Design Note · Espionage

Spy systems across 4X

How ten major 4X games model espionage — what the agent is, what defence costs, how a mission resolves, and what happens diplomatically when you get caught. Compiled as raw material for designing an original system.

September 2026 · fourth in a series with Board Game References, Ship Progression Models and Online Play Considerations

The ROTP baseline, in eight sentences

Starting from the Seaworth write-up, because it is the most precisely documented system on the list and the one everything else can be measured against.

One spy network per directed relationship, funded separately per target — but a single empire-wide security slider defends against everyone at once. Attacking costs per enemy; defending costs once.
A spy is three integers, re-rolled every turn. No name, no experience, no history. Losing one costs the purchase price, not an accumulated asset — which is why a 30% per-turn casualty rate does not feel punitive.
Two independent rolls each turn: an infiltration roll (do you survive contact with their security) and a mission roll (do you succeed at the job). Conflating them is the standard misreading.
Hiding still delivers full passive intelligence — colonies, fleet composition, tech tree — at a −0.30 detection bonus. Reconnaissance is nearly free; only the offensive act is priced.
Steal depth scales with the thief's own tech level. Espionage rewards the leader, not the laggard. This is the opposite of most later games.
Security has three separate effects — it shifts the detection table, shrinks the pool of techs visible to steal, and scales down the winning steal roll — so it is far stronger against theft than the detection numbers suggest, and does nothing at all against sabotage magnitude.
Sabotage damage keys off weapons tech, not computers. The spy gets you in the door; what you can destroy depends on what explosives you know how to make.
Espionage cannot start a war on its own. Incidents only become a valid casus belli while a warning timer is live, and issuing a warning is what charges it. The loop is: spy, get caught, get warned, keep spying, get invaded.

The one idea worth stealing outright

That last point is the most transferable thing in the entire system. It converts espionage from a hidden dice game into a negotiation with an explicit decision point. The dramatic moment is not being caught — it is choosing to continue after the complaint. Almost no other 4X models this; most treat discovery as a flat relations penalty and leave the player nothing to decide.

Five archetypes

Every system on the list answers the same first question — what, mechanically, is a spy? — in one of five ways. Everything else follows from that answer.

A · The funded network MOO1 / ROTP · Civ IV: Beyond the Sword

No units and no characters. A percentage of production buys an abstract headcount aimed at one rival, and a second slider buys defence. Missions are standing orders, resolved silently between turns; the player reads a report rather than watching a roll.

Strength: Zero micromanagement at any empire size, and it scales cleanly — the tenth rival costs the same interface as the first. Resolves entirely server-side, which matters enormously for turn density.
Weakness: Nothing to do. Espionage becomes a budget line you set once and revisit rarely, so it produces few memorable moments and reads to some players as weather rather than decision.

B · The produced unit MOO2 · Master of Orion (2016) · Galactic Civilizations III/IV

Spies are built like ships, out of the same production that buys everything else, then assigned to a rival empire or held home for counter-intelligence. They occupy no map position and cannot be intercepted in transit; assignment is a slot in a menu.

Strength: Real opportunity cost against military and infrastructure, expressed in the currency the player already understands. Stacking several agents on one target creates a legible commitment.
Weakness: Scales badly. By the late game the player is reassigning a column of interchangeable agents every few turns, which is exactly the complaint levelled at the 2016 remake. The assignment screen becomes a chore list.

C · The map unit Alpha Centauri probe teams · Civ IV spy units

The agent is a physical unit with a chassis and movement that must walk or sail to the target and can be intercepted en route. Defence is another probe team standing in the base; when one enters, the two fight, and the higher morale usually wins.

Strength: Espionage becomes geography. Distance, chokepoints, escorts and speed all matter, and the whole system reuses the movement and combat rules the player already knows — no separate subsystem to learn. It also makes espionage visible, which makes it dramatic.
Weakness: Only works when the map is the primary interface. In a game where fleets are abstracted or ranges are radial rather than path-based, a walking spy has nothing to walk through. Alpha Centauri's mind control — buying an enemy base outright with energy credits — is also the single most swingy espionage effect in the genre.

D · The placed agent on a timer Civ V · Civ VI · Endless Legend: Shadows

The agent is assigned to a specific enemy place — a city, or in Civ VI a specific district — and then runs a mission with a duration. The outcome resolves at the end of the timer, into a graded result: undetected success, undetected failure, detected, captured, killed.

