Mechanics

The AI opponent

Opponents do not cheat. They use the same rules you do, see the same fog and play under the same resource constraints. Their decisions derive from the seed: the same seed and the same command sequence reproduce the same match.

Personalities

A personality is not a hidden behaviour script — it is a set of weights.

• Expansionist — weights founding cities and region value highly. Its economy grows, its army comes late. Weakness: exposed to early pressure.
• Defensive — prioritises Walls and Spearman, is strong on its border, expands slowly. It fortifies its units: inside its territory, a unit that is wounded or under threat digs in (Combat). Once a set share of the match has passed — after turn 22 in a Short match, 36 in Standard, 44 in Long — if it is behind on score, it multiplies its aggression weights by 2.2 — it will not sit still and make the match dull. This wake-up belongs to the Defensive personality alone.
• Opportunist — tracks power ratios; attacks whoever weakens and avoids fights with the strong. It will fortify a wounded unit inside its territory and wait for it to recover, but it does not hold a line the way the Defensive one does. Weakness: it can open two fronts.

Fortification is one of the visible personality differences: the Expansionist never fortifies — an expansion army does not dig in, it walks.

The growth focus (Cities and population) is not an opponent difference: none of the three personalities sets a focus, so food follows the default rule — the least populous eligible city grows. The Defensive used to make its capital the focus, and measurement took that back. The rule has changed since: a city at the population cap can no longer receive food, so nothing flows into a full store any more (Food). The arm was measured again after that fix and the Defensive's win share still dropped — feeding one strong centre leaves the border cities small. That is why setting a focus is a tool in your hands; the opponent does not use it.

The opponent does see the cap: in a city whose population sits at the cap it wants the building that opens it more (Farm, and Almshouse once the quota is full); with the quota full, a capped city choosing research puts the cap-raising nodes (Healing, Cavalry, Land Title) first. When none of its cities can take food, its surplus goes to its granary too — the rule lives in one place for both sides.

Truce appetite (Diplomacy and truces) is the second visible difference: the Defensive pays for time, the Opportunist uses a truce to close one front, and the Expansionist never begs for peace — it accepts one when crushed but never offers. An opponent only weighs the force it sees: it never asks peace of a player it has not met.

What an opponent builds

When a city's queue empties, it asks a single question: which item returns the most value per unit of production spent? The score is the personality's weight, scaled by a situational multiplier, divided by the item's cost — so a cheap item beats an expensive one unless the personality wants the expensive one distinctly more. The multiplier is a condition read off the board ("my city quota is full", "an enemy is near", "I know where a rival city is"), and it is what moves a personality from one item to the next. For the location-bound buildings (Farm, Workshop, Market) the score is also multiplied by a location factor: the income the building would yield in that city, read through the very same income calculation the engine uses, divided by the building's middle income. So the opponent leans toward putting a Workshop in its forested city and a Market in its coastal one; it does not forbid the building in a city without matches, it just wants it less. The Pioneer's rising cost enters the same division: as the city count grows, the Pioneer's score drops on its own. The Pioneer's score is also multiplied by the match's remaining-turn share — (last turn − this turn) / last turn, the same share research uses: a settler is an investment that pays back in the turns after the city is founded, and a city founded near the end of the match does not see enough turns to repay its cost. So the opponent expands early, and wants the Pioneer less and less as the match nears its end. Units and buildings compete in the same queue (The production queue), so a city filling a building slot is a city not raising an army.

Here too the difference between personalities is not a hidden script but the result of their weights:

• All three open with one Scout: while no rival city is known, the cheapest item buys the most information. With a scout already in the field the second one loses its value and never reaches the queue — but kill the Scout and a replacement goes in, because the test is "do I have a scout out there", not a counter. Once a rival city is found the Scout falls back to its base value too: they stop building them not by prohibition but because every other item outscores it.
• The Expansionist, once its opening Pioneer has settled, raises fresh Pioneers and switches to Siege once its city quota (6) is full and it knows where an enemy city sits: with nowhere left to expand, it takes its points from a rival instead of the ground. Once the quota is full it also builds an Monument — the only personality that raises one.
• The Defensive fills its slots with Farm, Workshop and Walls, and builds its army around Spearman. It never raises a Monument itself, but it inherits one along with a city that falls to it.
• The Opportunist raises Raider and gives one of its slots to a Market — the only personality that builds one, because it spends its gold rather than hoarding it (see below). A Monument is its last choice: its late game is finding a weakened neighbour, not erecting a landmark.

The prerequisite binds opponents too

The strategic resource rule (The production queue) applies to opponents exactly as it does to you: an opponent with no Horse pasture in its territory builds no Raider, and with no Iron deposit no Siege. It never even queues those items — it does not produce commands the engine would reject.

