricts AI-generated code: bans substantial AI-authored code + autonomous agents + "vibe coding" (auto-ban from the GitHub org), requires human authorship + disclosure, confines AI to menial help (completion/regex/find-replace), and forbids AI text in maintainer comms; plus new-contributor gates (≤3 merged PRs → need maintainer permission before features/refactors, start with bugfixes/docs) and reaffirmed human review. Rationale: PR backlog + reviewer burnout, and two human-side reasons — machines can't learn from feedback (mentorship drain) + contributors must be able to maintain their code. Updated godot-foundation (now governs contribution intake, not just direction) and synthesis: extended the vision-as-governance axis with this enforceable sequel to the Vision Statement, and made the "AI reshaping engine development" thread two-directional (adopted in-editor vs resisted at the contribution layer — coexisting in one engine). Cross-spoke: agentic-tooling (counter-current to autonomous-coding) + ai-governance (org AI-use policy). +1 page. Ran avoid-ai-writing. Per-route verify deferred (routine ingest, no structural page move).

[2026-07-05] ingest | Goblins' Den — Godot 3D dungeon-crawler tutorial (80.lv) → game-engines-wiki

Routed from hub. 80.lv write-up (T3) of a free GameDev Tavern YouTube series building one full Godot 3D dungeon crawler ("Goblins' Den") over 20 episodes / ~20 hrs — node-based state machines, composition, ragdoll physics, shaders, pathfinding, lighting, level design + the full asset pipeline (models/rigs/anim/music/SFX). MIT code+assets on GitHub, playable on itch.io; intermediate-level. New source page goblins-den-dungeon-crawler-tutorial (@type Course) + new gamedev-tavern Organization (entity discovery). Synthesis: extended the onboarding/knowledge-scaffolding thread with a community pillar (project-based series teaching end-to-end integration the topic-docs skip) beside the first-party manual. Godot version unstated; 4.7 was current stable at publication.

[2026-07-12] ingest | Why space games struggle with the scale of the universe (Space.com) — hub-routed

Hub-routed Telegram drop (msg 863). Fetch-blocked stub: WebFetch (×2) returned only page chrome + the Douglas Adams framing quote ("[Content truncated due to length...]"), and the firecrawl fallback also got only newsletter boilerplate (37 lines, no body). Per the HUB edge rule, routed by title/beat and recorded an honest stub — no fabricated specifics (games cited / techniques named are unverified). Placement rationale: the "struggle with scale" is at root an RT3D-engine problem (floating-point precision at astronomical coordinates, LOD, procedural generation), the technology corner of this spoke; no other spoke touches games or real-time 3D, so no runner-up. Created space-games-universe-scale (NewsArticle, url-only, T3 space.com). Synthesis: new "representing astronomical scale" open question (the precision/LOD/procgen frontier the engine pages haven't covered). +1 page. Provisional — re-file if a recovered body proves it a pure game-design essay.

[2026-07-16] ingest | Oxide Infinite World Generator (Lite) — Godot Asset Library (hub-routed, Telegram)

Ingested oxide-infinite-world-generator — a free (MIT) Godot 4.0+ 2D procedural world generator asset by RedActor (Hans Ritter), Asset Library category 2D Tools. FastNoiseLite noise terrain + automatic chunk load/unload for infinite maps + slider-tuned biomes (forest/swamp/lake/river/hills) with real-time editing; bundled survival scaffolding (day/night, hunger/thirst, minimap, odometer). Free Lite funnels to a paid Pro (Boosty: procedural civilizations, roads, save/load, hazard fog). Typed SoftwareApplication + source (T3 — author/asset-page self-description, promotional). New concept procedural-generation (noise + chunking; grounds the procedural thread space-games-universe-scale opened, at world-building scale — astronomical scale still open). Synthesis: added the procedural-gen "first concrete step" note + an Asset-Library-as-commercial-marketplace point (free-as-funnel on the open Asset Store). Author node deferred (thin, single T3 asset). avoid-ai-writing self-pass (clean). +2 pages, 2 spine files updated.

