Emergent quests from living worlds, and an open blueprint for the next generation of MMOs.
Questless is an open-source, deterministic quest-economy engine for MMO-style worlds. A settlement of 100 rule-based NPCs farms, mines, forges, hauls and trades. When the economy leaves needs unmet, NPCs, guilds and the town post funded player contracts. Every contract can answer the question "Why does this quest exist?" with a causal chain built from recorded data, and each number in that chain can be re-measured to prove it.
Questless has two halves:
| What it is | Status | |
|---|---|---|
| The engine | A small, tested Python engine (one settlement, 100 NPCs, 4 goods) with a dashboard, API and CLI. Coin and goods are conserved every simulated day, every quest is funded and explains itself, and every run replays exactly. | Working today (v0.1.0, 153 tests) |
| The blueprint | A public roadmap and build guide for the systems a 2030s megaserver MMO would need: futures and central banks, courts, elections, diplomacy, ecology, fair automation instead of bots, privacy by design, AI steward NPCs, and real-world skills taught through play. | Planned and open for contributors (v0.2 to v0.6, about 60 milestones) |
It is not a game: there's no 3D client, no combat and no multiplayer server. It's the groundwork: working proofs of the hard systems at small scale, and a detailed plan for the rest. Built as a systems-design study by one person with AI-assisted development; MIT licensed.
The late 1990s and early 2000s were a golden era for massively multiplayer games. Ultima Online, EverQuest, Star Wars Galaxies, EVE Online and World of Warcraft gave millions of people persistent shared worlds with player-driven economies, guilds and politics, and friendships that lasted years. Economists studied their markets; EVE still runs a single shared world for all its players.
Since then, very large new MMOs have become rare: they're expensive and risky to build, hard to balance, and easy to break with bots, inflation and exploits. Many of the problems that sank them are the ones Questless works on:
| 2000s problem | What the 2030s make possible | Where Questless works on it |
|---|---|---|
| Economies balanced by patch notes, after the damage | Simulate and test balance before release, on every change | Deterministic replay; economy checks on pull requests (M23) |
| Hand-written quests that ignore the world | Quests caused by the world, that explain themselves | The engine today |
| Bots, gold farmers, multiboxing | Fair automation for everyone, economic detection, privacy-respecting identity | M31, M37, privacy design |
| Servers split by size and language | One megaserver with regions and simulation level of detail; live translation | Blueprint, M34, M43 |
| NPCs as vending machines | AI steward NPCs that keep the world fun and push players together | M44 |
| Games as pure escapism | Worlds where players also learn economics, law, politics and data literacy | Skill tracks, M26, M47 |
The settings are interchangeable: the same systems are designed to run a fantasy continent or a science-fiction galaxy (M33).
A mine collapses on day 15. Ore output falls, smiths run out of ore, and on day 29 the Blacksmiths Guild posts a contract. This is the engine's own explanation:
Day 15: The mine: mine collapse factor set to 0.20 by an external shock.
Day 29: Iron ore production output fell to 5/day (normal 23.80/day).
Day 29: Market for iron ore: 51 units wanted by buyers who could pay went unsold.
Day 28: Elan Anvil went 2 days without iron ore (patience 2 days).
Day 29: Blacksmiths Guild: 88% of members who use it are short (emergency at 60%).
Day 29: Blacksmiths Guild could fund 31 of the 127 units needed.
Day 29: Blacksmiths Guild escrowed 363 coin (had 914, limit 365).
Nothing in that chain is written by hand. In a longer disaster the walk goes further:
Mine disaster → ore output down → smiths idle → tools output down → farmers without tools → food output down → a food contract.
That food contract is linked to the earlier ore contract as its parent.
Requires uv, which installs the pinned Python (3.12) for you.
git clone https://github.com/Adiezc/questless.git && cd questless
uv sync --extra demo
uv run questless demoThat opens the dashboard at http://localhost:8501. It starts on a live quest; press Why?, Accept, send a caravan, advance 3 days and Deliver.
Headless:
uv run questless run scenarios/iron_shortage.toml # contracts and why they exist
uv run questless run scenarios/drought.toml --json # machine-readable
uv run questless run scenarios/bandits.toml --save s.json # save the run
uv run questless replay s.json # rebuild it and verify the state hash
uv run questless repro bug.json # replay a bug report and check what it saw
uv run questless diff --base origin/main # economy impact of your changes
uv run questless export scenarios/iron_shortage.toml --seeds 20 --out data/ # CSV for research
uv run pytest # 153 tests, ~14 s NPC activity ──► economic conditions ──► unmet needs ──► funded contracts
▲ │
└──────────── world consequences ◄──── player actions ◄────┘
- NPCs (34 farmers, 16 miners, 8 blacksmiths, 14 merchants, 28 guards) follow priority rules: eat, keep their tools working, produce, trade. There's no LLM and no scripting; behaviour is seeded and deterministic.
