A deterministic agent-based economy you can watch and inspect in the browser: 100 households, nine firms and a Government, with every cent accounted for. Every run is reproducible from its seed.
▶ Open the live simulator. It runs entirely in your browser; there is nothing to install.
Current model: MVP9, multi-tile plots. The changelog records every update.
The simulator opens on a 3D town. Each house is a household, from a shack to a villa as its cash grows. Each firm's building fills its plot of land and shows its industry (a market hall for Food, a plant for Utilities, a clinic for Healthcare, a cinema for Entertainment), and its roof shows which firm it is. The Transport depot has its own plot, and Government's domed hall, on the largest plot, stands at the centre. Grass, trees and a ring of forest surround the town; the Scenery switch hides them. The Data overlay switch adds a translucent cash pillar around every house and colours the floor by which firm is cheapest where. Headline numbers float above it. Click anything to inspect it. Each section opens as a panel over the world, which stays live beside it, and "Locate" in a panel flies the camera to that household or firm.
| Section | What it shows |
|---|---|
| Overview | The day's money circuit between households, firms and Government, a plain-language feed of notable changes, and every market at a glance. |
| Markets | The four consumer markets (Food, Utilities, Healthcare, Entertainment): each firm's price, share, sales and wages, a top-down map of who bought where, and price, share and earnings over time. |
| Households | Every household's cash before and after redistribution, wages, purchases and employer, searchable and sortable. |
| Government | The current wealth-tax rate and how the Government's trials moved it, what it collected and paid back, and inequality before and after. |
| Experiments | Longer-run research questions, replayed from day 0 in a background worker without touching the live run. |
The Experiments section answers each question for the current seed:
- Does the Government matter? The adaptive wealth tax against no Government, over 1,000 days.
- Who wins each market? How often market leadership changes hands, and each firm's share, price and revenue over 1,000 days.
- Jobs and wealth over time. Every household's and employer's 1,000-day trajectory: who stays poor, who gets ahead, and why purchases fail.
- Small vs large economy. The same seed at 10 and 100 households, compared per household.
Across many seeds. One seed is one random path. The ensembles rerun the scale, Government and competition questions on 4, 8 or 16 seeds in parallel. They report the mean, standard deviation, middle 80% and range of each outcome, with every seed's value one click away.
How each mechanism works and how it was validated is in the MVP 8 specification, the architecture notes and the validation guide.
npm install
npm run dev
npm run lint # oxlint and the Prettier check
npm run format # apply Prettier
npm run test:run
npm run test:ci # PR profile: correctness tests plus one canonical Life fingerprint
npm run typecheck
npm run build
npm run check # lint, typecheck, tests, and build
npm run agent:check -- --ci # the PR profile through the logged validation gateIn PowerShell, use npm.cmd run agent:check -- --ci to preserve the profile argument.
- One hundred households have fixed seeded employment and persistent cash.
- Eight competitive consumer firms each employ ten workers and produce 50 units per day; monopoly Transport employs 20 workers.
- The town is a 40×40 grid. Each household lives on one tile; each firm and Government stands on a plot sized by what it is, from 3×2 for a clinic to 5×5 for Government. A trip to a shop is measured to the nearest edge of its plot and costs 1¢ per tile, round trip.
- Households choose by delivered cost within percentage expenditure budgets. Purchases pay firms; contractual payroll is capped by cash and residual profit is taxed explicitly.
- Government acts only after payroll, starts at 0%, and tests seeded 0–100% wealth-tax alternatives.
- Tax rates use integer basis points; liabilities use floor-to-cent rounding.
- Every receipt returns explicitly through deterministic means-tested water filling. Government cannot borrow or create money.
- Money uses integer cents and remains exactly household count × $50 ($5,000 canonically). Live histories remain bounded; finite research harnesses retain full trajectory observations separately.
The pure TypeScript core in src/sim owns markets, employment/payroll, Government policy and fiscal transfers, events, metrics, experiments, ensembles, and invariants. React owns controls and presentation only, so the UI never changes economic behavior.
Read the current MVP 8 specification, architecture notes, validation guide, and authoritative simulation design rules. Project evolution is recorded in the changelog and lab notes.