[2026-07-17] ingest | MonoGame (monogame.net, hub-routed, Telegram)

Ingested the first-party MonoGame site. C#/.NET cross-platform game framework (code-first, no visual editor), the open-source successor to Microsoft's discontinued XNA (reimplements XNA's API on modern .NET). Platforms: Win/mac/Linux + Android/iOS/iPadOS + PS4/PS5/Xbox One/Switch (console auth required). License Microsoft Public License (MS-PL), "free forever — no subscription/royalty/licensing/runtime fees." Owned by the MonoGame Foundation (501c3, EIN 93-3803929). Ships Stardew Valley, Celeste, Streets of Rage 4, Barotrauma, TMNT: Shredder's Revenge. v3.8.5 (Jul 2026). New pages: monogame (SoftwareApplication, url-only, T3); monogame-foundation (Organization). Cross-wiki reuse: microsoft (../agentic-tooling-wiki) for XNA/MS-PL heritage. Synthesis: added a framework-cohort paragraph — C#/.NET is the 6th framework language (Dart/JS/Lua/Rust/Java + C#); a new lineage flavor (community reimplementation of a discontinued proprietary framework's API, distinct from source-opened Carbon/O3DE and clean-room openage); 3rd foundation steward; and the cohort's production-credibility anchor (Stardew/Celeste answer the "framework = hobby-tier?" open question). Updated the framework-cohort open question (+C#/.NET). avoid-ai-writing applied. +2 pages, synthesis+index updated. T3 first-party site.

[2026-07-27] ingest | Godot MCP Toolkit (NPGameDev, Godot Asset Library asset/5375)

Routed here by the hub (runner-up: ../agentic-tooling-wiki — it owns MCP as a standard, but this spoke's domain explicitly covers "the AI tooling reshaping development" and Godot is its deepest corner; the MCP mechanism is cross-linked, the Godot plugin files here). T3 — Asset Library listing + the author's own description; v1.0.0, rating 0, no reviews, nothing exercised. Fetched via the Asset Library API (/asset-library/api/asset/5375) for exact metadata. New: godot-mcp-toolkit (source), engine-agent-surface (pattern). Updated: godot-engine (the AI surface is a plugin, not a feature), godot-contribution-policy-2026 (where the line actually falls), unity-cli (the same conclusions reached independently), synthesis (AI-front thread extended + a new flagged tension), index. Three things this adds:

1. The third engine gets an agent surface, and it's a community plugin. Unreal ships a first-party MCP plugin, Unity productizes an execution surface, Godot's arrives MIT on the Asset Library from a solo dev. Each engine's AI-integration strategy mirrors its governance model — and Godot's is exactly what godot-vision-statement prescribes (lean core, plugins for the rest).
2. The pattern is now nameable. engine-agent-surface collects four properties the three converge on: runtime self-description (discover_tools ≈ unity command), extension in the project's own language (GDScript MCP tools ≈ [CliCommand]), the agent's context budget as a design constraint (a 112-tool surface that stays small until asked), and token-gated localhost-only access. Two unaffiliated projects landing on identical security posture is the strongest evidence this is a real pattern and not one vendor's framing.
3. It makes the Foundation's AI line legible. Godot auto-bans autonomous agents from its GitHub org and its Asset Library distributes an agent toolkit three weeks later. Not a contradiction: contributing to the engine vs using it. Recorded as a flagged tension, because the plugin's listing goes out of its way to say it is unaffiliated and unendorsed — the author clearly expects the question. Entity NPGameDev deferred (solo community author, one asset, thin evidence beyond a handle and a site) — page it if a second asset or a recurrence lands. Verify deferred per hub policy (content-only). avoid-ai-writing run.

[2026-07-29] ingest | Unity PolySpatial visionOS (docs introduction)

Routed in from the hub. First-party Unity package docs, so T1 — with a freshness problem worth naming: the URL is pinned to 0.1.2, generated 18 Oct 2023, months before the Vision Pro shipped. Fetched the current 2.0.4 introduction alongside it, and the delta carried most of the value.

One page, unity-polyspatial. No separate visionOS / spatial-computing page: one source, folded in per the ENTITIES recursion discipline and flagged there — a second XR-platform source makes one warranted, and the spoke already has godot-xr-game-jam-v waiting on that axis. unity-engine updated with a visionOS section and relinked.

