How the units were balanced

The Dojo: the game’s own combat fought thousands of times at equal gold. Who beats whom, how it was tuned, what went wrong on the way, and what is still open.

Before the cavalry joined the army (October 2026), the four soldiers (warriors, lancers, archers and riders) were tuned against each other in the Dojo: the game’s own combat, fighting alone, thousands of times. This page shows how it was done, what it measured, what went wrong on the way, and what is still open. The Dojo is open to you too: fight these scenes yourself in the Dojo. Where a result falls short of what was aimed for, it says so.

Tip: Today’s numbers come from the Dojo’s reports of 28 October 2026: each scene on 20 seeds, each seed fought both ways, so 40 fights a scene. The story’s numbers are the ones measured at each step, between 26 and 28 October 2026, each table with its own count of fights.

Before and after, fought again

Two of the changes below, fought again in the Dojo: on the left the fight as it was, on the right as it is today. The same scene and the same seed on both fields, played ×3.75. Under each field, every fight of that scene fills in: 20 seeds, each fought both ways.

Both films run on today’s engine with the old number put back. They are not the runs of that day: the engine has changed since (chapter 2, chapter 3 and chapter 6), so the archer’s “before” comes out 30 %, not the 47 % of chapter 4. In the second, the dodge alone had taken this scene from 5 % to 10 % (measured 27 October); the price, which buys a seventh rider for the same gold, did the rest.

The Dojo

The Dojo is a test ground for the game’s makers. It takes the combat that plays out every match on the server (hit points, blows, shields, charges, arrows, even the dice) and runs it with nobody at the controls: two armies on an open field, each side with a simple order. Charge goes for the nearest enemy; hold stands and fights whatever comes into reach. A fight ends when one side has nobody left standing, or at a time limit: a draw.

So “do lancers beat warriors?” is not worked out on paper from damage and hit points: it is fought out, many times, by the same rules as a match.

What the Dojo cannot do is play like you: flank by hand, pull a wounded unit back, pick the fight. It measures how units trade blows, not who wins a match.

A fair fight

• Equal gold. Each side spends the same. A warrior costs 20 gold, a lancer and an archer 25 gold each, a rider 45 gold: 600 gold buys 30 warriors, 24 lancers, 24 archers or 13 riders. Some duels are equal numbers instead (12 against 12). Lancers and archers cost the same, so theirs is equal gold too; where warriors meet them in equal numbers, a second row gives the fight at equal gold.
• Many seeds. A seed sets the dice of a fight: who strikes first, which blow lands, who sidesteps. One fight proves nothing; every scene is fought on many seeds.
• Both colours. Every seed is fought again with the armies swapped: the field mirrored, blue’s army playing red. An edge one colour has cancels out, and the gap between the two runs shows whether the colour itself matters. Once, it did: see chapter 2.
• All against all. Eight armies of 600 gold, pure and mixed, each against every other, both ways.
• Scenes with a plan. Some questions need one: riders going round a screen of warriors to the archers behind it, lancers in a line or in ranks. Those groups follow a script: a route, then a target.

The aim: in each pair of the triangle, the stronger unit wins about 70–80 % of the fights at equal gold, and a mirror (the same army on both sides) comes out close to half.

Who beats whom today

The triangle

Lancers beat warriors, warriors beat archers, archers beat lancers:

Two of these miss the 70–80 % aimed for: lancers beat warriors every time, and in an open fight archers beat lancers only 65 % of the time (still open).

The cavalry

The rider stands outside the triangle on purpose: the dearest and fastest soldier, made to hunt archers and raid, and to break on a braced spear. So it is not measured by beating everything at equal gold, but by scenes of its role (600 gold a side unless the numbers say otherwise):

Riders win where archers are caught walking or have only monks beside them, and lose to a mass of archers who stand their ground and to lancers in ranks. A thin line of lancers is another matter: riders break it 63 % of the time. Against warriors it is close. Behind a screen of warriors, at today’s price, riders win whichever way they go in: the flank is barely better than the front (98 % against 95 %).

Does the colour matter?

