RMH Studios — Full-Website Optimization & Dead-Code Removal Plan¶

Generated 2026-06-20 from a full read of the codebase. Scope: app/, components/, lib/, hooks/, stores/, server/, scripts/, public/, build config, and deps.

Execution status (updated 2026-06-20)¶

Everything except the CDN migration has been executed. Several items turned out to be already handled by the codebase/framework — recorded here so they aren’t re-attempted.

Item	Status
0.1 Library asset audit	✅ Audited — 0 orphans (all 44 files referenced); size reduction deferred to CDN
0.1 CDN migration of `public/`	⏸️ Deferred (per request)
0.2 Gitignore generated vibe-packages	✅ Already gitignored & untracked
0.3 Delete legacy `server/socket-server.ts`	✅ Deleted + tsconfig entry removed
0.4 Remove `threejs-brain-animation`	✅ Removed
1.1 Lazy-load heavy routes	✅ `autoCodeSplitting` is the framework default (on) — components already split; no change needed
1.2/1.3 Large-module + vendor chunk splitting	✅ Verified: `mockData.ts`/research papers isolated to their route chunks; vendor chunks already configured
1.4 Strip client `console.log`	✅ Added `esbuild.pure` (drops `console.log`/`debug` at minify; keeps warn/error)
2.1 Server infra dedup	✅ Already factored into `server/shared/`; `auth.ts`/`config.ts` left intentionally divergent (rmhbox has a Discord-auth path; security-sensitive)
2.1b recap/kowloon wiring	✅ `recap` added to `build`/`start`; orphaned `kowloon-knockout/relay.ts` deleted
3.1 Consolidate formatters	✅ `formatCount`/`timeAgoShort`/`formatRelativeTime` in `lib/utils.ts`; 7 call-sites updated
3.2 Standardize on `OptimizedImage`	✅ Build-thumbnail call-sites converted (avatars/blob-previews/external art left as-is)
3.3 Markdown stack	✅ Reviewed — both libs legitimately used; no change
3.4 TODO triage	✅ 4 markers, all legit feature notes; nothing to remove
4 Repo hygiene	✅ `src/` deleted; `clean` script added; `.DS_Store` already untracked
5 Unused-dep sweep	✅ Removed `@radix-ui/react-popover`, `@tanstack/react-virtual`, `embla-carousel-autoplay`; knip file/export sweep left as manual follow-up (too many false positives to auto-act on)

Verification: pnpm lint (0 errors) + tsc --noEmit (0 errors) on changed files; lockfile re-synced.

Read this first — two project-specific traps that make “obvious” cleanups wrong:

lib/rmhvibe/vibe-packages.ts is a curated registry. Packages like gsap, d3, konva, p5, chart.js, matter-js, lodash-es have zero import sites in the app but are installed on purpose so the RMHVibe sandbox can host/inline them for user-generated pages. Do not delete them. A dep is only truly unused if it’s absent from both the app and VIBE_PACKAGES.

*.server.ts files are stubbed on the client by the stubServerFiles() Vite plugin (vite.config.ts:15). Don’t “fix” a client import of a server file — that’s by design.

Codebase at a glance¶

Area	Files	Lines	Notes
`components/`	612	127,924	Largest surface; per-feature subtrees
`lib/`	430	108,170	Game engines, data, schemas
`server/`	110	36,490	8 standalone Node services + handlers
`app/`	269	31,416	130+ TanStack routes (all eager)
`go-services/`	85	18,490	Separate Go microservices
`public/`	—	743 MB	git-tracked assets (see §1)

The app is healthy and already does several things right (manual vendor chunks, server-file stubbing, heavy 3D/editor libs externalized). The wins below are concentrated in (a) the asset/repo weight, (b) eager route loading, and (c) duplicated server infrastructure.

Priority 0 — Highest impact, low risk¶

0.1 Reduce the 743 MB of git-tracked public assets¶

public/ is 743 MB and fully committed to git (none of it is gitignored):

Dir	Size
`public/library`	488 MB
`public/sprites`	123 MB
`public/music`	95 MB
`public/models`	20 MB
`public/vibe-packages`	8.7 MB (generated — see 0.2)
`public/images`	7.5 MB

This bloats every clone, the Docker build context, and image layers.

Actions (in priority order):

Move large binary media (library, music, models, large sprites) to object storage (S3/R2/Spaces) or Git LFS and serve via CDN. The app already has an image proxy (app/routes/api/image-proxy.ts) and lib/image-optimize.ts to build on.
Audit public/library (488 MB) for orphaned/superseded covers and PDFs against data/library-metadata.json; delete anything not referenced.
Confirm .dockerignore excludes whatever moves to CDN so the build context shrinks.
Verify before deleting: grep each asset path against the codebase; library/build assets are often referenced by string-built URLs, not literal imports.