Two things went to synthesis rather than staying on the page. The renderer handoff is a third position on the architecture axis: PolySpatial apps are simulated by Unity and drawn by RealityKit, because visionOS composites apps into a shared space and won't hand any one of them the frame. Flame gives up the editor and keeps the loop; this gives up the renderer and keeps the rest. 2.0.x has been walking it back — the Metal path takes the pipeline home for immersive apps, and "PolySpatial MR" as a category you pick is gone. The licensing gate is the other: 2.0.4 is Pro/Enterprise/Industry only, so Unity Personal cannot ship to visionOS. That gives the subscription pole a dimension the MIT and royalty poles structurally cannot have — a tier that withholds a platform, not just features. 0.1.2 states no such restriction; the gate arrived with maturity.

Opened one question: which Unity features actually break on the handoff. Unity ships a validator to manage the gap and neither intro enumerates it, so the wiki records that the gap exists and not its contents. FAQ + project-validation pages are the next fetch. avoid-ai-writing run. Verify deferred per hub policy (content-only, no page moves).

[2026-08-12] lint | Quality cycle: one structural repair, five stale pages cleared, zero new pages

The structural defect. synthesis.md had two ## Open questions headings — one in its proper place and a second stranded below ## Flagged contradictions / tensions, holding two live questions (astronomical scale, the Asset Library as a marketplace). Anything reading the spine for the gap backlog saw the first list and stopped. The two questions are now inside the real section and the duplicate heading is gone. Nothing was rewritten in the move.

Freshness: 5 past the 60-day window → 0, four of them by regrade.

• godot-4-7-rc1 and godot-4-7-rc2 → stable. Dated release-candidate announcements; re-reading the URL returns the same text, so the flag could never be cleared by anyone (../QUALITY.md).
• love2d and phaser → stable. Encyclopedic descriptions of the frameworks, carrying no current-state quantity a reader would act on.
• bevy re-verified against the live front page and kept volatile. The compile-time claim holds ("0.8-3.0 seconds with the 'fast compiles' configuration") and the licence now reads "100% free. Forever and always." Two changes: bevy.org no longer mentions an editor, so the page's "editor still in progress" parenthetical had lost its source and was dropped rather than updated; and the feature list has grown to skeletal animation with blend shapes, flexbox UI, scene save/load with hot reloading and audio. New open question written for the first: on this spoke's own library-versus-engine line, Bevy's claims now sit at the engine end.

Also dropped a stock phrase from love2d ("punches above its weight") per the hub's prose rule.

No new sources. The spoke's depth and integrity are fine; what it lacked was a synthesis that could be read in one pass.

[2026-08-15] lint | freshness queue worked — 7 stale pages, one wrong claim, one new engine

Quality-cycle pass (hub quality-log.md). Seven volatile pages all carried updated: 2026-06-15 and crossed the 60-day window together, so the queue was one batch edit coming due, not seven independent decays.

Star counts re-verified against the GitHub API and refreshed in place: babylonjs 25.6k → 25.9k, libgdx 25.2k → 25.3k, openage 14.2k → 14.4k, pixijs 47.4k → 48.0k, playcanvas 16k → 16.5k, gdevelop 23.8k → 25.7k. Movement is ordinary two-month drift in every case; GDevelop's +1.9k is the largest.

spring-rts failed re-verification on a claim, not a number. The page called it an "active project"; spring/spring was last pushed 2024-03-31, ~17 months ago, at 3,963★. Development continues in a fork, and the fork says so in its own README — created recoil-engine (T1, first-party repo, 654★, pushed the day of this check): "Recoil is a fork and continuation of an RTS engine version 105.0", linking to the Spring repository. Zero-K, one of the three titles the Spring page cites, describes itself as running on "the Spring/Recoil engine".

Both pages record the correction rather than overwriting the history: Spring keeps its genre and licensing analysis and is now framed as the origin of a live lineage. synthesis.md's general-purpose-vs-genre-specialized section carries the correction inline, including the point it adds — the liveness attribute Superpowers introduced gains a third value beyond maintained/defunct, namely succeeded, and with it a measurement trap: stars accumulate on the dormant repository while the work happens on the fork. index.md updated.