In a mirror the same army fights on both sides, so blue should win about half. Blue’s wins in the first 20 fights of each mirror (the mixed hosts measured 27 October 2026):

Twenty fights are few: a fair coin tossed 20 times lands heads 13 times or more about one time in eight, so 65 % may be luck. The joust’s 70 % would come up by luck about one time in seventeen: it needs more fights to tell. The mixed host fought 100 times a colour came out 42 % / 58 %. Either way, every other number on this page is the average of both colours, so no colour’s edge is in them.

The big charge

Eight armies of 600 gold, each against every other, both charging, the melee in ranks in front:

Here the triangle does not show yet. Pure warriors win 95 % of their fights on average, and beat pure lancers 65 % of the time, though lancers beat warriors in every duel; pure archers beat pure lancers only 5 %. The riders’ low average is expected (this head-on charge is not their fight), but they still beat an army of archers 80 % of the time.

How we got here

Seven steps, between 26 and 28 October 2026. The numbers are the ones measured at the time: some were measured again later, and the story says where.

1. A first look

About 5,600 fights: the duels on 100 seeds each, flanks and formations, and eight armies all against all. The duels met their marks, but at scale warriors ruled: pure warriors, and warriors with archers, won 82 % of their fights on average. Riders beat archers 98 % in the duel and lost to nearly everything else. And one number looked wrong: the mixed mirror, the same army on both sides, came out 30 / 70.

Nothing was changed yet. First, the Dojo had to learn to measure what this look could not: both colours, flanks, armies all against all.

2. The colour that won

A mirror should come out even. The Dojo found why the mixed one did not, in the game’s engine itself: three of the steps that move units every moment (following orders, pushing bodies apart, the crowd) always moved blue first, then red. That fixed order decided the mixed mirror: the same army won 18 % of its fights as blue and 83 % as red (100 seeds).

Now the two sides take turns in a random order every moment, rolled with the battle’s own dice, as the blows already did. The same mirror: 42 % / 58 % (100 seeds). The engine is the one every match runs on, so the fix reached matches too.

Measured again without it, some results moved: archers no longer beat lancers (70 % → 48 %), and lancers beat warriors at equal gold 73 % instead of 85 %. Part of what had looked like balance had been the colour.

3. The lancer looked the wrong way

With the colour gone, 30 warriors beat every other army of 600 gold (100 % on average). The cause was not the warrior but the lancer’s guard, which pushes back a foe that strikes it from the front. It read which way the lancer faced on the wrong axis: it pushed whoever stood above the lancer on the screen, not whoever came at it. In ranks the warriors came straight at the lancers, and the guard never saw them.

Fixed, and still sharing the warrior’s shield’s 2 s cooldown, the guard made lancers win every time, so it got a cooldown of its own. The sweep (the lancers’ wins against warriors):

Chosen: 3.75 s. At 600 gold lancers now beat warriors 65 % (0 % before), and in the duel at equal gold 100 % (73 %), above the band: bringing the duel down takes a cooldown of 15–20 s, which leaves lancers no chance at scale. A dearer warrior (22 or 24 gold) was tried too: it tipped the pair the other way without weakening warriors against anything else.

4. The archer’s arrows

Archers and lancers were even in the duel (archers 47 %), and at 600 gold archers lost 85 %. A slower lancer and a longer bow (140 → 165) were tried first: archers won even less at 600 gold (10 % and 5 %). Then four more (the archers’ wins, except the riders’ column; the duels on 20 seeds both ways, the armies on 10):

No number suits every fight: a result jumps when a unit needs one blow more or less to fall. A cheaper archer won at scale but broke the duels (13 archers who hold always beat 15 warriors, and riders stopped beating archers). Chosen: the archer’s damage, 8 → 9 (the towers’ archers unchanged). Archers then beat lancers 78 % in the duel. The table shows the side effect: archers who hold their ground now beat warriors at equal gold (58 %), which chapter 6 found was partly the Dojo’s own fault.

5. The rider’s charge

The army matrix said 15 riders beat 24 archers 98 %; a scene said 24 archers who stand beat them 60 %. What decides it is the archers’ stance, not the riders’ formation (the riders’ wins, 600 gold, 10 seeds both ways):

Archers who walk forward run into the charges and fall; archers who wait and shoot win. The levers tried:

More hit points or damage made the rider beat everything, lancers and warriors too. A stronger charge only hurts whoever does not stop it, and a raised shield or a braced spear does. Chosen: the charge’s blow, ×1.2 → ×1.5. 24 archers who hold against 15 riders: the riders’ wins 40 % → 60 %, lancers and warriors against riders unchanged.

