WAHNAHBE SYSTEM

A Solo Leveling-style life-RPG dashboard that runs permanently on my machine and holds my whole life: learning progress, quests, health, agenda, and Japanese study — fed live by my second-brain Obsidian vault, with an uncapped rank ladder that unlocks premium visual effects as I level.

The idea: files are the API

There is no database, no cloud, no auth. The dashboard is a local Express server + React SPA whose entire state is plain files:

• Read-only vault sources — my learning report card, tutoring gradebook, concept-mastery tracker, wiki activity log, and Japanese progress.md are markdown files maintained by Claude tutoring sessions. The server parses them live into an 8-stat mastery octagon, cumulative XP + level, a review queue, and activity feeds.
• Dashboard-owned state — quests, health, agenda, XP ledger, and settings live as schema-validated JSON inside the vault. Any Claude Code session (or any editor) can update the dashboard just by editing a file.

A chokidar watcher pushes every file change to the browser over SSE, so the dashboard reflects a finished tutoring session — or a hand-edited JSON file — within about a second.

What it does

XP & levels — uncapped	One unified pool: report-card XP (parsed from the vault) + an append-only XP ledger. Named ranks climb Apprentice → Engineer → Forward-Deployed Master → Architect → Netrunner → Ghost → Sovereign → Monarch (1,650 XP), then procedural ★ tiers forever, each 28% wider than the last.
Rank identity	Each rank carries an SVG insignia beside the level number and shifts the header's accent — violet at 6–7, spectral white at 8, gold at 9, white-gold at Monarch+.
Rank FX unlocks	Every rank permanently unlocks a page-level effect (see below) — lazy-loaded only when earned, paused on hidden tabs, all killed by reduced-motion or one settings toggle.
Ascension ceremony	Crossing a threshold fires a GSAP master timeline: the rank name scramble-decodes, the insignia draws itself stroke-by-stroke, and a shockwave + particle burst settles into LEVEL UP — {RANK}. Exactly once per crossing, recorded in the append-only ledger.
8-stat octagon	The radar plots real mastery scores (1–5) from graded tutoring sessions — not self-reported numbers.
MEM://MASTERY	Solid/forming/shaky concept bands as a stacked meter plus a review queue of my three weakest concepts, straight from the vault — with STALE tags when a concept hasn't been touched in 30 days.
Quests	Dailies (midnight reset), one-off side quests (+10/25/50 XP), and long-arc main quests with progress sliders. Every award is stat-tagged and rolls up into per-stat mileage next to the octagon's mastery averages.
Health & agenda	Weigh-in log with sparkline + delta from baseline; quick-add agenda with Google Calendar sync (via a Claude skill that merges by gcalId, never touching manual entries).
Japanese	Level, streak, and rolling comprehension parsed from the tutoring vault's frontmatter, with a staleness nudge after 14 idle days.

The rank FX ladder

Lv	Rank	Unlock	Technique
2	Practitioner	Panel focus glow	GSAP micro-tweens
3	Analyst	XP bar energy current	masked gradient sweep
4	Engineer	Momentum ring draw-in	DrawSVG
5	FD Master	Terminal scramble-decode on titles	ScrambleText
6	Architect	Staggered boot cascade	GSAP master timeline
7	Netrunner	Katakana data rain backdrop	pooled canvas, tab-aware
8	Ghost	Organic turbulence displacement	SVG feTurbulence + feDisplacementMap
9	Sovereign	Gilded light sweep across the header	measured px tween, blend-mode screen
10	Monarch	FBM aurora shader behind all panels	OGL WebGL fragment shader, half-res; ★ tiers deepen it

Architecture

second-brain vault (markdown + JSON)          THESYSTEM
┌─────────────────────────────────┐   ┌─────────────────────────────┐
│ report card / gradebook / logs  │──▶│ parsers (read-only)         │
│ concept mastery / jp progress   │   │                             │
│                                 │   │ Express API  ──▶  React SPA │
│ system/*.json (dashboard state) │◀─▶│ zod + atomic writes         │
└─────────────────────────────────┘   │ chokidar ──▶ SSE live push  │
                                      │ rank fx: GSAP + OGL, gated  │
                                      └─────────────────────────────┘

• Server (server/) — Express 4, the only component that touches disk. A single thresholdFor(level) generator drives both level math and ascension detection, so they can never disagree. Vault sources degrade per-source ([SYS] DATA LINK LOST) instead of crashing the endpoint; state writes are zod-validated then written atomically (temp + rename). The XP ledger is append-only and awardXp is its sole writer — completions award XP before persisting quest state, so a failed write can never strand a completed quest with no XP.
• Web (web/) — React 18 + Vite. A faithful port of a Claude Design HUD mock: clip-path panels, scanlines, boot sequence, Rajdhani/Share Tech Mono type. FX modules live in web/src/fx/* behind a gated lazy-loading hook (useRankFx) — no shader bytes ship to a level-2 player, and every module restores the DOM it touched on unmount (named clearProps only; GSAP never wrestles React for style ownership).
• Deployment (scripts/) — a Windows Task Scheduler task launches the production server hidden at logon. The launcher is synchronous, so the task instance lives and dies with the node process, and a 5-minute watchdog trigger restarts it unattended after a crash — measured recovery: 53 seconds. Binds 127.0.0.1 only.
• Claude integration (.claude/skills/) — skills for calendar sync, banking XP from any conversation, and rebuild-and-restart. CLAUDE.md documents the file contracts so any session can edit state safely.

Testing

107 tests: parser units against real-format fixtures (including malformed variants), XP-engine boundary math across the procedural ladder (559/560, 2099/2100 ★2), FX gating with stale-async race coverage, supertest API integration on a fixture vault, and a Playwright E2E that boots a sandboxed production server, completes a quest, and proves the SSE live-update loop by editing a vault file on disk mid-test. Visual tiers verified by staged browser smoke at levels 5, 8, and 12.

npm test      # server + web suites
npm run e2e   # playwright smoke on an isolated fixture vault (run from a worktree — dist is live-served)

Running it

Built for my machine, but the paths are env-driven:

npm install
npm run build
WAHNAHBE_VAULT=/path/to/vault WAHNAHBE_JP_VAULT=/path/to/jp npm start   # http://localhost:4777

Dev mode: npm run dev (server) + npm run dev --workspace=web (Vite on :5173, proxied).

Provenance

Spec-first build across three shipped iterations — v1 dashboard, purpose pass (every element must earn its place: HP/MP deleted, mastery surfaced, mileage wired), and uncapped levels + rank FX — each executed as a TDD plan with a fresh implementation agent per task and independent review gates that caught real bugs before deploy (a level-math crash, an XP-eating write order, a layout-wiping clearProps:'all', a LAN-exposed bind). The visual source of truth is the original design mock.

Screenshots show staged demo data (the hero is MONARCH ★3 at a simulated 2,749 XP).