[2026-08-16] lint | Freshness — the 4.7 RC page resolved and regraded

godot-4-7-rc3 came due on the volatile window. Godot 4.7 stable shipped 18 June 2026 — three days after this RC, so RC3 was the last candidate and its busier-than-expected fix count did not delay the release. The branch has since produced 4.7.2 RC 1 (3 August) and mainline is at 4.8 dev 3 (7 August).

Regraded volatile → stable: this is a dated announcement whose URL returns the same text forever, so a timer on it can never fire usefully. The live-version question belongs on godot-engine, and the page now says so — the 2026-08-04 dated-publication rule applied to a release-candidate post for the first time in this spoke.

[2026-08-18] ingest | IMMENS — an engine that does not exist yet

Routed by the hub from Telegram. https://www.immens.dev/ — T3, freshness: volatile.

New pages: immens (source), arjan-brussee, michal-valient, sjoerd-de-jong (Person), guerrilla-games (Organization). Updated: engine-agent-surface (new instance + a closing "Retrofit versus foundation" section), synthesis (the AI-tooling open question), index.

Fetch note. WebFetch got HTTP 403 — a bot block, not a transient failure, so the retry rule did not apply. curl with an ordinary browser user-agent returned the page at 26 KB on the first try. Cheaper than the firecrawl fallback and worth trying first on a marketing site that fronts a CDN.

What it is. No engine. No repository, licence, price, date or benchmark — "Request access", "Read the specs", and "IMMENS is still in its early days." The claims: text-based open-standard formats, "full API and MCP interoperability", "structured, machine-readable context with full provenance" on every asset, an editor and pipeline "built extensively on Python", real-time GI, neural super-resolution, diffusion rendering and raytracing "as a baseline on future consoles", massive-scale simulation, boots in seconds.

Why it is still worth a page. The founders are the evidence. arjan-brussee co-founded guerrilla-games and spent eight years at epic-games helping ship UE5; michal-valient was Epic's Senior Director of Engineering over core Unreal components and engineering director on The Matrix Awakens and Lumen in the Land of Nanite, after a decade leading Decima's technology and time at Apple on Metal 2 and 3; sjoerd-de-jong spent 27 years on Unreal and twelve at Epic running developer experience. Three people whose collective experience is the incumbent, arguing the incumbent's foundations are wrong.

Where it lands. Fourth instance on engine-agent-surface, and the first of a different kind. Unreal, Unity and Godot all retrofitted a surface onto engines built for human tools, and that page's four convergent properties read as accommodations to the retrofit. IMMENS claims the surface is the architecture. Recorded as a stated position, not an instance — and the new closing section on that page says plainly that nobody knows whether the shared constraint was the retrofit or the problem, because the only project claiming otherwise has not shipped.

Tier honesty. T3 throughout, marked on the page: usable for what is claimed and for the public record of who is claiming it, useless for whether any of it works. Nothing from it is repeated elsewhere in this wiki as a property of a shipping engine. The only unmediated detail on the site is an unfilled news template still in the markup ("Outlet name", "Headline", 1 Jan 2026) beside a press kit advertising 98 brand assets — noted on the page, since it is the one thing there that was not written to persuade.

Entity budget: 4 of 8, all confirmed new by entity-index. guerrilla-games is paged because it is the shared past of all three founders, and it opens a real gap the spoke should want: Decima has no page — a shipped, proprietary, studio-built 3D engine known mostly through conference talks.

[2026-08-22] quality | Cycle: netcode, the axis the spoke could not compare on

Coverage edge 5 half-closed with two sources and a concept page. ggpo (T1, the repo and its Developer Guide) gives rollback: inputs on the wire, prediction plus speculative execution, and the three requirements a game must already satisfy — a fully deterministic simulation, a fully serializable state, and load/save/step of one frame without rendering. game-networking-fiedler (T2, Gaffer On Games) gives the lineage and the genre split: Doom lockstep, Quake client-server at an assumed 200 ms against a real 300+ ms, QuakeWorld prediction, Unreal/Halo state sync, and RTS keeping lockstep because the state is too large to send.

netcode written as the concept page. The finding it records: two of the three families demand properties of the whole simulation, so unlike rendering or tooling this is a comparison that cannot be deferred.