Strength: Target selection is a genuine strategic decision, because places differ. Civ VI's district targeting means "which of their cities, and which building in it" is the whole game. Endless Legend couples defence to city approval, so an unhappy empire is a leaky one — a superb feedback loop that costs no new systems.
Weakness: Prompt-heavy. Civ VI adds a second, player-chosen escape roll after the mission — faster escape route, higher risk — which is great drama live and a round-trip in async play. Agent counts are small and hand-managed.

E · Infiltration as a stock Stellaris (Nemesis) · Endless Legend infiltration points

Presence generates a resource. An envoy assigned to a rival accumulates infiltration toward a cap; operations have an infiltration threshold to unlock and a cost to run, so completing one spends down the access you built. Operations then resolve as a staged sequence of rolls against a difficulty, with random events during preparation that can raise the difficulty or demand you burn an asset to avoid exposure.

Strength: Espionage becomes an economy rather than a dice game — you save up for a big operation the way you save for a wonder. Because envoys are the same limited pool used for diplomacy, spying and diplomacy compete directly for one resource, which is the cleanest expression of "shadow channel versus open channel" on the list.
Weakness: Long feedback loops and a lot of interface. The staged-roll model produces a stream of small event prompts, and the numbers governing it — encryption against codebreaking, sprawl, asset skill — are hard to hold in the head at once.

Game by game

Master of Orion 1 / Remnants of the Precursors

Covered above. Worth adding: the spending model caps a single target at 10% of production and all targets combined at 50%, scaling everything down proportionally past that, so a wide espionage programme is self-limiting without a hard rule. Race modifiers are three numbers, and only one race is non-neutral — half-price spies, double base mission success, and free maximum counter-intelligence. That single row is the entire Darlok identity.

Master of Orion 2

Spies become built units, and each mission type carries a fixed base risk — community reverse-engineering puts stealing charts near 0.3, stealing tech near 0.5, and sabotage, assassination and inciting revolt at 1.0 — modified by the target's empire security and your own espionage bonuses. Empire security is diluted across colonies, so a wide empire is inherently leakier per world. Governments modify counter-intelligence, and racial picks let a player buy an espionage specialism at creation rather than receiving one.

The detail most worth noting is anti-save-scum: the failure roll is derived from the turn number and the galaxy seed, so reloading and re-running the same mission on the same turn reproduces the same outcome. Retrying requires spending a different turn. That is a deliberate answer to a problem every probabilistic spy system has.

Master of Orion (2016)

Spies gain levels; colonies carry a security rating; counter-espionage agents add their levels to empire security, scaled by a racial factor. More missions and more agent character than MOO2. Its instructive failure is late-game volume: captured spies are returned, new ones arrive, and the player ends up reassigning a large roster every few turns. The community complaint is not that the system is shallow but that it is clerical — several players note they end up parking everyone on defence purely to stop dealing with it.

Alpha Centauri

Probe teams are units built from a modular chassis, so a fast probe on a hovertank frame is a different tool from a foot team. They can steal research, sabotage facilities, incite drone riots, assassinate researchers, and — most dramatically — mind-control a base or subvert a unit outright for energy credits. Defence is a probe team garrisoned in the base, resolved as a contest between the two, usually won by the higher morale. A social-engineering rating modifies both attack and defence empire-wide, so espionage capability is a policy choice, and one secret project shuts enemy probes out of your empire entirely.

Two ideas stand out: espionage strength as a governance setting rather than a spending line, and a single wonder that ends the entire subsystem for one player. The second is a blunt instrument, but it does mean a player who hates the minigame has an exit.

Civilization IV: Beyond the Sword

A hybrid: an espionage slider accumulates points per rival, and those points are spent on discrete missions — investigate a city, steal a technology, sabotage production, incite a revolt — with costs scaled by distance, era and the target's own counter-espionage. Spy units exist as well and must reach the target. Closest ancestor of the Stellaris stock model, but with the currency exposed as a plain number.

Civilization V (Gods & Kings / Brave New World)

Spies arrive by era, are placed in a city, and grind. In a foreign city they steal technology over time at a rate driven by that city's science output, slowed by defensive buildings that reduce the acquisition rate without reducing the eventual success. In a city-state they rig elections or attempt a coup. In your own city they run counter-intelligence and may kill intruders. Ranks — Recruit, Agent, Special Agent — are earned by successful thefts or kills, and improve theft speed, rigging odds, coup odds and detection.

Brave New World adds the Diplomat: the same unit, reassigned to a purely political role, which reveals how a rival intends to vote in the World Congress and enables vote trading. That is the strongest example on the list of espionage aimed at a victory condition rather than at the economy.