0.2 Stop committing generated `public/vibe-packages/`¶

These are built by scripts/build-vibe-packages.ts (run in pnpm build). If committed, they’re 8.7 MB of regenerable output. Gitignore the build outputs, keep the build step. Confirm the Docker/deploy flow runs build-vibe-packages (it does, via the build script).

0.3 Delete the orphaned legacy socket server¶

server/socket-server.ts (1,164 lines) is superseded by the server/socket-server/ directory. It is imported nowhere — its only reference is tsconfig.server.json.

Action: delete server/socket-server.ts and remove its entry from tsconfig.server.json. Confirm server/socket-server/index.ts is the one wired into package.json and Dockerfile (it is).

0.4 Remove the one genuinely-unused dependency¶

threejs-brain-animation appears only in package.json — not in the app, not in scripts, not in VIBE_PACKAGES. Remove it.

Re-confirm with grep -rl "threejs-brain-animation" --include=*.ts --include=*.tsx . (excluding node_modules) before removing. Run the same check for any other dep you suspect, and cross-check against VIBE_PACKAGES first (see top-of-file trap #1).

Priority 1 — Performance: client bundle & route loading¶

1.1 Lazy-load heavy routes (biggest perf lever)¶

All 130+ routes use createFileRoute and app/routeTree.gen.ts statically imports every one. createLazyFileRoute usage is 0. The Vite config itself documents the fallout:

“With 130 eager routes the crawl never settles… the page never hydrates.” (vite.config.ts:249, the holdUntilCrawlEnd: false workaround)

Heavy game/editor routes (altair, velum2099, void-breaker, kowloon-knockout, rmhcode/Monaco, temple-of-joy, slice-it, dream-rift, synapse-storm) are pulled into the initial graph even when a visitor only wants a marketing page under _site/.

Action: split route code from route definitions using TanStack Router’s lazy pattern (*.lazy.tsx / createLazyFileRoute) for the heavy game/tool routes first. Several routes already do component-level lazy() (e.g. altair/index.tsx, velum2099, neon-driftway) — extend that consistently and push it up to the route boundary so the route module itself is deferred, not just its canvas. Measure with pnpm build chunk output before/after.

1.2 Verify per-feature splitting of the largest data/engine modules¶

Large modules that should each live in their own lazily-loaded chunk (confirm they aren’t in a shared eager chunk):

lib/rmh-eats/mockData.ts — 7,234 lines (mock data shipped to client?). If this is demo data, gate it behind a dynamic import or move it server-side.
components/research/papers/*.tsx — three files at ~3,000 lines each. These are static article content; lazy-load per paper (route is already research.$slug.tsx).
lib/temple-of-joy/data/upgrades.ts (2,991), lib/altair/engine/*, components/velum2099/ game/scene/CyberpunkScene.ts (2,297). Ensure each is reached only through its route chunk.

1.3 Confirm the existing manual chunks still hold¶

vite.config.ts already isolates three, monaco (sub-split by area), tiptap, pixi, recharts, framer-motion. Keep this — just verify after route-splitting that these vendor chunks aren’t being pulled into the entry chunk by an eager route import.

1.4 Strip `console.log` from production client code¶

55 console.log calls in app//components//lib/. Replace with a gated logger or strip via build (esbuild drop: ['console'] for the client build), keeping intentional console.warn/error. Don’t blanket-drop on the server services — they use real logging.

Priority 2 — Server: deduplicate per-service infrastructure¶

Each standalone service (socket-server, rmhbox, rmhtube, rmhmusic, discord-bot) re-implements the same plumbing. server/shared/ already exists and is the right home, but is under-used:

File	Copies	Canonical home
`prisma-client.ts`	6	`server/shared/prisma-client.ts` (already 45 lines, fuller)
`logger.ts`	6	`server/shared/logger.ts` (already 52 lines)
`rate-limit.ts`	4	`server/shared/rate-limit.ts` (already 93 lines)
`auth.ts`	3 (rmhbox/socket/rmhtube, 100–154 lines)	extract shared core
`config.ts`	4	extract shared env-parsing helper, keep per-service values

Action: the per-service prisma-client.ts/logger.ts files are mostly 7–22 line re-stubs. Re-export from server/shared/ instead of duplicating, then delete the duplicates. For auth.ts and config.ts, factor the shared validation/env-parsing into server/shared/ and leave only service-specific values behind. This is the single largest maintenance-redundancy win in the server tree. Each service is bundled independently (esbuild --packages=external), so sharing a module has no bundling downside.

