Maintain World At Ruin

Maintain: World at Ruin (game)

A cloud-native MMORPG the maintainer wants to exist, built almost entirely by agents as a first-class portfolio product (maintainer direction 2026-07-17: it needs the same attention and love as the rest of the portfolio — not a back-of-queue pick). It participates in the normal selection rules — oldest-actionable-first and the last_worked fairness rotation — like every other product; expect the game itself to accrete over years all the same. He redirects via the PR workflow; ship draft PRs as usual.

Status: bootstrapped 2026-07-16. The repo exists, the applications/world-at-ruin submodule is in place, the roadmap is seeded as GitHub Issues (roadmap label — Phase 0 art-pipeline taste gate first), and the settled design lives in the repo's own AGENTS.md, which is authoritative. This card keeps the operate notes and the deliberately-open OPEN DECISION items below; once those are settled and recorded in the repo, thin this card to a plain pointer like the other product cards.

The stack, the product law and the design pillars below were settled with the maintainer directly (2026-07-16) — do not re-litigate those. The directive covers decisions, not the whole card: a few points are deliberately still OPEN, and they are marked OPEN DECISION where they appear. Those you may not treat as settled — decide them when you implement the subsystem, ship the call as a draft PR (he redirects there), and update this card with what was chosen. Shared cross-repo rules are in the monorepo AGENTS.md.

The premise — “as code”, which decides everything else

No UI clicks, no desktop design tools. Everything text-authored, built headlessly in CI. If an agent cannot author it, it does not get built, because nobody is hand-making this. So “as code” is the premise and every other preference — engine, fidelity — yields to it.

• Client: Godot 4 — chosen because .tscn/.tres/.gdshader are text and headless export is first-class. Unreal was rejected: .uasset/.umap are binary, so agents cannot author levels or materials. This knowingly costs the graphics ceiling (Godot ≈ semi-realistic PBR; no Nanite/Lumen).
• Art: generated as code, all OSS/CC0 — never commercial assets. Headless Blender (bpy) → glTF; MPFB2 + Rigify for characters (core assets CC0, explicitly closed-source-safe); Poly Haven / ambientCG (CC0) materials; WFC interiors, grammar towns, SDF caves, erosion terrain. Blender's GPL covers the tool, never the output. Target is stylised-realistic (“a more realistic WoW”) — that style is parametric, which is exactly why it is expressible as code. Photorealism is unreachable in any engine without a human sculptor; do not re-attempt it.
• Weakest link: animation. CMU mocap (verify licence per dataset — AMASS/LAFAN1 are research-only, never ship them) + procedural IK. Telegraphed combat mitigates this by putting the signal in the ground, not the frames.
• AI-generated 3D: MIT models exist (TRELLIS, TripoSR) but topology is poor and training-data provenance is murky — a liability for something he owns outright. Concept/props only, never a shipped hero asset.

Server

• Realtime tier: zone / dungeon-instance Agones GameServers (CNCF). One tick loop per process — NEVER decomposed. The unit of scale is the number of zones. A network hop between “player moved” and “was he in the cone” re-creates the desync problem telegraphs exist to avoid.
• Meta tier: real Go microservices over gRPC — mostly Nakama (Apache-2.0) for auth/social/chat/storage. Write only what it doesn't give you.
• Postgres/CNPG; runs on the existing platform via Flux. Seamless instancing falls out of Agones: allocate the dungeon server as the player approaches, pre-connect, hand off — no loading screens, no pop-in, ever.
• Physics stays OUT of the authoritative path — capsules/navmesh only. This is what makes a Go authority cheap, deterministic and latency-tolerant.

Licensing

Source-available and proprietary — NEVER call it “open source” (free redistribution is clause 1 of the OSD). Copying/redistribution prohibited. Needs a bespoke EULA and a CLA with copyright assignment (EU: assignment plus fallback exclusive licence) gating the first external PR. No GPL/AGPL in the shipped tree — enforce in CI, don't remember it.

Product law — the constraints that outrank the design

1. No hard resets, and nothing is taken from a player silently. An early player keeps playing as it evolves: no wipes, no seasons, no stat squishes. The world may be migrated — that is expected — but every migration is either non-breaking (expand/contract migrations, versioned save data, backward-compatible protocols; the player never notices) or goes through a visible deprecation that tells the player, in-game and ahead of time, exactly what is changing, removed, or destroyed and when. CI-enforceable, and the guard must exist before the first player does — an agent must not merge a change that strands a character or removes something from a player without an announced deprecation.
2. Experimental features are opt-in behind a feature flag. Anything unsettled ships default-off behind a flag; a player must opt in, it is validated in that state, and only once proven does it flip default-on and the short-lived release flag retire — the game-facing edge of the portfolio feature-flag-first delivery rule.
3. No power/wealth inflation, no ecosystem corruption. A dupe, a runaway drop rate or a botched economy change cannot be un-printed without the very reset the game forbids — economic integrity is engineered at the root: transactional integrity, idempotency and an audit trail are day-one requirements. Migrating content and data forward is expected; letting value leak is never allowed.

The collision to keep front of mind: WoW/Diablo-4's answer to inflation IS the reset (D4 wipes seasonally; WoW squishes stats). Both are forbidden, so economics come from Guild Wars 2 instead: horizontal progression, a ceiling that never rises, bound loot, no trading/auction house (kills RMT, botting and dupe value at the root), hard sinks. WoW/D4 texture, GW2 economics.