Civilization VI

The most tactile implementation. Spies are capped units that travel between cities with real travel times, then target a specific district: steal gold from a commercial hub, take a technology boost or great work from a campus, sabotage a spaceport to slow a science victory, disrupt an industrial zone. A mission has a duration and a difficulty, then a second and separate escape decision — the player chooses an extraction route, faster routes being riskier. Outcomes are graded, and a captured spy is imprisoned rather than killed and can be traded back diplomatically. Successful offensive missions grant a promotion chosen from three, which snowballs: early success makes later success likelier.

Stellaris (Nemesis)

Covered in archetype E. Additional detail worth carrying forward: infiltration growth is modified by the target's sprawl, so large empires are structurally easier to infiltrate; assets are recruited characters that add skill to matching operations and can be sacrificed mid-operation to avoid a bad outcome; and a sleeper-cell operation freezes a network's level so the envoy can be withdrawn for other work. The operation list is explicitly tiered by infiltration requirement, which turns the stock into a visible progress bar toward named rewards.

Endless Legend: Shadows

The agent is a hero — the same scarce asset that could otherwise govern a city or lead an army — sent into a visible enemy city for an influence cost. While infiltrated the hero leaves the map entirely, you see the city as though you owned it, and infiltration points accrue each turn according to the hero's spying level. Points buy actions against the city or the empire. Defence is the city's security rating, boosted by governor skills, items and buildings, and — critically — modified by approval: a content city is hard, an unhappy or rebelling city is easy.

Galactic Civilizations III / IV

Two parallel channels: assign a spy to a civilization, which raises an abstract espionage level that eventually produces a technology drip, or assign one to a planet, where it sabotages specific buildings. Unassigned spies form a counter-espionage pool. The design is instructive mostly as a warning — the abstract channel produces theft with almost no feedback, and players report losing technology repeatedly without a reliable notification of who took it or whether their defence is working.

Distant Worlds 2

Named characters with skills — espionage, concealment, counter-intelligence — plus traits that suit them to particular mission types. Missions display a success percentage and a duration, and resolve into success, partial success, failure or capture, with a reputation penalty when caught. Difficulty scales with the target empire's size and with the tier of the technology being stolen.

Its well-documented problem is a good lesson: the displayed success percentage does not include enemy counter-intelligence. Players see 90%, lose the agent, and conclude the game is lying to them. Patch notes across its first year show repeated rebalancing toward more partial successes and fewer captures — the developers converging on the view that in a system with named, levelled agents, total loss is too harsh a default outcome.

Comparison — structure

Game	What a spy is	What it costs	Target granularity	Defence model	Resolution
MOO1 / ROTP	Abstract headcount, stateless	% of production, per target, capped 10% each / 50% total	Empire (sabotage picks a system)	Asymmetric one empire-wide slider	Two dice per spy per turn, silent
MOO2	Built unit, levelled	Production, like a ship	Empire	Symmetric agents held home + colony security	Per-mission risk value, seeded roll
MOO (2016)	Built unit, levelled	Production	Colony	Symmetric agents + colony security tier	Percentage roll per mission
Alpha Centauri	Map unit with a chassis	Minerals + travel time	Base or single unit	Symmetric garrisoned probe + policy rating	Opposed contest, morale-weighted
Civ IV: BTS	Point pool + optional unit	Slider %, points banked per rival	City	Asymmetric own point total + buildings	Spend points, cost-versus-defence check
Civ V	Placed agent, 3 ranks	Free by era; buildings for defence	City / city-state	Symmetric counter-spy + rate-slowing buildings	Progress timer, then outcome roll
Civ VI	Capped unit, promotions	Production + travel time	District — the finest on the list	Symmetric counter-spy covering adjacent districts	Mission roll, then a separate escape roll
Stellaris	Envoy + recruited assets	Envoy slot, influence, energy upkeep	Empire (some target a planet or starbase)	Asymmetric encryption vs codebreaking, empire-wide	Staged rolls over months, with events
Endless Legend	A hero — a scarce shared asset	Influence + the hero's opportunity cost	City	Symmetric per-city security, tied to approval	Spend accumulated infiltration points
GalCiv III / IV	Citizen assigned to a slot	A citizen — traded against other roles	Civilization or planet	Asymmetric unassigned spies pool	Background drip, weak feedback
Distant Worlds 2	Named character with skills	Character upkeep; recruitment capped by population	Empire / colony / research project	Symmetric counter-intelligence agents	Single roll after a chosen duration

