Multi-Agent Space Simulation: A Reconstructed Thesis
Filed under: space, space-studies, simulation
Simulations in which one agent flies under honest physics exist, and so do simulations in which many agents share one world. No known product does both. The reason is not incompetence. Each product serves its own purpose well, and honest travel times are hard to share among many people. Nothing in physics forbids a simulation that did both. It could teach true physics to the people who used it, and how those people behaved in it would be indicative evidence of how people might trade, move cargo and fight in near-term space, the solar system under known physics and the engineering that can be specified from it. Whether it would pay is the open question.
That argument reconstructs the thesis of a senior seminar paper that the author wrote at the end of an undergraduate space studies programme. The paper reviewed space simulations meant for the general public and asked why none of them combined a physical model that does not cheat with many independent agents in one shared world. The paper itself is lost. The author kept meticulous notes, and a hardware failure took the notes and the paper together, with no backup of either. What follows is therefore a reconstruction from memory. It does not quote the paper, give its title or reproduce its figures, because none of those survive for anyone to check.
The loss may be no great misfortune. In the author’s view, an undergraduate class is a poor format for writing a good paper, because undergraduate classes prioritise grading, the display of knowledge and the verification of academic honesty above the quality of the written material, and they grade writing against somewhat arbitrary criteria. This reconstruction is free of those requirements. Its aim is simply to be a good paper.
The article has four parts. The first sets out the thesis and what a model is. The second reviews four products as space exploration and space exploitation simulations for the general public, namely Kerbal Space Program, Children of a Dead Earth, EVE Online and Star Citizen, and explains why they look as they do. The third catalogues the conventions of soft science fiction, meaning fiction that puts the story before the physics, which these products and the wider media present as if they were physics. It gives the correct physics and a remedy for each. The fourth turns the reviews and the remedies into a blueprint for a high-fidelity multi-agent predictive simulation that is approachable enough to educate the public and deliberately refrains from miseducating it. Three short sections close the article. They sort its claims by their support, name what it leaves out and conclude.
The article uses four terms in a fixed sense throughout. It calls the people and programs that act inside a simulation agents, whether a human or an automaton stands behind them. It calls the products simulations, because the question it asks of each is how faithfully it simulates. A multi-agent simulation means many independent agents, thousands and not a group of friends, in one persistent world on one shared timeline, in a product built for the general public. A model that does not cheat may omit physics and may rescale it, but it never contradicts it, in the sense the What Is a Model? section defines.
The whole article is the author’s position. A claim that carries a citation is a sourced fact, a claim named as an inference is one, and every other judgement is the author’s. The Epistemic State section sorts them.
This is the first article in a short series. An earlier article on this site treats the background mechanics of delta-v, the change in velocity that every manoeuvre costs, which recur below Orbital Manoeuvres. Two others treat propulsion that needs no propellant Solar Sailing and propulsion that leaves the reaction mass behind Mass Drivers.
The Thesis
The thesis makes six claims.
- Strong single-agent models exist. A single agent can already fly a spacecraft through a faithful model of orbital mechanics.
- Multi-agent models have poor physical models. The reviewed simulations in which many people act together in one world simplify or replace the physics of space travel.
- A multi-agent simulation with a physical model that does not cheat is possible, and none is known. None of the products the paper reviewed is such a simulation, and the author knows of none since, so the claim is that none is known at that scale, not that none exists. No obstacle in physics is known. The open problems are problems of engineering and design, and the hardest of them is time. The Possibility section gives the argument and its conditions.
- The existing products have not failed. Their developers are generally competent, and each one’s model successfully expresses its own product thesis, given its product-market fit and its development constraints, including staffing. The gap lies between the products, not in any one of them.
- People adapt their behaviour to conditions. A sufficiently high-fidelity multi-agent simulation should therefore have significant predictive value about how people would behave under the conditions it models, to the extent that its agents are subject to the same incentives and constraints that reality imposes. That hinges on a scenario whose incentives and constraints are close enough to those of real prospective problems. Predictive value is then largely a function of being able to formulate realistic scenarios that agents are willing to take part in exploring. The Predictive Value section gives the reasoning, says why the competence of the agents matters less than the scenario and names the three things that bound the value.
- There is market demand for such a simulation, and meeting it would educate the public. This is the least supported of the six. Both halves are conjectures, and the Market Demand section says why.
The paper drew on several purchased texts on modelling and simulation, which covered subjects such as n-body simulation, in which every body attracts every other, and the simulation of fluids and fuel tanks. The author no longer has those books to hand and no longer recalls their titles, so they go uncited here. The author has also forgotten some of the detailed articles that the paper cited. The sources below are public ones, and the argument is a broad synthesis that leans on space agency sources wherever one exists. A later article in this series dates the work that has appeared since the paper and tests the third claim against it. Whether such a simulation is commercially viable is a separate question from whether it is possible, and the Time and Market Demand sections below treat it honestly.
Two near misses deserve naming at the outset, because a reader who knows the field will think of them. Community modifications let small groups of agents share Kerbal Space Program, and its sequel announced the same for the general public, and the studio closed without delivering it. The first is a proof of concept at small scale, and the second is an attempt that failed. The reviews below say why neither is the simulation the third claim describes.
What Is a Model?
A model is an intentional simplification. Its purpose is to help a person reason about something. The test of a model is whether it supports the reasoning a problem needs, not whether it is true in every detail.
The Bohr model of the atom is the standard example. In 1913 Niels Bohr placed the electron of hydrogen on discrete circular orbits and allowed it to radiate only when it jumped between them Bohr 1913. Bohr himself wrote that “the ordinary mechanics cannot have an absolute validity, but will only hold in calculations of certain mean values of the motion of the electrons.” The model gives the energy levels of hydrogen. Let $E_{n}$ be the energy of level $n$ in electronvolts and $n$ a positive whole number.
\[E_{n} = -\frac{13.6\ \mathrm{eV}}{n^{2}} \tag{1}\]The transition from $n = 3$ to $n = 2$ releases $13.6 \times \left( \tfrac{1}{4} - \tfrac{1}{9} \right) = 1.89$ electronvolts, the red Balmer line of hydrogen near 656 nanometres. The model is wrong about what an electron is and where it goes, and the shells it describes are not the shells of quantum mechanics. Yet it gets that number right, and teachers still give it to students meeting atomic structure for the first time. Physics education researchers have debated whether it should be, and one study of the debate found that “Although the evidence for this claim is weak, many have accepted it,” of the claim that it should not McKagan et al. 2008. The Bohr model survives because it fits the problem of a student who needs energy levels and not orbitals.
Fidelity should fit the problem. A simple model can be too complex for one task, and a complicated model can be too simple for another. A model with every perturbation included is too complex for a student asking how long a transfer to Mars takes, because the perturbations hide the answer. The same model is too simple for a navigator planning a flyby to within a kilometre, if it omits the one perturbation that matters. A problem that needs several models may need a different fidelity in each. A spacecraft simulation might need a precise model of gravity and a coarse model of crew morale, or the reverse, depending on what it is for.
This principle is the measuring stick for the reviews that follow. The review does not fault a product for simplifying, and it judges each product twice. It judges first against the product’s own purpose, where the only fault is a simplification that defeats that purpose. It judges again against the purpose of the simulation the thesis describes, namely to educate the public and to predict behaviour in space. For that purpose a stricter rule follows from the same principle, and it turns on the distinction between omission and contradiction. A model that leaves a phenomenon out can still teach truthfully about everything it keeps. A model that replaces a phenomenon with its opposite teaches the opposite. A third case needs naming. Changing a parameter, such as the size of a planet or the rate at which time passes, keeps the laws of physics and teaches them truthfully. Changing a law, such as adding drag to a vacuum, contradicts them. A model that does not cheat, in the words of the thesis, omits and rescales but never contradicts.
The Bohr model does contradict. It is wrong about the electron, and it is still right-sized for its classroom. The sizing principle allows that, because the teacher tells the student that the model is provisional, and the student replaces it within a few years. A simulation for the public carries no such warning, and no later course replaces it. Its agents learn what it shows them, and anyone who reads their behaviour as evidence about real space inherits every contradiction in the model. For a simulation that teaches and predicts, the right size therefore excludes contradiction. The rule is not a law of modelling. It is the sizing principle applied to those two purposes.
Review Criteria
The reviews form the second part of the article, and they ask the same seven questions of each product. Each review answers them under labels that fit the product, in the order its story needs.
- The physical model. How are gravity and orbits computed?
- Travel and propellant. Does moving cost propellant, and does travel take the time physics says it takes?
- The agents. How many act in one world, and in what roles?
- Approachability. How much must an agent already know to act at all?
- The product’s own thesis. What is the product trying to be, and does its model serve that aim?
- Education. Does a member of the public leave knowing more true physics, or more false physics?
- What the blueprint takes and refuses.
Kerbal Space Program
Kerbal Space Program is a single-agent spaceflight simulation by the studio Squad, and its store page carries the label “Single-player” and none for more than one agent Kerbal Space Program. Its first public release, version 0.7.3, came in 2011 Squad 2011. It entered Steam Early Access in March 2013 and reached version 1.0 in April 2015 Take-Two 2017. Its store page describes it as “based on realistic aerodynamic and orbital physics” Kerbal Space Program.
The physical model. Kerbal Space Program computes orbits with patched conics. At any moment a craft feels the gravity of exactly one body, the one whose sphere of influence it is inside, and its path is a conic section about that body. When it crosses into another sphere, the program switches bodies and patches a new conic onto the old one. The community wiki on the program’s official domain states the consequence plainly. The design “allows simplifying the n-body problem to a one-body problem,” and such simplifications mean that trajectories “such as Lagrange points” do not exist in the program Sphere of Influence. The community edits that wiki, but the developer said the same in 2012. Posting as TheHarvesteR, Felipe Falanghe wrote that “The planetarium only simulates two-body gravity, and switches gravity sources for spacecraft based on the SOI they are in” Falanghe 2012. SOI is the sphere of influence.
The sphere of influence comes from Laplace’s treatment of the three-body problem, and standard texts derive it Curtis 2020 Battin 1999. Let $r_{\mathrm{SOI}}$ be the radius of the sphere, $a$ the semi-major axis of the smaller body’s orbit, $m$ the mass of the smaller body and $M$ the mass of the larger.
\[r_{\mathrm{SOI}} = a \left( \frac{m}{M} \right)^{2/5} \tag{2}\]For Earth about the Sun, with $a = 1.496 \times 10^{8}$ kilometres and a mass ratio of $3.003 \times 10^{-6}$, the radius is about 925,000 kilometres. Inside that sphere the patched-conic method treats the Sun’s pull as absent. The method is accurate enough for the first plan of most real missions, so textbooks teach it, and it fails where neither body’s pull can be neglected. That region contains the Lagrange points, the five points at which a small craft can hold its place relative to two larger bodies. The distance from Earth to its first and second Lagrange points is close to the Hill radius. With $r_{\mathrm{H}}$ the Hill radius and the other symbols as before,
\[r_{\mathrm{H}} \approx a \left( \frac{m}{3M} \right)^{1/3} \tag{3}\]Kane 2017. For Earth the Hill radius is about 1.5 million kilometres. That is the distance of the second Lagrange point, where the James Webb Space Telescope of the National Aeronautics and Space Administration, NASA, orbits the Sun NASA Webb Orbit, and it lies outside the sphere of influence computed above. So a patched-conic simulation cannot host the orbit of one of the most important observatories ever flown. This is an omission and not a contradiction, since the program leaves the third body out and puts nothing false in its place. For an agent learning the Hohmann transfer, the cheapest two-burn path between two circular orbits, the omission is sound. For an agent who goes looking for that orbit, the model is too simple for the task, and the sizing principle counts that as a fault.
Scale and approachability. The program’s home planet, Kerbin, has an equatorial radius of 600 kilometres Kerbin, against Earth’s mean radius of 6,371 kilometres NASA Earth Fact Sheet. The community wiki puts Kerbin at “approximately one-tenth the radius of Earth.” A smaller planet with the same surface gravity needs a lower orbital speed, so a beginner reaches orbit with a small rocket. Let $v_{\mathrm{s}}$ be the speed of a circular orbit just above the surface, $g$ the surface gravity and $R$ the planet’s radius.
\[v_{\mathrm{s}} = \sqrt{g R} \tag{4}\]NASA Flight to Orbit. The same wiki gives Kerbin a surface gravity of 9.81 metres per second squared, so the speed is about 2.4 kilometres per second, against about 7.9 for Earth. This is an approachability choice, and a good one, because it changes a parameter and keeps every law. The change has a price. The same wiki gives Kerbin “1/113th the mass” of Earth. With $\bar{\rho}$ the mean density of a body of mass $M$ and radius $R$,
\[\bar{\rho} = \frac{3 M}{4 \pi R^{3}} \tag{5}\]A body with that share of Earth’s mass, at a radius of 600 kilometres against 6,371, has about 10.6 times Earth’s mean density, and no ordinary matter is so dense. The planet is impossible as matter and correct as mechanics, and the mechanics are what the program teaches. The program still teaches that orbit is a matter of speed and not of height.
Travel and propellant. Every manoeuvre spends propellant, and a transfer takes the time its orbit takes. The wiki puts the cost of reaching low orbit about Kerbin at “approximately 3400 m/s” of delta-v Kerbin. A single agent passes the waiting by accelerating time, a device that the multi-agent modifications below must work around.
Higher fidelity through modifications. Agents who want more fidelity can install community modifications. Principia “implements N-body and extended body gravitation” so that a craft is “influenced by all the celestials,” and its documentation calls the stock behaviour “the toy physics of the stock game” Principia. Real Solar System “Changes KSP’s solar system to make it like the real one” Real Solar System, and Realism Overhaul gives parts “realistic stats and sizes” Realism Overhaul. The product thus lets an agent choose fidelity, and so puts the sizing principle of the What Is a Model? section into practice.
The agents. The stock program has no multi-agent mode Multiplayer. Community modifications add it. Dark Multiplayer “supports subspace style (and master controlled warp) warping” Dark Multiplayer, and Luna Multiplayer “uses some ideas from DMP like the concept of subspaces,” DMP being Dark Multiplayer Luna Multiplayer. A subspace is a private timeline. When one agent accelerates time, that agent leaves the shared present, and the others see the agent’s craft as it was. The subspace is an honest answer to a hard problem, namely that one agent’s wish to skip ahead cannot be imposed on everyone else, and a later article in this series treats it in depth.
These modifications are the nearest thing known to a counterexample to the third claim of the thesis, so the reason they are not one needs stating. Their physics does not cheat, since it is the program’s own. Under subspaces the world is shared and its present is not. The community wiki lists Dark Multiplayer under an approach that, by its account, introduces “paradox-inducing abnormalities” unless the agents share the same time frame. Under the alternative, in which all agents advance time together, the clock is shared, and the wiki calls that arrangement “great for small network games” Multiplayer. That second mode is the blueprint’s shared clock in miniature. What the modifications lack is the scale in the definition of a multi-agent simulation. They are community additions to a single-agent product, their own community describes shared time as suited to small groups, and no source found shows them holding thousands of agents in one persistent world. They are a proof of concept at small scale. The thesis concerns what happens at large scale, where one agent’s wish to skip ahead meets thousands of others, and it states its claim as none known at that scale.
Education. Kerbal Space Program has a recorded place in public education. Squad released an Asteroid Redirect Mission pack in 2014 that it said “was made in collaboration with NASA, to make sure our Kerbal version of the Mission was not only true to its real-life counterpart, but that it was also fun, educational” Squad 2014. NASA later described through its Goddard Space Flight Center a scientist on the asteroid sample return mission Origins, Spectral Interpretation, Resource Identification, Security, Regolith Explorer, known as OSIRIS-REx, challenging the program’s users to fly a virtual version of the mission NASA 2016. Its publisher described a school edition, KerbalEdu, as “an official school-ready standalone remix” of the program KerbalEdu 2016. The original program had sold more than two million units by 2017 Take-Two 2017 and more than five million by 2023 Take-Two 2023.
The product thesis. Kerbal Space Program set out to make rocketry approachable and fun, with a textbook physical model as its core. Its model expresses that thesis well. It is a great textbook sandbox for a single agent, and its fidelity is right-sized for what it teaches.
Takes. The blueprint takes the textbook core, the planning tools that let an agent see a future orbit before committing to it, and the principle that an agent may choose how much fidelity to handle.
Refuses. It refuses the single-agent world, and it refuses a fixed patched-conic model where the agents’ purposes need more.
Children of a Dead Earth
Children of a Dead Earth is a single-agent space warfare simulation by Q-Switched Productions, released on 23 September 2016 and labelled “Single-player” only Children of a Dead Earth. Its developer kept a blog through 2016 that explains the physics behind each system, and the original paper drew on it, as this article does.
The product thesis. The developer states it directly. The simulation “was developed primarily to answer the question: What would space warfare actually be like?” It was “designed with no vision in mind,” because the aim was “to discover what space warfare would be like, rather than to say what space warfare would be like” How Realistic Is It Actually?. The developer adds that it is “a simulation first, and a game second. No amount of realism was compromised to make things more fun.” That is the right method for a predictive experiment. It imposes no preconceived notion on the scenario and lets the equations decide what happens.
The physical model. The simulation uses n-body gravity. The developer writes that it “uses an N-Body Simulation to simulate its orbital mechanics,” that it was “Originally … supposed to use Patched Conics,” and that it “simulates all manner of Lagrange Point orbits” Fun with Orbital Mechanics. Propulsion, weapons, armour, sensors and heat rejection are each derived from engineering equations in the developer’s posts, so travel costs propellant and takes the time its trajectory takes Burn Rockets Burn Space Guns The Photon Lance Raw Steel Stealth in Space. The developer’s post on propellant tanks discusses boil-off, pressurisation, ullage and anti-slosh baffles as matters of tank design and mass Slosh Baffles. That post does not establish whether the simulation integrates the motion of the liquid itself.
What the physics produced. The result of letting the equations decide is a tempo that almost no science fiction shows. The developer’s own example is two fleets in the same orbit, one running retrograde, which “may experience five seconds of combat where the enemy zooms by at ridiculous speeds, and then ten days of downtime while the crew prepare for another five seconds of combat” The Essence of Space Warfare. In the author’s experience, an engagement in the simulation is a great deal of waiting followed by a deadly high-speed exchange of about thirty seconds. The arithmetic of closing speed shows why. With $t_{\mathrm{e}}$ the time two craft spend within range of each other, $R$ the effective range of their weapons and $v_{\mathrm{rel}}$ their relative speed, a pass straight through gives
\[t_{\mathrm{e}} = \frac{2R}{v_{\mathrm{rel}}} \tag{6}\]Two craft in opposite circular orbits at the height of the International Space Station each move at about 7.66 kilometres per second, so they close at about 15.3 kilometres per second. With an effective range of 100 kilometres, the exchange lasts about 13 seconds. This equation is a simplification of the author’s, a straight pass at constant speed, and it shows only the order of magnitude.
Approachability. The developer’s posts do not discuss difficulty or the intended audience. The closest statement is that designing lasers “is often far more difficult than designing any other system, so there are plenty of factory-made options” The Photon Lance. The fidelity that makes the simulation valuable also makes it a poor fit for the general public in practice. Its users joke that using it needs several science degrees. The author’s experience is milder. Anyone with a proper science degree can work out the parts that their own programme did not cover. But a simulation that needs a degree to begin with is not a simulation for the general public. It is simply too complex for most people.
The meta. The simulation has a meta, a set of designs and practices that its agents converge on because they work well against the model as built. A meta is the agents’ adaptation to the model, and it is exactly the behaviour the fifth claim of the thesis predicts. It also bounds the experiment’s predictive value. Agents optimise against the simulation’s assumptions, its scenarios and its omissions, not against reality, so where the model departs from reality, the meta departs with it. This is not a flaw peculiar to this product. It is a property every predictive simulation shares, and it is the reason the blueprint below treats fidelity and approachability together.
Education. The simulation is a wonderful instrument for the people who can work with it and a poor educational utility for everyone else, because a lesson nobody takes teaches nothing.
Takes. The blueprint takes the first-principles method, the n-body core, derivation from engineering equations, and the refusal to start from a conclusion.
Refuses. It refuses to expose all of that fidelity to every agent at once.
EVE Online
EVE Online is a multi-agent space simulation launched in May 2003 by CCP Games CCP 2003, which rebranded as Fenris Creations in 2026 after separating from its former parent company, Pearl Abyss Gematsu 2026. In 2013 the developer reported that it “had surpassed the 500,000 subscriber mark worldwide for the first time,” and that its agents “compete for riches, power, glory and adventure in a single universe” CCP 2013b. The developer calls its human agents capsuleers Capsuleers.
A measured economy. The simulation’s market is its strongest model. It is “Supported by a vast player-run economy,” and in 2007 the developer appointed an economist, Eyjólfur Guðmundsson, who “will publish quarterly reports on the state of the EVE Online economy” CCP 2007b. The developer still publishes monthly economic reports CCP 2026. A real economy of real agents, measured by a professional, is a model of high fidelity for the thing it models.
The physical model. The simulation’s physical models are so far from the physics they depict that they are wrong. Ships fly as submarines move through water. The developer’s own account of warp gives every ship a “maximum speed” under its own engines CCP 2013c. A top speed under continuous thrust is already a departure from mechanics. In vacuum the acceleration from a given thrust does not fall as speed rises, and a ship gains speed for as long as it has propellant to burn. The developer appears to have published no velocity law of its own. The community’s own guide gives one. With $v(t)$ the speed after $t$ seconds of full propulsion, $v_{\max}$ the ship’s top speed, $I$ an inertia modifier in seconds per kilogram and $M$ the ship’s mass in kilograms, the time constant is $\tau = I M \times 10^{-6}$ seconds and
\[v(t) = v_{\max} \left( 1 - e^{-t/\tau} \right) \tag{7}\]Acceleration. That is the law of a body pushed through a viscous fluid, in which drag grows with speed until it balances the thrust. The developer’s statements agree with the guide’s law on two points, the top speed above and the alignment time below. The same guide says that when propulsion stops, the ship decelerates to a standstill, because the model “assumes that space has some substance to it and thus some friction.” The author has watched ships in the simulation slow to rest in just that way. Its opposite is Newton’s first law, which NASA states as “an object in motion remains in motion at constant speed and in a straight line unless acted on by an unbalanced force” Newton’s Laws of Motion. The simulation’s model is therefore not an omission of orbital mechanics but a contradiction of inertia. The community name for it, submarine physics, belongs to the community, not to the developer.
Travel. A ship crosses a solar system by warp. The developer explains that a warp “is split in to three phases - Acceleration, Cruise and Deceleration,” with speed “measured in AU/s,” and that a ship must first reach “the alignment threshold of 75% of its maximum speed” CCP 2013c. Setting $v(t) = 0.75\,v_{\max}$ in the velocity law gives the alignment time. With $t_{\mathrm{a}}$ that time,
\[t_{\mathrm{a}} = \tau \ln 4 \tag{8}\]That is about $1.39\,\tau$, and it agrees with the alignment time given in the developer-hosted documentation for third-party developers Useful Formulae. Warp speeds run from about 1.5 to 8 astronomical units per second by the community’s figures Warp, so a crossing that would take a real spacecraft months takes seconds. Stargates link systems. A ship spends no propellant, and the simulation tracks no delta-v. Ships draw on a capacitor that recharges from an onboard generator, which powers modules and starts each warp, and only capital jump drives burn fuel Capacitor Jump Drives.