- Market and transport: goods are traded once a day at prices that follow production, consumption and stock cover. Merchants haul goods from the mine and farms to the market, with limited daily capacity.
- Guilds (Miners, Merchants, Blacksmiths) collect dues, keep members able to work, and pool members' needs into one contract when many share them. The town does the same for its guards.
- Contracts are posted only for persistent scarcity: the NPC needed the good, could pay
for it, and nobody could sell. Poverty is tracked separately. Before posting, a contract must be:
- funded: the reward is escrowed by the NPC, its guild or the town. A guild that can't afford the whole job posts what it can.
- deduplicated: one live contract per need.
- cooled down: a need whose contract expired or was withdrawn waits before it can post again.
- capped: limits on live contracts and on new contracts per day.
- feasible: someone could actually obtain the goods.
- Provenance: a bounded causal walk explains each shortage back to its root cause, using
production drops against a pre-shock baseline, equipment and input shortages among producers,
price spikes and shocks. Measured facts carry their metric key, so
verify()can recompute them. - Players source goods honestly: they buy from NPCs (which moves prices) or import by caravan (finite stock, takes 3 days).
Details: Architecture · Simulation spec · Demo guide · Original plan
- Conservation: coin and goods are conserved on every simulated day. All changes go through one ledger, and no balance ever goes negative.
- Every contract is funded: escrow always equals the rewards of live contracts.
- Every contract explains itself, verifiably: the counterfactual test (same seed, no shock) produces no such contract, and forged evidence is detected.
- Bounded: there's at most one live contract per need, plus cooldowns, a global cap and a per-day cap. These hold under three simultaneous extreme shocks and against a random "chaos player".
- Deterministic: the same scenario, seed and player actions give the same SHA-256 state hash, across processes and Python hash seeds. Save files replay exactly.
Ran 180 paired runs (3 scenarios × 6 setups × 10 seeds). Each shocked run is measured against the no-shock run on the same seed. Full write-up: docs/RESULTS.md.
| Finding | Number |
|---|---|
| Invalid contracts across all runs | 0 of 180 runs |
| Iron shortage: days with >10% demand unmet, no player vs scripted player | 37 → 18 |
| Iron shortage: recovery after the mine reopens, no player vs player | 3 → 2 days |
| Mine disaster: seeds that recover (any arm) | 0-2 of 10: an economic trap contracts can't fix |
| Hungry NPC-days with guilds on vs off (iron / disaster / drought) | 6 vs 176 · 843 vs 1423 · 466 vs 573 |
| Speed | ~2 ms per simulated day for 100 NPCs (M4 MacBook Air), invariants checked every day |
Quests also have side effects, which the engine makes visible. Player profit leaves circulation, and importing finished goods takes sales from local producers.
questless export runs a scenario and any number of policy arms over many seeds and writes plain
CSV tables, with a generated codebook.md that defines every column:
uv run questless export scenarios/iron_shortage.toml \
--policy experiments/stress/policies/progressive_levy.toml --seeds 20 --out data/
uv run python experiments/analyse_export.py data/ # a worked example, standard library only| Table | One row per | For questions like |
|---|---|---|
daily.csv |
arm × seed × day | how fast coin concentrates (Gini, top 10% share), shortages, prices |
npc_days.csv |
arm × seed × day × NPC | who gets rich and why, wealth mobility, output per worker |
events.csv |
shock or changed rule | studies around a shock: what moved, and when |
runs.csv |
arm × seed | a policy's effect, paired by seed so the starting world is identical |
player_days.csv |
arm × seed × day, with --player |
what a player earns, and how a player changes the town |
transactions.csv |
ledger entry, with --transactions |
who paid whom, and why: flows of funds |
Without --player nobody can deliver a quest, so every quest expires. Arms share seeds, so every comparison is a controlled experiment: the same world with one rule
changed. Every run's state hash is in runs.csv, so anyone can rerun it and check every row.
docs/VALIDATION.md lists which real-world patterns the model reproduces,
and where it doesn't.
