The Dungeon Under the House

Any lot owner in New Britannia can hang a dungeon under their house: craft or find the pieces, click a doorway, and assemble halls, crypts, cisterns and boss chambers room by room, then flip the whole thing into Adventure Mode and fight what lives there. That system is deeper than most players realize, and almost nobody visits it. This document tours what shipped, surveys every serious attempt at player-made dungeons the genre has produced, and proposes twenty-five improvements, from one-evening client fixes to the prize chest that would make a stranger's basement worth raiding.

Executive summary

Twenty-five years of player-authored dungeon systems tell one story in three acts. Act one: systems that let authors mint server rewards melted down. City of Heroes' Mission Architect produced three-hour max-level farms within days of launch in 2009; EverQuest 2's Dungeon Maker had experience surgically removed from it in 2014; Neverwinter's Foundry lost its daily rewards in 2015 and was then killed entirely in 2019, when the studio admitted the knowledge required to maintain its bespoke editor pipeline no longer existed. Act two: systems that fixed farming by isolating rewards died of irrelevance instead. EQ2's play-as-an-avatar phase, where your real character never entered and a token store held the payout, proved that nobody grinds content that does not touch their real character. Act three: the survivors externalized cost and made the dungeon itself the economy. Neverwinter Nights persistent worlds are still setting concurrency records twenty-two years on because the community owns everything; and Old School RuneScape's player-owned-house dungeons have run un-nerfed for twenty years because the monsters are furniture bought with gold, the treasure chest holds the owner's stake, and the server mints nothing inside.

SotA's system already made the two hardest calls correctly. Dungeons are built from live-game primitives: the pieces are decoration items, the layout lives in the same storage as house deco, the rooms are real scenes. There is no bespoke editor to rot, which is the exact failure that killed the Foundry. And totem spawns were pre-emptively tuned to modest loot, so the farm meltdown never happened here. But the system stopped at the cost structure and never shipped the social loop. There is no way to find a dungeon, no reason to enter one, no record that anyone ever did, no map of your own halls, no traps, six creature families, and an acquisition grind that punishes the few who build. Measured against the genre, our gaps cluster in three places:

1. The ghost-town loop — no directory, no ratings, no visit or clear records, no first-clear names, no prize layer. Builders build for self-satisfaction and stop.
2. A shallow adventure vocabulary — one creature family per room, fight rooms that forbid furniture, zero traps or triggers, no live map, and blueprints gated behind rare drops.
3. Trust and stewardship debt — the point budget is client-enforced only, there is no beatability standard for anything shared, and daily friction (mode flips from outside only, budgets frozen at the founding lot's size, silent placement failures) wears down the faithful.

Twenty-five proposals follow: six client-only quick wins, eight core upgrades, and eleven server-backed flagships. The flagships share one design doctrine, taken from the only system that has survived two decades untouched: status pays the author, players pay each other, and the server mints nothing. Every server-backed card rides the two dungeon operation groups that already exist; none of them adds new network plumbing categories.

What we already built

The R61 player-dungeon system is one of the deepest pieces of housing tech in the game, and this document deliberately builds on it rather than re-proposing any of it. What exists today: a dungeon entrance placed as lot or basement decoration; six foyer types; roughly 130 room and passage pieces across five material sets, including secret breakaway walls and encounter rooms modeled on famous dungeons of Novia; a doorway-click builder with a live snapped preview that refuses to orphan rooms from the foyer; a dungeon point budget by lot size, upgradeable with gold or Crowns; encounter totems in six creature families with tiers up to boss level; Adventure Mode with an occupant-safe countdown and reload; a Random Dungeon Generator chest that assembles a seeded layout from whatever pieces you own; an open-PvP flag with an honest warning on the doorway tooltip; custom dungeon names that always display beside the owner's name; and per-room baked pathing stitched together at runtime so the monsters can actually chase you through whatever you built.

• 1The name is the entire public identity. No author line beyond the hover tooltip, no stats, no history, no record that anyone ever descended. Proposals S2, S16.
• 2Enter is all a visitor can do. Nothing says whether the dungeon is worth entering, how dangerous it is, or whether anyone has ever cleared it. Proposals S16, S17, S18.
• 3Sample dungeons are the teaching seam. Five dev-built demos ship today; the button and per-entrance override already exist. Proposal S6.
• 4Every owner feature hides in an unlabeled flyout. All five options work today; the composed window promotes them to a visible Setup tab. Section 4.
• 5Adventure Mode is the build/play switch and can only be flipped from this window, outside the dungeon. Proposal S1.