The agents. Every agent that matters is a person. The interactions between them have full fidelity, because the simulation does not simulate them at all. It only sets the rules of a sandbox around them, which the developer itself calls “sandbox-style gameplay” CCP 2013b. The sum is greater than the parts. Poor individual models, joined by a real market and real people, produce alliances, wars, espionage and economic history that no single component could.
Approachability. An agent needs no physics to act, since the simulation contains none to know. What an agent must learn is the economy, the politics and the other agents, and that difficulty is of the kind the simulation exists to offer.
The product thesis. The developer’s pages read for this review state no aim beyond the descriptions quoted above, a single universe in which agents compete, run as a sandbox. The thesis attributed to the product here is an inference from them and from the product itself. It is one world whose content is the dealings of its agents with one another.
The Economy
EVE Online’s economy is fascinatingly odd. Its individual models are terrible, yet they interlock so comprehensively that the whole reproduces real economic behaviour, and its predictive value lies there. The commodity market is the one place where the model is strong and accurate, because nothing simulates it. Real agents set every price, so the market behaves as markets do.
This subsection is long, because the economy is the best evidence available here that agents in a simulation behave as people do. It describes the supply chain and its rules, with the developer’s formulas for readers who want them, then the money, and then the real economic behaviour that appeared with nobody designing it. A reader in a hurry can pass over the formulas, which sit under the labels Industry costs and The market, and resume at Logistics without losing the argument.
EVE Online’s mechanics change from one release to the next, so each figure below carries the date of its source. A figure that is out of date still describes how agents acted at the time, and the conclusions drawn from it do not depend on its being current.
The supply chain. The community guide describes industry as “a catch-all term for all the industrial activities within EVE: mining, gas huffing, manufacturing, researching, invention, reprocessing, and reverse engineering” Industry. Agents mine ore, reprocess it into minerals, haul the minerals to a facility, manufacture ships and modules from blueprints, and sell them. Every ship destroyed in combat is gone, so destruction drives demand back down the chain. In the parts, this is complete science fantasy. The minerals have invented names, a blueprint turns them into a warship without any physics of manufacture, and nothing about the process resembles a real shipyard. As a system, it is the best multi-agent supply chain the author knows of, logistics included.
In some cases the incorrectness serves the soft science fiction thesis itself. The invented names are one. EVE’s base mineral is Tritanium, which the developer describes as “The main building block in space structures” Tritanium. The word is Star Trek’s. In the episode “Obsession” of 1967, Spock reports “Pure tritanium” and Kirk replies “20 times as hard as diamond” Obsession Transcript. A reporter noted in 2008 that EVE’s tritanium “hearkens back to references made in Star Trek” Engadget 2008. A search found no statement by the developer on the name’s origin, so the borrowing is the author’s reading, of technobabble pulled into the base of a supply chain.
Industry costs. The developer publishes its formulas. They appear here to show what kind of model the parts are. Each is a rule of bookkeeping with no physics in it, and one of them is a sound piece of economic design. With $C_{\mathrm{job}}$ the cost of installing an industry job, $V_{\mathrm{EIV}}$ the estimated item value of its output, $I_{\mathrm{SCI}}$ the cost index of the solar system, $\beta$ the product of the facility’s bonuses, and $t_{\mathrm{f}}$, $t_{\mathrm{SCC}}$ and $t_{\alpha}$ the facility tax, a fixed surcharge and a surcharge on accounts without a subscription, the developer’s own statement, “TIF = EIV * ((SCI * bonuses) + FacilityTax + SCC + AlphaClone),” reads
\[C_{\mathrm{job}} = V_{\mathrm{EIV}} \left( I_{\mathrm{SCI}}\, \beta + t_{\mathrm{f}} + t_{\mathrm{SCC}} + t_{\alpha} \right) \tag{9}\]CCP 2023. The cost index is congestion pricing. When the developer introduced the index in 2014, it defined it from “the number of job-hours of that activity type in that system over the last 28 days, divided by the number of job-hours of that activity type in the universe as a whole over the last 28 days, and then square-rooted” CCP 2014b. With $h_{\mathrm{sys}}$ and $h_{\mathrm{all}}$ those two totals, the index in that form grows as
\[I_{\mathrm{SCI}} \propto \sqrt{\frac{h_{\mathrm{sys}}}{h_{\mathrm{all}}}} \tag{10}\]A system that hosts four times another’s share of production pays twice the index, so production spreads out across the map instead of piling into one system. The community guide warns that a later patch changed the index and that the square-root form is probably no longer exact.
Materials obey a rule of their own. The same 2014 account states that “Each step of ME reduces materials by 1% (always rounded up),” that rounding now happens per job and not per run, and that “every run requires a minimum of one unit of every listed material” CCP 2014b. With $n_{\mathrm{mat}}$ the quantity of one material a job consumes, $r$ the number of runs, $n_{0}$ the blueprint’s base quantity per run, $E$ its material efficiency level, $\gamma$ the facility’s material modifier and $q$ the requirement after rounding, those rules assemble into
\[\begin{aligned} n_{\mathrm{mat}} &= \max\left( r,\ q \right) \\ q &= \left\lceil r\, n_{0} \left( 1 - \frac{E}{100} \right) \gamma \right\rceil \end{aligned} \tag{11}\]A blueprint calling for 1,000 units per run at a material efficiency of 10, run 10 times with no facility bonus, consumes 9,000 units. Nothing in that rule is manufacture. It is an entry in a ledger, and agents have built a real industry on it.
The market. Agents set prices in regional markets, and the developer’s own economist has noted “the unrelenting dominance of Jita as a central hub” CCP 2015a. Every sale pays a sales tax, raised in 2025 “from 4% to 7.5%” CCP 2025, and every order pays a broker’s fee. Structures owned by agents set their own fees, and the developer observes that “It is not uncommon for structure owners to give themselves and their friends a 0% fee” CCP 2020. A haul between two hubs is worth making only when the margin survives the fees. With $p_{\mathrm{A}}$ the price paid at the source, $p_{\mathrm{B}}$ the price received at the destination, $\lambda_{\mathrm{s}}$ the sales tax rate, $f_{\mathrm{b}}$ the broker’s fee rate and $c_{\mathrm{h}}$ the cost of hauling one unit, a trade placed as an order at each end pays when
\[\begin{aligned} & p_{\mathrm{B}} \left( 1 - \lambda_{\mathrm{s}} - f_{\mathrm{b}} \right) \\ & \quad - p_{\mathrm{A}} \left( 1 + f_{\mathrm{b}} \right) - c_{\mathrm{h}} > 0 \end{aligned} \tag{12}\]The developer set the Accounting skill in 2019 to reduce the sales tax by 11 percent per level and Broker Relations to reduce the broker’s fee by 0.3 percentage points per level CCP 2019, and the community guide gives the base broker’s fee at a station of the simulation’s own as 3 percent Trading. Taking the sales tax of 2025 with the skill effects of 2019, which assumes that those effects did not change in between, the rates at the fifth level of each are about 3.4 and 1.5 percent. Buying at 100 and selling at 110 then leaves about 3.1 per unit before the cost of hauling, so a gross spread of 10 percent shrinks to a net margin near 3.
Logistics. Hauling is a profession. The developer’s profile of one hauling corporation, Red Frog Freight, says that “It works a little bit like a real life logistics company, with around 600-700 contracts daily” CCP 2013a. Courier contracts carry collateral held in escrow, so a hauler who loses the cargo pays for it.
Money. The currency of the simulation is the ISK. The developer described its money supply in 2006 in terms an economist would use. “We call anything that introduces money into the economy an ISK Faucet and anything that takes it out again an ISK Sink” CCP 2006. In 2012 it reported that bounties were the largest faucet, “33 trillion ISK” in one month, and that three of four price indices showed “considerable inflation” CCP 2012a.
PLEX and real money. In 2008 the developer let subscribers convert paid subscription time into an item, the “30 Day Pilot’s License Extension (PLEX),” which a subscriber could put “directly on to the market” CCP 2008. A PLEX “can be bought via the in-game market for ISK and then applied to add 30 days of game time” CCP 2009b, and in 2017 the developer converted each “into 500 PLEX” CCP 2017. PLEX gave the economy a legal exchange rate with real money. The developer’s own announcement said it “Opens up for a larger supply, lowering the ISK exchange rate in the long run” CCP 2008, and its economist later called it “EVE’s equivalent of the gold standard” CCP 2015a. A press report of 2014 records the developer selling PLEX from a reserve to stop a price spike Engadget 2014, which resembles a central bank’s intervention to steady a price. The developer forbids trade outside that channel. The licence the developer quotes says that agents may not “transfer, sell or auction, or buy or accept any offer to transfer, sell or auction” anything in the simulation, “including without limitation characters, character attributes, items, currency, and objects” CCP 2016, and its security team reported banning 70,492 accounts in 2021 for all causes CCP 2022. The exchange rate also prices destruction. Of the battle of B-R5RB in 2014, the developer reported losses of “11 TRILLION ISK” and that “According to some PLEX conversions that could equate to approximately \$300,000-\$330,000 USD” CCP 2014a.
Real phenomena that appeared unbidden. Because the interlocking parts reproduce real incentives, real economic behaviour appeared that nobody designed.
- Tax havens. The simulation’s starter corporations once offered “war declaration immunity and 0% tax,” which the developer said gave them “an inherent advantage over player corporations which was never intended” CCP 2009a. The developer responded in 2009 by taxing the starter corporations at 11 percent, a rate it chose after researching the average taxes of corporations run by agents. Its economist had already measured corporate taxation, finding that “The average tax rate is 8.6% but the median is tax rate is 5%” CCP 2007a.
- Embezzlement at a bank. In 2009 an agent who ran a bank owned by agents, EBank, embezzled “roughly 200 billion ISK, or 8.6% of EBANK’s 2.3 trillion ISK in deposits” Engadget 2009, and, as Reuters reported, exchanged it for cash through the black market Reuters 2009, about five thousand United States dollars The Register 2009. The developer’s economist said the bank would have to solve the problem itself, since it was “a player driven entity” IEEE Spectrum 2009.
- Cartels. A scarce resource found “completely in one distinct region” caused, in the developer’s words, “wealth transfers that are simply too large for the effort and risk involved” CCP 2012b, and one of the cartel’s own organisers described it in the trade press as “a price-fixing cartel which controls the Technetium supply” The Mittani 2012.
The developer answers in kind. Faucets, sinks, inflation indices and sales of PLEX from reserve are monetary policy under other names, practised by the developer, not by the agents.
Skills that transfer. A simulation that demands commitment can hold its agents to very high expectations, and what they learn under those expectations can carry over into their working lives. A survey reported in an article presented by the developer found that “56% of players said that skills they learned through playing EVE had proven to be valuable in their lives outside of the game and in their careers” Yancy 2020. The developer sponsored the source, and the underlying study did not turn up, so the figure stands here as the developer’s claim. Peer-reviewed work is more cautious but points the same way. One study examined leadership in online environments shared by many participants, EVE Online among them Lisk et al. 2011, and another found evidence that leadership practised online spills over into work Xanthopoulou and Papagiannidis 2012.
The Verdict on EVE Online
Right-sized fidelity. EVE Online’s wrong models have the right fidelity for the problem they actually serve, namely driving meaningful interaction between agents. By the sizing principle, that is a success. A supply chain that exists and works matters more to that problem than one that is accurate. Being able to navigate at all, and to operate from and in different places, matters more than navigating realistically. EVE Online therefore passes the fidelity test for its own purpose and fails it for education. The blueprint below must pass both. The author’s conjecture is that more realistic simplified models could have worked in EVE Online, and that the developer lacked the specialists to distil them for the product. No source consulted confirms or refutes it.
Education. EVE Online largely simulates the science fiction of film and television, and its soft science fiction sandbox borders on miseducating the public, because it shows its agents, year after year, that ships coast to a stop and that a solar system is a few seconds wide.
Takes. The blueprint takes the market, the single shared world, real people as agents, loss that lasts, and the lesson that systemic interaction can matter more to agents than the realism of any one component.
Refuses. It refuses drag in vacuum, free travel, and the absence of propellant.
Star Citizen
Star Citizen is a multi-agent space simulation by Cloud Imperium Games. It remains in alpha testing as of this writing. The developer’s own crowdfunding counter reported funds of 106,210,587,748 and a head count of 6,757,438 on 12 November 2026 RSI Crowdfund Statistics. The counter’s data labels the head count as fans, who need not all be paying backers, and it does not state the units of the funds. Read as cents, the funds come to about 1.06 billion United States dollars. Its Alpha 4.0 release introduced server meshing, which spreads one world across several server processes, with shards of up to 500 agents, a shard being one copy of the world and the agents in it Star Citizen Alpha 4.0.
The product thesis. The developer’s chairman, Chris Roberts, describes the aim as “a First Person Universe that allows you to live a virtual life in a compelling futuristic setting” Roberts 2018, and as the ability to “immerse you as a player into your own adventure to rival any big budget movie” Roberts 2020. The product thesis is that agents can act out adventures as if they were the stars of their own films. Its roles are rich and asymmetric, including “exploration, hauling, trading, mercenary, outlaw, bounty hunting and mining,” with more planned Roberts 2020.
Star Citizen appears to be broadly aspiring to a higher-fidelity version of EVE Online’s soft science fiction thesis, with its own spin. Where EVE Online’s agents command ships from a capsule, Star Citizen’s agents act at the tactical level of a human avatar, on ships, on stations and on the surfaces of planets and moons. That makes the two very different in practice. A large ship has several stations that do very different things, flown together by a crew, and a pilot may fly with no gunners or several, depending on the ship’s configuration and the people available. A ship can also carry smaller craft and ground vehicles in its hold. The developer’s own description of the Carrack promises several “seats with their own dedicated actions” and “a Pisces-class launch (design redacted) and an Ursa-class rover for ground exploration” RSI 2014a.
Its characteristic scale of interaction is the crew and the landing party, where EVE Online’s is the fleet and the market. Most of its activities, which the developer and the community call loops, are of two kinds. Ship-based loops include bounty hunting and cargo, present by Alpha 3.0.0 in December 2017 Star Citizen Alpha 3.0.0 Patch Notes, mining by ship, added in Alpha 3.2.0 in June 2018 to “the moons of Crusader (Daymar, Yela, and Cellin)” Alpha 3.2.0 Patch Notes, and salvage, whose “first implementation” arrived in Alpha 3.18 in March 2023 RSI Roadmap. Personnel-based loops include first-person combat in bunkers, added as “Hurston underground bunker FPS AI missions” in Alpha 3.4.0 in December 2018 Alpha 3.4.0 Patch Notes, hacking at terminals, exploring on foot, and driving ground vehicles.
Some loops belong to neither kind. Engineering arrived as a station of its own in Alpha 4.5.0 in December 2025, when “Ships now feature an engineering terminal providing comprehensive oversight of all ship systems” Star Citizen Alpha 4.5.0. Medicine became a loop in Alpha 3.15 in 2021, in which “ambulances (such as the Cutlass Red) can treat minor wounds and stabilize patients in transit to hospital” Deadly Consequences, so search and rescue is a nominal experience in the simulation. The release notes consulted list no dedicated search-and-rescue missions. Other loops remain planned. Crafting arrived in a first tier for personal equipment in 2026, while base building, farming and data running remain announcements.
The product has been a buggy mess in perpetual alpha, and it has been getting better. Its management and its controversies aside, the product as delivered is interesting enough to care about, perpetual alpha and all. It is directionally promising. EVE Online receives more attention in this article only because it is the more mature product, with two decades of a running economy to study.
Travel and propellant. Ships cross a system by quantum travel, a warp in all but name, and the same handwave of travel inconvenience that EVE Online makes. Unlike EVE Online, Star Citizen tracks propellant. The developer explains that “The ship’s thrusters consume fuel stored in the Hydrogen Fuel Tank,” and that “When in Quantum Travel, thrusters use the fuel contained in the Quantum Fuel Tank” The Shipyard: Fuel Mechanics. The fuel model is of very low fidelity, and it would be surprising if the simulation modelled propellant slosh at all. The author understands Kerbal Space Program to treat each tank as a point mass that varies between its wet and dry masses, which also leaves slosh out. Slosh matters in real craft. A NASA monograph of 1966 devoted to the behaviour of liquids in moving containers treats it as a problem of space vehicle technology Abramson 1966, and the catalogue below says why.
The physical model of the system. The planets and moons rotate in place at fixed positions and do not orbit. The evidence for that finding comes for now from the community, not from the developer. The developer’s release notes for Alpha 3.0 state that “Planets and moons now rotate to give a day/night cycle” Star Citizen Alpha 3.0.0. The developer designed for orbits. In 2016 Chris Roberts said that day and night would come from a planet rotating on its axis, and that the system “will also be able to rotate the planet around the sun” GamersNexus 2016. No developer statement read here says whether that capability is in use. The community wiki states that the planets “spin, have a day and night cycle, and are planned to eventually orbit” Planet. So the simulation, though designed for orbits, does not simulate them. That matches the author’s experience. The author recalls noting, while using the simulation, that its celestial bodies were not subject to celestial mechanics, and that a body sits at fixed coordinates to which a ship travels.
Gravity. Ships and stations carry artificial gravity. The developer’s design brief for it says that it “must be able to be turned on and off,” and that “‘handwavium’ is completely acceptable if it makes sense” Lore Builder Twenty. Gravity of that kind is the gravity of film and television, and the catalogue below treats it.
Launch and reentry. Since 2020 the simulation has modelled lift and drag in atmosphere. The developer states that “Ships with wings are now able to stall” and that “In atmosphere, thrusters now lose efficiency and become much weaker” Flight and Fight Improvements. The developer’s notes read here give no sign of reentry heating as a hazard or of any requirement to reach orbital speed in order to reach orbit. Agents fly through launch and reentry, but the simulation does not model them as the challenges they are.
Approachability. An agent needs no physics to act here either. An agent enters the simulation as a person in a body, who walks, flies and shoots by the conventions of film, and that is as approachable as an interface can be.
Education. Star Citizen is a multi-agent soft science fiction reenactment, and its model expresses that thesis well. Like EVE Online, it is a soft science fiction sandbox that borders on miseducating the public.
Takes. The blueprint takes asymmetric roles, the crew as the unit that flies a ship, and tracked propellant.
Refuses. It refuses artificial gravity without a mechanism, travel without time, and bodies that do not orbit.
Why the Products Look As They Do
None of the four products has failed. Kerbal Space Program wanted approachable rocketry and got it. Children of a Dead Earth wanted to discover space warfare and did. EVE Online wanted a single universe of real people with a real economy, and Star Citizen wants agents to live their own films.
That is the stance from which the review reads them, and it is an interpretation as much as a finding. The four are prominent, so it describes products that succeeded, not every attempt. Where the review infers a product thesis from the product, as it does for EVE Online, the product cannot fail to express it. Star Citizen is the case that strains the stance, as the account of its finances below shows.
This section compares the four and then explains the gap between them by staffing and by management. Two cases, Kerbal Space Program 2 and Star Citizen, show what management and money do to an ambitious simulation.
Two tables set the four reviews side by side, the single-agent products first.
| Kerbal Space Program | Children of a Dead Earth | |
|---|---|---|
| Orbits | Patched conics | N-body |
| Travel | True time, skipped by accelerating time | True time |
| Propellant | Tracked | Tracked |
| Agents in one world | One | One |
| Physics needed to begin | Little | A great deal |
| What it shows | True physics, with sound omissions | True physics, to few |
| EVE Online | Star Citizen | |
|---|---|---|
| Orbits | None, ships fly under a drag law | None found, bodies rotate in place |
| Travel | Warp, seconds per system | Quantum travel |
| Propellant | None, except jump fuel | Tracked at low fidelity |
| Agents in one world | Many, in a single world | Many, up to 500 per shard |
| Physics needed to begin | None | None |
| What it shows | Contradicted physics | Contradicted physics |
The first table holds the honest physics, and the second holds the agents. Neither holds both.
The developers are generally competent. Each product reflects a combination of its product thesis and its product-market fit, the match between a product and an audience willing to pay for it. Where one of the four falls short of the standard this article sets for high-fidelity space simulation, it is because it had a different product thesis and the developer achieved it. Many other titles succeed on their own terms as multi-agent soft science fiction simulations. Falling short of that standard in that way is not a failure of craft. It is a choice of target. The other way to fall short is to aim at the standard and miss, as Kerbal Space Program 2 did in the account below.
Staffing is part of the explanation. A model of high fidelity costs specialists to build and to maintain. A small studio can afford it only in a narrow product, as Children of a Dead Earth shows, and a large studio can afford it only if its market rewards it. The two multi-agent products drew their audiences with the simplest physics, and the product that drew a large audience with honest physics, Kerbal Space Program, drew it one agent at a time. That is the gap the thesis names. No known product has yet paid for physics that does not cheat in a product built for many agents and the general public at once.
Management is the other part, and two products show it. Kerbal Space Program 2, a sequel by a new studio, Intercept Games, entered early access in February 2023. Its publisher’s announcement listed a multi-agent mode among the features to come, saying that “Players will be able to both cooperate and compete with friends to explore space together” Take-Two 2023. The announcement speaks of friends and states no scale. It still made the sequel the one known attempt at a multi-agent simulation on the textbook core in a product for the general public, and it is a development since the paper.
The press recorded that the sequel, first due in 2020, “limped into early access” on Steam, that “Bugs and performance issues were major problems,” and that it “launched without features included in the original” Chalk 2024. The mode never arrived. In November 2026 the product’s store page still lists “colonies, interstellar travel, and multiplayer” as planned, under a notice from the store that the developers last updated the product more than two years ago Kerbal Space Program 2.
In 2024 Take-Two filed a notice with Washington’s Employment Security Department of a permanent closure in Seattle affecting seventy workers, which did not name the studio Washington WARN Notices, and the press reported, citing Bloomberg, that Take-Two was shutting Intercept Games down GeekWire 2024. The author’s judgement is that mismanagement sank the project, and the studio with it. What the record shows is narrower. The attempt failed in execution before it could test whether the public wanted the result.
Star Citizen is the harder case. It has stayed in alpha for well over a decade, yet it has found an odd market equilibrium in which a nominally unfinished product earns large recurring revenue.
The terms on which that revenue arrives explain how the equilibrium can last. The developer collects its revenue as pledges, not purchases. The developer’s terms state that “You do not purchase anything, you make a pledge towards the development of the Game” and that pledge funds “are a deposit to be used for the development and production cost of the Game” RSI 2026. A pledge is nonetheless contractually binding. The same terms make any unearned portion non-refundable “until and unless RSI has ceased development and failed to deliver,” with an arbitration clause and a withdrawal period of thirty days added in 2023. The refund condition has tightened over time. The original terms of 2012 allowed a refund twelve months after the estimated delivery date, and the press reported the change in 2016 to the condition of ceased development at the time Massively Overpowered 2016.
The sales model is grey, in the author’s view, and the project is controversial. Most of that controversy is the sentiment of customers and spectators, which primary sources do not record, and the developer has no incentive to acknowledge unhappiness publicly. The press records some of it. A backer who sued for a refund and lost said of the developer that “They figured out how to sell scope creep itself” Grayson 2018, the developer told another that “Refunds are always discretionary and are considered on a case-by-case basis” Polygon 2017, and the developer added a disclaimer to its ship sales after the United Kingdom’s advertising regulator called them misleading PC Gamer 2021.