The shortage shows up in two ways:

• In its marches. A tile carrying a resource the opponent is missing picks up a second weight on top of an ordinary node tile's value, so its units walk toward that land more eagerly. The weights add, they do not multiply — the tile keeps its ordinary node value and the missing resource stacks onto it.
• In its production. While a strategic resource is missing, the Pioneer gains value: the way to reach the resource is a new city. This only applies while the city quota is not full and no other settler is already in the field.

The practical consequence is an opportunity for you: take an opponent's horse pasture or iron deposit first and you deny it not just income but a whole unit.

Research and imar buildings

Research is an item in the same queue too, and the AI weighs it after units and buildings: at an equal score, the side that ends up with something concrete wins. Every node in the grid follows the same pattern: the base is zero and none is a candidate until one of two gates opens. The first is a full city quota (6); the second is a production pool surplus: if the pool still has at least 36 production left after covering what the cities' queues still need, the AI spends production on research before its quota fills. Both gates carry the same weight; once either opens, the nodes of imar, muhafaza, harp and the doctrine-free iskan branch together reach a positive score (Research and the tech tree). The mastery nodes (kanun, kapikulu, seferberlik) are not exempt from this rule, and carry one more condition on top: they are only a candidate for a rival who has chosen their own branch's doctrine.

Wisdom is the lowest-weighted research candidate: it is only weighed while there is a pool surplus, and its weight stays far below that of the branch nodes. Once the tree is done, or the remaining nodes are closed off by doctrine, it is the only research outlet left for surplus production. The rival also keeps the empty-queue default on (The production queue): a city whose queue runs out flows into Wisdom on its own, and if the rival finds the setting off it turns it back on.

The three personalities do not choose a doctrine the same way: each one takes the doctrine that carries the highest score among the doctrine branches' weights, and there is no tie, because each personality has one branch distinctly ahead (Research and the tech tree). The personality that carries harp as its main branch and actually reaches all five of the branch's nodes is the Opportunist; it is also the one that chooses gaza and builds Forlorn Hope.

Five imar buildings follow the same pattern: base zero, and the multiplier kicks in once the quota is full. None of the six gets built while an enemy is near — a border city produces soldiers, an interior city builds. The Bedesten's quota multiplier is higher than the other four buildings'; the Divan's base is above zero and it is already a candidate before the quota fills (Divan).

The iskan branch takes no part in the doctrine choice — tapu is not a doctrine — and its branch weight is the same for all three personalities. Its buildings follow the imar pattern too: base zero, kicking in once the quota is full; the Mill, Foundry and Land Registry are not built while an enemy is near. The Watchtower carries no such guard: it is a border city's building and stays a candidate with an enemy near.

How an opponent spends gold

An opponent can also finish the head of its queue with gold (Economy), and it pays the same rate you do — there is no separate price for it. It treats the decision as a trade: the gold in its purse turns into score at the end of the match (Gold), so rushing gives up points in exchange for turns saved. It prices those forfeited points with an internal rate separate from the scoreboard's (identical for all three today); what differs by personality is how much a saved turn is worth; if the item would finish next turn anyway, none of them rush.

That is the third visible personality difference: the Defensive and the Opportunist convert their purse into time, while the Expansionist never rushes — in an expansion economy gold waits in the treasury.

Difficulty

Difficulty means better decisions, not more resources:

• Easy — candidate tiles are sampled sparsely, with a 1-turn planning horizon: it only looks at some of the tiles a unit could reach.
• Normal — the candidate limit effectively never binds (the tiles a unit can reach almost always stay under it) and the planning horizon is the same as Easy's (1 turn): it picks a tile on position alone, then — having moved — strikes if a target has come within its reach. Easy skips that second look. Two things separate Normal from Easy: how many tiles it scans and that second attack look.
• Hard — full candidate generation and a 2-turn horizon: a tile's value includes the best attack available from it, so the unit moves towards an opportunity on purpose. On top of that it gets a one-off +5 production, +5 food bonus at the start of the match.

Setup: every opponent is picked on its own

Personality and difficulty are chosen per opponent; the setup screen gives each opponent its own card. A card's personality options are the three personalities plus Mixed.

Mixed is the default. A mixed opponent's personality comes from the seed: the personality pool is shuffled with the match seed and handed out to the mixed opponents in order. Two consequences:

• A match built from the same seed produces the same roster. Hand the seed to a friend and they face the same opponents.
• Because the pool is shuffled and dealt, mixed opponents do not repeat a personality until the pool runs out — a default match is never three copies of one personality.