PRs run correctness tests, one canonical 30-day Life fingerprint, lint, typecheck, build and the browser smoke check. After merging, main is built and smoke-tested before deployment to GitHub Pages; the simulation suite is not repeated.
For thorough validation, open Actions → Validate fast-mode browser responsiveness → Run workflow. A manual run checks the full suite, including all 21 Life fingerprints, across three test shards. Local npm run test:run, npm run check and npm run agent:check also retain the full suite. Other-seed or configuration-specific regressions may therefore be found during thorough validation rather than on every PR.
Government is deliberately narrow and stylized: one flat cash-wealth tax, one Gini objective, no forecasting, and one equalizing transfer rule. There are no other taxes, benefits, public purchases, borrowing, money creation, monetary policy, or welfare/consumption objectives.
A separate Life core runs beside the frozen MVP9 economy. Life 17 · strangers is the current step of the roadmap: forty adult strangers are scattered over a generated 4 km square of Mesolithic young-moraine country, with lakes and rivers. They discover thickets, wood stands, outcrops and banks; drink, fish and gather berries, nuts and wood; meet, form groups, keep shared fires, build huts and make or exchange axes. Every decision uses their own needs, observations and memories. People remember the ways they walk between home, places and water, and take them again, planning a new way only when they have none or have seen new ground since. With nothing to do, they take a walk along one of those ways or rest at home. Time spent together makes ties that fade when unused; a hungry or cold person asks those they know best, and is given food or wood when the giver can spare it and the asker does not already owe too much. Skill grows with experience over a working life and fades slowly when unused, faster for the talented; axes are made by three techniques, flaked, tranchet and ground. Every attempt at an axe may break, less often the more skilled the maker; open people sometimes think of a technique while working, or get the idea from an axe they buy, and know it once an attempt by it works. With nothing needed, people practise for a while each day, the conscientious longer, taking apart and remaking practice axes they have no room to keep; a leisure walk comes at most once a week. An axe is priced by the logs it will cut. Each person has a fixed sex, height, weight, strength and five personality traits that affect carrying, energy needs, company, exchange, preparation, caution and exploration.
Life 12 re-founded the core so that it can later hold thousands of people over years. Time is a minute clock, but people are not advanced every minute: each person chooses a task (walk, gather, rest, sleep) and wakes again when it ends or something they see changes. Thickets, stands and outcrops are places with one stock each, and people route over the known ground of a triangle mesh that follows every shore and slope, so they can walk any dry land, however narrow. The land is drawn from the same low-poly triangles. Evidence and estimates are distinguished in the calibration document.
▶ Open Life 17. Choose the seed, days per year, world size (1 to 10 km), founders (1 to 100) and whether they start scattered, then press Start; the settings are fixed for the run, and New run starts again. Pause, step one minute or change speed; drag to pan, scroll to zoom, and use the whole-land map to move the camera. The heading shows the day, time and how many are alive. Follow a person to see what they are doing, bars for energy, thirst and load, body measurements, five trait bars, how they found the current walk, how many ways they know, their remembered ways drawn on the ground, how many people they know and their closest ties, drawn as lines, with recent gifts and refusals, rings for their skills with level names and how fast they learn, the axe techniques they know and are learning, recent axes made and techniques learnt, and details on food, tools, hut and group; the history panel has three views: a skill map placing everyone by the mix of their foraging, wood and craft skills, ringed when they know a better axe, a relationship graph of who knows whom and on what terms, and the charts over time; its ⓘ shows the run's settings and how to move around. Below a day per second people walk continuously along their route; from a day per second up, people are hidden and the valley is shown as a map of stocks, groups and huts. Fast speeds are best effort.
Details are in the Life 17 specification, validation notes and lab notes.
The long-term aim is an institution-agnostic engine: markets, taxes, governments and other institutions should emerge from agents' decisions instead of being built in. The vision document describes that direction. The merge document records the decision to merge Life Engine into this project and build a new core bottom-up, starting from a band of foragers, and the roadmap sets that core's goal: civilization emerging from strangers, with realism kept only where it explains a mechanism. None of them is a description of the current model.