Cloud Imperium publishes its annual financials, which make the equilibrium measurable. The reports for 2020 to 2024 state income and costs including capital expenditure, in millions of United States dollars, for the whole company, Star Citizen and Squadron 42 together Cloud Imperium 2020 Cloud Imperium 2021 Cloud Imperium 2022 Cloud Imperium 2023 Cloud Imperium 2024. For 2020 to 2022 the income is the sum of the stated pledges or sales, subscriptions and other income, and the cost is the sum of the stated costs and capital expenditure. Let $\rho_{y}$ be the ratio of spending to income in year $y$, $C_{y}$ the costs including capital expenditure and $R_{y}$ the income.
\[\rho_{y} = \frac{C_{y}}{R_{y}} \tag{13}\]| Year | Income | Costs | Ratio |
|---|---|---|---|
| 2020 | 88.2 | 80.9 | 0.92 |
| 2021 | 100.4 | 100.5 | 1.00 |
| 2022 | 130.7 | 129.4 | 0.99 |
| 2023 | 143 | 163 | 1.14 |
| 2024 | 146.6 | 161 | 1.10 |
Over the five years the company took in about 609 million dollars and spent about 635 million, a ratio of 1.04. It spends what it raises, and in the last two years more.
The figures support two readings, and they cannot decide between them. The author leans to the second. On either reading the case shows that an ambitious space simulation is expensive and risky, and that is the reason for setting it out here.
On the first reading, the perpetual alpha is the business model working as intended. The alpha label lowers expectations, selling ships ahead of what agents can do with them depends on nothing being final, and the terms keep pledges non-refundable for as long as development continues. That reading fits the principle that developers are generally competent and achieve their product thesis. Star Citizen has achieved product-market fit while still in alpha, which is highly irregular. Under those circumstances the author might stay in perpetual alpha too, since a release would trade a working market for a set of expectations the product would then have to meet.
On the second reading, the project is mismanaged. The author is not unhappy with Star Citizen, but understands why some people are, and leans to the second reading from professional experience in the game industry. The author’s judgement is that spending of about 635 million dollars in five years, on top of the eight years before them, would have produced a version 1.0 release under ordinary control of scope. That judgement rests on the combination of cumulative expenditure and a perpetual alpha, and a reader who weighs the scale of the ambition differently may reach the first reading. When the author was following Star Citizen, the developer was largely selling dreams, ships and features either not yet built or only partially implemented, and delivering at a pace sufficient to keep its backers engaged.
The figures sharpen rather than settle the question. A ratio near one is consistent with what the terms promise, since a pledge is a deposit towards development, and the figures show the company banking no profit. The reports read here do not show what covered the excess of spending over income in the last two years. Income that recurs without a release imposes no deadline, so a project can grow for fourteen years without one. What the figures cannot show is what the spending bought.
Setting the ethics of that model aside, Star Citizen is plainly doing something right even as it does a number of other things wrong. Neither Kerbal Space Program 2 nor Star Citizen is evidence against the competence of developers in general. Both show that an ambitious space simulation is expensive and risky. The fourth claim of the thesis names exactly that constraint.
Soft Science Fiction and Its Remedies
The two multi-agent products share their physics with film and television. Star Citizen says as much of its aim, and EVE Online’s borrowed mineral shows the lineage in one case. That the developers inherited the conventions and did not arrive at them independently is otherwise an inference from the resemblance. This third part of the article traces where the conventions came from, catalogues the ones a space simulation is likely to repeat, and gives a remedy for each. The remedy is usually a version of do not do that, or use real physics.
Where the Conventions Came From
Star Trek began as a western. Gene Roddenberry’s pitch of 1964 opened with the line “STAR TREK is a ‘Wagon Train’ concept,” built around characters who travel to worlds “similar” to Earth Gizmodo 2011, and the format reprinted in the 1968 account of the production reads “Wagon Train to the Stars” Whitfield and Roddenberry 1968. A western brings a fixed cast to a new town each week. A space western needs the same cast on a new world each week, whatever the distance. The same account records the production reasons behind two devices. Of landing the ship, Roddenberry asked whether they could “LAND A SHIP FOURTEEN STORIES TALL ON A PLANET SURFACE EVERY WEEK?” and answered that the budget “FORCED US INTO CONCEIVING THE TRANSPORTER DEVICE.” Planets like Earth were likewise “a compromise with production costs.” A search found no primary source saying that the production chose warp drive for the same reason. That it served the same purpose, literally a wave of the hand that delivers the cast to the next adventure, is the author’s inference from the format.
EVE Online and Star Citizen use that model of travel. They can justify it for the same reasons the series did. Both put agents into adventures in a shared world, and the adventures need the agents to be in the same place at the same time. The justification is real, but it is a justification for the format, not a description of space.
Three causes kept these conventions alive. The first is the format, as the western shows. The second is that effects budgets limited what a production could show faithfully, as the transporter shows. The third, in the author’s view, is that writers often lacked training in the physics they depicted and audiences seldom knew the difference. The second cause weakens with every year since the arrival of computer-generated imagery. The Last Starfighter of 1984 carried about twenty-five minutes of computer-generated imagery in about 230 scenes, rendered on a Cray X-MP SIGGRAPH. On Babylon 5, J. Michael Straczynski credited Ron Thornton as the one who “first suggested we use CGI” Slashdot 2013. The Expanse later showed that a large audience will accept spin gravity and thrust gravity. Its co-author Daniel Abraham described ships “having thrust gravity and spin gravity and ships that are built like towers,” while conceding that its drive is handwavium Space.com 2015. Rendering a coasting ship now costs no more than rendering a banking one. What remains is a choice.
The Science Fiction Loop
Science fiction inspires children to become scientists, and science inspires the writers of science fiction. Michio Kaku recalls that at about eight “I got hooked on the Saturday morning TV shows. In particular, Flash Gordon” Big Think 2015. Stephen Hawking co-wrote a series of children’s adventure novels with his daughter Lucy, beginning with George’s Secret Key to the Universe in 2007 Hawking and Hawking 2007. NASA chose Nichelle Nichols, who played Uhura in Star Trek, to record a recruiting video and speak to audiences nationwide in the late 1970s, and NASA records that astronaut Mae Jemison “credits Nichelle Nichols as her inspiration to become an astronaut” NASA 2024. A peer-reviewed survey of 575 Doctor Who viewers found reported effects ranging from fact-checking the series’ science “to inspiring them to pursue a science career” Orthia 2019.
The loop runs whatever the hardness of the fiction. What the hardness changes is the fidelity of what passes through it, the ideas the children carry and the training they seek. A child inspired by a ship that coasts to a stop must later unlearn it. A simulation is a participatory medium. It does not only show a convention, it trains its agents to act on it. Education research gives two reasons to take that seriously. Fiction does teach false science. A study of middle school students who watched the film The Core found that “a single viewing of a science fiction film can negatively impact student ideas regarding scientific phenomena” Barnett et al. 2006. The study was small, the 82 students of one teacher, and its authors say that its findings may not hold for the population at large. And errors about motion are among the hardest to give up. Clement reported in physics students a mistaken view of force and acceleration that “is highly resistant to change” and traced it to “a deep-seated preconception that makes a full understanding of Newton’s first and second laws very difficult” Clement 1982. A later review of the field summarised the finding. In its words, “many students believe that motion implies a force, both before and after the study of introductory mechanics” McDermott and Redish 1999. That belief is the flight model of EVE Online, in which a ship moves only while its engine pushes. It is an inference that an agent who flies under that model for years is practising the preconception a physics course must then undo. A search found no study of what a simulation’s flight model does to its agents’ physics, so the miseducation this article warns of is a hypothesis with indirect support. If it holds, solving the problem described here is of direct relevance to educating the public for the future benefit of humanity.
The Catalogue
Each entry gives the convention, the correct physics and the remedy, and says where the convention appears when a reviewed product or a source shows it. The entries run in order through motion and travel, gravity, shields and heat, vacuum and sound, weapons and asteroid fields, communication, and reentry and propellant, and they end with aliens, tractor beams and black holes.
Ships that stop when the engine stops. EVE Online’s velocity law above is the clearest case in a simulation, and film and television do the same whenever a ship slows without firing a thruster. In vacuum a ship keeps its velocity Newton’s Laws of Motion.
Remedy. Use Newton’s first law. A ship that wants to stop must spend as much delta-v as it spent to start.
Banking turns, wings in vacuum and dogfights. A banking turn uses air. In vacuum a craft turns by reaction control, a set of small thrusters that the Space Shuttle used “to provide propulsion thrust for precise attitude control and three-axis translation” Blevins and Hohmann 1975. Rotating a craft does not change the direction it moves. Only thrust along the new heading does. Star Citizen’s atmospheric model is correct to give wings lift in air. The error, wherever it occurs, is to carry the behaviour into vacuum.
Remedy. Separate attitude from velocity, and give wings effect only where there is air.
Rockets that push against air. NASA’s educational pages address the question of “how they can work in space where there is no air for them to push against,” and answer that “The answer to this question comes from the third law” Rocket Principles. A rocket in fact does better without air. NASA lists the thrust of its RS-25 engine as “512,300 lbs. (vacuum)” against “418,000 lbs. (sea level)” NASA 2025.
Remedy. Show exhaust momentum as the source of thrust, and let a rocket perform better in vacuum than in air, as it does.
Travel without time. Warp in EVE Online and quantum travel in Star Citizen deliver a ship in seconds. Real travel takes time set by orbital mechanics, and the delta-v of a transfer sets what it costs Orbital Manoeuvres. The rocket equation connects delta-v to propellant. Let $\Delta v$ be the change in velocity, $I_{\mathrm{sp}}$ the specific impulse in seconds, $g_{0}$ standard gravity at 9.80665 metres per second squared, $m_{0}$ the initial mass and $m_{\mathrm{f}}$ the final mass.
\[\Delta v = I_{\mathrm{sp}}\, g_{0} \ln \frac{m_{0}}{m_{\mathrm{f}}} \tag{14}\]NASA Ideal Rocket Equation. An engine with a specific impulse of 450 seconds and a mass ratio of four gives a delta-v of about 6.1 kilometres per second.
Time is the other cost. The cheapest two-burn transfer between two circular orbits in the same plane is the Hohmann transfer, half an ellipse that touches both Curtis 2020. With $t_{\mathrm{H}}$ the transfer time, $a_{\mathrm{t}}$ the semi-major axis of the transfer ellipse, the mean of the two orbital radii, and $\mu_{\odot}$ the Sun’s gravitational parameter at $1.327 \times 10^{11}$ cubic kilometres per second squared,
\[t_{\mathrm{H}} = \pi \sqrt{\frac{a_{\mathrm{t}}^{3}}{\mu_{\odot}}} \tag{15}\]From Earth at 1.000 astronomical units to Mars at 1.524, the transfer ellipse has a semi-major axis of 1.262 astronomical units, or $1.888 \times 10^{8}$ kilometres, and the transfer takes about 259 days. Nor can it begin at any moment. The planets return to the same relative position once each synodic period. Let $S$ be the synodic period and $T_{1}$ and $T_{2}$ the orbital periods of the inner and outer planets.
\[\frac{1}{S} = \frac{1}{T_{1}} - \frac{1}{T_{2}} \tag{16}\]For Earth at 365.25 days and Mars at 686.98, the synodic period is about 780 days, so a window for the cheapest transfer to Mars opens about once every 26 months Curtis 2020.
Remedy. Track propellant and delta-v, and make travel take its true time. The blueprint below gives travel time a job instead of hiding it.
Bodies that do not orbit, and ships that hover. A planet or moon in a fixed position cannot teach orbital mechanics. A ship that holds position above a point on a planet needs a geostationary orbit, which exists only above the equator and is circular at an altitude of 35,786 kilometres, where a satellite moves at about 3 kilometres per second, as the European Space Agency, ESA, describes it ESA 2020. Let $r_{\mathrm{G}}$ be the radius of the orbit, $\mu$ Earth’s gravitational parameter at 398,600 cubic kilometres per second squared and $T$ the sidereal day of 86,164 seconds.
\[r_{\mathrm{G}} = \left( \frac{\mu T^{2}}{4 \pi^{2}} \right)^{1/3} \tag{17}\]The radius is 42,164 kilometres, or 35,786 kilometres above the equator. Everywhere else, holding position costs thrust continuously.
Remedy. Put every body on its true orbit, and let a ship hold position only where physics allows it.
Space is far away. Space begins low. The Fédération Aéronautique Internationale describes “the 100km altitude used by FAI and many other organisations” as the boundary of space FAI 2018. What makes orbit hard is speed. Let $v_{\mathrm{c}}$ be the speed of a circular orbit and $r$ its radius from Earth’s centre.
\[v_{\mathrm{c}} = \sqrt{\frac{\mu}{r}} \tag{18}\]NASA Flight to Orbit. At the height of the International Space Station, about 420 kilometres up, the radius is 6,791 kilometres and the speed is about 7.66 kilometres per second.
Remedy. Teach that orbit is a speed, which Kerbal Space Program already does at its smaller scale.
Zero gravity in orbit. Orbiting astronauts are in free fall, not out of gravity. NASA’s own explanation for students says that at the station’s altitude “Earth’s gravity is about 90 percent of what it is on the planet’s surface,” and that microgravity “is sometimes called ‘zero gravity,’ but this is misleading” NASA 2012. Let $g(h)$ be the acceleration of gravity at altitude $h$, $g_{0}$ its value at the surface and $R$ Earth’s radius.
\[g(h) = g_{0} \left( \frac{R}{R + h} \right)^{2} \tag{19}\]At 420 kilometres the ratio is about 0.88.
Remedy. Call it free fall or microgravity, and show that the station is falling around Earth.
Magic artificial gravity. Floors that pull downward by an unexplained mechanism, switched on and off at will, are the gravity of film and television and of Star Citizen’s design brief Lore Builder Twenty. The physics offers two kinds of gravity on a ship, rotation and thrust. NASA’s 1977 design study of a space settlement chose rotation, writing that “To simulate Earth’s normal gravity the entire habitat rotates at one revolution per minute about the central hub,” that people living permanently in space “should live between 0.9 g and 1 g,” and that rates “greater than 1 rpm should be avoided” Johnson and Holbrow 1977. Let $a_{\mathrm{c}}$ be the acceleration felt at the rim, $\omega$ the rate of rotation in radians per second and $r$ the radius.
\[a_{\mathrm{c}} = \omega^{2} r \tag{20}\]At one revolution per minute, $\omega$ is about 0.105 radians per second, and the study’s radius of 830 metres gives about 9.1 metres per second squared, or 0.93 of standard gravity.
Remedy. Give gravity a mechanism. Use spin for habitats and thrust for ships under way, as The Expanse does, and let free fall be the condition of a coasting ship.
Shields. The force field of film and television has no basis in physics. Spacecraft do carry shields, of a different kind. NASA’s Hypervelocity Impact Technology group records that “The Whipple Shield is the first spacecraft shield ever implemented” and “is still in use today” NASA Whipple Shield, after Fred Whipple’s proposal of 1947 Whipple 1947. A Whipple shield is a thin bumper held off the hull, which breaks up an impacting particle before it reaches the wall. Researchers have studied active magnetic shielding against radiation Westover et al. 2019, but it remains a concept.
Remedy. Model armour, Whipple shields and radiation shielding as mass, and do not add force fields.
Stealth, waste heat and radiators. Every powered spacecraft must reject its waste heat, and in vacuum it can do so only by radiation. NASA’s small spacecraft guidance states that “In a vacuum, heat is transferred only by radiation and conduction with no convection” NASA Thermal Control, and the International Space Station’s external thermal system has “a total capability of 70 kW” ISS Active Thermal Control. Let $P$ be the power radiated, $\varepsilon$ the emissivity of the radiator, $\sigma$ the Stefan-Boltzmann constant at $5.670 \times 10^{-8}$ watts per square metre per kelvin to the fourth, $A$ the radiating area and $T$ its temperature in kelvin.
\[P = \varepsilon \sigma A T^{4} \tag{21}\]Liles and Amundsen 2023. A radiator at 275 kelvin with an emissivity of 0.9 rejects about 292 watts per square metre, so 70 kilowatts needs about 240 square metres facing cold space, before any allowance for sunlight absorbed. That glowing area is visible to an infrared sensor, and one optics study observes that “By observing the radiator, we can catch a glimpse of a spacecraft’s inner information” Li et al. 2017. The developer of Children of a Dead Earth goes further and argues that there is “no stealth in space” The Essence of Space Warfare. No agency or peer-reviewed source makes that claim in general, so it stands here as the developer’s argument.
Remedy. Model heat as a resource that a craft must reject, give every craft a thermal signature, and do not let a ship vanish.
Space is cold. Space has almost nothing in it to be cold. Vacuum is an insulator, as the thermal guidance above implies. A body in vacuum loses heat only by radiation, which is slow at ordinary temperatures, so nothing freezes instantly. Sunlight matters as much as shadow. NASA notes that without thermal controls the Sun-facing side of the space station “would soar to 250 degrees F (121 C), while thermometers on the dark side would plunge to minus 250 degrees F (-157 C)” NASA 2001.
Remedy. Model thermal balance, in which a spacecraft’s harder problem is often getting rid of heat.
Exposure to vacuum. Bodies do not explode, and nobody freezes solid at once. A NASA account of pressure suits describes a 1965 chamber accident in which the subject “remained conscious for about 14 seconds, then lost consciousness,” and states that “You would not instantly freeze solid, as some movies have shown” NASA 2017.
Remedy. Show the documented effects, which are serious enough without exaggeration.
Sound in space. NASA states that “Sound waves cannot travel in the vacuum of space because there is no medium to transmit these mechanical waves” Anatomy of an Electromagnetic Wave.
Remedy. Silence outside, and sound only where it can travel, through the hull and the air inside.
Visible beams and slow bolts. A laser beam is invisible from the side in vacuum. A manufacturer of optics notes that “In the absence of any type of scattering medium, the beam will only be visible as a spot when it reaches its target” Edmund Optics, and NASA notes that the infrared light used for its laser communications “is invisible to human eyes” NASA 2021. A bolt of light that travels slowly enough to watch is not light.
Remedy. Show beams only by their effect, and give projectiles their true speeds.
Fireball explosions. An explosion in space has no air to carry a blast wave and no oxygen to sustain a fire. The nuclear effects literature records that the fraction of energy “converted into blast and shock is less and decreases with increasing altitude” Glasstone 1964, though a high-altitude nuclear test can still form a fireball, so the physics is subtler than film and simpler than fire. The developer of Children of a Dead Earth puts the practical result bluntly, that in space “explosions are pitifully weak” The Essence of Space Warfare.
Remedy. Model the energy delivered, not the spectacle, and let fragments and radiation carry the damage, since there is no air to carry a blast.
Dense asteroid fields. A ship weaving among tumbling boulders is a set piece of film. NASA’s Dawn mission team answers that if the main belt’s asteroids of a mile across or larger were spread evenly, “the average distance between individual asteroids 1 mile in diameter or larger would be about 1.9 million miles,” nearly eight times the distance between Earth and the Moon, and that the science fiction depiction of boulders “jostling for space” is “not the case at all” NASA Dawn FAQ.
Remedy. Put the asteroids where they are, almost everywhere far apart.
Instant communication. Light takes time. NASA’s Goddard communications staff note that “Movies and television shows can make communicating with space look easy” NASA 2020, and a NASA technical memorandum gives Mars an “up to 22-minute one-way (44-minute round-trip) delay at maximum distance from Earth” Landon et al. 2025. Let $t_{\mathrm{L}}$ be the one-way delay, $d$ the distance and $c$ the speed of light.
\[t_{\mathrm{L}} = \frac{d}{c} \tag{22}\]At its closest, about 0.37 astronomical units, Mars is about 3.1 light-minutes away, and at its farthest, about 2.67 astronomical units, about 22 light-minutes.
Remedy. Delay every message and every sensor reading by its light time. This single rule changes command, markets and warfare, and Children of a Dead Earth’s developer notes that “seconds or even minutes of lag in command would prove fatal in combat” How Realistic Is It Actually?.
Reentry by friction. Popular accounts describe reentry heating as friction. The dominant mechanism for a blunt body is compression in the shock layer ahead of it. Allen and Eggers showed in 1958 that “for any truly blunt body, the bow shock wave is detached,” which is why capsules are blunt Allen and Eggers 1958. NASA’s own history of reentry uses both descriptions, the heat of “shock wave compression” and “aerodynamic heating, a product of the friction of the high-speed airflow” Launius and Jenkins 2012, so friction is not wrong so much as incomplete.
Remedy. Model ascent and reentry as the delta-v and thermal problems they are, which no reviewed multi-agent simulation does.
Propellant without slosh. A tank of liquid propellant is not a fixed mass. The liquid moves, shifts the craft’s centre of mass and couples into its attitude control. NASA’s monograph of 1966 on liquids in moving containers exists for that reason Abramson 1966. Leaving slosh out is an omission and not a contradiction, and it appears here because the remedy is a matter of sizing.
Remedy. Model slosh where a subsystem needs it, in docking, landing and fine pointing, and size it out where it does not.
Humanoid aliens. The humanoid alien of film and television was often a prosthetic on an actor, and the same production account that records the transporter puts worlds like Earth down to production costs Whitfield and Roddenberry 1968. Science disagrees about whether evolution elsewhere would converge on human forms. Simon Conway Morris argues for strong convergence Conway Morris 2003, and Levin and colleagues argue from natural selection that alien life is predictable in its processes rather than its shapes Levin et al. 2017.
Remedy. For a simulation of near-term space, do not include aliens. None is known.
Tractor beams and inertial dampeners. A NASA study of optical trapping reports that “tractor beams do exist but can only move a limited number of targets over micron-level distances” Stysley 2011. Inertial dampeners have no source of any kind. By Newton’s second law a crew feels every acceleration of its ship, and that limit on a crewed craft’s manoeuvres is part of the physics worth teaching.
Remedy. Do not include either.
Black holes that suck. NASA states that black holes do not suck in other matter, and that “From far enough away, their gravitational effects are just like those of other objects of the same mass” NASA Black Holes.
Remedy. Use the gravity of the mass, which an n-body model does unprompted.
A Blueprint
This fourth part turns the reviews and the remedies into one design. It describes a high-fidelity multi-agent predictive simulation for the general public that is approachable enough to educate and refrains from miseducating.
Its nearest concrete form is a real-time, multi-agent simulation built on the method of Children of a Dead Earth. Real-time here means continuous and persistent. The world runs around the clock whether or not a given agent is present, and it does not proceed in turns. It does not mean that the clock runs at the rate of a wall clock, which the Time section treats. Near-term space, the solar system under known physics and engineering that can be specified from it, is the standard Children of a Dead Earth set itself. That simulation already has the physics, derived from engineering equations with no vision imposed on the result. What it lacks is the shared world. The design below keeps the method and the physical core, runs them in real time on a shared clock, puts many human agents into one persistent world with a market and asymmetric roles, and makes the whole approachable by tools rather than by fiction. It takes a method, not a product, and its builders would derive every system afresh from first principles.
The subsections follow the argument. They state the principles and the physical core, argue that the design is possible, and then take up time, the agents, the economy, approachability, predictive value, scenarios and market demand. The equations an implementer needs come last.
Principles
Five principles govern the design.
- Omit, never contradict. The simulation may leave a phenomenon out when the agents’ purposes do not need it, and it says so. It never replaces a phenomenon with its opposite. It may change parameters, such as scale and the rate of time, but never laws. This is the sizing principle of the What Is a Model? section applied to a simulation that teaches and predicts, stated as a rule of design.
- Size fidelity per subsystem. Gravity, propulsion, heat, sensors, communication, propellant slosh and the economy each get the fidelity the agents’ decisions need.
