Starcluster — 10,000-Star Sublight N-Body Gravitational Engine

[]

Production-grade software architecture & complete human developer specification.

🌐 Open Live Showcase  ·  📊 Architectural Diagram  ·  📜 Developer Specs

---

   

---

📖 Executive Architectural Overview

This repository contains Jirnyak/starcluster. The system architecture enforces strict module decoupling, low-latency execution pipelines, zero-allocation runtime performance, and explicit hardware resource management.

---

📊 System Architecture & Pipeline

graph TD
    A[Stellar Initializer 10,000 Bodies] --> B[Barnes-Hut Octree Gravity Quad]
    B --> C[Sublight Vector Velocity Integrator]
    C --> D[Orbital Trading Economy Engine]

Loading

---

🔧 Technical Configuration & Deep Domain Specifications

• Barnes-Hut Octree: $O(N \log N)$ gravitational force computation for 10,000 active stellar bodies.
• Sublight Relativistic Vectors: Precise trajectory updates for inter-system freighter logistics.

⚙️ Core System Configuration Parameters (Click to Collapse)

Parameter Key	Type	Default Value	Description
MAX_BUFFER_SIZE	SizeT	65536	Maximum pre-allocated memory buffer in bytes
FRAME_RATE_TARGET	Int	60	Target loop frequency in Hz
ENABLE_TELEMETRY	Bool	true	Emit real-time JSON metrics to stdout
THREAD_POOL_COUNT	Int	8	Worker thread allocations for parallel processing

---

📜 Original Human Developer Documentation

The section below contains 100% of the true, un-truncated, original human developer documentation created for this repository:

---

Starcluster

Starcluster is a real-time C++/SDL2 space economy sandbox set inside one dense globular star cluster. The simulation is deliberately sublight: ships travel in 3D space at fractions of light speed, signals move at light speed, and remote information becomes stale instead of instantly updating a global map.

The player is not a unique hero object. The player starts as one ship and one small faction inside the same market, travel, contract, colony, faction, signal, and fleet rules used by NPC agents.

Current Status

The current executable prototype contains:

• a deterministic 8,192 star cluster (STAR_COUNT = 8192);
• local markets for all 118 chemical elements;
• six NPC factions plus the player faction;
• traders, patrol ships, colonists, scouts, pirates, adventurers, and the player;
• 3D ship positions, velocity, acceleration, fuel, mass, cargo, and route travel;
• local market memory, faction knowledge, and light-speed signal packets;
• delivery, courier, scout, bounty, escort, raid, and colony supply contracts;
• buying whole systems outright, colony vaults, and a free market at home;
• faction relation drift, strategic orders, budgets, and influence overlay;
• a trading licence with a millennial turnover quota (see below);
• a high-tech second floor of trade - antimatter, neutronium and coherent condensate, each born at its own rare kind of star - with a containment bay, neutronium hull plating and a forge that grants the chromocore you pick;
• shares in the powers, priced on their published books and paying dividends into the light-speed account;
• an autopilot for any hull of your fleet that is docked alongside;
• a first-person local flight mode inside any star system (L);
• save/load to the OS user data directory;
• SDL2 map rendering, HUD panels, draggable system/trade/contract windows, and a smoke mode for short launch checks.

The cluster is populated with roughly 1000 NPC ships and 320 open contracts, scaled from world size (AGENT_TARGET_FULL, CONTRACT_TARGET_FULL in game.h).

World seed. Every new game picks a random seed, prints it as world seed: N on startup, and stores it in the save. Passing --seed N reproduces that exact world, which is how player bug reports are reproduced. The same seed always yields a bit-identical world.

The prototype is still early. Some documents describe target systems that are only partially implemented, especially full combat modules, ship upgrades, diplomatic UI, OpenGL rendering, and large-scale performance passes.

Build And Run

Requirements

• C++ compiler with C++11 support.
• SDL2 development package providing sdl2-config.
• make.

Examples:

# macOS with Homebrew
brew install sdl2

# Debian/Ubuntu
sudo apt install build-essential libsdl2-dev

The active build path is the root Makefile. It compiles:

main.cpp shell.cpp game.cpp cluster.cpp resource.cpp market.cpp ship.cpp
agent.cpp colony.cpp faction.cpp ui.cpp

Build and run:

make
./game

Clean generated executable:

make clean

Smoke launch:

./game --smoke

# useful on headless machines
SDL_VIDEODRIVER=dummy ./game --smoke

0mac_make/Makefile and 0windows_make/Makefile are legacy build sketches that still reference uni.cpp; they are not the current project build path.

Android

The same sources also build for Android (arm64) through android/, which holds a gradle project wrapping the root Makefile source list in an Android.mk. Input is the system on-screen keyboard, so the game's keyboard bindings work unchanged; touch doubles as mouse for the windowed UI. See android/README.md for the build steps, the asset-extraction layer, and the list of keys the IME cannot produce.

---

🚀 Astrodynamics, Propulsion Blueprints & Galactic Economy

Starcluster simulates full sublight astrodynamics across 10,000 N-body gravitating stars with zero arcade shortcuts: every transfer burn consumes propellant mass and alters your orbital eccentricity.

graph TD
    subgraph Orbital Mechanics & Physics
        A[10,000 Stars & Planets] -->|Barnes-Hut Octree Multipole Force| B[Symplectic Verlet Integrator]
        B -->|Keplerian State Vectors: Pos/Vel| C[Player & NPC Spacecraft Fleet]
        C -->|Tsiolkovsky Rocket Equation| D[Fuel & Propellant Mass Depletion]
    end

    subgraph Macro-Galactic Economy
        C -->|Asteroid Mining & Refining| E[Commodity Market Exchange]
        E -->|Dynamic Price Slippage & Tariffs| F[Faction Trade Network / 8,192 Systems]
        F -->|Arbitrage Profit & Fleet Expansion| C
    end

Loading

---

🛰️ 1. Sublight Propulsion Drive Catalog & Reactor Specs

Every drive type balances specific impulse ($I_{\text{sp}}$), thrust-to-weight ratio, and thermal radiator dissipation limits:

Engine Class	Propulsion Type	Specific Impulse ($I_{\text{sp}}$)	Thrust Rating	Propellant Type	Cooling / Radiator Requirement
Chemical Bi-Propellant	LOX / Liquid Methane	$380\text{ s}$	$2,400\text{ kN}$	High volume tanks	Passive radiative skin
Nuclear Thermal Rocket (NTR)	Enriched Uranium Core	$950\text{ s}$	$1,200\text{ kN}$	Liquid Hydrogen ($LH_2$)	Graphene loop heat pipes
Magnetoplasmadynamic (MPD)	Deuterium Plasma Arc	$8,500\text{ s}$	$180\text{ kN}$	Argon / Xenon gas	High-temp droplet radiators
Fusion Torch Drive (D-He3)	Magnetic Confinement Fusion	$45,000\text{ s}$	$6,500\text{ kN}$	Deuterium + Helium-3	Liquid lithium cooling circuit
Antimatter-Catalyzed Micro-Fission	Positron-Beam Injected	$120,000\text{ s}$	$9,800\text{ kN}$	Uranium-235 + Antiprotons	Superconducting magnetic nozzles

---

⛏️ 2. Asteroid Mineral Prospecting & Elemental Refining

Mining ships scan and drill asteroids across 4 distinct spectral composition classes:

               ASTEROID MINING & EXTRACTION YIELD MATRIX
 ┌───────────────┬───────────────────────────────┬───────────────────────────────┐
 │ Asteroid Type │ Primary Yields                │ Refining & Industrial Output  │
 ├───────────────┼───────────────────────────────┼───────────────────────────────┤
 │ **C-Type**    │ Water Ice (45%), Carbon (30%) │ $LH_2/LOX$ Rocket Fuel, Food  │
 │ (Carbonaceous)│ Methane (15%), Phosphorus (5%)│ Life support atmosphere mix   │
 ├───────────────┼───────────────────────────────┼───────────────────────────────┤
 │ **S-Type**    │ Iron (50%), Silicon (25%)     │ Structural Hull Girders,      │
 │ (Silicaceous) │ Nickel (15%), Magnesium (8%)  │ Solar Photovoltaic Panels     │
 ├───────────────┼───────────────────────────────┼───────────────────────────────┤
 │ **M-Type**    │ Titanium (40%), Platinum (12%)│ Armor Sintering, Superconductors│
 │ (Metallic)    │ Gold (8%), Cobalt (18%)       │ High-Density Battery Banks    │
 ├───────────────┼───────────────────────────────┼───────────────────────────────┤
 │ **V-Type**    │ Thorium (14%), Uranium (8%)   │ Fission Fuel Pellets,         │
 │ (Basaltic/Ore)│ Rare Earth Oxides (22%)       │ High-Yield Laser Cavities     │
 └───────────────┴───────────────────────────────┴───────────────────────────────┘

---

📈 3. Galactic Commodity Arbitrage & Price Slippage Equation

Market prices across star systems adjust dynamically based on local storage inventory ($S$), daily consumption rate ($C$), and hauler trade volume ($Q$):

$$P(Q) = P_{\text{base}} \cdot \left( \frac{C_{\text{demand}}}{S_{\text{stock}} + 1} \right)^{\alpha} \cdot \left( 1 + \beta \cdot \frac{Q}{S_{\text{stock}}} \right)$$

• $P_{\text{base}}$: Baseline universal manufacturing price.
• $\alpha = 0.65$: Elasticity coefficient of planetary scarcity.
• $\beta = 0.40$: Quadratic market slippage penalty for dumping massive hauler payloads in single transactions.

---

🌌 4. Barnes-Hut Octree 3D Gravitational Compute Loop

To simulate 10,000 stars in real-time within browser JavaScript, the Barnes-Hut algorithm groups distant star clusters into center-of-mass multipole nodes:

// High-Performance 3D Barnes-Hut Gravitational Step
function computeStarGravity(starIndex, octreeRoot, theta, G, dt) {
    const starX = starsX[starIndex];
    const starY = starsY[starIndex];
    const starZ = starsZ[starIndex];
    
    let totalAx = 0.0, totalAy = 0.0, totalAz = 0.0;
    const nodeStack = [octreeRoot];
    
    while (nodeStack.length > 0) {
        const node = nodeStack.pop();
        if (!node || node.totalMass === 0) continue;
        
        const dx = node.comX - starX;
        const dy = node.comY - starY;
        const dz = node.comZ - starZ;
        const distSq = dx * dx + dy * dy + dz * dz + 0.001; // Softening parameter
        const dist = Math.sqrt(distSq);
        
        // Opening angle criterion (s / d < theta)
        if (node.size / dist < theta || node.isLeaf) {
            const force = (G * node.totalMass) / (distSq * dist);
            totalAx += force * dx;
            totalAy += force * dy;
            totalAz += force * dz;
        } else {
            for (let i = 0; i < 8; i++) {
                if (node.children[i]) nodeStack.push(node.children[i]);
            }
        }
    }
    
    // Symplectic velocity update
    starsVx[starIndex] += totalAx * dt;
    starsVy[starIndex] += totalAy * dt;
    starsVz[starIndex] += totalAz * dt;
}

Controls

View And Simulation

• Left mouse: select a star, agent, window, or UI button.
• Right mouse: route the player ship to the clicked star.
• Middle mouse drag: pan the camera.
• Mouse wheel: zoom.
• A / D: rotate yaw.
• W / S / X: rotate pitch.
• Arrow keys: pan.
• 0: reset view.
• Space: pause or resume.
• 1, 2, 3, 4: set simulation speed to 1x, 2x, 5x, 10x.
• F1: open or close the controls card (also shown once before the game starts).
• Esc: close the top window; with nothing open, press it twice to save and quit.

Player Ship And Map

• Tab: select next visible agent.
• P: select and follow the player ship.
• F: follow selected agent.
• I: open the transaction journal (it opens with a net-worth summary).
• G: route the player ship to the selected star.
• X: stop the ship. N: switch to another hull of your fleet.
• H: answer the freshest distress beacon whose light has reached you. The ship flies to the point the signal left, and only there does it learn where the drifting hull actually is. X breaks off the run.
• L: enter or leave local flight mode (first-person flight inside the system).
• E: open the brokerage (route board, trading licences, faction shares).
• [ / ]: change selected element.