6. The Dojo’s own mistakes

Measuring the rider’s role, riders against archers who held kept ending in draws at the time limit. It was not the game: it was the Dojo’s commander, the part that gives a charging side its orders. Three faults:

• The empty middle. Every 2 s it sent idle troops to the middle of the enemy army. Once that army broke, the middle was nobody’s ground: the troops got there, saw no one, and stood still. Now they go to the nearest enemy.
• The target out of sight. A scripted attack left a unit alone while it had a target, even one gone out of sight; in the fog it stood still. Now it is ordered again.
• The unit already fighting. An archer stepping back as it shoots has no order, so the commander sent it straight back at the lancers, into the melee. Now it leaves alone a unit with an enemy within reach. (A first try at this rule left two warriors facing each other until the time limit; that was measured and corrected too.)

Chapters 3 to 5 had been measured with these faults: lancers stood idle in the empty middle and archers were sent into melee, so some of their numbers were partly the commander’s. Four new tests fail with the old commander, so it cannot slip back. The triangle, measured again (20 seeds both ways):

The triangle held, but the archer’s edge over the lancer shrank below the band: before, the lancers had been standing still.

7. The rider’s price

The rider was given a role of its own: the dearest and fastest soldier, a raider and archer hunter that braced spears stop. Its measure became the scenes of that role, not the big charge. It cost 40 gold then.

50 gold. The first try: 50 gold and 2.5 upkeep, and the spy and the pawn slowed down (now 111 and 108) so that the rider (115) is the fastest. 600 gold bought 12 riders instead of 15, and they lost nearly everything: against archers 0 % (holding) and 5 % (charging), by a flank 38 %, and lancers and warriors beat them every time. The mounted charge then got its own rules (it throws people as a troll’s slam does, stops on a raised shield unhurt, runs onto a braced spear, and two riders miss each other half the time), and its hit points and damage were swept (the riders’ wins, 10 seeds both ways):

Every gain against archers came after the rider had started to beat the lancers and warriors that should stop it.

The dodge. Counting where riders were when an arrow hit them showed why: only 9–12 % of the arrows reached a rider on the move; 57–61 % hit it while it stood striking one archer, in reach of the others. So a new rule: a rider on the move sidesteps 60 % of an archer’s arrows (a tower’s are not dodged). By a flank the riders’ wins doubled (40 % → 80 %); against archers who hold nothing changed (0 %); lancers and warriors, who shoot no arrows, were not touched.

45 gold. Last, the price (the riders’ wins, 10 seeds both ways, each scene with the riders its gold buys):

At 40 gold, and at 45 with damage 10, lancers no longer stopped it. At 45 it does its job: lancers beat it, warriors trade about evenly with it (45 % for the riders here; 60 % for the warriors in today’s duel), it sweeps archers by a flank and archers left unguarded, and a mass of archers who hold their ground is its counter. Chosen: 45 gold and 2.25 gold/min upkeep.

What is still open

• Two pairs off the band. Archers beat lancers 65 % in an open fight, under the 70–80 % aimed for; lancers beat warriors 100 % at equal gold, over it.
• The big charge. With all eight armies charging, pure warriors win 95 % on average and the triangle does not show.
• Arrows aim where the target was. An arrow lands where its target stood when it was loosed, so about 40 % of the arrows miss a charging lancer. Aiming ahead would raise the archer’s damage by more than half, and it touches everything that shoots (towers, monsters): it will be measured on its own.
• The rider’s last rules. Little damage to buildings, and horses that shy from an awake troll: after the launch.
• Some mirrors. Twenty fights a colour cannot tell a small edge from luck (above).
• You. The Dojo cannot flank by hand, pull the wounded back or pick its fights. Only matches show what players do with these numbers.

What comes next

A change to combat is measured in the Dojo before it reaches a match: before and after on the same seeds, both colours, and the reports it moves written again. The makers’ own rule: a change keeps the triangle, or says why. When the next one lands, this page gets its chapter, with its numbers: the good ones and the others.

Every price and blow named here is the game’s own (Units): it changes on this page when it changes in the game.