2.1 Fix the recap/kowloon build inconsistencies (found while mapping services)¶

server/recap/index.ts (309 lines) is built in the Dockerfile (line 94) and run in docker-compose (line 141) but is absent from package.json’s build/start/dev scripts. Local builds never produce it. Add it to the npm scripts or document why it’s Docker-only.
server/kowloon-knockout/relay.ts (159 lines) is built/wired nowhere (gameplay appears to run through server/socket-server/handlers/kowloon-knockout.ts). Confirm it’s live; if not, delete it.

Priority 3 — Code redundancy & consistency¶

3.1 Consolidate duplicated date/number formatting¶

formatDate/formatNumber/formatDistanceToNow helpers are re-defined across at least: lib/temple-of-joy/numbers.ts, lib/lights-out/seed.ts, lib/daily-puzzles/seed.ts, components/user-builds/BuildDetail.tsx, components/feed/ConversationView.tsx, components/feed/ProfileColumn.tsx, components/rmhcode/TokenGenerator.tsx.

Action: add formatters to lib/utils.ts (already the home of cn) and replace local copies. date-fns is already a dependency — use it consistently rather than hand-rolled relative-time logic.

3.2 Standardize on the existing image component¶

components/ui/OptimizedImage.tsx + lib/image-optimize.ts exist, but several call sites still build <img>/URLs by hand (components/builds/OfficialBuildCard.tsx, components/user-builds/BuildCard.tsx, etc.). Route all user/library imagery through OptimizedImage so sizing/lazy/loading="lazy" is consistent.

3.3 Markdown stack — both libs are used; keep but scope¶

marked (3 sites) and react-markdown (4 sites) coexist. This is acceptable (server-side string rendering vs React rendering), but confirm neither is pulled into a shared eager chunk, and pick one for any new markdown rendering to avoid a third path.

3.4 Address the 10 TODO/FIXME/HACK markers¶

Low volume (10 across app/components/lib/server). Triage in one pass: fix the trivial ones, convert real work into tracked issues, delete stale ones.

Priority 4 — Repo hygiene / build artifacts¶

dist-server/ exists on disk with stale Feb/Mar build output. It is gitignored (good), but clean it locally (rm -rf dist-server) so it’s regenerated fresh.
.output/ is 1.5 GB locally (gitignored). Add a clean script (rimraf .output dist-server .tanstack tsconfig.tsbuildinfo) and document it.
src/ is empty — delete the directory.
.DS_Store files are committed throughout (server/.DS_Store, components/.DS_Store, lib/.DS_Store, public/.DS_Store, docs/.DS_Store, cli/.DS_Store, db/.DS_Store, root). Add .DS_Store to .gitignore and git rm --cached them.

Priority 5 — Dependency audit (do carefully)¶

Run a tooling-assisted pass, but filter every result against VIBE_PACKAGES (trap #1).

# Candidates: in package.json, but check VIBE_PACKAGES before removing
for p in <suspect>; do
  echo "$p: app=$(grep -rl "['\"]$p['\"]" --include=*.ts --include=*.tsx app components lib server scripts hooks stores | wc -l)"
done

Already verified in this pass:

threejs-brain-animation → truly unused, remove (§0.4).
gsap, d3, konva, p5, chart.js, matter-js → 0 app sites but in VIBE_PACKAGES (hosted). KEEP.
lodash-es → 0 app sites but VIBE_PACKAGES inline tier (must stay a runtime dep). KEEP.
react-player (1), pixi.js (2), jszip (1), turndown (1) → low but real usage. Keep.

Then consider depcheck/knip for a fuller unused-export and unused-file sweep across components/ and lib/ (the two largest trees), which static greps here can’t fully cover. Run knip and treat its output as candidates, not commands — verify each before deleting.

Suggested execution order¶

P0 — assets to CDN/LFS, gitignore generated/.DS_Store, delete server/socket-server.ts
- empty src/, remove threejs-brain-animation. (High impact, near-zero risk.)
P1.1 — lazy-load heavy routes; re-measure bundle. (Biggest runtime win.)
P2 — collapse duplicated server infra into server/shared/; fix recap/kowloon wiring.
P1.4 + P3 — strip client console.log, consolidate formatters/image usage.
P5 — knip/depcheck sweep for unused files/exports.

Verification per change¶

pnpm lint and a typecheck (tsc -p tsconfig.json --noEmit) after each batch.
pnpm build and diff the client chunk list / sizes before vs after route-splitting.
pnpm epic:test / vitest where the touched area has coverage.
For asset removals: grep the path across the repo (literal and string-constructed URLs) before deleting; assets are frequently referenced via interpolated paths.
For dep removals: install clean (pnpm install) and run pnpm build to catch transitive use.