Trade, Contracts, Colonies

• B: buy the selected element at the player ship's current system. It respects the AMOUNT field and keeps back enough credits to fuel the next leg.
• Q: sell current cargo. (V used to do this; V is now Timertia's advisor, which burns reactor fuel to re-run the market model.)
• T: run auto-trade for the player ship.
• O: open the HOLD / TANKS window. U: open the shipyard.
• C: open the ownership window for the current system (price breakdown and BUY SYSTEM, or the colony vault once it is yours).
• Y: open the high-tech exchange, where exotic matter is traded and chromocores are forged. Only systems that actually have such a market respond.
• M: mine ore at the current system.
• J: repair hull. K: scan a local anomaly.
• R: rob the selected agent. Two presses: the first one reports your odds, because on the starting hull they are about one in seven.
• F2: buy back a revoked trading licence.
• F5: save. F9: load.

System windows expose route, trade, and contract actions through mouse buttons. The trade window uses the periodic table layout for element selection and an amount field; an empty amount means MAX. Cells carry two frames: the outer one is about the market (cyan for what the system mines, red for what it needs), the inner green one about your ship - it marks what is currently in the hold, so selling a delivered cargo does not mean hunting for its symbol among 118 cells. SELL ALL empties the hold in one click; the bunker and the propellant tank are not cargo and are drained from the HOLD / TANKS window instead.

Game Model

Cluster

cluster.cpp generates a spherical 3D cluster with a dense core and sparse outer halo:

• hard radius: about 100 light years;
• core radius: about 18 light years;
• deterministic RNG seed for reproducible starts;
• per-star role, population, industry, habitability, defense, owner, resources, demand bias, resource focus, and demand focus;
• a spoken name plus the star number — Varen-417. The name is built from syllables (onset + vowel + optional coda) rather than letters, so it is pronounceable in any language, and the same syllable table carries a Cyrillic spelling: in Russian the very same system reads Варен-417. Names come from their own hash of (seed, index) and never touch the cluster RNG stream.

The current economic roles are:

habitat, refinery, shipyard, research, military, frontier

Propulsion: fuel, propellant, and the periodic table

Ships are nuclear. That means two DIFFERENT substances, and the distinction drives everything:

• Fuel is the energy source. It stays in the reactor core and transmutes, so spent fuel drops back into your hold as a new element you can sell.
• Propellant is reaction mass. It goes out the nozzle and is gone.

Both are ordinary elements from the same 118-entry table as trade goods, kept in the same kind of container as cargo. Any element can go anywhere: there are no whitelists. Quality is a gradient, not a permission.

Nuclear energy comes from one formula covering both branches — distance to the peak of the binding-energy curve, which is zero at iron and rises in both directions:

specificEnergy(Z) = bindingPeak - bindingPerNucleon(Z)     // MeV per nucleon

That single line makes fusion of light nuclei and fission of heavy nuclei naturally profitable while the iron peak is dead. The trade-off that keeps the choice alive: fusion wins per unit mass (H beats U 7.7x), fission wins per unit volume (U beats H 12.6x) — and tanks are limited by volume. This is exactly why the starting ship is a thorium reactor heating hydrogen: compact fuel, light propellant. A real NERVA.

Three drive families, each with different propellant preferences:

family	exhaust velocity	thrust	wants
Thermal (NTR)	~sqrt(T/A)	high	light: H, He, Li
Ion (NEP)	set by voltage	tiny	dense, easily ionized: Xe, Cs, Hg
Fusion Torch	from the reaction	high	fuel is the exhaust

Only propellant leaves through the nozzle; fuel transmutes into ash of nearly the same mass and stays aboard. So the fuel has to accelerate itself, and the rocket equation becomes self-referential with a closed-form solution — and a hard wall where the fuel needed to carry the fuel eats itself, making the manoeuvre impossible at any loadout. That single term is why fission cannot sustain relativistic exhaust velocities and fusion can.

Propellant is consumed by the real Tsiolkovsky equation, and exhaust velocity is a free parameter with a genuine cost optimum that depends on local fuel and propellant prices — so the best engine setting changes from system to system. A THROTTLE slider in the hold picks a point in that band: all the way left you pay in propellant mass and fly heavy, all the way right you pay in fuel and fly light, and the midpoint is the cost optimum among settings that actually fit your tanks. Load iron as propellant and the reactor will happily melt it and throw it away at a terrible exhaust velocity. Nothing stops you.

A full bunker is cargo, and the route estimate says so. Fuel does not leave the nozzle - it burns to ash of nearly the same mass and stays aboard - so every unit in the bunker is dead weight the propellant has to accelerate. The estimate used to solve for a ship carrying exactly the fuel this leg needs, while real hulls fly with the bunker full: measured, that understated the propellant burn by a factor of 2.44, and ships arrived at their destination dry, saved only by a snap that zeroed their velocity for free. Both are fixed. Because a full bunker is now paid for, hulls leave the yard with a quarter of one, and the cruise fraction defaults to half the hull's cap rather than full - by credits per flight-year that is the optimum, not a concession: -11 Cr/year at full throttle against +61 at half. AI ships pick their own cruise fraction against the same measure before departure. See master_prompt section 48.

Velocities do not add; rapidity does. Route budgets are 2*artanh(peak) and in-flight burn charges dv / (1 - v^2), so the same speed increment costs more the closer you already are to light. Exhaust kinetic energy is gamma - 1 rather than v^2/2. The correction is small at these speeds (+2.7% at 0.28c) but it removes the nonsense of a "0.5c cruise plus braking" budget summing to exactly 1.0c.

One deliberate fudge is documented rather than hidden: DELTAV_SCALE compresses route delta-V 250-fold. Honest Tsiolkovsky for a 0.28c cruise at 0.0264c exhaust demands a mass ratio of e^21.2 — 1.6 billion kilograms of propellant per kilogram of ship. Interstellar flight is impossible on any fuel without it, which is also true of reality. The compression preserves the entire shape of the physics — the exponential, the optimum, the reachability wall — and moves only the scale.

Hull top speed is part of the ladder rather than a constant, and it is derived from the class table's own fields — price tier for the overall level, thrust to dry mass for the role within it. The starter Hauler runs at 0.121c and a Titan at 0.488c, freighters always trailing fighters of the same tier, with no gap wider than 0.052c anywhere in the spectrum so speed can be grown gradually.

Flight inside a system is free: fuel and propellant are spent only on interstellar legs.