- Make the physics approachable, not soft. Tools that explain and compute supply approachability. Fiction that removes the physics never does.
- Make loss last. Other agents can destroy what an agent builds, and what they destroy stays destroyed, as in EVE Online. Agents who risk nothing they value give no evidence of how people behave when something is at stake.
- Be open to checking. The simulation publishes its model, its economy and the record of what happened, because evidence nobody can check is not evidence.
The Physical Core
The physical core computes the world. Each of its elements gets the fidelity that the agents’ decisions need.
- Scale. The world is the real solar system at its true scale. Kerbal Space Program’s small planets are a sound change of parameter for a single agent learning to reach orbit. A simulation meant to predict behaviour in near-term space needs the distances and the delta-v budgets of near-term space. Rescaling and time compression both shorten a voyage. Rescaling also lowers every delta-v budget and shortens every light time, so it changes the prices and the delays the simulation exists to study. Compression changes only how long those delays feel. The design prefers compression for that reason, and the rate of time is the one parameter it changes. A reduced system at a lower compression is an alternative it has not weighed further.
- Gravity. The planets and moons follow ephemerides, precomputed positions from a real solar system model, such as the planetary and lunar ephemerides that the Jet Propulsion Laboratory fits to observations Park et al. 2021 and publishes for download JPL 2020. That is an omission of their response to the agents’ craft, which is negligible, and not a contradiction. Of a planet that gives a spacecraft a gravity assist, NASA writes that “The massive planet’s loss is too small to be measured” NASA Basics of Spaceflight. The simulation integrates craft numerically under n-body gravity, so Lagrange points, the perturbations of third bodies and gravity assists emerge unscripted. The equation in the implementation subsection treats each body as a point mass. The oblateness of a planet, the decay of a low orbit in thin air and the pressure of sunlight are further terms of the same equation, and one account of the standard method lists solar radiation pressure, atmospheric drag and corrections for general relativity among the forces on a spacecraft Berry and Coppola 2007. By the test the review applied to Kerbal Space Program, that a model must not be too simple for the task, agents need those terms wherever they work in low orbit. An orbit that keeps pace with the Sun exists because “The Earth’s equatorial bulge imparts an out-of-plane force on the orbit which causes a gyroscopic precession of the line of nodes” Mesarch 2018, and NASA notes that “Satellites in a low Earth orbit are also pulled out of their orbit by drag from the atmosphere” Riebeek 2009. The design adds those terms there. An integrator whose energy error stays bounded over long runs, such as a symplectic method, keeps orbits stable through months of simulated time. Yoshida constructed symplectic methods of higher order Yoshida 1990, and Wisdom and Holman built symplectic maps for n-body problems with a dominant central mass and followed the outer planets with them for a billion years Wisdom and Holman 1991.
- Gravity at scale. Honest gravity need not be expensive gravity. Three practical techniques keep it affordable without contradicting it. The first is the ephemeris itself. An ephemeris fitted to observation is more accurate than anything the simulation would integrate, it costs a lookup, and every machine reads the same values. It also cannot go wrong in the way a system integrated live can. The makers of the Principia modification found that under n-body gravity the fictional system of the stock program was not stable, and that “the Jool system breaks down within 19 days,” so they altered it to keep its moons Principia Jool Stability. Kerbal Space Program avoids the problem as the blueprint does, and its wiki notes that “the orbits of the celestial bodies can’t be changed” Sphere of Influence. The second is to bind each craft, and each small natural object, to a parent body. The simulation integrates the craft in the parent’s frame, where it feels the parent in full and every other body as a perturbation, and it drops a perturbation that falls below a declared threshold. The implementation subsection gives the equation and the sizes. A craft in low orbit about Earth needs Earth, the Moon and the Sun, and Jupiter at its nearest adds about one part in a trillion. The third is to divide the world along the same lines. The neighbourhood of each body is a natural shard, here a division of one world that its own machines run, a craft passes from one shard to the next at the sphere of influence, and the cruise between planets is a shard of its own. Light time helps between distant shards. Events at two planets cannot affect each other faster than light crosses the gap, which takes minutes, so the machines that run those shards have that long, divided by the compression of the clock, to agree. It does not help at a boundary, where two craft on either side may be metres apart and the hand-off between machines must be exact. Nor does the division relieve a crowded region, since the busiest places, low orbits and stations, lie inside one shard. Each technique is an omission of stated size. The design leaves out the momentum a craft exchanges with the body it flies past, the slight effect of a flyby on that body’s rotation and the pull of one craft on another, and it says so. It keeps the behaviour that matters at the scale of a mission, namely transfers, captures, gravity assists, Lagrange points and the slow work of perturbations on an orbit.
- Clocks. Each craft keeps its own clock, which runs at the rate that relativity gives it. The difference is slight within the solar system and becomes decisive only in interstellar travel.
- Reproducibility. Every machine that computes the world reaches the same state from the same inputs. A world that anyone can replay from its recorded inputs is open to checking, and a planning tool that runs the same step as the world forecasts exactly what the world will do, for as long as no other agent intervenes. The implementation subsection says what that demands of the arithmetic.
- Propellant and delta-v. Every manoeuvre costs propellant by the rocket equation, and the interface shows each agent the delta-v remaining.
- Heat. A craft must reject its waste heat, and the radiators that reject it are a signature that others can see.
- Light time. Every message, order and sensor reading arrives late by its light time. The rule binds craft and instruments. It cannot bind the people behind them, who talk to one another outside the simulation and may act through several characters at once. The design enforces light time on what it controls, namely what a craft senses, when an order reaches it and when a trade settles, so that knowing a thing early does not let an agent command a distant craft early. It does not stop an agent who hears a distant price from trading where that agent already stands. What agents learn outside the simulation remains a leak, and it is a declared limit on what their behaviour shows.
- Slosh and structure. The simulation models them where docking, landing or pointing need them, and sizes them out elsewhere.
Possibility
The third claim of the thesis says that such a simulation is possible. The claim rests on two observations and carries one condition.
The first is that each half already exists. Children of a Dead Earth runs n-body gravity and systems derived from engineering equations on a personal computer, and EVE Online holds a large population in one persistent world. That second proof has a caveat. EVE Online’s record is 65,303 agents connected at once, which the press reported in 2013 Drain 2013, and those agents occupied a world divided into systems joined by gates. The developer balances its load by moving whole systems between server processes, and it has written of an overloaded system that “We cannot split the solar system up into more units and spread them out” CCP 2010. When a battle outgrows its process, the developer slows that process’s clock, down to what it calls “the artificial 10% cap on the time dilation factor” CCP 2014c. Star Citizen’s shards hold 500 agents each. Neither product shows that one continuous solar system can hold as many agents, and a world whose orbits all run on one clock cannot slow one region without breaking the timing of the rest. The Physical Core section above proposes a division that the physics itself supplies.
The second is that the halves do not obstruct each other. A craft’s pull on other craft and on the planets is negligible, so each craft is a separate problem, a small body moving among massive ones whose positions are already known. Trajectory software for real missions works the same way. One account of the standard method says that it ignores the forces of the spacecraft on the celestial bodies, so that “the motion of the celestial bodies can be determined independently from the motion of the spacecraft” Berry and Coppola 2007. So the cost of the gravity model grows in proportion to the number of craft not to its square. Let $t_{\mathrm{step}}$ be the time one processor core needs to advance every craft by one step, $N_{\mathrm{c}}$ the number of craft, $\bar{n}$ the mean number of bodies each craft feels and $t_{\mathrm{f}}$ the time of one evaluation of the inverse-square law.
\[t_{\mathrm{step}} = N_{\mathrm{c}}\, \bar{n}\, t_{\mathrm{f}} \tag{23}\]A published benchmark of gravitational force loops measured two billion evaluations a second on one processor core with vector instructions, twenty times the rate of plain code Tanikawa et al. 2013. Plain code then needs about 10 nanoseconds for each evaluation. The benchmark used single precision and a solar system needs double, so the estimate allows 20. One hundred thousand craft, each feeling thirty bodies, then need three million evaluations and about 0.06 seconds for each step. At a step of ten simulated seconds and a compression of one hundred, both of them assumptions, the world needs ten steps in each second of real time, or 0.6 seconds of one core’s work. Binding each craft to a parent body, as the Physical Core section describes, cuts the bodies a craft feels from thirty to about three in low orbit and about ten in cruise between planets, and the step to between 6 and 20 milliseconds. On either count, gravity for one hundred thousand craft costs less than one processor core. The work divides cleanly among cores, because no craft waits on another. A craft in interplanetary cruise also needs far fewer steps than one in low orbit.
That estimate is the author’s and shows an order of magnitude only. It covers the propagation of the live world and nothing else. The number of steps in each second of real time grows with the compression of the clock. The planning tools may cost more than the world does, since each plan propagates a craft across months and each edit repeats the work. And the estimate leaves out collisions, sensors and everything else that happens when craft meet, where a real implementation would spend its effort.
The condition is time. The hardest problem is not the computation of orbits. It is the sharing of honest travel times among many people, and the Time section treats it. The claim of possibility is therefore conditional. No obstacle in physics is known, and the gravity model is cheap. What happens when many craft crowd one region is an open problem of engineering, and it goes uncosted here. Whether a design can make the result palatable to the general public depends on the problem of time, which this article names and does not solve.
Time
This section treats the hardest of the open problems. It settles that a shared world needs one shared clock and plans committed in advance. It leaves open how fast that clock should run and whether the public will accept the result.
One clock. Travel takes its true time, and the simulation gives that time a job. The job is to be the interval in which agents plan, trade and prepare, as it is in real spaceflight. A time warp chosen by one agent cannot be imposed on thousands of others, since one agent’s acceleration of time would move every other agent’s world. A shared clock may still run fast, provided it runs fast for everyone. Let $t_{\mathrm{w}}$ be the wall-clock time an agent waits, $t_{\mathrm{s}}$ the simulated time that passes and $\kappa$ the compression factor shared by every agent.
\[t_{\mathrm{w}} = \frac{t_{\mathrm{s}}}{\kappa} \tag{24}\]The same factor shortens everything. One more duration helps to show it, a single orbit at the height of the space station. With $T_{\mathrm{o}}$ the period of a circular orbit of radius $r$ and $\mu$ Earth’s gravitational parameter as before,
\[T_{\mathrm{o}} = 2 \pi \sqrt{\frac{r^{3}}{\mu}} \tag{25}\]NASA Flight to Orbit. At a radius of 6,791 kilometres the period is about 93 minutes. The table applies the compression to that period and to four durations worked earlier.
| Duration | True | At 10 | At 100 |
|---|---|---|---|
| A pass in low orbit | 13 seconds | 1.3 seconds | 0.13 seconds |
| Longest light time to Mars | 22 minutes | 2.2 minutes | 13 seconds |
| One low orbit | 93 minutes | 9.3 minutes | 56 seconds |
| Transfer to Mars | 259 days | 26 days | 2.6 days |
| Synodic period | 780 days | 78 days | 7.8 days |
No single compression suits both the voyage and the engagement. At one hundred the voyage becomes bearable, and the exchange becomes too short for any person to act in.
What compression costs. Compression also has a cost that is easy to miss. It keeps every ratio between physical durations, so the physics stays true, but it does not compress the agents. A person needs the same seconds to read a display and decide, whatever the clock does. At a compression of one hundred, the light time to Mars shrinks from a wait that reshapes command to a pause of seconds, and the delay that the catalogue’s remedy imposes all but disappears from the agents’ experience. Compression is therefore a change of parameter that leaves the physics intact and still moves the simulation away from the conditions it means to predict. The less of it a design needs, the more its agents’ behaviour is worth as evidence. The blueprint does not fix the factor. It states a preference, the lowest compression its agents will tolerate, and leaves the value to trial.
The rest of this section addresses the tension between the compression that agents will tolerate and the evidence that compression costs.
Plans committed in advance. The inference drawn here is that a shared world needs a shared clock, compressed by the same factor for everyone, and agents who plan and commit manoeuvres in advance and then do something else while the plan executes. That is how real spaceflight operates. NASA’s guide to the subject says that “Planned sequences of commands” “are typically less risky than real-time commands because they benefit from a long process of extensive debate and selection, testing and checking and simulation prior to uplink” NASA Basics of Spaceflight, Cruise. It is also the tempo Children of a Dead Earth discovered, long waits followed by short decisive exchanges. A private timeline of the kind the Kerbal Space Program multi-agent modifications use is a workable compromise when agents must act alone, and a later article in this series compares the two.
Long journeys. Travel that takes its true time in the lives of the participants is not a new idea. In the tabletop role-playing game Traveller, a starship that jumps between stars “always stays in Jump Space for roughly one week” Traveller System Reference Document, and the author recalls overhearing the notion of running that game in real time, so that a journey from one world to the next takes a week of the participants’ lives.
Long journeys change what an agent is. Under any compression a design can afford, a transfer to Mars still takes days or weeks of an agent’s real time. An agent has two ways to live with a long journey. One is to act through several characters, one in transit while another works at the destination. A character can leave earlier and arrive shortly before the role it is to fill begins, much as a real programme sends a crew ahead of a mission. The other is simply to attend to real life while the journey runs. Balancing life against a high-commitment simulation is a problem its dedicated agents already face, and some welcome an escape valve built into the design. An agent who expects to be away has good reason to begin a long voyage first, so that the absence costs nothing. That is a matter of the agent’s own management of time, outside the simulation. Inside it, the effect is a population of individuals who set out for distant places for reasons the simulation never sees. That is a fair model of people who travel far out of restlessness or for causes of their own. It is also a bias. In reality the time away is the largest cost of a long voyage, and here it can cost an absent agent nothing, so the simulation should expect more long voyages than reality would see. The incentives bound below must count that.
A long journey also touches the clock of the character who makes it. Relativistic time dilation is nominally relevant, though it is chiefly a concern of interstellar travel. It is the dilation of physics, not the slowing of a server’s clock to which EVE Online gives the same name. At the speeds of near-term spaceflight it amounts to a fraction of a second in a year, as the clock equation in the implementation subsection shows. At interstellar speeds the exact factor of special relativity replaces the first-order term for speed.
\[\frac{d t_{\mathrm{c}}}{d t} = \sqrt{1 - \frac{v^{2}}{c^{2}}} \tag{26}\]Shapiro 1978. At a tenth of the speed of light a traveller’s clock would lose about 1.8 days in every year, and at higher speeds a character in transit would live through less time than the world it left. Conclusions about colonisation, exploration and exploitation that reach to the stars must reckon with it.
The schedule. A world that runs around the clock also imposes a schedule. EVE Online’s agents, as a class, already work to taxing schedules across time zones when an event matters, even though no individual can attend every pivotal development. The developer designs around the problem. It calls time zone safety “an absolutely critical part of any capture system in a worldwide game like EVE Online” and holds that “Nobody should feel the need to play the game 24/7 in order to be competitive” CCP 2015b. Any high-fidelity simulation that runs continuously may subject its agents to the same realities, inconvenient as they are. The brief exchange of a pass between two fleets happens when the orbits say it happens, which may be the middle of the night on a Tuesday or the middle of an important meeting. Agents must then balance their real lives against their obligations in the simulation, and that balance is itself a kind of stake.
Two answers follow. The first is organisational. Agents may need to find other stakeholders who can cover the inconvenient shifts, as an organisation with members in several time zones does. Run around the clock, a single simulated spaceship may need three shifts of real people, as a real ship at sea or a mission control room does. NASA’s payload operations centre for the space station “is staffed 24 hours a day, 365 days a year, by three shifts of flight controllers” NASA 2019. That sounds absurd to anyone who would never touch such a simulation, and it is exactly what its most committed agents would organise.
The second answer is a mechanism that resolves events from orders committed in advance. The mechanism has two forms, and each has a cost. In the first, the event still happens when the orbits say, and the craft acts on standing orders with nobody at the controls. That keeps the shared present and takes the person out of the moment, so it yields evidence of how people plan and none of how they act under fire. In the second, the simulation computes the outcome early for craft whose agents have all committed, which locks those craft from that moment and is a private timeline of the kind the Kerbal Space Program modifications use. It has a further cost. The committed agents learn an outcome before the shared timeline reaches it, and they can act on that knowledge elsewhere. Neither form moves an event. What an agent can choose, at the time of planning, is a trajectory whose decisive moment falls within waking hours. An attacker has the opposite incentive and will choose the hours when the defender sleeps, as raiders always have. Committed orders suit the physics, since a planned transfer is already a committed order.
Encounters. True scale brings one more difficulty, which this article only names. A reader may take the fast pass for the rare event. It is not. In a crowded shell of low orbit, craft cross within range of one another often, at the closing speeds worked earlier. Let $\nu$ be the rate at which one craft passes within range $R$ of another, $N$ the number of craft in a shell of volume $V$ and $v_{\mathrm{rel}}$ their typical relative speed.
\[\nu = \frac{N}{V}\, \pi R^{2}\, v_{\mathrm{rel}} \tag{27}\]Ten thousand craft in a shell 200 kilometres thick at the height of the space station, with a range of 100 kilometres and a relative speed of 10 kilometres per second, give each craft a pass about every 37 seconds. The figures are illustrative, and the estimate treats the craft as moving at random, which craft in orbit do not. What honest delta-v budgets make rare is a meeting slow enough to trade, board or talk, since one craft must pay to match the other’s orbit, and any meeting at all between craft in different regimes. The western’s problem returns. Adventures need agents in the same place at the same time, and honest physics makes that expensive. Meetings would concentrate where traffic must pass, in low orbits, at stations and in the weeks around a transfer window. Whether that is enough to sustain a shared world is open, and a later article in this series takes it up.
The risk. The problem of time is where the thesis is most at risk, and the article should say so plainly. True travel times may make multi-agent scenarios that the general public finds palatable intractable. Or the problem may be hard enough that solving it is a questionable exercise for a studio that must earn a return on its investment. Most studios must, because for a studio the return on investment is generally an existential problem. The claim that such a simulation is possible is a claim about physics and engineering. It is not a claim that building one would pay.
The Agents
The agents are the other half of the design. Four choices define them.
- Real people, in roles. As in EVE Online, agents are people, so the simulation bounds their interactions and does not simulate them. As in Star Citizen, a crew may fly a ship, and the roles are asymmetric, among them navigation, propulsion, trade, mining, construction, science, logistics and defence.
- One shared world. A single persistent world, as EVE Online demonstrated is possible for a large population, so that the consequences of one agent’s actions reach the others.
- A real market. An economy of real agents, measured and reported as EVE Online’s is, in which propellant, delta-v and time have prices. In such a market, the rocket equation becomes an economic fact that every agent learns by paying it.
- Automated agents. Programs may fly craft for people, provided they obey the same physics and the same light time. The predictions the simulation offers concern its human agents, so automated agents are tools of those people, not subjects of the experiment.
An Economy Under Physics
EVE Online shows that agents will build and run a supply chain that spans a world. The blueprint keeps that and puts physics under it. Delta-v prices freight. With $m_{\mathrm{p}}$ the propellant needed, $m_{\mathrm{pay}}$ the payload, $m_{\mathrm{s}}$ the structure and engines that make the trip with it, and the other symbols as in the rocket equation,
\[\begin{aligned} m_{\mathrm{p}} = {} & \left( m_{\mathrm{pay}} + m_{\mathrm{s}} \right) \\ & \times \left( e^{\Delta v / \left( I_{\mathrm{sp}}\, g_{0} \right)} - 1 \right) \end{aligned} \tag{28}\]At a specific impulse of 450 seconds, a delta-v of 4 kilometres per second needs about 1.5 kilograms of propellant for every kilogram that makes the trip. Geography becomes orbital mechanics. A cheap route is a low delta-v route, and the synodic period worked in the catalogue opens the cheapest route to Mars once every 26 months, so an honest economy would have trading seasons, with prices rising and falling as windows open and close. Prices inside the simulation would also travel at the speed of light. A quote from another planet is minutes old when it arrives, so a trader there acts on the past. That holds only for an agent with no channel outside the simulation, the leak that the light-time rule already declares. The arbitrage condition from the review of EVE Online’s market still governs every haul, with the propellant now part of its cost. Congestion pricing of the kind EVE Online applies to production fits scarce real facilities too, such as docking ports and shipyards in a crowded orbit.
The simulation should also publish its economy. EVE Online already does. Its economists report on the money supply every month CCP 2026, and its official interface for third parties, which the developer calls “the official RESTful API for EVE third-party development,” serves every region’s market orders and price history, some of its routes needing no authentication at all ESI Overview. Anyone can study the economy from outside. A predictive simulation needs that openness, because evidence nobody can check is not evidence. Openness and light time pull against each other, since a public feed of every market would let agents see prices before the light arrives. So the simulation must publish its record late, by at least the longest light time in the world, or after the fact.
Approachability Without Miseducation
The design answer to Children of a Dead Earth’s difficulty is layered disclosure. Interface designers know the idea well. Nielsen describes progressive disclosure, which “defers advanced or rarely used features to a secondary screen, making applications easier to learn and less error-prone” Nielsen 2006. An early experiment blocked off the advanced functions of a word processor for beginners and found that the training version “afforded substantially faster learning coupled with better learning achievement” Carroll and Carrithers 1984. The physics stays complete. What changes from one agent to the next is how much of it the agent must handle by hand. A beginner sets a destination, and a planning tool computes the transfer, the delta-v and the arrival time, and explains why. An expert inspects and edits every burn. In one shared world, agents cannot each choose their own physics, as Kerbal Space Program’s modifications allow a single agent to do. What each agent chooses is how much of the one model to handle by hand. The planning tool may think in patched conics, which are accurate enough for a first plan, as the review of Kerbal Space Program noted, provided it then shows the correction that the full model makes. The tool can be simple while the world stays true.
Layered disclosure has a cost in a shared world that it does not have in a sandbox. A beginner who relies on the planning tool competes with experts who edit every burn, and by the fourth principle what the beginner loses stays lost. The account of the meta in the review of Children of a Dead Earth predicts who wins. The blueprint does not solve this. It records it as an open problem of design, in which approachability and lasting loss pull against each other. Layered disclosure also bears on teaching. A beginner who hands every burn to the tool may learn little physics from it. That such a simulation would teach is a hypothesis, the mirror of the one about miseducation, and no measurement found here supports or refutes it. Kerbal Space Program’s own planning tool works in this spirit. Its wiki describes a manoeuvre node as “a planned velocity change along an orbit” that “shows the velocity change needed to reach the next new orbit” Maneuver Node. Kerbal Space Program’s sales and its use in schools suggest that the public will take up planning tools of this kind, though no measurement of what its users learn came to light. An autopilot that obeys physics is an approachability tool. A warp drive that disobeys it is miseducation.
Predictive Value
This section says why such a simulation should have predictive value, what evidence bears on the claim, what bounds it and what lies beyond its reach. It claims indicative evidence, not prophecy.
People adapt their behaviour to conditions. If the conditions of a simulation are those of real near-term space, the agents’ adaptation to them is evidence about how people would behave there, in markets, in logistics and in conflict.
The mechanism. A sufficiently high-fidelity multi-agent simulation should have significant predictive value because its agents are subject to the same incentives and constraints that reality imposes. People who face the same costs, the same delays and the same scarcities tend toward the same choices, whoever they are. The history of wargaming offers an example. Between the world wars the United States Naval War College gamed a war against Japan, with its own students taking the Japanese side. A historian of those games observes that “The fact that the Japanese players used tactics in games resembling those they actually used during the Pacific War demonstrates that circumstances rather than some form of perceived Japanese military culture determined much of what they did” Friedman 2017. The students behaved as the Japanese later did because they stood where the Japanese later stood.