from questless import Simulation
sim = Simulation.from_scenario("scenarios/iron_shortage.toml")
cid = sim.run_until_contract() # day 30: the Blacksmiths Guild posts C-0001
print("\n".join(sim.explain(cid))) # the causal chain above
assert sim.verify(cid) == [] # every measured fact re-checks
c = sim.contract(cid)
sim.accept(cid)
arrives = sim.caravan("iron_ore", c.qty) # import the ore; costs coin, takes 3 days
sim.step(arrives - sim.day)
sim.deliver(cid) # goods to the smiths, escrowed reward to you
sim.save("run.json") # tiny: scenario + action log + state hashScenarios are TOML and can extend each other (scenarios/). Every rate, threshold
and premium is a designer knob, with no code changes. uv run python experiments/calibrate.py
reports each scenario against its targets across 10 seeds.
What it is: one settlement, 100 NPCs, 4 goods, rule-based agents, a Streamlit dashboard and a Python API.
What it is not: a game client, multiplayer, an LLM agent system, or production-ready engine integration.
What's next: the roadmap runs from v0.2 to v0.6:
- v0.2: relief and loan quests, futures, smuggling, a central bank, run reports, a clearer dashboard.
- v0.3: economy checks on every pull request, an AI-agent training environment, a self-writing world wiki with developer-only private pages, lessons drawn from each run, bug reports that replay themselves, courts, a tamper-evident log, anti-multiboxing safeguards, elections and a campaign-finance register.
- v0.4, megaserver foundations: fantasy and sci-fi themes, regions with simulation level of detail, a map and ecology shaped by the economy, standing orders instead of bots, player-posted quests, treaties and embargoes, spoilage, know-how, hiring, tool rental, recycling, fair days, and worker co-operatives against companies.
- v0.5, people, privacy and learning: privacy as code, a language bridge with English as the common language, steward NPCs, a skills passport and eleven skill tracks, voice check-ins and real breaks, a weekly newspaper, accessibility with one-switch play, offline basics, guilds on prescription for lonely or isolated people, and career tasters.
- v0.6, a world connected to the real one: real weather, clean-energy simulation, cross-game federation, science from play, a tabletop AR map, worlds that never die, volunteer computing on clean power, and a policy lab for real towns.
Background reading: blueprint · crypto and blockchain assessment (why there are no tokens or NFTs) · privacy and identity · real-world skills · research evidence (peer-reviewed findings behind the levers).
Known limitations:
- Producers in the same trade pause in sync when unsold stock builds up.
- Feedback loops (ore, then tools, then ore) appear in a chain but aren't drawn as loops.
- Reputation only nudges rewards.
- See docs/RESULTS.md for the experiment's caveats.
Quests from NPC needs and NPC-driven economies both exist already. Questless combines them, with contracts that must be funded, validated and explained from data.
- Economies: J. Doran & I. Parberry, Emergent Economies for Role Playing Games. L. Doucet, BazaarBot.
- Quests from NPC goals: Breault et al., Let CONAN tell you a story: Procedural quest generation (2018). Procedural Quest Generation by NPC in MMORPG, Journal of Korea Game Society.
- Background: Doran & Parberry, structure of 750+ MMORPG quests (PCG 2011). Kybartas & Bidarra, story-generation survey (IEEE TCIAIG 2017). J. Ryan, Curating Simulated Storyworlds.
Links and notes are in docs/EVALUATION.md.
Most of the roadmap is unbuilt, and it's designed to be built in the open.
- Start here: CONTRIBUTING.md explains what gets merged, the checks every pull request must pass (tests, lint, security scans), and how contributors are credited.
- Pick a task: docs/BUILD_GUIDE.md breaks every milestone into issue-sized tasks with the files to touch, tests to write and pitfalls; look for good first issue.
- Not a programmer? Economists, lawyers, teachers and political scientists can write lessons and skill tracks without touching the engine.
- Have an idea? Suggest it in Discussions or with the Suggest an idea issue template.
- Security issues: report privately, see SECURITY.md.
Created by Adrian Diez Cuadrado. The fastest way to get in touch is LinkedIn. To cite this work, see CITATION.cff.
Built with AI-assisted development (Claude); design decisions and responsibility rest with the creator.
MIT © 2026 Adrian Diez Cuadrado. You may use, modify and distribute Questless, including commercially, provided the copyright and permission notice is kept in copies or substantial portions of the software. A credit or a short note if you use it is appreciated, but not required.