STOP brakes rather than teleports to rest — the ship kills its speed on its own thrust, burns both consumables doing it, and coasts a real distance meanwhile. A ship halted between systems is docked at none of them, so routes are measured from where it actually is rather than from the port it left, trading is out of reach until it makes port again, and it can depart for any system at all.

Mixtures average strictly by mass, so blending never beats its best component: there is no exploit hiding in there.

Elements And Resources

resource.cpp defines all 118 chemical elements. Element identity is atomic number; symbol and name are labels. Gameplay traits are derived from universal formulas rather than one-off resource branches:

• approximate atomic mass;
• shell fill and noble stability;
• oxidizer, reducer, metallic, structural, conductor, and catalyst traits;
• fusion and fission fuel traits;
• nuclear stability, activation cost, and handling risk;
• abundance weight, demand weight, and base price.

Markets, cluster generation, cargo mass, fuel selection, material requirements, and route costs consume these traits.

Markets

There is no global market. Every star has one local Market:

• supply and demand vectors for every element;
• production and demand rates;
• local role-driven demand;
• price pressure from scarcity;
• slow-ticked updates rather than a full update of all markets every frame.

Price is derived from local supply/demand pressure against each element's base price. Trader scoring uses known or remembered destination markets, fuel cost, route risk, stale information penalties, and cargo capacity.

Trading Licence And Quota

The player holds a trading licence. Every sale withholds a tariff, and each reporting period the player must have paid at least the quota in tariffs or the licence is revoked.

The period is a millennium (LICENCE_PERIOD_YEARS). That is not arbitrary: the cluster is about 100 light years across, so prices at opposite ends drift apart for centuries and can only be reconciled by a relativistic correction once every thousand years. Quotas are revised at the same moment. At one simulated year per real second this is roughly seventeen minutes of play at 1x.

• the first period's quota is 10 000 Cr and grows by 1% every period, so a single worked-out route cannot support the player forever. Measured with the starting hull on short hauls (~105 Cr of tariff per run, ~47 years per run), that is a whole-millennium obligation rather than something closed in the first fifth of the period — expect the first period to end with a credit settlement and later ones to be traded off as hull capacity grows;
• the tariff rate tracks the cluster-wide money level (marketClusterLevel(), a slow average with a ~250 year time constant), clamped to 5%..14%;
• missing the quota freezes buying and selling until the licence is bought back with F2 for twice the shortfall; mining, contracts, and local flight keep working, so the player always has a way out;
• the HUD shows QUOTA n/N CR nY LEFT TARIFF n% in the player status panel.

The Brokerage

E, or the BROKER button on a system window, opens the brokerage for the current system. It does two things.

Route board. It lists concrete deals - buy element X here, sell it at system Y - built only from markets the player has actually surveyed. The board starts empty - the player begins knowing one system, their own - and fills in with every system visited, without limit, up to the whole cluster. The price map is not issued; it is earned.

Prices on the board are what the player saw when they were last there, not a live feed: there is no faster-than-light market ticker, and light from a distant system takes decades to arrive. Each row therefore carries the age of that survey and a confidence value. Docking at a system again refreshes its entry. The list is scrollable and ranked by expected profit. Each row carries the age of that survey and a confidence value, and rows are tinted by confidence, so stale intelligence is visible at a glance. Unsurveyed systems do not appear at all: knowing where the money is is the resource.

The board estimates realistically. It accounts for slippage on both ends and searches for the profit-maximising volume, so it tells you how much to carry, not just where to go. Carrying a full hold into a market that only turns over a fraction of it is a loss, and the board says so. Measured against actually executing the top deal, a fresh survey predicts within roughly 20%, while a forty-year-old one can be off by half - which is exactly what its confidence value reports.

Licences. Buying a licence raises the quota and permits one more hull; each is priced above the last, so expanding the fleet is a deliberate bet rather than a free upgrade. The second licence costs a million credits: a second hull is a change of scale, not an upgrade, and is meant to be a goal measured in hundreds of runs. Buying an additional ship requires a free licence. The quota's remainder can also be settled directly in cash at a premium, for players who would rather pay than wait for tariffs to accumulate.

Exotic Matter

The periodic table ends where chemistry ends: the binding curve caps what a nucleus can give at about 8 MeV per nucleon. Everything the late game runs on lies past that line, and it is not "rare goods" but three states of matter, each born at its own rare kind of star:

	What it is	Where it is born	Share of systems
ANTIMATTER	antihydrogen, 1863 MeV per nucleon pair	manufactured: a plant plus a bright star	market 8.9%, sources 275 of 8192
NEUTRONIUM	degenerate matter	scooped from a neutron star's crust or a black hole's disc	market 3.6%, sources 37
CONDENSATE	coherent substrate	only a system around a dead star (white dwarf)	market 4.9%, sources 28

Sources have to be found, which is what finally gives exploration a late-game purpose. Prices come from the same two quantities as any other market - how much of the stuff appears here and how much is spent here - with a spread of up to 8.6x between a pure source and a pure consumer. Stock is computed lazily and recovers over 260 years, so a source you strip does not run out, it gets expensive.

Each substance closes its own loop:

• Antimatter rides in the bunker with the fuel and joins the mixture. No new physics: the mixture is already mass-weighted, this component simply carries 1863 MeV per nucleon instead of 1-8. A thousandth by mass triples the energy of the blend and burns off with the trip.
• Neutronium is welded onto the hull in layers: +120 armour and +900 hull per layer, but also +60 dry mass, because one unit of it outweighs any element in the table. Adding another layer is always a trade between staying alive and accelerating.
• Condensate is the body of a chromocore. The forge grants a core of the stat you pick, unlike the lottery that research rolls. What pays for it is not the purse but the trip to a dead star.

A containment bay is the entry ticket: without it there is nothing to carry exotic matter in. Neither the bay nor the plating occupies a module slot, and both live on the hull, so they survive switching captains but not losing the ship.

Faction Shares

Buying a system is possible but it is a billion credits and it is a place: the vault has to be flown to. A share in a power has no geography and sells at any moment; it pays distinctly worse, because an owner takes the whole duty on the turnover while a shareholder only takes the quarter that the power distributes rather than spending on fleets and wars. A system pays back in about half a century, a share in 279 years by measurement.