The scenario. The claim hinges on the scenario. The incentives and constraints of the simulation must be close enough to those of real prospective problems. Where they are not, the agents adapt just as faithfully to the wrong conditions. Predictive value is largely a function of two abilities, to formulate a realistic scenario and to find agents willing to take part in exploring it. The first is the work of fidelity and the second of approachability, which is why the bounds below begin with those two. The Scenarios section below says how the blueprint supplies the first.
The same demands hold outside the armed forces. Herman, Frost and Kurz describe wargames designed and staged for the United States Army, for global corporations and for nonprofit groups Herman et al. 2009. A reviewer of the book notes that its authors “several times offer caveats about the predictive value of all this” while holding that such exercises “do allow plausible scenarios to be explored” Brynen 2010. The author’s reading of such accounts is that a wargame is only as good as its scenario and the willingness of the right people to come to it.
The agents. The competence of the agents matters less than it might seem. Most people are of average ability for their station. That holds for the participants in a wargame, and it holds for the people who occupy the real stations that the wargame simulates. People of clearly less than average ability take part in simulations too, and such people also hold real stations. Brilliant people are few, in a simulation as in the world, and their scarcity matters little, because circumstance and station drive a great many decisions whoever makes them. An agent needs only enough competence to do something reasonable, and a competent person of less than average ability asks for advice.
The United Kingdom’s defence handbook on wargaming sets that floor. By its account, “Wargames are only as good as the participants” and “An uninformed, unqualified or overconfident wargame team is unlikely to add value” UK Ministry of Defence 2017. Competence for the station is the floor, and nothing in the handbook asks for brilliance above it. A station here means a role together with the selection that fills it. The claim is not that anyone will do. It is that a simulated station filled by people chosen roughly as its real holders are, or by people of ordinary competence, yields decisions like the real ones, because the station leaves less room for talent than it seems to.
The games at Newport are consistent with this, and they do not prove it. Their participants came from the body of officers that went on to command the fleet. The same historian records that “Of the flag officers commanding combat formations in the 1941 fleet, only one had not attended the War College” Friedman 2017. Newport shows circumstance driving the decisions of selected professionals. It does not show that ability is unimportant, and the agents of a public simulation are volunteers, not selected officers. Fleet Admiral Nimitz told the college in 1960 that “the war with Japan had been re-enacted in the game rooms here” “in so many different ways that nothing that happened during the war was a surprise” except the kamikaze Hattendorf 2014. The exception is a limit of the argument from circumstance, since the one surprise was the behaviour that circumstance did not predict.
The evidence and its limits. Research on forecasting points the same way, on thinner evidence than the claim would like. Green and Armstrong compared forecasts of the decisions made in real conflicts and found that “The forecasts of experts who used their unaided judgment were little better than those of novices.” A form of role playing did far better. It reduced error by 47 percent against the forecasts of game-theory experts, and its participants “were mostly undergraduate students” Green and Armstrong 2007. Armstrong’s review of role playing drew the same lesson about who takes each role, that “the little evidence available suggests that casting is not critical” Armstrong 2001. That research concerns small groups acting out a single decision, not a wargame.
The profession of wargaming itself warns against the word predict. An article in the college’s own review holds that “a war game can never be truly predictive” and that what a game teaches, “being merely indicative in itself, must be combined with judgment in order to have useful predictive value” Rubel 2006. A history of the method from the college’s own press concludes that “the large majority of games do not correlate to later events” Caffrey 2019. The British handbook says flatly that “Wargames are not predictive” and that they “are good at showing when something is plausible (as opposed to probable)” UK Ministry of Defence 2017.
The author’s claim is the narrower one that this record supports. Participants of average ability for their station, placed under realistic incentives and constraints, have produced conclusions with surprising predictive potential. That the result depends on the scenario at least as much as on the talent in the room is the author’s reading of limited evidence, and no source quoted here compares the two directly.
The wider claim, that behaviour in a simulated world can map to behaviour in the real one, is not new, and research on virtual worlds supports it with conditions. Castronova and colleagues tested whether aggregate economic behaviour maps from the real world to a virtual one, using the transaction records of a large commercial virtual world, and reported that “virtual economic behavior follows real-world patterns” Castronova et al. 2009. They also found the virtual economy less stable than a real one, and wrote that “virtual economies are not perfect analogs of real economies at the aggregate level.” Two epidemiologists made a similar argument after “the accidental inclusion of a disease-like phenomenon” in an online virtual world, proposing that such worlds could help to validate epidemic models and supply what they lack, namely “important human behaviours” Lofgren and Fefferman 2007.
Williams named the condition on which all such work depends, a “mapping principle,” defined as “the extent to which behaviors in one space are consistent with behaviors in another” Williams 2010. Williams is also the sharpest critic of the idea. The same paper counts the epidemic study among those that “took the mapping principle on faith” and warns that “the risks and rewards of virtual worlds do not always map to those offline. The most obvious example is pain and death.” None of that work concerns spaceflight. It supports the general claim that behaviour in a simulated world can map to behaviour in the real one, and it insists that researchers show the mapping and not assume it. The three bounds that follow give an account of what the mapping depends on.
Three bounds. The first bound is fidelity, because agents adapt to the model, not to reality, as Children of a Dead Earth’s meta shows. Castronova and colleagues reached the same caution from the other side, suggesting that “macroeconomic outcomes in a virtual world may be explained largely by design structure.” Fidelity includes the scenario. A simulation must posit technologies, prices and starting conditions that the real future will not copy, and whatever it yields is conditional on them. Time compression belongs under this bound too, for the reason the Time section gave.
The second is approachability, because a model only an expert can use yields the behaviour of a small and unrepresentative population of experts. The design for time pulls against this bound, since agents who organise three shifts for one ship are a small and unrepresentative population of another kind.
The third is incentives. An agent who risks nothing it values has no reason to behave as it would where the stakes are real. Wargamers know the effect, and Rubel notes that “players tend to be more aggressive than they would be in the real world with real lives at stake” Rubel 2006. The long voyage that costs an absent agent nothing, noted in the Time section, is a bias of the same kind.
A simulation that maximises one bound at the expense of the others predicts poorly outside the conditions it happens to model well. The blueprint’s principles exist to raise all three together. The published wargames reviewed elsewhere on this site show the same dependence, since their value rests on how faithfully their rules capture the conditions they model Published Wargames.
EVE Online as the test. EVE Online shows that a simulation can approach the third bound. Some of its agents take it more seriously than their real jobs, and a saying among them holds that internet spaceships are more fun than work. People who defend a ledger of fantasy wealth that fiercely are displaying the incentive feedback that a predictive simulation needs, even if the enthusiasm is questionably placed. The review of its economy recorded how. A destroyed ship is gone, and the exchange rate puts a price in real money on what its owner lost. The author has placed significant effort in the EVE Online sandbox, questionably, despite being a fairly tempered participant.
The same case shows what high commitment brings with it. A multi-agent simulation that people take seriously can hold its agents to wildly high expectations, and what they learn there carries over into their working lives, as the review of EVE Online recorded. Burnout is just as real. A press account of EVE Online’s wars says that their pace “can cause massive stress on a player’s life, and can lead to a lot of fatigue, and over time, quite a bit of burnout” Yancy 2017. Obligations that run around the clock wear people down, and a simulation that depends on committed agents loses them when the commitment costs more than it returns. Yet a simulation compelling enough will still draw people to participate of their own volition, and that voluntary commitment is what the third bound asks for.
EVE Online adds a refinement. Its predictive value comes less from the fidelity of any component than from how comprehensively the components interlock, so terrible parts produce tax havens and embezzlement. It is an inference that predictive value has two sources, the fidelity of the parts and the completeness of their interlock. Interlock is not a fourth bound. It is the first bound seen at the level of the system, since the couplings between the parts are themselves part of the model. EVE Online is strong in the second source and weak in the first, and Children of a Dead Earth the reverse.
That raises an obvious question. If EVE Online reproduces real behaviour without honest physics, why does physics matter at all? The answer is that predictive value extends only to the conditions a simulation models. EVE Online models an economy, a polity and a society of real people, so its evidence concerns economic, political and social behaviour, and there its fidelity is high, because the market and the people are real. It models no delta-v, no light-time and no launch windows, so it gives no evidence of how people would behave under them. A blockade that physics makes impossible, a market that the synodic period opens and closes, an order that arrives twenty minutes late, and a war decided by who can afford the propellant to reach the battle lie outside its scope, however good its economy. A simulation for near-term space needs EVE Online’s interlock and honest physics together, because the questions it exists to answer are the ones the physics shapes.
One caution applies to the whole case. What EVE Online has demonstrated is reproduction. Behaviour known from real economies reappeared in it, and nobody designed it. A model that reproduces known behaviour unprompted has passed the first test of a predictive model. It has not yet made a prediction that was later confirmed, and neither has anything proposed here. The gap is widest exactly where the physics adds something. Behaviour under delta-v, light time and launch windows has no real counterpart yet to compare with, so nobody can show the mapping that Williams requires until people live under those conditions.
What lies outside the design. One limit lies outside the design altogether. The agents of any such simulation are volunteers with leisure. The actors of real near-term space are states, firms and crews, who act under law and at mortal risk. The simulation could show how people adapt to physical and economic conditions. It could not show what institutions would permit.
Two practical problems. The wargaming record also names two practical problems that such a simulation would face. The first is getting the stakeholders who would most benefit to take part. Perla and McGrady hold that a game’s effect on “key decision makers” is an essential source of an organisation’s adaptation Perla and McGrady 2011, yet “Blue players are typically junior to the individuals who will make the actual decisions” Caffrey 2019.
The second is the reflex to wave away a result that nobody expected. Rubel calls such results whispers, “tempting to dismiss” though they are, and holds that “these whispers are frequently the most important outcomes of war gaming.” The Japanese games before Midway are the standard case. The part usually told is an umpire’s reversal of a lucky hit, and Rubel judges that ruling justified. The failure lay elsewhere. Someone asked what would happen if an American carrier force ambushed the Japanese carriers, and “that uncomfortable question seems to have been ignored” Rubel 2006. Sixty years later the United States did much the same in its Millennium Challenge exercise. The official report records that the free play of the opposing force “was eventually constrained to the point where the end state was scripted” Joint Forces Command, and the general who led that force recalled that “A phrase I heard over and over was: ‘That would never have happened,’” Borger 2002.
A simulation of the kind proposed here inherits both problems. Those with the most to learn from it, the agencies, firms and investors who will make the real decisions, are the least likely to spend their evenings in it. And its most valuable results will be the ones that look wrong. A blockade that fails, a market that seizes or a colony that its own agents abandon will invite the verdict that the simulation is unrealistic, when the purpose of building it was to find out.
Scenarios
A persistent world is one run of one scenario. A scenario is a stated set of starting conditions, namely the technology agents may build, what each agent begins with, what the world beyond the simulation buys and sells, and what rules stand in for law. The thesis makes the formulation of realistic scenarios the main source of predictive value, so the blueprint must say where scenarios come from.
One history is thin evidence. The United Kingdom’s handbook on wargaming warns of “a risk of false lessons being identified from a single run of a wargame” UK Ministry of Defence 2017. The games at Newport earned their reputation by replaying one war many times over, and the research on role playing rests on many short sessions. A single persistent world yields a single path through its possibilities, however many agents walk it. The blueprint therefore treats a world as one instance of a scenario among several. The same starting conditions can open more than one world, at once or in succession, and a variant can change one condition and hold the rest. The reproducible step and the published record make such worlds comparable, since the starting state of each is known exactly. What several worlds agree on is worth more than what one of them shows.
A scenario also has a span. At a compression of ten, a world that runs for one real year covers ten simulated years, and at one hundred it covers a century. A world that stood as long as EVE Online has, some twenty-three years, would cover 230 simulated years at a compression of ten. Nobody can plausibly hold near-term technology fixed across such a span, so a world needs an end, or a stated rule for how its technology advances. An end has its own cost. Loss that lasts only until the world closes is a weaker stake than loss that lasts for good, and agents who know when the end comes will behave differently as it nears. Lasting loss and repeatable scenarios pull against each other. The blueprint records that tension and does not resolve it.
Market Demand
The thesis holds that there is market demand for such a simulation, and the reviewed products support the claim only indirectly. Kerbal Space Program sold more than five million units Take-Two 2023. EVE Online passed half a million subscribers in one shared world in 2013 CCP 2013b. Star Citizen’s counter shows about a billion dollars raised before release, if its units are cents RSI Crowdfund Statistics. The public pays for spaceflight, for shared worlds and for both together. It is an inference that it would pay for both with honest physics.
The inference is the weakest link in the thesis, and two objections need stating. The audiences of EVE Online and Star Citizen may be paying for exactly the conveniences that honest physics removes, so nobody can simply add their numbers to Kerbal Space Program’s. And Kerbal Space Program 2, the one known attempt, lost its studio before it delivered its multi-agent mode. That closure shows a failure of execution. It cannot show an absence of demand, since the product that would have tested the demand never shipped, and it cannot show its presence either. The claim therefore stands as an inference that no product has yet tested.
It is the market half of the thesis, and it ties to the educational half. A product the public will not use is a poor educational utility, however good its physics, as Children of a Dead Earth shows. A product the public does use teaches whatever physics it contains, as EVE Online shows. The opportunity is a product the public will use whose physics is true.
Equations for an Implementation
This subsection is a reference for implementers, and a reader interested only in the argument can pass to the Epistemic State section. The physical core needs a small set of equations, and an implementer needs them in one place. An earlier article on this site works the manoeuvres an agent plans with numbers, among them the Hohmann transfer, plane changes and the Oberth effect Orbital Manoeuvres. The equations below are the ones the simulation itself must integrate.
Gravity and thrust. Each craft moves under the gravity of every massive body and its own thrust. In an inertial frame, with $\ddot{\mathbf{r}}_{i}$ the acceleration of craft $i$, $G$ the gravitational constant, $m_{j}$ and $\mathbf{r}_{j}$ the mass and position of massive body $j$, and $\mathbf{a}_{\mathrm{T},i}$ the acceleration of its own thrust,
\[\ddot{\mathbf{r}}_{i} = -G \sum_{j} m_{j} \frac{\mathbf{r}_{i} - \mathbf{r}_{j}}{\lvert \mathbf{r}_{i} - \mathbf{r}_{j} \rvert^{3}} + \mathbf{a}_{\mathrm{T},i} \tag{29}\]Battin 1999 Hughes 2007. The sum runs over the planets and moons on their ephemerides, so a craft feels every body at once, and Lagrange points and perturbations appear unscripted. Thrust consumes propellant. Let $\dot{m}$ be the rate of change of the craft’s mass and $F$ the thrust.
\[\dot{m} = -\frac{F}{I_{\mathrm{sp}}\, g_{0}} \tag{30}\]NASA Specific Impulse. An engine of 100 kilonewtons at a specific impulse of 450 seconds burns about 22.7 kilograms of propellant each second. Integrated over a burn, this equation is the rocket equation of the catalogue.
Binding to a parent body. The same motion has a second form, relative to one body, the craft’s parent. With $\mathbf{r}$ the position of the craft relative to the parent, $\mu_{\mathrm{p}}$ the parent’s gravitational parameter, $\mu_{k}$ and $\mathbf{r}_{k}$ the gravitational parameter of another body $k$ and its position relative to the parent, and $\mathbf{a}_{\mathrm{T}}$ the acceleration of thrust,
\[\begin{aligned} \ddot{\mathbf{r}} = {} & -\frac{\mu_{\mathrm{p}}}{\lvert \mathbf{r} \rvert^{3}}\, \mathbf{r} + \mathbf{a}_{\mathrm{T}} \\ & + \sum_{k} \mu_{k} \left( \frac{\mathbf{r}_{k} - \mathbf{r}}{\lvert \mathbf{r}_{k} - \mathbf{r} \rvert^{3}} - \frac{\mathbf{r}_{k}}{\lvert \mathbf{r}_{k} \rvert^{3}} \right) \end{aligned} \tag{31}\]Hughes 2007. The first term in the bracket is the pull of body $k$ on the craft. The second is its pull on the parent, which the equation needs because, as one account puts it, “the integration frame, which has its origin at the center of mass of the primary body, is not inertial” Berry and Coppola 2007. The two terms nearly cancel, and what remains is a tide. For a craft at distance $r$ from its parent and a third body at the much greater distance $d$, the size $a_{k}$ of the bracket has a bound.
\[a_{k} \lesssim \frac{2 \mu_{k}\, r}{d^{3}} \tag{32}\]A craft at the height of the space station feels 8.64 metres per second squared from Earth. With NASA’s values for the other bodies NASA Moon Fact Sheet NASA Sun Fact Sheet NASA Jupiter Fact Sheet, the Moon adds at most about $1.2 \times 10^{-6}$ metres per second squared, the Sun about $5.4 \times 10^{-7}$, and Jupiter at its nearest about $8 \times 10^{-12}$. Those are about 1.4 parts in ten million of Earth’s pull, 0.6 parts in ten million and one part in a trillion. A simulation that keeps every perturbation above one part in a billion gives that craft three bodies to feel in place of thirty, and the threshold is a declared omission of known size. The tidal bound is an approximation made here, good to first order in the ratio of the two distances.
The step. A symplectic method can advance the equations of motion of a coasting craft in fixed steps. The simplest is the leapfrog in its kick-drift-kick form. Let $\Delta t$ be the step, $\mathbf{v}$ the velocity and $\mathbf{a}(\mathbf{r})$ the acceleration at position $\mathbf{r}$.
\[\begin{aligned} \mathbf{v}_{n+1/2} &= \mathbf{v}_{n} + \tfrac{\Delta t}{2}\, \mathbf{a}(\mathbf{r}_{n}) \\ \mathbf{r}_{n+1} &= \mathbf{r}_{n} + \Delta t\, \mathbf{v}_{n+1/2} \\ \mathbf{v}_{n+1} &= \mathbf{v}_{n+1/2} + \tfrac{\Delta t}{2}\, \mathbf{a}(\mathbf{r}_{n+1}) \end{aligned} \tag{33}\]This second-order method is the base from which Yoshida built higher orders Yoshida 1990. For a craft coasting about one dominant body, its energy error stays bounded instead of drifting, so the orbit does not slowly spiral in or out over a long simulation. Hairer, Lubich and Wanner show the behaviour on the two-body problem, where “there are only small errors in the Hamiltonian along the numerical solution, but no drift” Hairer et al. 2003. The property has conditions. It holds for a constant step, and it bounds the error in energy, not the error in position along the orbit. Among moving planets a craft’s energy is not constant even in exact physics, a fact that a gravity assist exploits, so the result does not carry over unchanged. A close flyby needs a shorter step than a cruise, and a burn or a pass through air is not a conservative force. Changing the step is itself a hazard, since “Adaptive step size control is difficult to combine with geometric numerical integration” Hairer and Söderlind 2005, though the same authors give a reversible rule that keeps the long-term behaviour. The choice of integrator stays open. Whatever the choice, every machine must apply the same rule for choosing steps, and a later article in this series treats it.
Determinism. A shared world does not strictly need a reproducible step. One authoritative server can compute the world and tell every client the answer, and nothing is simpler. The blueprint asks for reproducibility anyway, for two reasons. A predictive simulation should be open to checking, and a world that anyone can replay from its recorded inputs is. And a planning tool that runs the same step as the world forecasts exactly what the world will do, for as long as no other agent intervenes. The idea is old in networked simulation. The developers of Age of Empires wrote in 2001 that “the expectation was to run the exact same simulation on each machine,” and that a recording of the commands could then replay a whole session Bettner and Terrano 2001. With $\mathbf{x}_{n}$ the full state of the world at step $n$, $U_{n}$ the set of agents’ inputs committed for that step in an agreed order, and $\Phi_{\Delta t}$ the step function,
\[\mathbf{x}_{n+1} = \Phi_{\Delta t}\!\left( \mathbf{x}_{n}, U_{n} \right) \tag{34}\]Every machine that starts from the same state and applies the same ordered inputs reaches the same state, provided the step function is exactly reproducible. That proviso is the difficult one. Floating-point arithmetic that a compiler may reorder or fuse can make two machines disagree in the last bit, and an n-body system amplifies the difference. Monniaux’s survey of the problem warns that “The result of computations can actually depend on compilation options or compiler versions,” and that “one should not expect consistent behaviour of transcendental functions across libraries, processor manufacturers or models” Monniaux 2008. The Age of Empires developers met the same thing in practice, where “very subtle differences would multiply over time” Bettner and Terrano 2001. An implementation therefore fixes the order of every operation, including those inside library functions and across processor cores, or uses fixed-point arithmetic for the state. The requirement applies to the machines that compute the world. It does not extend to the agents’ own machines, which must not hold what their sensors cannot see. This equation is the article’s statement of that requirement, and a later article in this series treats it in detail.
Attitude. Pointing is a separate problem from moving. With $\mathbf{I}$ the craft’s inertia tensor, $\boldsymbol{\omega}$ its angular velocity and $\boldsymbol{\tau}$ the torque of its reaction control thrusters, Euler’s equations of rigid-body rotation give
\[\mathbf{I}\, \dot{\boldsymbol{\omega}} + \boldsymbol{\omega} \times \left( \mathbf{I}\, \boldsymbol{\omega} \right) = \boldsymbol{\tau} \tag{35}\]Stoneking 2007. The torque changes where the craft points. It does not change where the craft goes. The remedy for banking turns now has its equation.
Atmosphere. Launch and reentry need drag and heating. Let $F_{\mathrm{D}}$ be the drag force, $\rho$ the density of the air, $v$ the speed, $C_{\mathrm{D}}$ the drag coefficient and $A$ the reference area.
\[F_{\mathrm{D}} = \tfrac{1}{2}\, \rho v^{2} C_{\mathrm{D}} A \tag{36}\]NASA Drag Equation. A simple model of the density, with $\rho_{0}$ its value at the surface, 1.217 kilograms per cubic metre, and $H$ the scale height, 8.5 kilometres NASA Earth Fact Sheet, is
\[\rho(h) = \rho_{0}\, e^{-h/H} \tag{37}\]Garcia-Llama 2007. Heating at the nose follows the stagnation-point correlation of Sutton and Graves Sutton and Graves 1971. With $\dot{q}$ the heat flux, $r_{\mathrm{n}}$ the radius of the nose and $k$ a constant for the atmosphere, about $1.74 \times 10^{-4}$ in units that give watts per square metre for Earth’s air,
\[\dot{q} = k \sqrt{\frac{\rho}{r_{\mathrm{n}}}}\, v^{3} \tag{38}\]At $10^{-4}$ kilograms per cubic metre, a speed of 7.5 kilometres per second and a nose radius of one metre, the flux is about 0.73 megawatts per square metre. The flux falls as the nose grows, which is Allen and Eggers’s argument for blunt bodies in a single term.
Heat. A craft’s temperature follows its energy balance. With $m$ its thermal mass, $c$ its specific heat, $T$ its temperature, $P_{\mathrm{in}}$ the power it generates and absorbs, and the radiator symbols as before,
\[m c\, \frac{dT}{dt} = P_{\mathrm{in}} - \varepsilon \sigma A T^{4} \tag{39}\]A craft that generates more heat than its radiators reject grows hotter, and that constraint makes heat a resource.