Not closed — no engine's own networking stack is paged, so the spoke still cannot say what Unreal, Unity or Godot supply. Successor edge 5a asks whether any general-purpose engine supports rollback in practice. index.md gains three entries. Entities: 0 created.

[2026-08-25] ingest | Cycle: the three engines' own networking stacks (via research pass)

Coverage edge 5 closed with three first-party T1 documents, hunted for hub ## Most wanted #25: unreal-replication (Epic's networking overview), unity-netcode-gameobjects (the NGO 2.13 manual, topologies and NetworkVariable pages), godot-multiplayer-api (Godot's high-level multiplayer tutorial). netcode now carries a seven-row comparison table instead of the paragraph admitting the spoke could not compare anything.

What the side-by-side reading found. All three ship the same family — client-server with an authoritative side — and none ships lockstep or rollback, which is edge 5a's prediction confirmed from the engines' own documentation rather than from inference. Unity is the outlier on authority: it offers client-server or distributed authority as a project decision, and states plainly that the cheaper one is more vulnerable to cheating. Unreal is the outlier on scarcity: it names four mechanisms for deciding what a client does not receive (relevancy, net priority, Replication Graph, Iris) where the other two manuals name none. Godot is the outlier on granularity: one topology, but authority is a per-node property reassignable at runtime, and the transport is a swappable peer.

Two smaller facts kept because they constrain gameplay code: NGO does not support plain string in a NetworkVariable (immutability would allocate on every update), and Godot's docs make argument validation an explicit developer duty because any_peer can be put on any function.

Not closed: no measurement. Three architectures, zero numbers, nobody comparing them under load — written as successor edge 5b. index.md gains three entries; netcode and edge 5 updated. Entities: 0 created (Epic Games, Unity Technologies and the Godot Foundation are all already paged).

[2026-08-28] ingest | vgpu — Vercel Labs' WebGPU library (hub-routed, Telegram, URL-only, T1)

vercel-labs/vgpu, MIT, TypeScript, 716★ / 25 forks / 1,037 commits, created 2026-05-05, last push 2026-08-28; npm vgpu first published 2026-05-26, 0.3.1 latest (2026-08-26), 20 versions. Read from the README, scripts/lib/bundle-budgets.mjs, the workspace package.json, vgpu.sh/agents.md, and the GitHub + npm APIs. Ingested with claude-opus-5 at high effort.

Created vgpu (T1 — the artifact's own facts: API surface, command inventory, licence, versions, repository metadata, all checkable against the artifact). Updated webgpu, engine-agent-surface, synthesis and index.md.

Why it belongs here. The spoke's web corner held four scene-graph engines over a bare API (three-js, babylonjs, pixijs, playcanvas on webgpu) and nothing between. vgpu is that rung: typed WGSL module imports, one explicit Gpu context threaded through every entry point, and a stated refusal of implicit scene-graph state. Not an engine, and the spoke's domain covers the real-time creation landscape's technology, so it files as the layer the matrix was missing rather than as a fifth engine.

Three runtimes off one API surface — browser, headless Node on Dawn (vgpu/node), and a deterministic software mock (vgpu/mock) for tests. Every engine in this wiki is tested by running its editor; a renderer that ships a no-GPU adapter as a first-class export is treating testability as an API decision.

The byte budget is machinery. scripts/lib/bundle-budgets.mjs: budgets snap to 512-byte steps (smallest step above the measured size), artifacts are classed client (hard gate, and the default for anything unclassified) or tooling (5% growth warns instead of failing). The README's 25 KB gzipped fullscreen effect is the advertised number; a 2026-07-26 commit raised the entrypoint budgets to 45 KB gzip and the CLI pack budget to 800 KB. Different artifacts, so not a contradiction — recorded because it shows what a pre-1.0 budget is: a ratchet reset on purpose and kept in history.