Comparison — behaviour and consequence

Game	Passive intel	Signature payloads	Agent progression	Getting caught	Favours	Async fit
MOO1 / ROTP	Free with presence	Steal tech; destroy factories or bases; incite rebellion; frame a third party	None — stateless	Compounding grievances; war only via a warning timer; allies break instantly	The leader	Excellent
MOO2	Charts and reports, as a mission	Steal tech; sabotage; assassinate; incite revolt	Levels	Relations hit; the victim may demand you stop	Either	Good
MOO (2016)	Per-colony intel	Steal tech; sabotage; revolt	Levels	Captured spies returned or executed; relations hit either way	Either	Fair — high volume
Alpha Centauri	Permanent datalink infiltration	Steal research; sabotage; incite riots; assassinate; mind-control a base	Morale ranks	Vendetta risk; probing a pact partner is a major breach	The rich	Poor — map movement
Civ IV: BTS	Investigate city as a mission	Steal tech; sabotage production; incite revolt; switch civics	None	Relations hit	The laggard	Good
Civ V	Surveillance after setup	Steal tech; rig elections; coup; vote intel via the Diplomat	3 ranks	Relations hit; forgiving a caught spy earns goodwill	The laggard	Fair
Civ VI	Listening post / gain sources	Steal gold, boosts, great works; sabotage spaceports and industry; recruit partisans	Levels + chosen promotions	Grievances; captured spies are tradeable hostages	The laggard	Poor — escape prompt
Stellaris	Infiltration raises intel passively	Steal tech; extort favours; sabotage starbase; arm privateers; discredit envoys	Assets accumulate and are spent	Relations hit and an incident; scaled by what was attempted	Either	Fair — event-heavy
Endless Legend	Full city view while infiltrated	Steal resources; reduce population; reveal enemy spies; empire-wide actions	Hero XP while infiltrated	Exposure can trigger war outright	The patient	Good — points accrue silently
GalCiv III / IV	Planet detail	Tech drip; disable buildings	Levels over time	Barely modelled	Either	Good but inert
Distant Worlds 2	Steal territory map as a mission	Steal research; sabotage; psy-ops to push a colony toward rebellion	Skills and traits, incl. going double agent	Reputation penalty; agent captured or killed	The leader	Fair

The design axes

Stripped of implementation, the genuine choices are these seven. Every game above is a point in this space.

1 · Symmetry of attack and defence

Per-target attack, single defence (MOO1, Stellaris) ←→ Both per-target (Civ V/VI, SMAC)

MOO1's asymmetry is the most economically interesting rule in any of these systems: attacking N rivals costs N times as much, defending against all of them costs once. It naturally makes espionage a specialist strategy — you cannot afford to spy on everyone — while making defence a sensible baseline for everyone. Symmetric systems instead make the espionage game a per-relationship arms race, which is more granular but multiplies the interface.

2 · Who does espionage reward?

The leader (MOO1, DW2) ←→ The laggard (Civ IV/V/VI)

This is the single largest balance decision and it is usually made implicitly. Civ's model — steal what you don't have from someone who does — is a rubber band, and players who are ahead experience it as a tax on success. MOO1's model, where a thief can only reach as deep as their own tech level, makes espionage an amplifier of a lead. A third option almost nobody uses: make the reward asymmetric by kind rather than by depth, so the leader steals different things than the laggard.

3 · Is intelligence free with presence, or a separate purchase?

Free while hidden (MOO1, Endless Legend) ←→ Its own mission (Civ VI, DW2)

MOO1's choice is the stronger one and it is barely noticed: hiding delivers the full intelligence package at a detection bonus, so a small quiet network is a permanent reconnaissance satellite and the dominant peacetime play. That means the espionage system earns its keep even for a player who never runs an offensive mission — which is exactly how you stop a subsystem from being ignorable.

4 · Do agents persist?

Fungible and stateless (MOO1) ←→ Named, levelled, irreplaceable (DW2, Civ VI)

Progression makes agents feel like characters and makes losing one hurt — which sounds like a virtue until you notice the consequences. It produces save-scumming, it snowballs (Civ VI: promotions from success make more success likelier), and it makes the punishing outcome — capture — feel unfair rather than dramatic. Distant Worlds 2's patch history is a year of walking that back. MOO1 tolerates a 30% casualty rate precisely because there is nothing to mourn.

5 · What licenses war?

A flat relations penalty (most) ←→ A warning timer you can ignore (MOO1)

The warning-timer gate is the best single mechanic in this whole survey and it is almost unique to MOO1. Getting caught is not the decision; continuing after a formal complaint is. It also produces an elegant property: an empire that never warns you cannot legally go to war over your spying, and one that stops warning has effectively forgiven you.

6 · Where does defence come from?