Slosh. Where a subsystem needs slosh, the standard practice is an equivalent mechanical model, in which the moving liquid behaves as a pendulum attached to the tank Abramson 1966. Let $\omega_{\mathrm{s}}$ be the natural frequency of the first slosh mode, $a$ the acceleration along the tank’s axis and $L$ the length of the equivalent pendulum.
\[\omega_{\mathrm{s}} = \sqrt{\frac{a}{L}} \tag{40}\]Under a thrust of a tenth of standard gravity, a one-metre pendulum swings at about one radian per second, a period of about six seconds, slow enough to couple into a docking manoeuvre. In a coasting craft the acceleration is nearly zero, the restoring force vanishes, and the liquid wanders. Hence a small ullage thrust settles it before an engine starts.
Clocks. A clock on a moving craft does not keep the time of a clock elsewhere. With $t_{\mathrm{c}}$ the time a craft’s own clock shows, $t$ the time of a clock at rest far from every mass, $\Phi$ the Newtonian gravitational potential at the craft, which is negative, $v$ the craft’s speed and $c$ the speed of light, the weak-field result that satellite navigation relies on is
\[\frac{d t_{\mathrm{c}}}{d t} = 1 + \frac{\Phi}{c^{2}} - \frac{v^{2}}{2 c^{2}} \tag{41}\]Ashby 2003. Ashby writes it with the reference clock on Earth’s surface, and observes of the Global Positioning System, GPS, that without such corrections “the system would not work.” NASA puts numbers on it. Uncorrected, “GPS satellite clocks would gain ~38 microseconds per day relative to clocks on the ground (~11 km range error)” Winternitz 2017. For a craft coasting at Earth’s distance from the Sun at 30 kilometres per second, the speed term is five parts in a billion and the term for the Sun’s potential is about ten, so its clock loses about half a second in a year against the distant clock. A clock on Earth loses nearly as much, and what a navigator must track is the difference between the two, which changes as the craft’s speed and distance from the Sun change. A crew would never notice it. The simulation keeps a clock for each craft all the same. The shared timeline of the world is the time $t$ of this equation, and each craft’s own clock is $t_{\mathrm{c}}$. Leaving the effect out would be an omission and not a contradiction, but the cost of including it is one line in the step. The clock binds the craft, not the person behind it, who lives by the wall clock, and that limit is of the same kind as the one on light time.
Epistemic State
The blueprint ends there. This section sorts what the article has claimed by the support each claim has. It lists the facts whose sources are weaker than a developer’s or an agency’s own page, then the author’s judgements, then the inferences, then the constructions and hypotheses, and it closes with the limits of the reconstruction.
Facts and their sources. The descriptions of each product’s models, travel, propellant, gravity, atmosphere, sales and subscribers come from the developers’ own pages, publishers’ releases and space agency pages cited in place. Where a fact rests on weaker ground, the text says so. The main cases follow.
- Community wikis and secondary reports. These support Kerbal Space Program’s scale, the account of how its multi-agent modifications handle time, EVE Online’s velocity law, its warp speeds and its capacitor, the base broker’s fee in EVE Online and the reception of Kerbal Space Program 2.
- Press reports. The 2009 bank embezzlement and the technetium cartel rest on press reports and a participant’s account. The controversy around Star Citizen rests on press reports of sentiment.
- A community wiki and the author’s own observation together. The author has watched EVE Online’s ships slow to rest, and recalls noting that Star Citizen’s celestial bodies were not subject to celestial mechanics.
- The developer’s counter. Star Citizen’s crowdfunding total comes from the developer’s own counter, whose units the review takes to be cents.
- Arithmetic on stated figures. The Cloud Imperium incomes for 2020 to 2022 and all five spending ratios are sums and quotients of stated figures.
- A sponsored article. The survey on transferable skills appeared in an article sponsored by the developer.
- A fan transcript. The Star Trek line about tritanium is from the episode as aired, located through a fan transcript.
- Abstracts. Of the studies of fiction, misconception and virtual worlds, the article cites two from their abstracts alone, Clement 1982 and Lofgren and Fefferman 2007. The summary of Clement’s finding, that motion implies a force, comes from a later review.
- Scanned documents. The equation of motion relative to a parent body comes from a scanned NASA specification. Three more standard equations come from scanned NASA documents, namely the period of an orbit, Euler’s equations and the special-relativistic clock rate. The mathematics of all four comes from page images, where a transcription can carry an error, and a reader can check each against its cited source.
- Memory and a remark. The account of Herman, Frost and Kurz rests on the catalogue record of their book, on a published review and on the author’s memory of reading it, since the author did not have its text to hand. That people have run Traveller in real time rests on a remark the author overheard in person and recalls imperfectly, which no source here supports.
- The physics of the remedies. The physics of each remedy comes from NASA, ESA, the Fédération Aéronautique Internationale, textbooks or peer-reviewed work, with three exceptions named in the text and a few statements that follow from first principles and carry no citation. The visibility of laser beams rests on an optics manufacturer, the claim that there is no stealth in space on a simulation developer, and the weakness of explosions on the same developer and a nuclear effects handbook.
The author’s view. Everything not cited and not named as an inference is the author’s judgement. The author offers these judgements as open to disagreement on the evidence.
- The thesis and the blueprint. The thesis and the blueprint are the author’s. So are the claim that ability average for a station suffices for agents, the view that predictive value is largely a matter of realistic scenarios and willing agents, and the treatment of a world as one instance of a scenario that its builders can open more than once. That participants in wargames are of average ability for their station is the author’s argument. Its support here is the record that the Newport participants went on to hold the commands they had gamed, and research on role playing by students, which is not wargaming.
- The four products together. The author judges that each product’s model serves its product thesis. The judgements on approachability, education and management are also the author’s, as is the expectation that no reviewed product simulates propellant slosh.
- Kerbal Space Program and Children of a Dead Earth. The author understands Kerbal Space Program to treat tanks as point masses and judges that Kerbal Space Program 2 was mismanaged. The observations about engagements and the meta in Children of a Dead Earth are the author’s.
- EVE Online. The author characterises its individual models as wrong and judges its economy the best multi-agent supply chain known. The author also judges that its wrong models have the right fidelity for interaction, and conjectures that the developer lacked the people to distil more realistic models. The characterisation of its most committed agents, their scheduling across time zones, the transfer of skills and burnout is the author’s. The author discloses an interest in this product, having participated in the EVE Online sandbox.
- Star Citizen. The author judges that it aspires to a higher-fidelity version of EVE Online’s thesis and is directionally promising, that its sales model is grey and that its project is mismanaged.
- The conventions. The training of writers as a cause of the soft science fiction conventions is the author’s view.
- The estimates. The estimate of what the gravity model costs to compute and the tidal bound on a third body’s perturbation are the author’s.
Inferences. The following are inferences.
- From the format, warp drive served Star Trek’s production as the transporter did. No primary source says it.
- From the resemblance, from Star Citizen’s stated aim and from one borrowed name, the two multi-agent products took their conventions from film and television.
- From the developer’s descriptions, EVE Online has the product thesis that the review attributes to it.
- An agent who flies for years under a model that contradicts inertia is practising a misconception.
- Time compression moves a simulation away from the conditions it means to predict.
- A shared world needs a shared clock.
- Layered disclosure would ease the conflict between fidelity and approachability.
- Several characters per agent would make long journeys workable.
- Resolution of events from committed orders would ease scheduling across time zones.
- Predictive value has two sources, the fidelity of the parts and the completeness of their interlock.
- Predictive value extends only to the conditions a simulation models.
- An honest economy would have trading seasons and prices that travel at the speed of light.
- Research on role playing by small groups carries over to a simulation of thousands.
- Voyages begun before an absence model individuals who travel for private reasons.
- The public would pay for honest physics. This inference is the weakest, and no product has tested it.
Constructions and hypotheses. The thirteen equations that the Equations section marks as bespoke are constructions made here, and the engagement-time, radiator, gravity-cost and pass-rate equations among them show orders of magnitude only. What EVE Online has demonstrated is reproduction of known behaviour, not a prediction later confirmed. The miseducation that the article warns of is a hypothesis with indirect support.
The reconstruction. The author reconstructed the thesis from memory. Where this article’s wording differs from the paper’s, nobody can consult the paper to settle it. This reconstruction names incentives as a third bound on predictive value. The author regards the reasoning behind it, that agents under the same incentives and constraints behave as real stakeholders would, as part of the thesis. The books the paper drew on stand on a shelf the author cannot reach, and the notes went with the paper, so most of its original sources go uncited here. The review takes the products as they stand in November 2026, not as they stood when the author wrote the paper, which, as far as the author recalls, leaves the conclusions unchanged.
Out of Scope
- Developments since the paper. Research harnesses that put many automated agents into Kerbal Space Program and into high-fidelity satellite simulators, and language models tested as spacecraft operators, are the subject of the next article in this series.
- The hard problems in detail. This article names shared time, the rarity of encounters under honest delta-v budgets, and deterministic networked physics, and a later article works them.
- Other products. The paper’s list held other simulations that it never reviewed for lack of time. Nor does the review cover Orbiter, Elite Dangerous, Space Engineers or the others, so the third claim reads none known.
- Hands-on testing. The reviews rest on the developers’ public statements, on first principles and, where the text says so, on the author’s earlier experience of the products, not on new hands-on testing.
- Reverse engineering. The author reverse engineered no product, and the series will not.
Conclusion
Strong single-agent models of spaceflight exist, and Kerbal Space Program and Children of a Dead Earth are two good ones, the first approachable and the second faithful. The multi-agent simulations, EVE Online and Star Citizen, have poor physical models shared with film and television, in which, between them, ships coast to a stop, travel takes almost no time, gravity has no mechanism and planets, on the evidence found, do not orbit. None of these products has failed. Each expresses its own thesis under its own constraints, and EVE Online in particular shows that a real economy of real agents can make the sum greater than the parts.
What no known product offers is the combination, a single shared world of many agents under physics that does not cheat, made approachable by tools rather than by fiction. No obstacle in physics is known. The open problems are of engineering and design, and true travel times are the hardest of them. Whether it would pay is another question. People adapt to the conditions they are given, and agents in such a simulation would face the incentives and constraints of real space, so it should yield evidence of how people would behave under the conditions it models, more a source of hypotheses than a forecast, bounded by its fidelity, by how many people can use it, and by how much its agents care. How much it yields depends chiefly on whether its builders can formulate realistic scenarios and whether agents are willing to explore them. The public has shown that it will pay for spaceflight and for shared worlds. A simulation that gave it both, truthfully, would also teach it, and through the loop between science fiction and science, the children it taught might help to build the future it depicted.
Equations
This section collects the 41 display equations of the article in the order they appear, each under its number and reproduced exactly as it stands in the text. A title in bold names an established result, and a title in italics beginning with the word Bespoke names a construction made for this analysis. Of the 41, 28 are established and 13 are bespoke. Each equation takes the units its section states. Distances are in kilometres unless the section says otherwise, the worked numbers are rounded to the precision of their inputs, and several letters are reused, each defined where it appears. The letter a is a semi-major axis in Equations 2 and 3 and an acceleration in Equation 40, R is the radius of a planet in Equations 4, 5 and 19 and a range in Equations 6 and 27, T is a sidereal day in Equation 17 and a temperature in Equations 21 and 39, the Greek letter tau is a time constant in Equations 7 and 8 and a torque in Equation 35, the Greek letter rho is a ratio of spending to income in Equation 13 and the density of air in Equations 36 to 38, and the capital Greek letter phi is a step function in Equation 34 and a gravitational potential in Equation 41.
Equation 1, Bohr Energy Levels of Hydrogen. The energy of each permitted level in Bohr’s model, which predicts the hydrogen spectrum despite a wrong picture of the electron. Bohr 1913
\[E_{n} = -\frac{13.6\ \mathrm{eV}}{n^{2}} \tag{1}\]Equation 2, Laplace Sphere of Influence. The radius within which the patched-conic method treats a smaller body’s gravity as the only gravity. Curtis 2020
\[r_{\mathrm{SOI}} = a \left( \frac{m}{M} \right)^{2/5} \tag{2}\]Equation 3, Hill Radius. The approximate distance of a body’s first and second Lagrange points, which lie outside its sphere of influence. Kane 2017
\[r_{\mathrm{H}} \approx a \left( \frac{m}{3M} \right)^{1/3} \tag{3}\]Equation 4, Circular Orbital Speed at the Surface. The speed of a circular orbit just above a body’s surface in terms of its surface gravity and radius, which falls as the body shrinks at fixed gravity. NASA Flight to Orbit Kerbin
\[v_{\mathrm{s}} = \sqrt{g R} \tag{4}\]Equation 5, Mean Density of a Sphere. The mean density of a body from its mass and radius, which shows what a small planet with Earth-like gravity must be made of. Kerbin
\[\bar{\rho} = \frac{3 M}{4 \pi R^{3}} \tag{5}\]Equation 6, Bespoke Engagement Time of a Straight Pass. The time two craft spend within weapon range on a straight pass at constant relative speed, an order-of-magnitude model of the article’s.
\[t_{\mathrm{e}} = \frac{2R}{v_{\mathrm{rel}}} \tag{6}\]Equation 7, EVE Online Velocity Law. The speed a ship reaches under full propulsion in EVE Online, the law of a body in a viscous fluid, as the community guide gives it. Acceleration
\[v(t) = v_{\max} \left( 1 - e^{-t/\tau} \right) \tag{7}\]Equation 8, EVE Online Alignment Time. The time a ship takes to reach three quarters of its top speed under the velocity law, which the developer-hosted documentation also gives. Useful Formulae
\[t_{\mathrm{a}} = \tau \ln 4 \tag{8}\]Equation 9, EVE Online Industry Job Cost. The cost of installing an industry job in EVE Online, as the developer states it, with the system cost index scaled by the facility’s bonuses. CCP 2023
\[C_{\mathrm{job}} = V_{\mathrm{EIV}} \left( I_{\mathrm{SCI}}\, \beta + t_{\mathrm{f}} + t_{\mathrm{SCC}} + t_{\alpha} \right) \tag{9}\]Equation 10, Bespoke Proportional Form of the 2014 System Cost Index. The square-root growth of EVE Online’s cost index with a system’s share of production, written from the developer’s description of 2014 as a proportionality, since the exact current form is not published in text. CCP 2014b
\[I_{\mathrm{SCI}} \propto \sqrt{\frac{h_{\mathrm{sys}}}{h_{\mathrm{all}}}} \tag{10}\]Equation 11, Bespoke Material Requirement from EVE Online’s Stated Rules. The quantity of one material a job consumes, assembled from the developer’s stated rules on material efficiency, rounding per job and a minimum of one unit per run. CCP 2014b
\[\begin{aligned} n_{\mathrm{mat}} &= \max\left( r,\ q \right) \\ q &= \left\lceil r\, n_{0} \left( 1 - \frac{E}{100} \right) \gamma \right\rceil \end{aligned} \tag{11}\]Equation 12, Bespoke Arbitrage Condition Between Two Hubs. The condition under which a haul between two markets pays after sales tax, broker’s fees and hauling cost, a construction of the article’s.
\[\begin{aligned} & p_{\mathrm{B}} \left( 1 - \lambda_{\mathrm{s}} - f_{\mathrm{b}} \right) \\ & \quad - p_{\mathrm{A}} \left( 1 + f_{\mathrm{b}} \right) - c_{\mathrm{h}} > 0 \end{aligned} \tag{12}\]Equation 13, Bespoke Ratio of Spending to Income. The ratio of a company’s costs including capital expenditure to its income in one year, used to test whether Cloud Imperium spends what it raises. Cloud Imperium 2023
\[\rho_{y} = \frac{C_{y}}{R_{y}} \tag{13}\]Equation 14, Tsiolkovsky Rocket Equation. The delta-v a rocket gains as its mass falls from the initial to the final mass. NASA Ideal Rocket Equation
\[\Delta v = I_{\mathrm{sp}}\, g_{0} \ln \frac{m_{0}}{m_{\mathrm{f}}} \tag{14}\]Equation 15, Hohmann Transfer Time. Half the period of the transfer ellipse, the time the cheapest two-burn transfer between two coplanar circular orbits takes. Curtis 2020
\[t_{\mathrm{H}} = \pi \sqrt{\frac{a_{\mathrm{t}}^{3}}{\mu_{\odot}}} \tag{15}\]Equation 16, Synodic Period. The interval between repetitions of two planets’ relative position, which spaces the windows for a transfer. Curtis 2020
\[\frac{1}{S} = \frac{1}{T_{1}} - \frac{1}{T_{2}} \tag{16}\]Equation 17, Geostationary Orbit Radius. The radius at which a circular orbit’s period equals one sidereal day, the only place a satellite holds position over the ground. ESA 2020
\[r_{\mathrm{G}} = \left( \frac{\mu T^{2}}{4 \pi^{2}} \right)^{1/3} \tag{17}\]Equation 18, Circular Orbital Speed. The speed of a circular orbit at a given radius, which is what makes reaching orbit hard. NASA Flight to Orbit
\[v_{\mathrm{c}} = \sqrt{\frac{\mu}{r}} \tag{18}\]Equation 19, Gravity at Altitude. The inverse-square fall of gravity with height, which leaves about nine-tenths of surface gravity at the space station. NASA 2012
\[g(h) = g_{0} \left( \frac{R}{R + h} \right)^{2} \tag{19}\]Equation 20, Centripetal Acceleration. The acceleration felt at the rim of a rotating habitat, the physical basis of spin gravity. Johnson and Holbrow 1977
\[a_{\mathrm{c}} = \omega^{2} r \tag{20}\]Equation 21, Bespoke Radiator Power Without Absorbed Sunlight. The Stefan-Boltzmann law applied to a radiator facing cold space with absorbed sunlight neglected, an assumption of the article’s that lowers the area needed. Liles and Amundsen 2023
\[P = \varepsilon \sigma A T^{4} \tag{21}\]Equation 22, Light-Time Delay. The one-way delay of any signal over a distance, which no message or sensor reading can beat. Landon et al. 2025
\[t_{\mathrm{L}} = \frac{d}{c} \tag{22}\]Equation 23, Bespoke Cost of One Gravity Step. The time one processor core needs to advance every craft by one step, linear in the number of craft because craft do not attract one another, an order-of-magnitude estimate of the article’s. Tanikawa et al. 2013
\[t_{\mathrm{step}} = N_{\mathrm{c}}\, \bar{n}\, t_{\mathrm{f}} \tag{23}\]Equation 24, Bespoke Wall-Clock Time Under a Shared Clock. The real time an agent waits for a span of simulated time when one compression factor applies to every agent.
\[t_{\mathrm{w}} = \frac{t_{\mathrm{s}}}{\kappa} \tag{24}\]Equation 25, Period of a Circular Orbit. The time one circular orbit takes at a given radius, Kepler’s third law for a circle. NASA Flight to Orbit
\[T_{\mathrm{o}} = 2 \pi \sqrt{\frac{r^{3}}{\mu}} \tag{25}\]Equation 26, Special-Relativistic Clock Rate. The exact rate of a moving clock against one at rest, which governs the time a traveller lives through at interstellar speeds. Shapiro 1978 Ashby 2003
\[\frac{d t_{\mathrm{c}}}{d t} = \sqrt{1 - \frac{v^{2}}{c^{2}}} \tag{26}\]Equation 27, Bespoke Rate of Passes in a Crowded Shell. The rate at which one craft passes within range of another when craft fill a shell at random, an illustrative estimate of the article’s.
\[\nu = \frac{N}{V}\, \pi R^{2}\, v_{\mathrm{rel}} \tag{27}\]Equation 28, Bespoke Freight Propellant From the Rocket Equation. The propellant a shipment needs for a given delta-v, the rocket equation rearranged to price freight by delta-v in the blueprint’s economy.
\[\begin{aligned} m_{\mathrm{p}} = {} & \left( m_{\mathrm{pay}} + m_{\mathrm{s}} \right) \\ & \times \left( e^{\Delta v / \left( I_{\mathrm{sp}}\, g_{0} \right)} - 1 \right) \end{aligned} \tag{28}\]Equation 29, N-Body Equation of Motion With Thrust. The acceleration of a craft in an inertial frame under the gravity of every massive body plus its own thrust, the equation the core integrates. Battin 1999 Hughes 2007
\[\ddot{\mathbf{r}}_{i} = -G \sum_{j} m_{j} \frac{\mathbf{r}_{i} - \mathbf{r}_{j}}{\lvert \mathbf{r}_{i} - \mathbf{r}_{j} \rvert^{3}} + \mathbf{a}_{\mathrm{T},i} \tag{29}\]Equation 30, Propellant Mass Flow. The rate at which an engine of given thrust and specific impulse consumes propellant, which integrates to the rocket equation. NASA Specific Impulse
\[\dot{m} = -\frac{F}{I_{\mathrm{sp}}\, g_{0}} \tag{30}\]Equation 31, Motion Relative to a Parent Body With Third-Body Terms. The acceleration of a craft in the frame of its parent body, where each other body enters as the difference between its pull on the craft and its pull on the parent. Hughes 2007 Berry and Coppola 2007
\[\begin{aligned} \ddot{\mathbf{r}} = {} & -\frac{\mu_{\mathrm{p}}}{\lvert \mathbf{r} \rvert^{3}}\, \mathbf{r} + \mathbf{a}_{\mathrm{T}} \\ & + \sum_{k} \mu_{k} \left( \frac{\mathbf{r}_{k} - \mathbf{r}}{\lvert \mathbf{r}_{k} - \mathbf{r} \rvert^{3}} - \frac{\mathbf{r}_{k}}{\lvert \mathbf{r}_{k} \rvert^{3}} \right) \end{aligned} \tag{31}\]Equation 32, Bespoke Tidal Bound on a Third Body’s Perturbation. The largest size of a third body’s perturbing acceleration on a craft near its parent, to first order in the ratio of the two distances, an approximation of the article’s.
\[a_{k} \lesssim \frac{2 \mu_{k}\, r}{d^{3}} \tag{32}\]Equation 33, Kick-Drift-Kick Leapfrog Step. The second-order symplectic step whose energy error stays bounded for a coasting craft at a constant step, the base of higher-order symplectic integrators. Yoshida 1990 Hairer et al. 2003
\[\begin{aligned} \mathbf{v}_{n+1/2} &= \mathbf{v}_{n} + \tfrac{\Delta t}{2}\, \mathbf{a}(\mathbf{r}_{n}) \\ \mathbf{r}_{n+1} &= \mathbf{r}_{n} + \Delta t\, \mathbf{v}_{n+1/2} \\ \mathbf{v}_{n+1} &= \mathbf{v}_{n+1/2} + \tfrac{\Delta t}{2}\, \mathbf{a}(\mathbf{r}_{n+1}) \end{aligned} \tag{33}\]Equation 34, Bespoke Reproducible State Transition. The shared world’s state advanced by one deterministic step from the agents’ ordered inputs, the article’s statement of the requirement that the step be reproducible.
\[\mathbf{x}_{n+1} = \Phi_{\Delta t}\!\left( \mathbf{x}_{n}, U_{n} \right) \tag{34}\]Equation 35, Euler’s Equations of Rigid-Body Rotation. The change in a craft’s angular velocity under the torque of its thrusters, which turns the craft without changing its path. Stoneking 2007
\[\mathbf{I}\, \dot{\boldsymbol{\omega}} + \boldsymbol{\omega} \times \left( \mathbf{I}\, \boldsymbol{\omega} \right) = \boldsymbol{\tau} \tag{35}\]Equation 36, Aerodynamic Drag. The drag force on a body moving through air, which governs launch and reentry. NASA Drag Equation
\[F_{\mathrm{D}} = \tfrac{1}{2}\, \rho v^{2} C_{\mathrm{D}} A \tag{36}\]Equation 37, Isothermal Exponential Atmosphere. The simple model of air density falling exponentially with height above sea level. Garcia-Llama 2007 NASA Earth Fact Sheet
\[\rho(h) = \rho_{0}\, e^{-h/H} \tag{37}\]Equation 38, Sutton-Graves Stagnation-Point Heating. The convective heat flux at the nose of a vehicle entering an atmosphere, which falls as the nose grows blunter. Sutton and Graves 1971
\[\dot{q} = k \sqrt{\frac{\rho}{r_{\mathrm{n}}}}\, v^{3} \tag{38}\]Equation 39, Bespoke Lumped Thermal Balance. A single-node energy balance in which a craft heats when the power it gains exceeds what its radiators reject, an assumption of uniform temperature.