Growth edge 4 sharpened, not closed. engine-agent-surface had three retrofitted engine surfaces and immens's unreleased claim that the surface should be foundational. vgpu is that construction and it ships: docs, examples, shader check and MCP all served from the installed package, agents.md / llms.txt / an OpenAPI examples API published, and agent-evals npm scripts that run agents against the library as a test task. Three of the four converging properties hold; the context budget does not — a library's API fits in an agent's attention and an engine's does not. So the answer to "differentiator or table stakes" depends on how big the exposed surface is, which is a better-posed question than the edge had. New residue on the edge: nobody publishes an outcome, and vgpu's eval results are not published either.

Recorded, not asserted. Contributor commits on 2026-08-28: matiasngf 1,014, claude 21, dev-tmg 1, vercel[bot] 1. The 21 are authored Claude <noreply@anthropic.com> and do ordinary maintenance. No repository policy states anything about this. Filed as a flagged tension beside godot-contribution-policy-2026, which auto-bans autonomous agents — with the asymmetry named (a decade-old foundation with a contributor ladder against a four-month-old library in a company labs org), because that is likely the whole explanation.

Not claimed: no benchmark against three-js or babylonjs — no frame times, no draw calls — so vgpu is absent from the spoke's performance axis. The repo description advertises "GPU tensors, neural networks, and math viz" that the README barely covers; recorded as claimed-not-shown.

Entity note: no new Organization node. npm run entity-index -- "Vercel" returned "nothing ≥ threshold; create new", which is the known false negative — [[vercel]] already exists in ../cloud-wiki and is the corpus's Vercel target (linked from aws and cloudflare-workers there). Linked, not duplicated. matiasngf left as a plain mention: one commit-count fact, no page-worth of evidence.

[2026-08-30] lint | superpowers renamed to superpowers-game-engine

At the curator's decision, closing the homonym item that had been carried on the hub's ## Outstanding list since 2026-07-28. Two unrelated projects were both slugged superpowers: Sparklin Labs' HTML5 engine here, and Jesse Vincent's agent-skills framework in agentic-tooling-wiki. Both pages carried a "do not merge" warning, and nothing was broken — but a bare cross-wiki [[superpowers]] had no single referent, so it would have resolved to nothing the first time anyone wrote one.

This spoke's page took the rename because the other project is the one that gets linked from outside it. The engine's name has not changed — Sparklin Labs calls it Superpowers and so does the page; only the filename moved. Display text is preserved everywhere through piped links ([[superpowers-game-engine|Superpowers]]).

Retargeted: synthesis.md, index.md, and the four pages that referenced it (three-js, babylonjs, playcanvas, pixijs). Including this log — the entries above predate the rename, and their link targets were repointed while their wording was left alone. That is a deliberate exception to append-only: had they kept the bare slug they would now resolve to an agent framework, which is a worse falsehood than an edited link target.

/game-engines-wiki/superpowers 301s to the new path, both with and without a trailing slash.

[2026-08-31] ingest | bgfx — the native rung under the engines

Routed from the hub (Telegram, github.com/bkaradzic/bgfx). T1 — first-party repository and README. New page bgfx, plus the entity branimir-karadzic (author and sole copyright holder, "Copyright 2010-2026").

Facts recorded from the README and the GitHub API: BSD-2-clause, C++, 17,450★ / 2,133 forks, repository opened 2012-04-04, pushed the day of ingest. Nine rendering backends (D3D11, D3D12, Metal, OpenGL 4.3+, OpenGL ES 3.0+, Vulkan, WebGL 2.0, WebGPU via Dawn Native only, GNM for licensed PS4 developers) over nine platform families. Bindings for ~15 languages, five of them in-repo.

Why it matters here. vgpu, routed three days ago, gave the spoke its first page on the layer between a graphics API and an engine — on the web. bgfx is that layer natively and has been since 2012, which reframes the rung from a young idea to a mature position. It also gives box3d's "middleware component" a second member, so the tier is a category now: physics core and rendering core, both swappable, neither owning the game loop.