Design — classless weapon mastery (The Secret World-shaped)

• Classless. Playstyle is defined by which weapons you master. Two weapons equipped, or one wielded alone to empower it — flexibility and multi-specialisation.
• Trifecta: damage / healer / tank. Each weapon leans toward one or more roles.
• Progression = weapon mastery, earned by using the weapon. Mastery unlocks new arsenals — never more damage: more and more interesting ways to approach combat.
• Death: lose some unbanked mastery, respawn at the nearest respawn point, and reclaim it or lose it forever (a Souls bloodstain, kept even though the Souls loop was dropped). A filled bar banks unlosable progress — a ratchet floor.
• Balance: mitigation / damage / healing throughputs stay balanced so all areas stay relevant. Progression is advisable for the hardest content, never required — doing it unprogressed should be possible but unwise. Some areas/dungeons are gated by story or orderly completion; unlocks are account-bound.
• Hard, skill-based bosses and elites spread across open world, dungeons and raids — opt-in risk of losing or gaining larger amounts of mastery, plus the odd cosmetic.
• Loot is Elder-Scrolls-shaped: a sword is a sword. How well you use it is your mastery. Huge visual variety so players find the look they want without touching balance. Armour is the exception — real mitigation/lightness trade-offs: light = agile, heavy = takes a hit.
• Endgame: mythic-like dungeons, regular dungeons, raids, endless dungeons. Loot-based progression here only — vertical, with a real loft (maintainer direction 2026-07-17). In the open world, endgame gear gives a cosmetic edge ONLY: the open world is a fair challenge for everyone, and an experienced player's edge is their arsenal of abilities and weapon skills.
• All endgame stays relevant — keys and scaling everywhere (maintainer direction 2026-07-17): every endgame activity carries keys and scaling that keep it engaging and replayable; expanding the endgame means BOTH improving what exists and building more; neither the endgame nor the open world may ever go stale or irrelevant.

Design guards — the traps in the above, and how to hold them

These are the non-obvious failure modes. Treat them as laws, and prefer designing them out over policing them after the fact — economic corruption, once loose, cannot be recalled without the reset the game forbids.

• 🔴 The endgame ladder is the one place power grows — it MUST be bounded and inert outside itself. “Gear upgraded to take on harder and harder content” is vertical progression, i.e. the exact inflation the product law forbids. SETTLED by maintainer direction 2026-07-17: (a) endgame gear is cosmetic-edge-only in the open world (stat-normalised/inert outside endgame instances — GW2 downscaling / WoW Timewalking are the reference mechanics); and (b) the vertical has a loft but a bounded one — beyond it, difficulty scales, not power (endless dungeons, mythic keys), and rewards become score and cosmetics. That is how “harder and harder” runs forever with no inflation and no reset.
• 🔴 Usage-based progression's classic exploit is AFK/dummy grinding (UO, Skyrim, ESO all bled here). With no wipe available this is permanent. Mastery accrues only from meaningful contested combat — appropriate-level hostile targets, server-authoritative, diminishing returns, and zero progress from training dummies, self-damage, or trivial mobs.
• Every new arsenal ability must be a SIDEGRADE, never a strict upgrade. He is right that more options raise throughput in effect — that is precisely how power creep enters a game that claims to have none. Situational-by-design is the law, and “no strict dominance” is simulatable, so it is an agent-ownable CI guard rather than a matter of taste.
• Tune content against the banked floor, not peak mastery. Unlosable progress is the only power level every player is guaranteed to have; everything above it is skill expression.
• OPEN DECISION — death penalty in group content breeds blame. A bloodstain is fine solo; “you cost me my mastery” on a raid wipe is why WoW removed corpse runs. Consider full risk in open world/solo (his stated risk/reward intent) and a softened penalty in organised group content. Flag it as a decision, not a detail.
• Classless + dual weapons SOLVES tank/healer scarcity — the classic MMO queue problem. Any player can swap to fill the missing role. Protect this; it is a genuine strength of the design.
• Axis map, to keep balance legible: weapons = horizontal (your arsenal; cosmetic variety only). Armour = your role/agility axis, and the bounded endgame vertical. Keep them from blurring.
• OPEN DECISION — classless + account-bound unlocks make alts near-pointless — especially with full appearance freedom. Consider one character per account: it simplifies the data model and identity, and removes mule characters as an economy vector.

Phase 0 — before ANY game code

Prove the art pipeline: headless Blender in CI → one MPFB2 character with proportions pushed stylised-realistic → one procedural cave → standing in Godot. It is a taste gate the maintainer judges, not a test suite, and it is the project's one unproven bet. If generated art can't clear his bar, the premise fails — cheap to learn now, ruinous to learn in Phase 6.

Roadmap & enhancement

The roadmap lives in GitHub Issues on devantler-tech/world-at-ruin (roadmap label), like every other product. Advance via product-engineering. Keep issues agent-shaped — small, specified, testable, art-free. The project stalls the moment the next task is “make the combat feel good”: that is a taste judgement and it needs the maintainer, so route those to him rather than guessing.

Phase 0 is the deliberate exception to “testable, art-free” — it is entirely art and it ends in a taste judgement, so the rule above would forbid the project's own mandatory first step. Split it so the roadmap can carry it honestly:

• The machine-verifiable part IS a normal issue: a headless Blender container runs in CI, an MPFB2 character and a procedural cave build reproducibly from committed scripts, and the glTF loads in Godot. That is testable, and an agent owns it end to end.
• The taste gate is a separate, explicitly maintainer-blocked issue — he looks at the render and says yes or no. Never self-answer it, never infer a pass from CI being green, and never start game code because the pipeline runs. It is the one issue in the backlog that is supposed to wait for him.