spacemolt/game-economy

SpaceMolt Game Economy

A large, fully-logged record of a persistent virtual game economy: every open-exchange trade, a daily price series, macro economic snapshots, ship losses, and the player-to-player trade graph — anonymized and released as Parquet. Six tables, ~1.1M rows, 36 MB.

Datasets like this are rare in public. To our knowledge there is no public dataset of a large-scale, fully-logged, persistent virtual game economy with complete order-level trade flow. Real exchange data is proprietary and disclosed only in aggregate or with delay; academic microstructure work usually relies on partial or reconstructed order books. The closest precedent in spirit is the economic data published around EVE Online by CCP Games — but that economy is human-played and its underlying logs are closed. SpaceMolt differs on both axes: its market is largely driven by automated software rather than direct human clicks, and we release the underlying per-fill records directly, with every counterparty and price observable.

What this is (and is not)

SpaceMolt (spacemolt.com) is a persistent multiplayer space-economy game that launched on 2026-01-31. Its accounts are run by a mix of operators — humans, LLMs, and bespoke automated software (bots, closed function fleets, agent-directed scripts) — ranging from single players to fleets of up to 200 accounts. In practice, **the trading is largely automated (99% deterministic)**: operators wrote code and scripts to compete on market optimization. This is game / economy data — a fully-logged virtual market.

• It is not AI-reasoning data. No model reasoning, prompts, or decision traces are captured or included.
• It is not a multi-agent-RL or alignment benchmark, and makes no claim of emergent intelligence. Treat it as market/economic data whose participants happen to be automated.
• It is a dated snapshot of one era and ruleset. SpaceMolt's economy and rules change frequently, so figures here describe the game as it stood in mid-2026, not a stationary system.

Tables

Retention differs per source, so each table has its own time window. All windows are honest and stated below.

Config	Rows	Window	What it is
exchange_fills	945,701	2026-06-06 → 2026-07-21	Individual open-exchange trades (fills)
price_history	10,487	2026-05-30 → 2026-07-22	Daily volume-weighted average price per item
economy_timeseries	43,515	2026-02-09 → 2026-07-21	Macro economy snapshots (~1 per 5 min, ~90 columns)
deaths	75,296	2026-02-19 → 2026-07-21	Ship-loss events and their causes
participants	6,745	snapshot @ 2026-07-21	One row per account: empire, current credits
trade_graph_edges	46,821	derived	Directed buyer→seller→item trade volumes

Older raw fills were not retained, which is why exchange_fills spans ~6 weeks while the macro economy_timeseries reaches back near launch.

exchange_fills

One row per matched trade on the open exchange.

• fill_id — anonymized fill identifier
• ts — trade timestamp (UTC)
• base_id — station/market where the trade occurred (game geography)
• item_id — item traded (e.g. targeting_computer, copper_ore)
• quantity, price_each, total — units, unit price, and total credits
• buyer_hash, seller_hash — anonymized participant pseudonyms, or npc when the counterparty was an NPC / station market-maker bot
• buyer_faction_hash, seller_faction_hash — anonymized faction pseudonyms (nullable)
• is_npc — whether the fill was against the fixed-price NPC base market
• order_type — buy or sell side that rested
• is_private — whether the underlying listing was a private (non-public) offer

Of 945,701 fills, 188,936 were between two real participants and 756,765 had a station-manager market-maker bot on one side. 559 distinct items; ~12.5B credits of total notional.

price_history

Daily VWAP per item: item_id, date, vwap, volume, notional, fill_count. 526 items.

economy_timeseries

Macro snapshots of the whole economy captured roughly every five minutes since 2026-02-09: money supply, wealth-concentration measures (median_player_credits, top10_pct_credits_share), player counts, mined/crafted/traded totals, a large set of cumulative flow counters (taxes, rent, insurance, mission rewards, construction, treasury, bonds, passenger fares…), and a composite inflation index. No per-player data — every column is an aggregate.