Cross-links written: box3d (middleware class), webgpu (WebGPU as one of nine backends, and used off the web through Dawn Native — the backend layer is not web-only), babylonjs (Babylon Native renders through it) and torque2d (Torque6 descendant).

Not claimed: no benchmarks, no frame times, no comparison against a hand-written Vulkan backend — the cost of the abstraction is unstated, the same hole vgpu left. Opened as growth edge 9. The user list is self-maintained by pull request, so it records who added themselves, not who still ships; the Minecraft entry is backed by Mojang's own attribution page, most others by nothing but the entry.

Entity note: one new Person node. npm run entity-index -- "Branimir Karadzic" returned "nothing ≥ threshold; create new" and the three weak hits were unrelated people in other spokes, so this one is genuine rather than the known false negative. No same-as: his GitHub account is already the page's url: (check:sameas rejects repeating it), and no Wikidata entity exists — a search returned zero results, and guessing one would have been worse than none. bx and bimg are recorded as plain mentions, verified against bgfx's own docs/build.rst ("the directories for bx, bimg, and bgfx should all be siblings of each other") rather than asserted from the README, which never names them.

[2026-08-31] ingest | OpenTS — a Tiberian Sun reconstruction, and a controlled pair with openage

Routed from the hub (Telegram, github.com/OpenTS-Developers/OpenTS), arriving in the same batch as bgfx. T1 — first-party repository, README and LICENSE.md. New page opents, plus the entity electronic-arts.

Facts: "A modern and faithful open source rebuild of the Tiberian Sun engine." C++, GPLv3-or-later, 263★ / 39 forks, repository opened 2026-06-28. Release 0.1.0 delivers the complete Tiberian Sun 2.03 Firestorm game — GDI and Nod campaigns, the Firestorm campaigns, skirmish, save/load, all play-through tested; LAN multiplayer tested more lightly; no known user-visible regression from the original. 32-bit Windows target, VS2022 + CMake, no native Linux build. Assets are not distributed: the player must own the game.

The seam with the other source in this batch. OpenTS's renderer is built on bgfx and reaches 4K including ultrawide because of it. Two sources arriving together, one the dependency of the other, is the first time this wiki holds both ends of a middleware relationship from primary evidence.

The finding worth keeping. openage and OpenTS are a controlled pair — same genre, same copyleft licence, same BYO-assets stance, same refusal to be a remake. They differ in one variable: openage reimplements the Genie engine with no released source to work from; OpenTS builds on EA's GPL-released Command & Conquer source plus reverse engineering against the original executable. Results after their respective efforts: openage ~14.4k★ and limited playability; OpenTS 263★ and a complete tested game two months in. The publisher's licensing decision outweighs a decade of community attention — and it also sets the derivative's licence, since the EA-derived files carry additional GPL Section 7 terms. Folded into synthesis; the copyleft pole of the licensing axis now records that it can be inherited rather than chosen.

Also noted: incremental modernization toward an entity-component architecture, with new work shaped so migration stays possible — the opposite of openage's rebuild-the-simulation-first choice, and the architecture bevy starts from.

Not claimed: no performance numbers, no independent verification of the play-through testing (the completeness claim rests on the maintainers' own testing), and no documented governance — funding is a Patreon for one maintainer, against the foundation stewards elsewhere in this wiki.

Entity note: one new Organization node (electronic-arts), evidenced by the licence text naming "the source code for Command & Conquer Red Alert, which Electronic Arts Inc. released under that same license." same-as verified by fetching both URLs — and the first Wikidata ID I reached for, Q26761, turned out to be a place in Gabon; the correct entity is Q173941, found by search. Westwood Studios was deliberately not paged: it appears only in the repository's topic tags, never in the README prose, and ENTITIES.md is evidence-only.

[2026-09-09] ingest | Entity Component System (Wikipedia)

Arrived via Telegram as the Wikipedia article link; routed here by the hub (runner-up: none — dev-tooling-wiki owns build/version/ship infra, not an in-engine architecture pattern, and no other spoke fits). Read the article's intro via the Wikipedia REST/extract API.