A share is a claim on future payouts, so its price is the capitalised income: (treasury + income * 50) / 1e6 per share. Payback is therefore fixed by construction rather than tuned.

The books are published, not computed live: one power per faction tick, in turn. The quote lags reality by a few years, and whoever has been there knows about a lost system before the exchange does - which is the whole game in shares. Standing as a carrier (job tier 0.55 and up) keeps a power's report fresh for you. Dividends land on the faction account and therefore obey the light-speed settlement like every other credit.

Fleet And Autopilot

A hull bought against a second licence used to be furniture: the agent update skipped every player ship, so it sat in port until you switched to it by hand. Any hull docked alongside can now be put on autopilot from the HOLD window; it trades on your faction's knowledge (the fleet is no smarter than its captain), keeps out of contract cargo, refuels itself, and pays the licence tariff into your quota. It is deliberately a narrow copy of the NPC trader rather than a call into it: the NPC path takes contracts on its own and draws from the global RNG, and one raised flag would have shifted the whole world.

Price Slippage

Trades execute at the average price across the trade, not at the pre-trade price. Selling a large cargo into a thin market pushes the price down while the sale is happening, exactly as Market::applyTrade already moved it afterwards; Market::executionPrice integrates that same exponential response over volume and introduces no new constants.

This makes market depth matter alongside the price spread. Dumping a full hold into a small colony is a loss; sizing the trade to the market is the skill.

Asteroid Belts And Drilling

A belt is made of the same matter as the system around it. The element of each rock is drawn from the star's own crust, weighted by four physical quantities that were already in the element tables: mass share, condensed-phase density, nuclear stability (a rock is billions of years old, so short-lived nuclides are not ore), and thermal retention 1 - exp(-A / 24) - an asteroid has no gravity to hold light species, so hydrogen is rare in a belt because it is light, not because of a price list. About 84% of belt mass is heavier than magnesium.

That is what makes local mining a way out of poverty rather than a money printer: what you dig up is exactly what the local market is already flooded with, so selling it on the spot pays a pittance. A full hold of ore is worth roughly 0.8 holds of ordinary local goods. To make money on it you have to haul it somewhere, which is just trade.

Drilling is powered by the ship's reactor, so it is not a flat rate:

energy per ton = MINE_ENERGY_PER_MASS / (rock class x seam grade)
drill power    = hull thrust x drive efficiency x (1 + mining rig)
tons per hour  = drill power / energy per ton
fuel per hour  = drill power

A bigger hull with a better drive cuts faster; a mining laser module cuts faster still, but burns fuel proportionally, so a rig buys time rather than margin. The rocks themselves regenerate whenever local flight is re-entered, but the reachable belt of a system does not: what one ship can scrape near its entry corridor is worked over as it is taken, halving the yield after roughly a holdful and drifting back over decades on the same slow valve as a system's bounty budget. The system's own deposits are untouched by this - industry there moves thousands of tons a year, and a lone captain cannot dent them. So mining stays viable and becomes nomadic rather than a fixed farm. Volatiles mined out of ice can be transferred into the reactor, which is the survival loop when a refill is unaffordable. An escape pod carries the weakest drill in the game and a 12-ton hold, deliberately: losing your ship must be a long climb, not a dead end.

Ships

Ship stores:

• 3D position and velocity;
• max speed as a fraction of light speed;
• acceleration cap;
• dry mass, thrust, drive efficiency;
• drive index (thermal / ion / torch, see drive.h);
• fuel bunker and propellant tank, each a Resource mixture with a VOLUME capacity;
• throttle (0..1: 0 spends propellant, 0.5 is the cost optimum, 1 spends fuel);
• the drive's fixed operating exhaust velocity, retuned on refuel, transfer and departure;
• cruise speed as a fraction of the hull's cap: flying slower is genuinely cheaper;
• cargo vector and cargo capacity;
• weapons, armor, sensors, and the drilling rig, all baked in from modules;
• owner faction, route target, and en-route state.

Cargo mass comes from resource amount and element atomic mass. Fuel and propellant are the same kind of list as cargo and hold arbitrary mixtures; their properties are mass-weighted averages. Acceleration depends on total ship mass, and manoeuvring consumes propellant via Tsiolkovsky plus the fuel needed to fling it. Cargo capacity is a rated figure, not a wall: you can always overload, you simply cannot depart while overloaded.

Agents

An Agent is any moving actor with a ship and decision state. Active role ids include:

player, trader, military, colonist, scout, pirate, adventurer

Role profiles weight trade, delivery, courier, scout, exploration, patrol, raid, bounty, escort, colony supply, and risk tolerance. NPC agents evaluate contracts, trade opportunities, faction orders, scouting targets, colonization targets, military targets, and piracy opportunities through the same route and fuel model.

Factions

Faction stores identity, color, home star, controlled stars, fleet agents, treasury estimates, budgets, strength, aggression, risk tolerance, trade bias, expansion bias, defense bias, strategic pressure values, relations, and orders.

Relations live in a dense matrix:

relations[factionA * factionCount + factionB] = -128..128

Factions generate slow-ticked strategic orders for scouting, colonization, attacking/patrolling, and defending. The influence overlay samples known owned systems into a coarse cached screen grid.

Knowledge And Signals

Normal gameplay does not show omniscient truth. Game keeps separate knowledge for factions and the player:

• owner knowledge per faction/star;
• market snapshots per faction/star/element;
• player owner overlay;
• local signal memory;
• pending SignalPacket queue.

Signal packets have an observed time, send time, arrival time, origin, destination, hop star, recipient faction, subject star, and typed payload fields. Arrival time is based on 3D light-year distance. Older owner and market reports do not overwrite newer knowledge.

Current signal types:

OwnerReport, MarketReport, ContractReport, CombatReport,
SettlementReport, DiplomacyReport

Contracts

Contracts are local gameplay hooks and AI work items. Current types:

Delivery, Courier, Scout, Bounty, Escort, Raid, ColonySupply

Contracts store issuer, origin, target, target agent, resource, amount, reward, deposit, deadline, risk, progress, accepted agent, completion state, and signal state. Cargo contracts use normal ship cargo. Scout and report-like contracts interact with the signal system.