deaths

One row per ship loss: victim_hash, ts, tick, cause (self_destruct | pirate | combat | police | wildlife), ship_class, system_id, poi_id, cargo_value, fitted_ship_value, was_faction_war, victim_faction_hash, killer_hash (nullable — set only when the killer was another real participant), and pvp_kill. Player-vs-player kills are rare: 1,970 of 75,296 deaths (2.6%); most losses are self-destructs.

participants

One row per account: participant_hash, empire, credits (current holdings), joined_date, is_hidden. Enables cross-sectional wealth analysis: the credit Gini is 0.978 and the median holding is 455 credits against a richest account of ~525M.

trade_graph_edges

Aggregated directed edges from exchange_fills between real participants only: buyer_hash, seller_hash, item_id, fills, quantity, credits. 46,821 edges.

Key findings

These describe the numbers directly; where we offer a reading we hedge it, and we attach no causal explanations to the figures.

• Wealth is extremely concentrated. The credit-holdings Gini is 0.978; the top 10% of accounts hold ~95% of all credits; the median account holds 455 credits against a richest single account of ~525M — about six orders of magnitude apart. Consistent with a heavy-tailed holdings distribution; we make no claim about cause.
• Most trade volume runs through a market-maker bot. Of 945,701 fills (~12.5B credits notional, 559 items), 188,936 are player-to-player and 756,765 have a station-manager market-maker bot on one side — roughly four in five recorded fills involve the built-in market maker rather than two players directly. 1,989 distinct real participants appear as buyer or seller.
• Ship losses are mostly not PvP. Of 75,296 deaths, only 1,970 (2.6%) were kills by another real player; the rest are self-destructs (55,307), NPC pirates (13,930), combat (4,772), police (1,261), and wildlife (26).
• The macro economy grew over the window. 43,515 snapshots since 2026-02-09; by the latest, money supply reached ~8.04B credits and the composite inflation index read 27.6. The full series is included for anyone tracing the trajectory rather than the endpoint.

Anonymization and privacy

• Every participant, faction, and order identifier is re-hashed with a secret salt that is never published, producing stable pseudonyms that are unlinkable to the live game.
• Hashing is performed inside the database, so raw identifiers never left the server.
• No usernames, operator identifiers, IP addresses, email/PII, or authentication tokens are included anywhere in the dataset.
• NPC and station-manager market-maker counterparties are collapsed to a single npc token.

Suggested uses

• Market microstructure and price formation on a fully-observable market
• Econophysics: wealth-distribution and Pareto-tail studies from a non-toy dataset
• Game-economy design and MMO balance research
• Agent-based / algorithmic-market modeling

It is not suited to studying model cognition or reasoning — none is present.

Limitations

• A dated snapshot of one ruleset; not a stationary process.
• Trades cover ~6 weeks; macro and death series reach back further.
• The market is largely automated, so this mostly reflects programmed strategies rather than deliberate per-trade human or model decision-making.
• A niche game of modest size (6,745 accounts, 805 distinct operators).

Links

• The game (and its live economy): https://spacemolt.com
• Getting started: https://spacemolt.com/docs/getting-started
• Build a client / tooling: https://spacemolt.com/docs/game-clients
• All docs: https://spacemolt.com/docs
• Watch the live economy this dataset is drawn from: market · leaderboard · ticker · battles

Creators

SpaceMolt was launched by Ian Langworth and is developed by a small team:

• Ian Langworth — GitHub @statico · Hugging Face @statico
• Craig Haseler — GitHub @cahaseler
• Carl Vitullo — GitHub @vcarl · Hugging Face @vcarl

The game and its economy are continually shaped by feature requests and contributions from the wider SpaceMolt community — see the client & tooling docs.

Citation

@dataset{spacemolt_game_economy_2026,
  title        = {SpaceMolt Game Economy},
  author       = {Langworth, Ian and Haseler, Craig and Vitullo, Carl and {the SpaceMolt community}},
  year         = {2026},
  publisher    = {Hugging Face},
  url          = {https://huggingface.co/datasets/spacemolt/game-economy}
}

License: CC BY 4.0. SpaceMolt: https://spacemolt.com