The machinery every proposal plugs into

A dungeon is decoration all the way down. The entrance is an interactive deco object on the lot; entering pushes the foyer scene with the owning lot carried along; the placed pieces load as deco items whose connection sockets hash into a room graph; the builder window validates a snapped preview against the piece sets, the budget, and a walk-back-to-the-foyer check before committing; and flipping Adventure Mode broadcasts to everyone inside, rebakes the stitched navmesh, and lights the totems, which pick the spawn table for their creature family and tier. Every proposal below extends one of these nodes. None forks the pipeline, and that is the system's crown jewel: there is no bespoke editor to maintain, which is the exact cost that killed the genre's biggest attempt.

flowchart LR
  A["Dungeon entrance\nlot decoration"] --> B["Entrance window\nenter / options"]
  B -->|Enter| C["Foyer scene loads\nwith the owning lot"]
  C --> D["Pieces load as deco\nconnection sockets"]
  D --> E["Room graph\nsnap + reachability"]
  B -->|owner doorway click| F["Piece picker\nvalidate + commit"]
  B -->|owner toggle| G["Adventure Mode\nbroadcast"]
  G --> H["Runtime navmesh\nstitch"]
  G --> I["Totems light\nspawn family + tier"]
  style F stroke:#c9a55a,stroke-width:2px
  style G stroke:#c9a55a,stroke-width:2px

Lessons: twenty-five years of player dungeons

Player housing invented the itch; player dungeons are the genre repeatedly trying to scratch it. Ultima Online proved players will pour years into spaces they own and show off, and its design contests, where winning house designs became official content, remain the cheapest author payment ever devised. Neverwinter Nights handed the whole cost structure to the community and became immortal. City of Heroes put the authoring tool inside a live MMO and learned, within days, what happens when authors can tune the rewards. EverQuest 2 built the closest system to ours, collect creatures and place them in rooms, and killed it twice, first with reward isolation and then with reward removal. The Foundry scaled quest authoring further than anyone and died of its own bespoke editor. Landmark shipped magnificent build tools with no audience loop and lasted seven months. And quietly, for twenty years, Old School RuneScape's construction skill has run dungeon rooms, monsters as gold-priced furniture, and an owner-staked treasure chest without a single economy scandal.

What each game contributes

Old School RuneScape: the whole reward doctrine. Guards are bought with gold on a curve from cheap to legendary, the treasure room chest holds the owner's own stake, challenge mode arms it, and house parties made the space a social institution. Twenty years, zero nerfs, because the server never mints anything inside. Super Mario Maker 2: the clear-check (you must beat your own course to publish it), First Clear and World Record names on every course page, fixed tags, and difficulty from real clear rates rather than author claims. Trackmania: the author's validation run seeds the bronze, silver and gold medal times automatically. Meet Your Maker: the raid ledger, every visitor's death and path reported back to the builder, was the strongest builder-retention ritual the genre has produced; its economy, paying builders per raider death, bred degenerate kill-boxes and killed the game inside a year. Neverwinter Nights: the DM client, a human running the dungeon live, is why its persistent worlds outlived four generations of MMOs. Dreams and LittleBigPlanet: editorial picks surface quality; algorithmic popularity gets farmed by reciprocal rings at any population size. Solasta: place-and-configure activators deliver most of trap scripting's value with none of free scripting's moderation and maintenance bill. City of Heroes and EverQuest 2: the two cautionary tales, author-minted rewards and isolated rewards, which between them bound the safe design space every card in this document lives inside.

Capability matrix

● native · ◔ partial · ○ absent. Reading the matrix: SotA today is already strong on the two columns that killed everyone else, moderation burden and the primitives-based stack, and weak only on the audience loop. This catalog is an audience-loop program, not a rebuild.

The twenty-five proposals

Tier 1 · Quick wins — client-only, small

Tier 2 · Core upgrades

Tier 3 · Server-backed flagships

Considered and rejected

• Boosted experience or loot from totem spawns. The genre's central catastrophe, three times over: City of Heroes' farm meltdown, EQ2's experience removal, the Foundry's reward strip. The modest-loot doctrine is the correct call; S20 replaces the incentive with player-staked prizes instead.
• A parallel dungeon currency or token store for runners. EQ2 proved reward isolation kills demand as surely as farming kills balance. Rewards here stay orthogonal but real: names, medals, titles, staked gold.
• Free-form scripting for all players. The Foundry died of its own bespoke pipeline, and free text bought City of Heroes a permanent policing bill. S14's configured activators are the proven middle.
• Per-visit or per-minute creator payouts. Fortnite's engagement pool spawned an AFK-farm industry. Authors are paid in plaques, features and jam wins.
• Kill-based builder rewards. Meet Your Maker paid builders per raider death, got kill-boxes, and died in a year. S17 pays in ratings; S19 pays in information.
• Play-as-a-standard-avatar visitor mode. EQ2 proved that without real character progression, nobody comes.
• Algorithmic popularity ranking as primary discovery. LittleBigPlanet's front page was destroyed by reciprocal-heart rings. Fixed rows and a human's weekly pick are the answer at this population.
• Full ghost replays. Death markers and times deliver most of the value without replay-capture infrastructure; revisit if S21 proves out.
• Author-tunable creature stats or difficulty knobs. Creativity in arrangement, never in stats: the precise line between RuneScape, safe for twenty years, and City of Heroes, melted in days.
• Resource-node spawners in dungeons. A standing rejection, kept rejected here for the same economy reasons.
• A bespoke external dungeon editor. The deco-primitive architecture is this system's crown jewel; everything in this document extends it, nothing replaces it.
• A volunteer curation corps with obligations. UO's Counselor program ended in a courtroom. S22 costs one staff hour a week, and community input stays advisory.
• Town dungeon gates and lot-free entrances. Not rejected; owned by the towns page, where those cards already live.
• A web directory or dungeon tour site. Owned by the companion-site program.
• Generic deco quality-of-life inside dungeons. Owned by the decorating page (points meter, undo, patrol routes, the dungeon deco pass).
• Totem auto-placement and generator chest acquisition. Owned by the random-generator roadmap follow-ups.