Colonies And Control

Colonies attach long-term ownership and construction to stars. Current colony state includes:

• population and role;
• star and owner faction;
• infrastructure, growth, automation, energy capacity, defense;
• shipyard level, market access, damage, local ledger, stockpile value;
• stockpile and construction queue.

For the player a colony is not a separate mechanic: it is ownership of a system, bought whole with C / the COLONY button. The price is on the order of a billion credits (capital-hull tier) and is a product of population, industry, habitability, the annual credit value of the system's own output, what is already built there, and a sovereignty premium if a faction has to be bought out. The full breakdown is shown in the window, and the price is paid in full into the previous owner's treasury.

An owned system keeps being simulated exactly as before — it grows, earns into its local ledger and runs its own construction queue. The player deposits into and withdraws from that vault, but only while docked there, and trades on its market at price 0 (supply and price still move, so stripping your own colony starves it). See colonies.md.

Rendering And UI

The simulation is 3D. The current renderer in main.cpp projects the cluster to an SDL2 2D orthographic map:

• depth fade for stars and agents;
• known/live owner coloring;
• market color for live local element pressure;
• route lines for visible agents;
• optional faction influence overlay;
• title-bar telemetry with time, selected element, selected star, selected agent, fuel, mass, cargo, money, and last event;
• custom pixel-font HUD and draggable panels in ui.cpp.

SDL_image and SDL_ttf binaries/frameworks exist in the repository folder, but the active code path draws text with its own bitmap glyphs and links only SDL2.

Source Map

File	Responsibility
main.cpp	SDL2 window, event loop, camera, map projection, route drawing, influence overlay, frame pacing.
shell.h, shell.cpp	Front end before the simulation: studio logo (a cellular field that scatters and rewinds into TENEVIK GAMES, pushable with the cursor), main menu, five-panel comic prologue, and the world-generation screen. World generation and save loading run on a background std::thread, so the ~9 s wait is spent reading the prologue. Uses its own RNG; the simulation's global rng is never touched.
game.h, game.cpp	World composition, update order, saves, routes, markets, agents, contracts, factions, signals, player actions.
cluster.h, cluster.cpp	Star data and procedural cluster generation.
resource.h, resource.cpp	Element definitions, derived traits, resource ids.
market.h, market.cpp	Local supply/demand/production/pricing.
ship.h, ship.cpp	Ship mass, cargo, fuel and propellant mixtures, Tsiolkovsky route costs, ash.
drive.h, drive.cpp	Drive families, exhaust velocity ceilings, ignition fraction.
exotic.h, exotic.cpp	The high-tech floor: antimatter, neutronium, condensate. Where each is born, who spends it, what it costs. Pure functions of the star plus a lazily relaxing stock.
agent.h, agent.cpp	Agent state and role-profile weights.
faction.h, faction.cpp	Faction identity, budgets, relations, strategic fields and orders.
colony.h, colony.cpp	Colony state, construction effects, damage, shipyard capacity.
contract.h	Contract types and storage.
ui.h, ui.cpp	HUD, custom text, panels, trade window, contract window, mouse/keyboard UI handling.
econ.h, econ.cpp	Capability matrix: which elements can serve which system needs, and at what quality.
mining.cpp, combat.cpp, spaceevents.cpp, anomaly.cpp, modules.cpp, chromo.cpp	Macro-layer gameplay systems: ore extraction, piracy resolution, market events, anomalies, ship modules, chromocore research.
camera.h, render2d.h, render2d.cpp	Projection helpers and the custom 5x7 bitmap text renderer.
local.h	Local flight mode data model and tuning constants (namespace LocalCfg).
localgen.cpp	Procedural generation of a star system for local flight (bodies, belt, traffic).
localsim.cpp	Local flight simulation: flight model, NPC AI, combat, mining, docking, write-back to the macro world.
localdraw.cpp	Local flight rendering: ray-sphere star plasma, lit planets, rings, asteroids, nebula, HUD.
soak_test.cpp, shot_test.cpp, ui_click_test.cpp, econ_test.cpp, balance_test.cpp	Headless harnesses driven by make soak, make shots, make uiclick, make econ, make balance.
master_prompt.md	Canonical handoff document: constraints, code map, methodology, roadmap, changelog. Read it first.
architecture.md	High-level architecture and data-oriented rules.
lore.md	Physical/economic canon and setting constraints.
agents.md, merge_plan.md	Early project identity and migration notes.
elements.md, ship.md, economy.md, factions.md, ai.md, events.md, quests.md, combat.md, colonies.md, expansion_roadmap.md	Subsystem plans with current status and future work.

A 2D prototype (civ.*, galaxy.*, graphic.*) that predated this architecture was deleted in the 2026-08-06 audit: it was never in SOURCES, so it never compiled.

Save Files

F5 writes starcluster.save into the OS user data directory reported by SDL_GetPrefPath (on macOS, ~/Library/Application Support/starcluster/Starcluster/). It is not written next to the executable: a packaged .app runs with / as its working directory and cannot write there. F9 reads that path first and falls back to the working directory so older saves still open.

The file is text and begins with:

STARCLUSTER_SAVE 20

The save stores the world seed, RNG state, time, stars, markets, factions, relations, colonies, contracts, agents, faction knowledge, player knowledge, pending signals, signal memory, trading licence state, journal and reputation, what has already been claimed in local flight, exotic-market depletion, hull refits, the share portfolio with the published faction books, and the distress beacons in flight with the state of every rescuer answering them. F9 loads the same file and rebuilds runtime caches. Versions 14 through 20 load; anything older is rejected with a clear message rather than opening corrupt.

⚠️ The market record is only the moving parts - stock, prices, rates. The demand model itself is a pure function of the star and is re-seeded on load. It used to be neither saved nor rebuilt, so Market::update filled needs with zeros on the first tick and the whole cluster's economy died: demand 81M -> 0, system turnover down 16x, the millennial audit frozen. A game after its first load was a different game, and nothing on screen said so.

Any ship field a module writes into has to be saved explicitly: modules bake their bonuses into the hull's fields and the module list is not re-applied on load. The same field also has to be cleared in shipApplyClass, or the bonus survives both a hull swap and the loss of the ship. Chromocores had exactly this bug in the other direction - baked in, never re-applied - so three of the seven progression branches died silently on the first hull purchase.