\[m c\, \frac{dT}{dt} = P_{\mathrm{in}} - \varepsilon \sigma A T^{4} \tag{39}\]Equation 40, Pendulum Model of First-Mode Slosh. The natural frequency of the equivalent pendulum that represents a tank’s first slosh mode under axial acceleration. Abramson 1966
\[\omega_{\mathrm{s}} = \sqrt{\frac{a}{L}} \tag{40}\]Equation 41, Weak-Field Rate of a Moving Clock. The rate of a craft’s clock against a distant clock at rest, slowed by the gravitational potential and by speed, to first order. Ashby 2003
\[\frac{d t_{\mathrm{c}}}{d t} = 1 + \frac{\Phi}{c^{2}} - \frac{v^{2}}{2 c^{2}} \tag{41}\]Notation
- $E_{n}$, the energy of hydrogen level $n$ in the Bohr model
- $r_{\mathrm{SOI}}$, the radius of the sphere of influence
- $r_{\mathrm{H}}$, the Hill radius
- $M$, the mass of the larger body, a ship’s mass in Equation 7, or a body’s mass in Equation 5
- $t_{\mathrm{e}}$, the time two craft spend within range of each other
- $v_{\mathrm{rel}}$, the relative speed of two craft
- $v_{\max}$, a ship’s top speed in EVE Online
- $\tau$, the time constant of EVE Online’s velocity law
- $I_{\mathrm{sp}}$, the specific impulse of a rocket engine
- $g_{0}$, standard gravity at the surface, 9.80665 metres per second squared
- $t_{\mathrm{H}}$, the Hohmann transfer time
- $a_{\mathrm{t}}$, the semi-major axis of the transfer ellipse
- $\mu_{\odot}$, the Sun’s gravitational parameter
- $S$, the synodic period
- $r_{\mathrm{G}}$, the radius of the geostationary orbit
- $\mu$, Earth’s gravitational parameter
- $v_{\mathrm{c}}$, the speed of a circular orbit
- $a_{\mathrm{c}}$, the centripetal acceleration at a habitat’s rim
- $\omega$, the rate of rotation in radians per second
- $\varepsilon$, the emissivity of a radiator
- $\sigma$, the Stefan-Boltzmann constant
- $t_{\mathrm{L}}$, the one-way light-time delay
- $\mathbf{a}_{\mathrm{T},i}$, the acceleration of craft $i$ from its own thrust
- $\dot{m}$, the rate of change of a craft’s mass
- $v_{\mathrm{s}}$, the speed of a circular orbit just above a body’s surface
- $\bar{\rho}$, the mean density of a body
- $t_{\mathrm{a}}$, the alignment time of a ship in EVE Online
- $a_{k}$, the size of the perturbing acceleration of a third body $k$
- $T_{\mathrm{o}}$, the period of a circular orbit
- $\nu$, the rate at which one craft passes within range of another
- $t_{\mathrm{step}}$, the time one processor core needs to advance every craft by one step
- $N_{\mathrm{c}}$, the number of craft
- $\bar{n}$, the mean number of bodies each craft feels
- $t_{\mathrm{f}}$, the time of one evaluation of the inverse-square law
- $\mu_{\mathrm{p}}$, the gravitational parameter of a craft’s parent body
- $\mu_{k}$, the gravitational parameter of another body $k$
- $t_{\mathrm{c}}$, the time a craft’s own clock shows
- $\Phi$, the Newtonian gravitational potential at a craft
- $c$, the speed of light
- $\Delta t$, the integration step
- $\Phi_{\Delta t}$, the deterministic step function of the shared world
- $U_{n}$, the ordered set of agents’ inputs committed for step $n$
- $\mathbf{I}$, a craft’s inertia tensor
- $\boldsymbol{\tau}$, the torque of a craft’s reaction control thrusters
- $F_{\mathrm{D}}$, the aerodynamic drag force
- $C_{\mathrm{D}}$, the drag coefficient
- $\rho$, the density of the air
- $H$, the scale height of the atmosphere
- $\dot{q}$, the stagnation-point heat flux
- $r_{\mathrm{n}}$, the radius of a vehicle’s nose
- $P_{\mathrm{in}}$, the power a craft generates and absorbs
- $\omega_{\mathrm{s}}$, the natural frequency of the first slosh mode
- $C_{\mathrm{job}}$, the cost of installing an industry job in EVE Online
- $V_{\mathrm{EIV}}$, the estimated item value of a job’s output
- $I_{\mathrm{SCI}}$, the cost index of a solar system
- $\beta$, the product of a facility’s bonuses
- $h_{\mathrm{sys}}$, the job-hours of an activity in one system over 28 days
- $h_{\mathrm{all}}$, the job-hours of that activity in the whole universe over 28 days
- $n_{\mathrm{mat}}$, the quantity of one material a job consumes
- $n_{0}$, a blueprint’s base quantity of a material per run
- $q$, a job’s material requirement after rounding
- $\gamma$, a facility’s material modifier
- $p_{\mathrm{A}}$, the price paid at the source market
- $p_{\mathrm{B}}$, the price received at the destination market
- $\lambda_{\mathrm{s}}$, the sales tax rate
- $f_{\mathrm{b}}$, the broker’s fee rate
- $c_{\mathrm{h}}$, the cost of hauling one unit
- $\rho_{y}$, the ratio of spending to income in year $y$
- $C_{y}$, the costs including capital expenditure in year $y$
- $R_{y}$, the income in year $y$
- $m_{\mathrm{p}}$, the propellant a shipment needs
- $m_{\mathrm{pay}}$, the payload of a shipment
- $m_{\mathrm{s}}$, the structure and engines that make the trip with the payload
- $t_{\mathrm{w}}$, the wall-clock time an agent waits
- $t_{\mathrm{s}}$, the simulated time that passes
- $\kappa$, the time compression factor shared by every agent
References
Primary Sources, Peer-Reviewed
- Ashby, Neil. 2003. “Relativity in the Global Positioning System.” Living Reviews in Relativity 6 (1): 1. https://doi.org/10.12942/lrr-2003-1
- Barnett, Michael, Heather Wagner, Anne Gatling, Janice Anderson, Meredith Houle, and Alan Kafka. 2006. “The Impact of Science Fiction Film on Student Understanding of Science.” Journal of Science Education and Technology 15 (2): 179–91. https://doi.org/10.1007/s10956-006-9001-y
- Bohr, N. 1913. “On the Constitution of Atoms and Molecules.” London, Edinburgh, and Dublin Philosophical Magazine and Journal of Science 26 (151): 1–25. https://doi.org/10.1080/14786441308634955
- Kane, Stephen R. 2017. “Worlds without Moons: Exomoon Constraints for Compact Planetary Systems.” Astrophysical Journal Letters 839 (2): L19. https://doi.org/10.3847/2041-8213/aa6bf2
- Li, Wenhao, Zhaohui Liu, You Mu, Rui Yang, and Xing Zhang. 2017. “Modeling and Research of a Space-Based Spacecraft Infrared Detection System.” Applied Optics 56 (9): 2428. https://doi.org/10.1364/ao.56.002428
- Lofgren, Eric T., and Nina H. Fefferman. 2007. “The Untapped Potential of Virtual Game Worlds to Shed Light on Real World Epidemics.” Lancet Infectious Diseases 7 (9): 625–29. https://doi.org/10.1016/s1473-3099(07)70212-8
- Monniaux, David. 2008. “The Pitfalls of Verifying Floating-Point Computations.” ACM Transactions on Programming Languages and Systems 30 (3): 1–41. https://doi.org/10.1145/1353445.1353446
- Park, Ryan S., William M. Folkner, James G. Williams, and Dale H. Boggs. 2021. “The JPL Planetary and Lunar Ephemerides DE440 and DE441.” Astronomical Journal 161 (3): 105. https://doi.org/10.3847/1538-3881/abd414
- Tanikawa, Ataru, Kohji Yoshikawa, Keigo Nitadori, and Takashi Okamoto. 2013. “Phantom-GRAPE: Numerical Software Library to Accelerate Collisionless N-Body Simulation with SIMD Instruction Set on x86 Architecture.” New Astronomy 19: 74–88. https://doi.org/10.1016/j.newast.2012.08.009
- Whipple, F. L. 1947. “Meteorites and Space Travel.” Astronomical Journal 52: 131. https://doi.org/10.1086/106009
- Wisdom, Jack, and Matthew Holman. 1991. “Symplectic Maps for the n-Body Problem.” Astronomical Journal 102: 1528. https://doi.org/10.1086/115978
- Xanthopoulou, Despoina, and Savvas Papagiannidis. 2012. “Play Online, Work Better? Examining the Spillover of Active Learning and Transformational Leadership.” Technological Forecasting and Social Change 79 (7): 1328–39. https://doi.org/10.1016/j.techfore.2012.03.006
- Yoshida, Haruo. 1990. “Construction of Higher Order Symplectic Integrators.” Physics Letters A 150 (5–7): 262–68. https://doi.org/10.1016/0375-9601(90)90092-3
Primary Sources, Non-Peer-Reviewed
- Abramson, H. Norman. 1966. The Dynamic Behavior of Liquids in Moving Containers, with Applications to Space Vehicle Technology. NASA-SP-106. NASA. https://ntrs.nasa.gov/citations/19670006555
- Allen, H. Julian, and A. J. Eggers Jr. 1958. A Study of the Motion and Aerodynamic Heating of Ballistic Missiles Entering the Earth’s Atmosphere at High Supersonic Speeds. NACA-TR-1381. NASA. https://ntrs.nasa.gov/citations/19930091020
- Armstrong, J. Scott. 2001. “Role Playing: A Method to Forecast Decisions.” In Principles of Forecasting. Springer US. https://doi.org/10.1007/978-0-306-47630-3_2
- Berry, Matthew M., and Vincent T. Coppola. 2007. Correct Modeling of the Indirect Term for Third-Body Perturbations. AAS 07-417. Analytical Graphics, Inc. https://www.agi.com/getmedia/eff08466-52ad-4483-958a-2e3ceb922eab/Correct-Modeling-of-the-Indirect-Term-for-Third-Body-Perturbations.pdf
- Blevins, D. R., and C. W. Hohmann. 1975. “Description of the Space Shuttle Reaction Control System.” In Society of Automotive Engineers. https://ntrs.nasa.gov/citations/19750061613
- Carroll, John M., and Caroline Carrithers. 1984. “Training Wheels in a User Interface.” Communications of the ACM 27 (8): 800–806. https://doi.org/10.1145/358198.358218
- Castronova, Edward, Dmitri Williams, Cuihua Shen, et al. 2009. “As Real as Real? Macroeconomic Behavior in a Large-Scale Virtual World.” New Media & Society 11 (5): 685–707. https://doi.org/10.1177/1461444809105346
- CCP. n.d.-b. “EVE Swagger Interface Overview.” developers.eveonline.com. Accessed November 13, 2026. https://developers.eveonline.com/docs/services/esi/overview/
- CCP. n.d.-c. “Tritanium Item Record.” esi.evetech.net. Accessed November 13, 2026. https://esi.evetech.net/latest/universe/types/34/?datasource=tranquility&language=en
- Clement, John. 1982. “Students’ Preconceptions in Introductory Mechanics.” American Journal of Physics 50 (1): 66–71. https://doi.org/10.1119/1.12989
- Curtis, Howard D. 2020. “Interplanetary Trajectories.” In Orbital Mechanics for Engineering Students. Elsevier. https://doi.org/10.1016/b978-0-08-102133-0.00008-8
- “DarkMultiPlayer.” n.d. GitHub. Accessed November 13, 2026. https://github.com/godarklight/DarkMultiPlayer
- “Earth Fact Sheet.” n.d. Accessed November 13, 2026. https://nssdc.gsfc.nasa.gov/planetary/factsheet/earthfact.html
- Garcia-Llama, Eduardo. 2007. “Analytic Guidance for the First Entry in a Skip Atmospheric Entry.” In American Inst. of Aeronautics and Astronautics. https://ntrs.nasa.gov/citations/20070025087
- Glasstone, Samuel. 1964. The Effects of Nuclear Weapons. Office of Scientific and Technical Information (OSTI). https://doi.org/10.2172/972902
- Green, Kesten C., and J. Scott Armstrong. 2007. “The Ombudsman: Value of Expertise for Forecasting Decisions in Conflicts.” Interfaces 37 (3): 287–99. https://doi.org/10.1287/inte.1060.0262
- Hairer, Ernst, Christian Lubich, and Gerhard Wanner. 2003. “Geometric Numerical Integration Illustrated by the Störmer–Verlet Method.” Acta Numerica 12: 399–450. https://doi.org/10.1017/s0962492902000144
- Hairer, Ernst, and Gustaf Söderlind. 2005. “Explicit, Time Reversible, Adaptive Step Size Control.” SIAM Journal on Scientific Computing 26 (6): 1838–51. https://doi.org/10.1137/040606995
- Hughes, Steve. 2007. General Mission Analysis Tool (GMAT) Mathematical Specifications. NASA Goddard Space Flight Center. https://ntrs.nasa.gov/citations/20080031744
- Johnson, Richard D., and Charles Holbrow. 1977. Space Settlements: A Design Study. NASA-SP-413. NASA. https://ntrs.nasa.gov/citations/19770014162
- “Jupiter Fact Sheet.” n.d. Accessed November 13, 2026. https://nssdc.gsfc.nasa.gov/planetary/factsheet/jupiterfact.html
- Landon, Lauren B., John A. Karasinski, Linda G. Morissette, et al. 2025. Assessment of the State of Communication Delay Research in Preparation for Missions beyond Low Earth Orbit. NASA/TM-20250003885. NASA. https://ntrs.nasa.gov/citations/20250003885
- Levin, Samuel R., Thomas W. Scott, Helen S. Cooper, and Stuart A. West. 2017. “Darwin’s Aliens.” International Journal of Astrobiology 18 (1): 1–9. https://doi.org/10.1017/s1473550417000362
- Liles, Kaitlin, and Ruth Amundsen. 2023. NASA Passive Thermal Control Engineering Guidebook. Langley Research Center. https://ntrs.nasa.gov/citations/20230013900
- Lisk, Timothy C., Ugur T. Kaplancali, and Ronald E. Riggio. 2011. “Leadership in Multiplayer Online Gaming Environments.” Simulation & Gaming 43 (1): 133–49. https://doi.org/10.1177/1046878110391975
- “Luna Multiplayer.” n.d. github.com. Accessed November 13, 2026. https://github.com/LunaMultiplayer/LunaMultiplayer/wiki
- McDermott, Lillian C., and Edward F. Redish. 1999. “Resource Letter: PER-1: Physics Education Research.” American Journal of Physics 67 (9): 755–67. https://doi.org/10.1119/1.19122
- McKagan, S. B., K. K. Perkins, and C. E. Wieman. 2008. “Why We Should Teach the Bohr Model and How to Teach It Effectively.” Physical Review Special Topics - Physics Education Research 4 (1): 010103. https://doi.org/10.1103/physrevstper.4.010103
- “Moon Fact Sheet.” n.d. Accessed November 13, 2026. https://nssdc.gsfc.nasa.gov/planetary/factsheet/moonfact.html
- “On the Dynamical Stability of Principia’s Modified Jool System.” n.d. GitHub. Accessed November 13, 2026. https://github.com/mockingbirdnest/Principia/wiki/On-the-dynamical-stability-of-Principia’s-modified-Jool-system
- Orthia, Lindy. 2019. “How Does Science Fiction Television Shape Fans’ Relationships to Science? Results from a Survey of 575 ‘Doctor Who’ Viewers.” Journal of Science Communication 18 (04): A08. https://doi.org/10.22323/2.18040208
- Perla, Peter P., and Ed McGrady. 2011. “Why Wargaming Works.” Naval War College Review. https://digital-commons.usnwc.edu/nwc-review/vol64/iss3/8/
- “Principia.” n.d. GitHub. Accessed November 13, 2026. https://github.com/mockingbirdnest/Principia
- “Real Solar System.” n.d. GitHub. Accessed November 13, 2026. https://github.com/KSP-RO/RealSolarSystem
- “Realism Overhaul.” n.d. github.com. Accessed November 13, 2026. https://github.com/KSP-RO/RealismOverhaul
- Rubel, Robert C. 2006. “The Epistemology of War Gaming.” Naval War College Review 59 (2). https://digital-commons.usnwc.edu/nwc-review/vol59/iss2/8
- Shapiro, I. I. 1978. “Down to Earth Relativity.” In Proceedings of the Tenth Annual Precise Time and Time Interval (PTTI) Applications and Planning Meeting. https://ntrs.nasa.gov/citations/19790016582
- Stoneking, Eric. 2007. “Newton-Euler Dynamic Equations of Motion for a Multi-Body Spacecraft.” In AIAA Guidance, Navigation and Control Conference. https://ntrs.nasa.gov/citations/20080044854
- “Sun Fact Sheet.” n.d. Accessed November 13, 2026. https://nssdc.gsfc.nasa.gov/planetary/factsheet/sunfact.html
- Sutton, K., and R. A. Graves Jr. 1971. A General Stagnation-Point Convective Heating Equation for Arbitrary Gas Mixtures. NASA-TR-R-376. NASA. https://ntrs.nasa.gov/citations/19720003329
- U.S. Joint Forces Command. n.d. Millennium Challenge 2002: Experiment Report. https://www.esd.whs.mil/Portals/54/Documents/FOID/Reading%20Room/Joint_Staff/12-F-0344-Millennium-Challenge-2002-Experiment-Report.pdf
- UK Ministry of Defence. 2017. Wargaming Handbook. Development, Concepts and Doctrine Centre. https://www.gov.uk/government/publications/defence-wargaming-handbook
- “Useful Formulae.” n.d. EVE: Developers. Accessed November 13, 2026. https://developers.eveonline.com/docs/guides/useful-formulae/
- Washington Employment Security Department. n.d. “WARN Notices.” Accessed November 13, 2026. https://fortress.wa.gov/esd/file/WARN/Public/SearchWARN.aspx
- Westover, Shayne C., Rainer Meinke, Shaun Nerolich, et al. 2019. Magnet Architectures and Active Radiation Shielding Study (MAARSS). HQ-E-DAA-TN65001. NASA Johnson Space Center. https://ntrs.nasa.gov/citations/20190002579
- Williams, Dmitri. 2010. “The Mapping Principle, and a Research Framework for Virtual Worlds.” Communication Theory 20 (4): 451–70. https://doi.org/10.1111/j.1468-2885.2010.01371.x
- Winternitz, Luke. 2017. “Introduction to GPS and Other Global Navigation Satellite Systems.” In Annual Time and Frequency Metrology Seminar. https://ntrs.nasa.gov/citations/20170004590
Secondary Sources
- Battin, Richard H. 1999. An Introduction to the Mathematics and Methods of Astrodynamics, Revised Edition. American Institute of Aeronautics and Astronautics, Inc. https://doi.org/10.2514/4.861543
- Caffrey, Matthew B., Jr. 2019. On Wargaming. Newport Paper 43. Naval War College Press. https://digital-commons.usnwc.edu/newport-papers/43/
- Conway Morris, Simon. 2003. Life’s Solution: Inevitable Humans in a Lonely Universe. Cambridge University Press. https://doi.org/10.1017/cbo9780511535499
- Friedman, Norman. 2017. Winning a Future War: War Gaming and Victory in the Pacific War. Naval History and Heritage Command. https://www.history.navy.mil/research/publications/publications-by-subject/winning-a-future-war.html
- Hawking, Stephen, and Lucy Hawking. 2007. George’s Secret Key to the Universe. Simon & Schuster Children’s Publishing. https://openlibrary.org/books/OL9523149M
- Herman, Mark, Mark Frost, and Robert Kurz. 2009. Wargaming for Leaders: Strategic Decision Making from the Battlefield to the Boardroom. McGraw-Hill. https://openlibrary.org/isbn/9780071596886
- Whitfield, Stephen E., and Gene Roddenberry. 1968. The Making of Star Trek. Ballantine Books. https://archive.org/details/makingofstartrek00whit
Commercial Sources
- “7.0 Thermal Control.” n.d. Accessed November 13, 2026. https://www.nasa.gov/smallsat-institute/sst-soa/thermal-control/
- “Active Thermal Control System (ATCS) Overview.” n.d. Accessed November 13, 2026. https://www.nasa.gov/wp-content/uploads/2021/02/473486main_iss_atcs_overview.pdf
- “Anatomy of an Electromagnetic Wave.” 2016. NASA Science, August 10. https://science.nasa.gov/ems/02_anatomy/
- “Black Holes.” 2024. NASA Science, March 13. https://science.nasa.gov/universe/black-holes/
- “Can a Laser Beam Be Seen from the Side?” n.d. Accessed November 13, 2026. https://www.edmundoptics.com/knowledge-center/frequently-asked-questions/general-questions/lasers/can-a-laser-beam-be-seen-from-the-side
- CCP. n.d.-a. “Capsuleers.” Accessed November 13, 2026. https://universe.eveonline.com/lore/capsuleers
- CCP. 2003. “Highly Anticipated Massively Multiplayer Online Game.” EVE Community. https://community.eveonline.com/news/news-channels/press-releases/highly-anticipated-massively-multiplayer-online-game-1
- CCP. 2006. “A Deal Is a Deal…” EVE Online, August 14. https://www.eveonline.com/news/view/a-deal-is-a-deal…
- CCP. 2007a. “Econ Dev Blog No. 3: Some Statistics on Corporations.” EVE Online, December 1. https://www.eveonline.com/news/view/econ-dev-blog-no.-3-some-statistics-on-corporations
- CCP. 2007b. “EVE Online Appoints in-World Economist.” EVE Community. https://community.eveonline.com/news/news-channels/press-releases/eve-online-appoints-in-world-economist-1
- CCP. 2008. “I Can Resist Anything but Temptation.” EVE Online, November 3. https://www.eveonline.com/news/view/i-can-resist-anything-but-temptation
- CCP. 2009a. “I Bring Gifts! (By Gifts I Mean Taxes, Sorry).” EVE Online, September 21. https://www.eveonline.com/news/view/i-bring-gifts-by-gifts-i-mean-taxes-sorry
- CCP. 2009b. “The Way of the PLEX.” EVE Online, August 11. https://www.eveonline.com/news/view/the-way-of-the-plex
- CCP. 2010. “Fixing Lag: Character Nodes.” EVE Online, August 28. https://www.eveonline.com/news/view/fixing-lag-character-nodes
- CCP. 2012a. “Price Indices, February 2012.” EVE Online, March 16. https://www.eveonline.com/news/view/price-indices-february-2012
- CCP. 2012b. “Tech Is Fine, L2P.” EVE Online, July 19. https://www.eveonline.com/news/view/tech-is-fine-l2p
- CCP. 2013a. “Community Spotlight: Red Frog Freight.” EVE Online, June 21. https://www.eveonline.com/news/view/community-spotlight-red-frog-freight