Pinning a personality does not constrain the deal: if you say "opponent 1 is Defensive, the rest mixed", one of the mixed ones may still come out Defensive.

Difficulty is not mixed — you pick each opponent's tier yourself. The tier a match is recorded under in your statistics is that of the hardest opponent at the table.

Threat indicator — "it sees you as a target"

An opponent's evaluation is readable on screen. A threat mark appears next to its badge in the score strip, and the selection panel writes one line per opponent: its personality, the power ratio and whether the late-game wake-up is in effect.

The power ratio is, as far as you can see, your attack strength over its strength. Below 1 means you are the weaker side — and that is exactly what an opponent's evaluation rewards: it turns on whoever weakens and pays a counter-damage penalty for attacking the strong (Combat). That term exists in all three personalities; only its intensity differs. If a profile has no appetite for attacking at all, the indicator never says it sees you as a target.

What the indicator does not say

The ratio rests only on the units you can see. An army behind the fog is not counted, so:

• No ratio is written for an opponent whose units you cannot see at all; the line says "under fog" instead. The absence of a mark does not mean "no threat" — it may mean "I cannot see", and the two are written as separate sentences.
• The indicator errs on the optimistic side: because a hidden army is not counted, it under-reports the threat. A ratio above 1 does not mean you are safe, only that you are the stronger side as far as you can see.
• The indicator does not say which city the opponent is marching on, what it is producing, or what it will do next turn. What it reports is the direction of its weights, not its plan.

An opponent comes with an army

An opponent does not send its units one at a time; it plays as a pack. At the start of every turn it builds a danger map from the board its own fog shows: every tile within reach of an enemy unit it can see is dangerous. Only task-free units (scouts, Pioneers, support) avoid that map — walking into a dangerous tile costs as much as the damage that unit would deal, and personality scales the avoidance: Defensive avoids danger the most, Opportunist the least. A Pioneer also will not found a city on a dangerous tile. Army units do not avoid the map; they march on their target. An army inside the fog is not on the map; an opponent cannot know your unseen army. It knows a city only as it last saw it: it does not read the health of a tile it saw long ago. It does not know a resource node or a ruin on a tile it has never seen either; it only walks toward those on tiles it has seen.

Combat units take one of three tasks:

• Gather — the opponent picks a target city (one it can win, that is valuable and near; a party it has a truce with and a neutral city are never picked) and pulls its army to a common tile near the city. Units do not trickle toward the target one by one.
• Attack — once the gathered units' combined damage is enough to bring the city down, the army marches and besieges together. If it is not enough, it waits.
• Defend — if the enemies that can reach a city within one turn can together deal 45% of that city's health or more, nearby army units drop their target and turn back to that city, and the city switches to defence production. It notices the preparation against its city that turn and answers; it does not wait for the city to be hit.

Pioneers, Scouts and support units do not join the army, and a wounded unit returns to its city. Target choice is weighted, with a tolerance band so it does not flip on small differences. Bandits do enter the danger map — walking in their reach is costly — but the army never picks them as a target (the bandits described below, see Bandit camp).

Opponents also play each other

In a three-opponent match there is a genuine three-way struggle; the AIs do not form a cartel against you. Whoever weakens, the Opportunist turns on them.

Opponents offer each other truces too: the side that sees itself as weaker makes the offer. Acceptance can come from either of two reasons — the receiving side also sees itself as weaker, or it is dealing with 2 or more fronts at once. Because of the second one a stronger opponent comes to the table too: closing one front is how it turns its advantage on the other rival. Bargaining, not a cartel (Diplomacy and truces).

Neutral bandits are not rivals

There is one more side on the map besides your rivals: the bandits inside a Bandit camp. They carry no personality, have no difficulty tier, found no cities and never enter the scoreboard — they are not a fourth archetype alongside the three above.

Their behaviour fits in one sentence: they stay within a fixed ring of their camp, hit the weakest adjacent unit and never touch cities. When they walk, they close on the nearest unit without leaving that ring. There is no personality weight, no difficulty tier and no risk calculus — the whole test is one line: hit the weakest one. The threat indicator does not show them either: the indicator reads the direction of a rival's weights, and a bandit has no direction to read.

To the opponents, though, a bandit is an ordinary target: every neutral unit they can see goes into the combat calculation, and they strike when it pays (Combat). So clearing a camp is not necessarily your job — in some matches a neighbour does it.

The same holds for the Neutral village: a village is not a rival but a settlement that can be taken. Rivals do treat it as a target and take it by war — annexing one with gold is, today, the human player's decision alone.

The numbers on this page are read from the game’s balance tables.