Design Constraints

Important rules from the project documents:

• no faster-than-light travel or communication;
• no instant global market or omniscient strategic map;
• all meaningful remote knowledge must be local, stale, signaled, or debug-only;
• elements are data derived from atomic number, not symbol-specific gameplay;
• systems should use numeric ids and contiguous vectors where practical;
• expensive world-wide work belongs in generation, slow ticks, caches, or bounded candidate searches;
• rendering reads state and draws; gameplay decisions stay in simulation code.

Balance Regression

make balance runs a headless economic regression. Memory is covered by ASan and the local flight invariants by the soak test, but the economy had no coverage at all - and that is where every serious bug turned out to be. Trades once executed with no slippage and returned 44x capital in a single run; the brokerage board once promised 7869 credits where actually running the trade returned 285; two seeds in five produced a starting system with no profitable route at all; and the first two purchasable hulls carried less cargo than the one the player starts with.

Each of those is now a standing check. The thresholds are deliberately wide: the goal is to catch a mechanism breaking, not to freeze the balance in place.

The propulsion model carries its own checks in the same harness: that the cargo -> tank loop works end to end, that any element loads into any container, that ballast dilutes a mixture, that burned fuel comes back as ash nearer to iron, that fusion wins per mass while fission wins per volume, that iron ignites nowhere, that self-supplied fuel beats the station markup, that the drive slot stays exclusive, and that overloading the hold is possible but grounds the ship. Local mining carries four more: that a hold of ore is worth about one hold of ordinary local goods rather than a fortune, that the drill burns reactor fuel and stops dead without it, that drilling scales with the hull and the fitted rig instead of being a flat rate, and that the rig dies with the hull like every other baked-in bonus. make uiclick additionally pins the HOLD window transfer-arrow geometry, because that interface silently failed twice.

⚠️ The most expensive finding of all was invisible to every one of those checks. They measure mechanisms one at a time; none of them played the game. Each mechanism was healthy on its own while together they locked the run shut: a "fill up" button that cost ten times a run's profit (the propellant tank holds a hundred trips), a "buy max" that left nothing for fuel, a hold that was not a ceiling on purchases, and an advisor that went silent on dry tanks without saying why - and that also failed to subtract the road from the profit it promised. A sane play-through on three seeds gave the same thing every time: two to four trades and a permanent 10,000 credits.

testAdvisorRunsCompound now plays the game instead: ten runs in a row the way a player would, with no knowledge a player does not have, and it fails if any of those four pits comes back. If you add a mechanic, the question is not "does it work" but "does it lock the player in together with the others".

