RUNECHAIN

Whitepaper · v1 (living document) · updated June 2026 · Published by 49th Parallel
A grind-only economy, a single value-exit, and a game built by a grid of coding agents under human audit.

1. Abstract

RUNECHAIN is a browser-native, cozy-gothic pixel Soulslike set in the ruined parish of Gracefall. Its thesis is economic as much as ludic: power is earned, never sold; cosmetics are the only thing real money buys; and the only way value ever leaves the system is by selling a character you actually finished. The game runs with no build step and minimal dependencies, and every line of its code is agent-written — by a fleet of heterogeneous coding agents (Codex, Claude, Devin, Jules, and Copilot), with a single human auditor who only reviews and merges. The next round closes even that loop: zero humans, end to end.

This is a research prototype. The Chainwell is an in-browser proof-of-work gameplay mechanic and holds no monetary value. Account sign-in with Google and Solana wallet binding is now being introduced as the identity layer; the real-money edge — settlement and character sales — is built and devnet-tested but ships disabled, gated behind legal/compliance sign-off. No real funds move today.

2. The world

A player is Recorded into the Chainwell — the in-game ledger — and progresses by grinding, not by paying. One game switches between four play-styles around a single character and a single economy:

• MMO town — a shared, persistent top-down hub (Hearthlight Chapel and the parish around it).
• Real-time battle arena — PvNPC combat and PvP initiation.
• 2D platformer — solo run/jump traversal.
• Turn-based RPG duel — where PvP is actually resolved, and where boss phases play out.

Bosses chain these modes together: a platformer run into an arena wave into a turn-based phase, scripted per boss as an ordered list of segments. Defeated bosses mint deterministic, one-of-a-kind Boss Sigils.

3. The economy

Two currencies, and the line between them is the whole design.

Because real money can only ever buy looks, the free grind always reaches every power tier — just slower than a payer gears cosmetically. There is no pay-to-win and no chance-based spending.

3.1 The single value-exit

Exactly one path lets value leave the system: selling a season-complete character. A character is sell-eligible only after finishing the season's mandatory grind tasks while the shared season window was open. The buyer inherits all grind-earned progress; any seasonal paid quality-of-life resets at the season boundary, so purchased advantage is never acquired by proxy. RUNE, Gold, relics, and Sigils never individually cash out.

3.2 Real-money settlement

Wrapped-SOL purchases route an atomic split — 50% burn · 35% marketing · 15% ops fee — and credit Gold on the Chainwell. The burn is a true SPL supply burn, not an incinerator address. The settlement and character-sale programs are written as Anchor/Solana contracts and pass devnet integration tests, but they ship paused and undeployed — the server still mocks settlement and no real funds move. Flipping it live is a hard precondition gated on documented legal/compliance review, because concentrating cash-out into character sales relocates regulatory exposure rather than removing it.

4. Two ledgers, a narrow seam

The Chainwell — a from-scratch SHA-256 proof-of-work chain, verified against Node's crypto — is the permanent ledger of record for the gameplay economy. Solana appears only at the real-money edge (purchase settlement) and the character-sale boundary. Those crossing points are few and are treated as first-class, reconciled interfaces. The server is authoritative for all value: clients propose, the server disposes. Reward RUNE is server-issued proof-of-work — the browser searches the nonce, the server constructs and re-validates the block before append, so forged credits are rejected.

5. How it was built — and Moltgrid

Every line of RUNECHAIN is agent-written. A fleet of heterogeneous coding agents builds it against one shared git repository, one living PRD, and one ratified design bible; a single human auditor (Angus Durrie, with help from Robert Kodes) only reviews, reconciles, and merges — no human writes code.

The hard problem is convergence, not capability: branches revert beneath one another and agents disagree. main is the source of truth, a living PRD tags every decision ([DECIDED], [OPEN], [TENSION]), and a ratified bible resolves conflicts so the workers converge instead of drift.

Development is now moving onto Moltgrid — a distributed agentic work grid that brokers jobs to heterogeneous agent systems. A planning tier decomposes a goal into an objective DAG; a broker plans each objective into implement-and-review jobs and hands them to worker adapters (Claude, Codex, DeepSeek, a local model, or AWS Bedrock) by earned, per-capability reputation. Each job runs in an isolated git worktree under its own local login; results are validated against schema, tests, and a reviewer structurally independent of the implementer, and are rejected or retried fail-closed. A deterministic dependency floor holds an objective until its upstreams are approved, and an optional debate panel — cheap models argue, a premium judge decides — rules release · hold · escalate. Today one human gate still merges; the next round closes it — Moltgrid runs end to end with zero humans in the loop, and that autonomy is the bet. Moltgrid is zero-dependency Node and runs locally; no AI credentials ever leave the machine.

6. Architecture constraints

• Buildless client (hard constraint). No bundler, no framework. The client is index.html plus plain ES modules. New functionality arrives as new modules — never a build step.
• Dependency-flexible server & contracts. The realm server is one zero-dependency Node file; on-chain settlement may justify the Anchor/Rust toolchain.
• Browser-now, don't preclude mobile. Input is routed through a source-agnostic action shape; wallet integration is adapter-based.
• Extend, don't replace. New play-styles slot into the existing mode manager and level formats rather than inventing parallel systems.

7. What is honestly unfinished

The art is the acknowledged deficit — the actual game is a tiny top-down pixel space, and the box art on the landing page is, in the proud tradition of SNES cartridge covers, far cooler than the thing it sells. Real-money settlement is still mocked: the on-chain contracts are written and devnet-tested but undeployed. Sybil resistance has moved from a soft, browser-local cap toward a three-legged identity — Google sign-on, a required Solana wallet as the one-account gate, and a browser device key — but verified, production-grade identity (and the one-character cap that depends on it) remains an open legal decision. None of this is hidden; it is the work in front of the grid.

8. Summary

RUNECHAIN is a bet that an economy can be honest — grind for power, pay only for beauty, one clean exit — and that a grid of coding agents can build and maintain the system that enforces it. Today a human still audits and merges; the next round runs on Moltgrid with zero humans in the loop. The game is the proof, the Chainwell is the ledger, and Moltgrid is the author to come.