A dedicated spending line (MOO1, Civ IV) ←→ A side effect of other systems (Endless Legend)

Endless Legend ties city security to approval, so espionage defence is a by-product of running a happy empire rather than a separate budget. That is worth more than it sounds: it makes espionage vulnerability a consequence of decisions made elsewhere, which is the same "no dead ends" property that makes the MOO1 systems map hold together. MOO1 does a smaller version of this — having no trade treaties grants a flat counter-espionage bonus, so opening your economy opens your borders.

7 · What is the target — economy, technology, or victory?

Tech and factories (MOO1, MOO2, GalCiv) ←→ Victory progress (Civ V Diplomat, Civ VI spaceport)

Almost every space 4X aims espionage at technology, because tech is the universal currency. But the most memorable spy plays in the genre are the ones aimed at a victory condition: sabotaging a spaceport to stall a science win, or reading a rival's council vote before it is cast. In a game with mutually exclusive victory conditions, this is by far the most interesting target class — and the least explored.

Failure modes, catalogued

Espionage is the most reliably disliked subsystem in 4X. The complaints are consistent enough to be treated as a checklist.

Clerical volume. Late-game rosters of interchangeable agents that must be reassigned every few turns. The 2016 remake's headline problem, and Galactic Civilizations' too.
Invisible defence. A displayed success percentage that omits the enemy's counter-intelligence. Players do not conclude "the odds were against me"; they conclude the game lied.
No feedback loop. Losing technology every few turns with no reliable notification of who took it or whether your defence is working. Defence spending becomes an act of faith.
Save-scumming. Any high-variance, high-stakes single roll invites reloading. MOO2's seeded roll is the only deliberate countermeasure found in this survey, and it is a good one.
Rubber-band resentment. A system whose main function is transferring the leader's advantage to whoever is behind reads as a punishment for playing well.
Weather, not decision. The opposite failure: a background drip with no moment of choice, which players correctly ignore.
Swinginess. Alpha Centauri's mind control can transfer a fully developed base for money. Memorable, and the reason several groups house-rule it away.

Hooks to brainstorm from

Not recommendations — starting points, chosen because they interact with commitments already made in this series (deterministic resolution, batched economic phases, standing orders, hidden information as a first-class system) and with seven factions whose victory conditions structurally undermine one another.

Aim espionage at victory progress, not at the tech web. With six mutually exclusive victory conditions, the natural spy target is a rival's path — the artefact they are about to recover, the vote they are about to cast, the pilgrimage route they depend on. This is the least-explored target class in the genre and it is exactly where the design's tension already lives.
Keep the warning timer. Discovery should not itself cause war. It should charge a clock that makes war legal, and the interesting decision should be whether to continue after the complaint. This survives translation to async play perfectly, because it is a decision the player makes on their own turn.
Make intelligence free with presence and price only the offensive act. This guarantees the subsystem is used by every player every game, not just by the one faction built around it, and it makes standing down a network a real cost rather than a saving.
Keep agents stateless. Progression buys character at the price of save-scumming, snowball and loss aversion. The network can have a level; the individual should be fungible. Any per-agent RPG layer also multiplies async prompt volume.
Source defence from somewhere else. Endless Legend ties security to approval; MOO1 ties it inversely to trade. Both make espionage vulnerability a consequence of decisions taken in another system, which is the property the whole design philosophy is built on. A dedicated counter-espionage slider is the boring answer.
Batch it into the economic phase. Per-turn detection rolls that generate prompts are exactly the density problem identified in Online Play Considerations. Espionage orders belong on the same screen as the build queue, resolving silently between turns into a single intelligence report — which happens to be what MOO1 already does.
Differentiate factions by vector, not by a multiplier. The Darlok solution is one row of numbers, which is admirably cheap but produces "the spy race" rather than seven espionage cultures. With factions defined by their relationship to AI and to biological modification, the more interesting axis is what can even be infiltrated: an empire that refuses AI has no network to hack and no way to hack anyone else's; one built on deep human–AI symbiosis has a vast attack surface and a native intrusion capability; one built on radical self-modification can place agents that are not detectable as foreign at all. Immunity to a vector is as strong a faction identity as a bonus to one, and cheaper to balance.
Give the system an exit. Alpha Centauri's one wonder that shuts out enemy probes entirely is crude, but it acknowledges that some players hate the minigame. A terminal defensive option — expensive, exclusive, and worth giving something up for — is worth considering, particularly if a faction's identity is refusing this whole channel.

The one to sit with

Item 7 is the only place in this survey where the standard answer looks clearly weak. Every game here treats espionage capability as a scalar — better spies, worse spies — and then hands one faction a large positive number. A system where factions differ in which channels exist for and against them would produce genuine rock-paper-scissors from asymmetry that already exists in the setting, without a table of unit types and without a balance number to tune.