The composed surfaces

The danger with twenty-five proposals is a game with twenty-five new widgets. The system has exactly two windows today, the entrance and the piece picker, plus the M-key map. The composition adds one new window and grows what exists: the entrance window gains a plaque header and an owner tab strip, the picker gains filter chips and one wiring row, the Delve Board arrives as the single new surface, and everything in-dungeon is a thin HUD layer. The plaque itself is one component rendered in three places: entrance window, Delve Board detail pane, and an optional physical plaque decoration by the door.

• 1The plaque header (S16) carries identity, danger and tags (S17, S18), records and medals (S21), and the bounty line (S20). Unpublished dungeons show a short header and none of the record rows.
• 2The action row keeps Enter exactly where it was and adds your personal best (S21).
• 3The owner strip promotes the old unlabeled flyout to a Setup tab, adds the Ledger (S19) with its death map, and the Publish tab (S15).

Sequencing

flowchart TD
  P1["Phase 1\nquick wins + tuning"] --> P2["Phase 2\nmap, budget, encounters"]
  P2 --> P3["Phase 3\npublish + record + medals"]
  P3 --> P4["Phase 4\nDelve Board + ledger + jams"]
  P3 --> P5["Phase 5\ntraps + prize chest + DM mode"]
  P2 -->|"budget is the trust floor"| P5
  style P3 stroke:#c9a55a,stroke-width:2px
  style P5 stroke:#c9a55a,stroke-width:2px

Appendix

A1 · Where each flagship's data lives

Every operation above joins one of the two dungeon-related operation groups that already exist and are already dispatched on the servers. No proposal adds a new network plumbing category, which removes the riskiest class of server change from the entire program. Schema changes land as one coherent migration wave in Phase 3 plus the S20 escrow in Phase 5, each with its migration-ledger entry in the same change.

A2 · Constraints every proposal honors

• Small-population-first: every card works at one owner and a handful of visitors. Nothing requires a critical mass, a moderation team, or an algorithm.
• Status pays the author, players pay each other, the server mints nothing: no card creates gold, experience, or loot beyond what a player voluntarily staked.
• Creativity in arrangement, never in stats: no card lets an author tune creature numbers, damage, or drops.
• The deco-primitive architecture is load-bearing: every card extends the existing piece, deco, and scene machinery; none introduces a parallel editor or content pipeline.
• All player-visible sample text in this document follows the plain-voice rules and can be lifted verbatim.

A3 · Test surface

The pure-logic pieces (placement denial causes, limit derivation, seed determinism, the map graph, medal thresholds, rating tallies, publish state machine, claim state machine, credit aggregation) are all EditMode-testable without scenes, extending the existing dungeon suites: the generator layout tests, the spawner placement ratchet, the totem spawn-table ratchet, and the entrance and map model tests.

A4 · Selected sources

Neverwinter Nights persistent-world histories and DM client documentation; City of Heroes Mission Architect launch coverage and the farm-nerf timeline; EverQuest 2 Dungeon Maker release notes and the 2014 producer's letter removing experience; Star Trek Online and Neverwinter Foundry retrospectives and the 2019 shutdown notice; Old School RuneScape wiki (Construction, dungeon rooms, treasure room, house parties); Super Mario Maker 2 course-world documentation (clear check, first clears, world records, tags); Trackmania medal system documentation; Meet Your Maker economy retrospectives; Dreams and LittleBigPlanet curation histories; Solasta Dungeon Maker documentation; community threads on visiting player dungeons, blueprint acquisition, spawn-room decoration, and the fan-made dungeon minimap tool. Full source list in the research notes accompanying this proposal.

Prepared 2026-08-04 against the unity-6.5 migration branch. Companion markdown: docs/plans/2026-08-04-player-dungeons-improvements-proposal.md. Twenty-five proposals: six client quick wins, eight core upgrades, eleven server-backed flagships, each grounded in the machinery that would carry it.

General comments

No account needed. Every proposal above also has its own Discussion box for talking about that one alone; this thread is for the plan as a whole. Comments are plain text, held to basic decency, and may be trimmed.