- CCP. 2013b. “EVE Online Surpasses 500,000 Subscribers Worldwide.” EVE Community. https://community.eveonline.com/news/news-channels/press-releases/eve-online-surpasses-500-000-subscribers-worldwide
- CCP. 2013c. “Warp Drive Active.” EVE Online, November 15. https://www.eveonline.com/news/view/warp-drive-active
- CCP. 2014a. “The Bloodbath of B-R5RB, Gaming’s Most Destructive Battle Ever.” EVE Online, February 1. https://www.eveonline.com/news/view/the-bloodbath-of-b-r5rb
- CCP. 2014b. “EVE Industry: All You Want to Know.” EVE Online, July 17. https://www.eveonline.com/news/view/eve-industry-all-you-want-to-know
- CCP. 2014c. “HED-GP Technical Retrospective: What a HED-Ache.” EVE Online, January 24. https://www.eveonline.com/news/view/what-a-hed-ache
- CCP. 2015a. “EVE Economy Update: EVE Vegas 2015 Report.” EVE Online, November 5. https://www.eveonline.com/news/view/eve-economy-update-eve-vegas-2015-report
- CCP. 2015b. “Summer 2015 Nullsec and Sov Status Report.” EVE Online, May 7. https://www.eveonline.com/news/view/summer-2015-nullsec-and-sov-status-report
- CCP. 2016. “Overlays, ISK Buyer Amnesty and Account Security.” EVE Online, February 24. https://www.eveonline.com/news/view/overlays-isk-buyer-amnesty-and-account-security
- CCP. 2017. “PLEX Changes on the Way!” EVE Online, March 16. https://www.eveonline.com/news/view/plex-changes-on-the-way
- CCP. 2019. “Updates to Sales Taxes & Brokers Fees.” EVE Online, July 29. https://www.eveonline.com/news/view/updates-to-sales-taxes-and-brokers-fees
- CCP. 2020. “Broker Relations.” EVE Online, February 24. https://www.eveonline.com/news/view/broker-relations
- CCP. 2022. “Team Security: 2021 in Review.” EVE Online, July 12. https://www.eveonline.com/news/view/team-security-2021-in-review
- CCP. 2023. “Viridian: Expansion Notes.” EVE Online, June 9. https://www.eveonline.com/news/view/viridian-expansion-notes
- CCP. 2025. “Patch Notes: Version 22.02.” EVE Online, May 15. https://www.eveonline.com/news/view/patch-notes-version-22-02
- CCP. 2026. “Monthly Economic Report: August 2026.” EVE Online, September 9. https://www.eveonline.com/news/view/monthly-economic-report-august-2026
- “Children of a Dead Earth on Steam.” n.d. Accessed November 13, 2026. https://store.steampowered.com/app/476530/
- Cloud Imperium. 2020. “Cloud Imperium Financials for 2020.” https://cloudimperiumgames.com/blog/corporate/cloud-imperium-financials-for-2020
- Cloud Imperium. 2021. “Cloud Imperium Financials for 2021.” https://cloudimperiumgames.com/blog/corporate/cloud-imperium-financials-for-2021
- Cloud Imperium. 2022. “Cloud Imperium Financials for 2022.” https://cloudimperiumgames.com/blog/corporate/cloud-imperium-financials-for-2022
- Cloud Imperium. 2023. “Cloud Imperium Financials for 2023.” https://cloudimperiumgames.com/blog/corporate/cloud-imperium-financials-for-2023
- Cloud Imperium. 2024. “CFO Comment on the 2024 Accounting.” https://cloudimperiumgames.com/blog/corporate/cfo-comment-on-the-2024-accounting
- “Drag Equation.” n.d. Glenn Research Center | NASA. Accessed November 13, 2026. https://www1.grc.nasa.gov/beginners-guide-to-aeronautics/drag-equation/
- ESA. 2020. “Geostationary Orbit.” https://www.esa.int/ESA_Multimedia/Images/2020/03/Geostationary_orbit
- FAI. 2018. “Statement about the Karman Line.” https://www-2025.fai.org/news/statement-about-karman-line
- “FAQ.” 2018. Dawn, NASA Science, October 18. https://science.nasa.gov/mission/dawn/faq/
- “Flight to Orbit.” n.d. Glenn Research Center | NASA. Accessed November 13, 2026. https://www1.grc.nasa.gov/beginners-guide-to-aeronautics/flight-to-orbit/
- JPL. 2020. “JPL Planetary and Lunar Ephemerides.” https://ssd.jpl.nasa.gov/planets/eph_export.html
- “Kerbal Space Program 2 on Steam.” n.d. Accessed November 13, 2026. https://store.steampowered.com/app/954850/
- “Kerbal Space Program on Steam.” n.d. Accessed November 13, 2026. https://store.steampowered.com/app/220200/
- Launius, Roger D., and Dennis R. Jenkins. 2012. “Coming Home: Reentry and Recovery from Space.” https://www.nasa.gov/wp-content/uploads/2015/04/695726main_cominghome-ebook.pdf
- Mesarch, Michael. 2018. “GDC Orbit Primer.” https://science.nasa.gov/wp-content/uploads/2023/05/GDC_OrbitPrimer.pdf
- NASA. n.d.-a. “Chapter 15: Cruise.” Basics of Space Flight. Accessed November 13, 2026. https://science.nasa.gov/learn/basics-of-space-flight/chapter15-1/
- NASA. n.d.-b. “Chapter 4: Trajectories.” Basics of Space Flight. Accessed November 13, 2026. https://science.nasa.gov/learn/basics-of-space-flight/chapter4-1/
- NASA. 2001. “Staying Cool on the ISS.” https://web.archive.org/web/20191215040200/https://science.nasa.gov/science-news/science-at-nasa/2001/ast21mar_1
- NASA. 2012. “What Is Microgravity?” https://www.nasa.gov/learning-resources/for-kids-and-students/what-is-microgravity-grades-5-8/
- NASA. 2016. “Gamers Tackle Virtual Asteroid Sampling Mission.” NASA, June 13. https://www.nasa.gov/centers-and-facilities/goddard/gamers-tackle-virtual-asteroid-sampling-mission/
- NASA. 2017. “Why Do We Really Need Pressure Suits?” https://www.nasa.gov/wp-content/uploads/2023/05/dressing-for-altitude-nov-2017.pdf
- NASA. 2019. “Payload Operations Integration Center.” https://www.nasa.gov/wp-content/uploads/2021/03/payload_ops_fact_sheet_190604a.pdf
- NASA. 2020. “Space Communications: 7 Things You Need to Know.” NASA, October 6. https://www.nasa.gov/missions/tech-demonstration/space-communications-7-things-you-need-to-know/
- NASA. 2021. “NASA Laser Communications Innovations: A Timeline.” NASA, May 26. https://www.nasa.gov/centers-and-facilities/goddard/nasa-laser-communications-innovations-a-timeline/
- NASA. 2024. “55 Years Ago: Star Trek Final Episode Airs, Relationship with NASA Endures.” NASA, June 4. https://www.nasa.gov/history/55-years-ago-star-trek-final-episode-airs-relationship-with-nasa-endures/
- NASA. 2025. “RS-25 Core Stage Engine.” https://www.nasa.gov/wp-content/uploads/2025/04/sls-4963-sls-rs-25-engine-fact-sheet-508.pdf
- “Newton’s Laws of Motion.” n.d. Glenn Research Center | NASA. Accessed November 13, 2026. https://www1.grc.nasa.gov/beginners-guide-to-aeronautics/newtons-laws-of-motion/
- Nielsen, Jakob. 2006. “Progressive Disclosure.” Nielsen Norman Group. https://www.nngroup.com/articles/progressive-disclosure/
- “Orbit.” 2022. James Webb Space Telescope, NASA Science, June 1. https://science.nasa.gov/mission/webb/orbit/
- Q-Switched. 2016a. “Burn Rockets Burn.” Children of a Dead Earth. https://childrenofadeadearth.wordpress.com/2016/05/02/burn-rockets-burn/
- Q-Switched. 2016b. “The Essence of Space Warfare.” Children of a Dead Earth. https://childrenofadeadearth.wordpress.com/2016/04/20/the-essence-of-space-warfare/
- Q-Switched. 2016c. “Fun with Orbital Mechanics.” Children of a Dead Earth, May 17. https://childrenofadeadearth.wordpress.com/2016/05/17/fun-with-orbital-mechanics/
- Q-Switched. 2016d. “How Realistic Is It Actually?” Children of a Dead Earth. https://childrenofadeadearth.wordpress.com/2016/04/15/how-realistic-is-it-actually/
- Q-Switched. 2016e. “The Photon Lance.” Children of a Dead Earth, July 2. https://childrenofadeadearth.wordpress.com/2016/07/02/the-photon-lance/
- Q-Switched. 2016f. “Raw Steel.” Children of a Dead Earth, August 4. https://childrenofadeadearth.wordpress.com/2016/08/04/raw-steel/
- Q-Switched. 2016g. “Slosh Baffles.” Children of a Dead Earth, May 24. https://childrenofadeadearth.wordpress.com/2016/05/24/slosh-baffles/
- Q-Switched. 2016h. “Space Guns.” Children of a Dead Earth, June 14. https://childrenofadeadearth.wordpress.com/2016/06/14/space-guns/
- Q-Switched. 2016i. “Stealth in Space.” Children of a Dead Earth, July 12. https://childrenofadeadearth.wordpress.com/2016/07/12/stealth-in-space/
- Riebeek, Holli. 2009. “Catalog of Earth Satellite Orbits.” https://science.nasa.gov/earth/earth-observatory/catalog-of-earth-satellite-orbits/
- Roberts, Chris. 2018. “Letter from the Chairman.” https://robertsspaceindustries.com/en/comm-link/transmission/16700-Letter-From-The-Chairman
- Roberts, Chris. 2020. “Letter from the Chairman.” Roberts Space Industries. https://robertsspaceindustries.com/en/comm-link/transmission/17805-Letter-From-The-Chairman
- “Rocket Principles.” n.d. Accessed November 13, 2026. https://www.grc.nasa.gov/www/k-12/rocket/TRCRocket/rocket_principles.html
- RSI. n.d.-a. “Crowdfund Statistics.” Accessed November 13, 2026. https://robertsspaceindustries.com/en/funding-goals
- RSI. n.d.-b. “Star Citizen Alpha 4.0.” Accessed November 13, 2026. https://robertsspaceindustries.com/en/comm-link/Patch-Notes/20360-Star-Citizen-Alpha-40
- RSI. 2014a. “Into the Unknown: A Carrack & Her Crew.” Roberts Space Industries. https://robertsspaceindustries.com/en/comm-link/transmission/14323-Into-The-Unknown-A-Carrack-Her-Crew
- RSI. 2014b. “Lore Builder Twenty: Journeys & Generators.” Roberts Space Industries. https://robertsspaceindustries.com/en/comm-link/spectrum-dispatch/14262-LORE-BUILDER-TWENTY-Journeys-Generators
- RSI. 2017a. “Star Citizen Alpha 3.0.0.” Roberts Space Industries. https://robertsspaceindustries.com/en/comm-link/transmission/16348-Star-Citizen-Alpha-300
- RSI. 2017b. “Star Citizen Alpha 3.0.0 Patch Notes.” https://robertsspaceindustries.com/en/comm-link/transmission/16349-Star-Citizen-Alpha-300
- RSI. 2018a. “The Shipyard: Fuel Mechanics.” https://robertsspaceindustries.com/en/comm-link/engineering/16517-The-Shipyard-Fuel-Mechanics
- RSI. 2018b. “Star Citizen Alpha 3.2.0 Patch Notes.” https://robertsspaceindustries.com/spectrum/community/SC/forum/190048/thread/star-citizen-alpha-3-2-0-live-796019-patch-notes
- RSI. 2018c. “Star Citizen Alpha 3.4.0 Patch Notes.” https://robertsspaceindustries.com/spectrum/community/SC/forum/190048/thread/star-citizen-alpha-3-4-0-live-1028053-patch-notes
- RSI. 2020. “Flight & Fight: Upcoming Improvements.” https://robertsspaceindustries.com/en/comm-link/transmission/17647-Flight-Fight-Upcoming-Improvements
- RSI. 2021. “Deadly Consequences: Alpha 3.15.” https://robertsspaceindustries.com/en/comm-link/transmission/18338-Deadly-Consequences-Alpha-315
- RSI. 2025. “Star Citizen Alpha 4.5.0.” Roberts Space Industries. https://robertsspaceindustries.com/en/comm-link/Patch-Notes/20934-Star-Citizen-Alpha-450
- RSI. 2026. “Terms of Service.” https://robertsspaceindustries.com/tos
- “Shield Development: Basic Concepts.” n.d. Hypervelocity Impact Technology, NASA Johnson Space Center. Accessed November 13, 2026. https://hvit.jsc.nasa.gov/shield-development/
- “Specific Impulse.” n.d. Glenn Research Center | NASA. Accessed November 13, 2026. https://www1.grc.nasa.gov/beginners-guide-to-aeronautics/specific-impulse/
- Squad. 2011. “First Release Is Up.” https://web.archive.org/web/20111013113946/http://kerbalspace.tumblr.com/post/6872084213/first-release-is-up
- Squad. 2014. “KSP: Asteroid Redirect Mission Is Officially Released.” Steam Community, April 1. https://steamcommunity.com/games/220200/announcements/detail/1315356550594321621
- “Star Citizen Roadmap.” n.d. RSI Roadmap. Accessed November 13, 2026. https://robertsspaceindustries.com/roadmap/release-view
- Stysley. 2011. “Laser-Based Optical Trap for Remote Sampling.” https://www.nasa.gov/wp-content/uploads/2017/07/niac_2011_phasei_stysley_laserbasedopticaltrap_tagged.pdf
- Take-Two. 2017. “Take-Two Acquires Kerbal Space Program.” https://ir.take2games.com/static-files/f86fae80-882c-4c68-b71c-07560c4b1183
- Take-Two. 2023. “Kerbal Space Program 2 Launches in Early Access.” https://ir.take2games.com/node/29461/pdf
- ““The Last Starfighter (Preview)” by Digital Productions.” n.d. Accessed November 13, 2026. https://history.siggraph.org/?p=94037
General Sources
- Bettner, Paul, and Mark Terrano. 2001. “1500 Archers on a 28.8: Network Programming in Age of Empires and Beyond.” Game Developer. https://www.gamedeveloper.com/programming/1500-archers-on-a-28-8-network-programming-in-age-of-empires-and-beyond
- Big Think. 2015. “A Universe in a Nutshell: The Physics of Everything, with Michio Kaku.” Big Think. https://bigthink.com/videos/universe-in-a-nutshell-the-physics-of-everything-with-michio-kaku/
- Borger, Julian. 2002. “Wake-up Call.” The Guardian, September 6. https://www.theguardian.com/world/2002/sep/06/usa.iraq
- Brynen, Rex. 2010. “Review: Wargaming for Leaders.” PAXsims, November 19. https://paxsims.wordpress.com/2010/11/19/review-wargaming-for-leaders/
- Chalk, Andy. 2024. “Kerbal Space Program 2 Developer Intercept Games and Rollerdrome Studio Roll7 Are Reportedly Closing as Part of Take-Two’s Big Layoff Plan.” PC Gamer, May 1. https://www.pcgamer.com/games/sim/kerbal-space-program-2-developer-intercept-games-may-be-closing-as-part-of-take-twos-big-layoff-plan/
- Drain, Brendan. 2013. “EVE Online Smashes PCU Record During Anniversary Celebrations.” Engadget. https://www.engadget.com/2013-05-06-eve-online-smashes-pcu-record-during-anniversary-celebrations.html
- Engadget. 2008. “Making/Money: The Origin of the Specie.” Engadget. https://www.engadget.com/2008-06-29-making-money-the-origin-of-the-specie.html
- Engadget. 2009. “New Perspective on EVE Online’s Latest Bank Embezzlement.” https://www.engadget.com/2009-07-02-new-perspective-on-eve-onlines-latest-bank-embezzlement.html
- Engadget. 2014. “EVE Fanfest 2014: Economy Talk Highlights PLEX Prices and Reveals Titan Production Statistics.” Engadget, May 2. https://www.engadget.com/2014-05-01-eve-fanfest-2014-economy-talk-highlights-plex-prices-and-reveal.html
- EVE University. n.d.-a. “Acceleration.” Accessed November 13, 2026. https://wiki.eveuniversity.org/Acceleration
- EVE University. n.d.-b. “Capacitor.” Accessed November 13, 2026. https://wiki.eveuniversity.org/Capacitor
- EVE University. n.d.-c. “Industry.” Accessed November 13, 2026. https://wiki.eveuniversity.org/Industry
- EVE University. n.d.-d. “Jump Drives.” Accessed November 13, 2026. https://wiki.eveuniversity.org/Jump_drives
- EVE University. n.d.-e. “Trading.” Accessed November 13, 2026. https://wiki.eveuniversity.org/Trading
- EVE University. n.d.-f. “Warp.” Accessed November 13, 2026. https://wiki.eveuniversity.org/Warp
- Falanghe, Felipe. 2012. “Patched Conics Trajectory from Initial Orbit State.” Physics Forums. https://www.physicsforums.com/threads/patched-conics-trajectory-from-initial-orbit-state.590896/
- GamersNexus. 2016. “Chris Roberts on Character Tech, Weather System, & Engine Architecture.” https://gamersnexus.net/gg/2617-chris-roberts-character-technology-weather-system-engine-arch
- GeekWire. 2024. “Take-Two Interactive Layoffs Hit Seattle Area.” https://www.geekwire.com/2024/take-two-layoffs-hit-seattle-area-70-workers-affected/
- Gematsu. 2026. “CCP Games Rebrands to Fenris Creations Following Split from Pearl Abyss.” Gematsu, May 6. https://www.gematsu.com/2026/05/ccp-games-rebrands-to-fenris-creations-following-split-from-pearl-abyss
- Gizmodo. 2011. “Read the First Draft of Gene Roddenberry’s Star Trek Pitch, Starring ‘Captain April’ and the ‘S.S. Yorktown’.” Gizmodo, January 29. https://gizmodo.com/read-the-first-draft-of-gene-roddenberrys-star-trek-pit-5746652
- Grayson, Nathan. 2018. “Star Citizen Backer Sues to Get $4,500 Back, Loses.” Kotaku, July 17. https://kotaku.com/star-citizen-backer-sues-to-get-4-500-back-loses-1827666550
- Hattendorf, John B. 2014. “Saving Nimitz’s ‘Graybook’.” Naval History Magazine, May 1. https://www.usni.org/magazines/naval-history-magazine/2014/june/saving-nimitzs-graybook
- “Ideal Rocket Equation.” n.d. Glenn Research Center | NASA. Accessed October 25, 2026. https://www1.grc.nasa.gov/beginners-guide-to-aeronautics/ideal-rocket-equation/
- IEEE Spectrum. 2009. “Eve Online Game Player Embezzles Virtual Money, Creates Real Trouble.” IEEE Spectrum, June 16. https://spectrum.ieee.org/eve-online-game-player-embezzles-virtual-money-creates-real-trouble
- “KerbalEdu.” 2016. Accessed November 13, 2026. https://web.archive.org/web/20160104215459/http://www.kerbaledu.com/
- KSP Wiki. n.d.-a. “Kerbin.” Accessed November 13, 2026. https://wiki.kerbalspaceprogram.com/wiki/Kerbin
- KSP Wiki. n.d.-b. “Maneuver Node.” Accessed November 13, 2026. https://wiki.kerbalspaceprogram.com/wiki/Maneuver_node
- KSP Wiki. n.d.-c. “Multiplayer.” Accessed November 13, 2026. https://wiki.kerbalspaceprogram.com/wiki/Multiplayer
- KSP Wiki. n.d.-d. “Sphere of Influence.” Accessed November 13, 2026. https://wiki.kerbalspaceprogram.com/wiki/Sphere_of_influence
- Massively Overpowered. 2016. “Star Citizen Changes Refund Clause for New Backers, Plans Procedurally Generated Planets for 2.7.” https://massivelyop.com/2016/06/21/star-citizen-changes-refund-clause-for-new-backers-plans-procedurally-generated-planets-for-2-7/
- PC Gamer. 2021. “Star Citizen Adds Disclaimer to Spaceship Preorders after Regulator Says They’re Misleading.” PC Gamer, September 8. https://www.pcgamer.com/star-citizen-adds-disclaimer-to-spaceship-preorders-after-regulator-says-theyre-misleading/
- Polygon. 2017. “Star Citizen Backer’s $25,000 Refund Has Taken Months, Still in Dispute (Update).” Polygon, December 14. https://www.polygon.com/2017/12/13/16767590/star-citizen-refund/
- Reuters. 2009. “Gamer Steals Online Loot to Pay Real-World Debts.” https://www.eeworldonline.com/gamer-steals-online-loot-to-pay-real-world-debts/
- Slashdot. 2013. “Interviews: J. Michael Straczynski Answers Your Questions.” https://entertainment.slashdot.org/story/13/04/15/1328251/interviews-j-michael-straczynski-answers-your-questions
- Space.com. 2015. “‘The Expanse’ Authors Talk Space’s Epic Size and Crazy Sci-Fi Tech.” Space.com, December 23. https://www.space.com/31437-the-expanse-crazy-sci-fi-tech.html
- “Spacecraft Operations.” n.d. Traveller SRD. Accessed November 13, 2026. https://www.traveller-srd.com/core-rules/spacecraft-operations/
- Star Citizen Wiki. n.d. “Planet.” Accessed November 13, 2026. https://starcitizen.tools/Planet
- “The Star Trek Transcripts: Obsession.” n.d. Accessed November 13, 2026. http://www.chakoteya.net/StarTrek/47.htm
- The Mittani. 2012. “Sins of a Solar Spymaster #78: The Extortion of Empire.” https://www.tentonhammer.com/columns/sins-of-a-solar-spymaster-78-the-extortion-of-empire
- The Register. 2009. “Gamer Embezzles Virtual Cash to Settle Real Debts.” The Register, July 3. https://www.theregister.com/security/2009/07/03/gamer-embezzles-virtual-cash-to-settle-real-debts/1491814
- Yancy, Lee. 2017. “War in EVE Online Is the Best, and Also the Worst.” Kotaku, August 18. https://kotaku.com/war-in-eve-online-is-the-best-and-also-the-worst-1797982059
- Yancy, Lee. 2020. “How EVE Online Is Changing Players’ Lives for the Better.” Den of Geek, December 14. https://www.denofgeek.com/games/eve-online-players-lives-better/
Related Posts
- Sechter, Brendan. 2026. “What Published Wargames Say About a War With China.” Brendan A R Sechter’s Development Blog, August 11. https://sgeos.github.io/geopolitics/military/war-gaming/2026/08/11/published_wargames_of_war_with_china.html
- Sechter, Brendan. 2026. “Orbital Manoeuvres in Terms of Delta-v.” Brendan A R Sechter’s Development Blog, August 17. https://sgeos.github.io/space/space-studies/2026/08/17/orbital_manoeuvres.html
- Sechter, Brendan. 2026. “Solar Sailing.” Brendan A R Sechter’s Development Blog, August 18. https://sgeos.github.io/space/space-studies/2026/08/18/solar_sailing.html
- Sechter, Brendan. 2026. “Mass Drivers.” Brendan A R Sechter’s Development Blog, August 19. https://sgeos.github.io/space/space-studies/2026/08/19/mass_drivers.html