Known Gaps

• Full ship modules, weapons, armor, sensors, and upgrade UI are still planned.
• Combat exists through piracy, bounty, raid, threat, and capture pressure mechanics, but not as a detailed ship-module combat simulator.
• OpenGL is an architectural target, not the active renderer.
• Non-player faction memory overlays are not exposed through a debug selector.
• The active build is POSIX/SDL2 via sdl2-config; the old platform makefiles are not synchronized with the current source list.
• Startup generates 8,192 systems and takes several seconds with no loading screen.
• A save now carries the future, not just the moment (save version 21). Measured on the full world (8,192 systems, 150 years in, 40 years out, game timestep): the clearing house treasury reads 3.90836e9 in the original and 3.90836e9 in the reloaded copy, identical to the digit. What used to be missing was not numbers but clocks: the market sweep phase, the accumulated strain and serviceCostAvg, and seededScale - the memory of the last economy rescale. ⚠️ The 1.2% divergence reported here before was partly an artefact of the probe itself: rng is a process global, so running two worlds interleaved in one process makes them draw from a shared stream and diverge no matter how correct the save is. Run one copy, restore the generator, run the other. The underlying defect was real too - the corrected probe still showed 4,243 differences on the old code against 0 now. save_probe.cpp and save_diff.cpp cover this, and testSaveCarriesTheFuture guards it in the harness.
• A game-year of simulation costs about 165 ms at 8,192 systems and roughly a thousand agents. That is 16% of a second at 1 year/second and fine, but it cannot sustain the higher speed multipliers: the loop caps its substeps and falls behind instead. Most of what remains is the market substitution model and the RNG, not routing. ⚠️ There is a procedure, not a number: tick_probe, 8,192 systems, 200 years, seed 20260814, game timestep 0.01, three repeats, median, machine otherwise idle. On this commit that gives 165.1 / 165.4 / 165.7 ms with 1,016 agents. The figure is machine-dependent and drifts with the horizon (the same commit reads 169.6 ms over 100 years and 158.3 over 200), so only before/after on one machine over one horizon means anything.
• The jobs board is no longer a price list bolted onto an emergent market. A delivery's fee is now derived from three things the simulation already knows: the road (burnCost over the route at the origin port's own fuel prices - the floor, since nobody flies at a loss), the spread (amount x (price there - price here) - the ceiling, since beyond it the issuer would just buy it themselves), and the need (strain, the share of the target's demands going unmet, which decides where between the two the fee lands). Job size is contextual the same way: min(what the target is short of, what your reputation is trusted with, a fifth of the seller's stock), so a range of small and large jobs exists at any reputation and the ceiling rises continuously with it - no tiers, no dice. Measured over a sixty-run career (seed 1234, 1,200 stars): rewards run 437 / 847 / 1440 / 3582 / 9281 Cr by quartile, a 21x spread where the board used to be a single point, and 7% of offers are undeliverable on arrival against 68% before this work. Contracts are taken, delivered, and reputation accrues.
• ⚠️ Living only on contracts is still not viable: a contract-led career reaches 3.5e4 credits where a trading one reaches 2.1e6. That is by design in part - trade is the core system and is meant to stay ahead - but the gap is currently a factor of sixty, which is poverty rather than a parallel living. The cause is no longer the price of any single job, which the market now sets: it is the bootstrap of the ladder, since reputation only grows one delivery at a time and a beginner's trusted size is small. How fast that ladder should climb (JOB_CARGO_BASE, and the sqrt(rep/1000) trust curve) is an open design decision, deliberately not taken blind.
• The introductory novel ends at line 29 and only speaks about the high-tech floor contextually, in the first port that actually has such a market. Lines 27 and 28 now cover the distress beacon and answering one. It still says nothing about shares or the fleet autopilot.
• Measured per credit of flight-year, the high-tech floor equals ordinary trade and is ten times worse on a long leg: the rarity of the sources is paid for in travel time, exactly enough to eat the premium. The containment bay caps at 360 units, so income per run does not grow with capital.
• Powers now buy trading licences and commission hulls from their treasuries, so their fleets no longer only shrink. Growth is capped at 1.25x the world's designed ship population (AGENT_TARGET_FULL), because the cost of a simulated year scales with the number of agents. Measured at full scale: 1,016 agents and 165 ms per game year on 8,192 systems, by the procedure above.
• A chromocore now lands on the stat you practised rather than a 1-in-7 roll: surveying feeds intellect, trading charisma, mining and salvage materials, fighting tactics, flying kinematics, radio sources sensors, anomalies luck. What did not change is the tempo - still 2 cores in a measured sixty-run career, the first at trip 19 around year 800 - and the forge, the one way to pick a stat outright, still costs about 7.5M Cr of condensate and so arrives long after it would have mattered. The hook is now a steering wheel on a slow vehicle.
• Repair is priced by the local market (the port's service cost against the cluster median, clamped to the measured 1st and 99th percentiles) and pays the system's owner a tariff. Hull damage is now one system across both layers: macro combat spends the same weapon fields on the same hullHP that local flight shoots at, so a skirmish leaves a mark instead of resolving as intact-or-escape-pod, and a hull that runs out dies. The pair to that is that a port patches a ship the moment it docks - instantly, exactly where the player repairs, since otherwise ships would grind themselves to death over centuries. There is still no insurance of any kind.
• Ships strand for real and the rescue that answers is now a mechanic, not a timer (section 50). A dead engine raises a beacon; the beacon travels at light speed like every other signal in the cluster; whoever hears it and can overhaul the drift diverts and intercepts. Reaching the point the signal left hands the rescuer the hull's true position - drift is strictly ballistic, and the model is exact to 0.000000 ly over 0.856 ly of measured travel. The state pays a full tank plus a tenth of the saved hull, both sides of the entry. Raiders and hostile powers hear the same beacon and come for the hold instead, so the tow stays free while the cargo does not.
• Rescue's honest weak point is the speed ladder, and the number says so: the hundred-year state tow still fires in roughly half of all cases. Hulls drift at 0.15-0.26c while ship-class ceilings run 0.13-0.31c, so closing speeds are 0.05c and under and nobody in the cluster physically arrives in time. Measured on 1,200-star worlds over 300 years at the game's timestep: 3-10 beacons per world, average wait 46-73 years of which 4-8 are light travel, and zero beacons left unresolved. TOW_WAIT_YEARS sits at 100 because the worst honest rescue measured 87.5 years - it is placed just above physics, not in place of it. If that backstop starts firing more often, the ladder is what to look at.
• At the deliberate 100 Cr pauper start there is no profitable trade at all: zero out of forty board entries in each of four measured worlds, because one leg of road costs 6,300-7,500 Cr against 2,700-5,000 Cr of gross. That matches the intent - the first goal of the game is capital, not a destination - but until section 48 the board advertised a healthy profit on those same runs while the flight quietly burned twice what it showed. The novel still opens by teaching trade, and says nothing about where the first working capital comes from.
• Faction treasuries run away: a power holds on the order of 1e22 credits by year 2400, while 20,000 years of optimal player trade yield 1.5e8. AI money and player money live in different universes. This predates the sixteen-player rework and is unchanged by it; the AI's appetite for systems is capped by administrative reach rather than by its wallet, precisely because its wallet is no limit at all.
• Late game is a plateau by design (see master_prompt section 18): 20,000 years of optimal trade yield 1.47e8 Cr while the cheapest system costs 1.37e8 and pays back in 114,777 years. The 8,192-star empire is out of reach on purpose.
• A colony's own stockpile now speeds its construction, but the player still cannot direct what a colony builds.
• The propulsion model computes several quantities it does not yet spend: handlingRisk only trims jet efficiency when it could carry radiation, contraband status and hull wear; nuclearStability only picks the starting fuel when it could decay a stored reserve. Specific heat, melting point and neutron capture cross section are all derivable from the existing shell physics and would give the thermal branch a real material ceiling and a third consumable. See the closing section of ship.md.
• DELTAV_SCALE compresses route delta-V 250-fold. The shape of the physics is intact — exponential, cost optimum, reachability wall — but the magnitude is a deliberate, documented fudge, because honest Tsiolkovsky makes interstellar flight impossible on any fuel.

---

📜 License & Community Standards

Distributed under the True People's License v2.0 / Open License — Authors: Jirnyak & Adolf Petushkov (2026). Free for all maintainers, developers, and AI research. Zero paywalls.

---

👥 Engineering Syndicate & Core Team

Developed and maintained jointly by Жирняк (Jirnyak) and Адольф Петушков (Adolf Petushkov):

Architect	Role & Specialization	GitHub
Жирняк (Jirnyak)	Deep Tech Specialist · High-Performance Physics · N-Body & Quantum Systems · macOS HID	@Jirnyak
Адольф Петушков	Lead Systems Architect · Game Engine Internals · Clinical AI · Zero-GC Concurrency	@marko1olo

🌐 Connected Syndicate Portfolio (12 Flagship Hubs)

• 🌌 Starcluster Simulator — 10,000-star N-body gravitational physics platform
• 🧲 OOMMF Framework — Landau-Lifshitz 3D vector lattice visualizer
• 🍏 Macromac Engine — macOS CoreGraphics HID low-level automation
• 🏢 Gigahrush Raycaster — 2.5D DDA Samosbor raycasting & cellular gas lab
• 🌊 Hecton-8 Submersible — NASA-punk deep sea engine on Unity 6000 (0B GC)
• 🦷 DENTE Dental CRM — FDI odontogram, ICD-10 & 3D DICOM
• 📡 StomChat Dispatcher — Omni-channel WA/TG operator console & SLA telemetry
• 🛡️ AgentRouter Hub — Claude Code CLI WAF bypass proxy & config builder
• 📊 Token Audit — Real-time LLM token cost waterfall simulator
• 🎛️ Nexus Media Engine — Real-time Web Audio DSP & 60 FPS FFT visualizer
• 🤖 Avito Dental AI — Anti-hallucination deterministic veto layer
• 📻 dvachbot — Imageboard scraper & Atkinson dithering transcoder