New concept page: entity-component-system (DefinedTerm, kind: architecture), grounded in the Wikipedia article (url:, encyclopedic T2) plus the spoke's existing bevy page; same-as Wikidata Q17089065 (checked). No tier: field — it is a concept node, not a source summary — with the T2 grounding stated in prose.

Why it earned a node now. The spoke had been discussing ECS without one: bevy is paged as the ECS exemplar and synthesis's open questions already name "ECS / data-oriented architecture" as an axis Bevy opened against the standalone engines. This page is that axis — the composition-over-inheritance pattern set against the scene-tree/object model (godot-engine nodes, unity-engine GameObjects), and the point where the engine argument meets data-oriented design and its cache-locality claim. Linked from bevy and the synthesis open-question passage.

Recorded as unmeasured: the corpus holds no engine's own ECS design doc and no test of the cache-friendly performance claim, so that claim is on the page as stated rationale, not demonstration. The pattern is "most associated with" games (hence this spoke) but is general — the source notes robotics simulators such as Gazebo use it too.

[2026-09-10] ingest | Kovati — Automation Forge for Unreal Engine

Routed in from the hub (Telegram). Next.js marketing site; fetched with curl and read as stripped text, including the plugin table, the architecture section and the vendor performance figures.

Source: kovati-automation-forge, T3 — vendor marketing for a commercial product. Every number is first-party with no methodology (~14 s prompt-to-clip, $0.35 for a 95-line voiced scene, 1:2 face-solve to audio on CPU, zero re-executed steps on an unchanged project), several media slots on the page still read "placeholder", and "open where it counts" is never resolved into what is actually open. No date anywhere, so no published:; freshness: volatile.

Recorded: the 8 sets / 34 plugins and what each makes; the PlayableOps four-step frame (make it playable · keep it traceable · graduate it deliberately · replace it safely); the playable gap argument; and the Trinity Architecture claim — every operation on a typed subsystem, editor UI and agent toolset as thin clients over it, "if a function needs a widget to work, the design is wrong."

Updated engine-agent-surface, and this is the substantive change. It becomes the page's fifth instance and its first from a third party — not an engine vendor's plugin and not a community MCP server. It also adds a fifth converged property: operations priced before they commit, with the run suspending where a human should decide. Unity, Unreal and Godot have no analogue because none of their operations spend money.

Synthesis: new section. The pitch concedes that generated assets are not shippable and sells iteration before authoring instead, which is a more engine-literate framing than this spoke's usual AI-tooling sources — it locates the work in the engine-side column (skeleton, montage slot, notify, binding) that no generator returns. Set against godot-contribution-policy-2026: two answers to the same provenance anxiety, one banning generated input, the other instrumenting it, neither measured.

No organization page. The site names the product family and one contact email, no company details beyond the name and a GitHub org; a node would repeat the source page. Revisit if a second source names the company.

[2026-09-14] ingest | glTF 2.0 Specification (Khronos) — via research pass

Hunted against growth edge 6 (the content pipeline). WebFetch got a 403 from the Khronos registry; fetched with curl and read locally. Source: gltf-specification, T1, published: 2021-10-11 (version 2.0.1). New concept: content-pipeline, linked from the four engine pages. Edge 6 half closed — the spec half; the studio pipeline account and USD are still owed. Synthesis section The way in gets a document.

[2026-10-06] ingest | Unity PolySpatial — Supported Unity Features & Components (via quality cycle research pass)

Hunted for growth edge 1 (what Unity loses on visionOS). Read Unity's first-party feature list for PolySpatial 2.0.4 (docs.unity3d.com/Packages/com.unity.polyspatial.visionos@2.0/manual/SupportedFeatures.html, T1). Folded into unity-polyspatial as a new section rather than a new page, since the page is about this package; struck the "doesn't enumerate the unsupported features" sentence with a dated correction. Finding: simulation (physics, animation, scripts, NavMesh) comes across; custom rendering (ShaderLab, post-processing, VFX Graph, Skybox, Fog, LOD, screen-space UI) does not. The page contradicts itself on UI masking; both statements kept. Synthesis: edge 1 closed, open question answered for Unity; the Godot/Unreal comparison (edge 2) remains. Dedup: rg over every spoke found no other PolySpatial or visionOS page.