TechnologyBalanceTools

Build Guides Calculated, Not Estimated

We wanted to publish build guides on the site. Instead of estimating figures by hand, we wrote a simulator that calls the server's actual formulas. Its very first run showed that four out of six armor slots did nothing.

·6 min read ·The Valenya team

Every game in the genre has build guides, and they all share the same problem: the figures are written by hand. Accurate on publication day, approximate the following week, wrong after the first balance patch. And nobody notices.

We wanted build guides on the site, so we started with the tool: a headless simulator that replays the server's real formulas and feeds its output directly into the page. This detour cost more than writing the guides by hand. It paid for itself before the first publication.

The rule holding it all together

The simulator has one strict rule, the one thing to remember from this post:

The simulator copies no balance constants.

Not one. Damage, progression thresholds, spell costs, synergies, class armor rules… everything is read from the server classes that govern the game. If a formula changes, the simulator changes with it without us touching a line.

The opposite approach is tempting: copying a constant takes three seconds, while importing a combat module into a site-generation tool takes three hours. But a copied constant is a future divergence: it will be right until the day it is not, and nothing will flag that day.

The simulator's only unique values are its assumptions: things the formulas do not determine, such as how many targets are in an area of effect, the assumed hit rate of aimed spells, or the proportion of combat spent in melee range.

Those assumptions appear on every guide. A damage-per-second figure means absolutely nothing without them; hiding them would turn the guide into a claim of authority rather than a measurement.

What the simulator measures

Six metrics, across nine archetypes spread over four classes, at four level tiers:

MetricWhat it answers
Single-target damage per secondCan it kill a boss?
Area damageCan it handle a pack?
Time to defeat a bossHow long under pressure?
Kills per minuteCan it farm?
SurvivalCan it take a hit?
HealingCan it sustain the group?

One methodological detail we like: combat duration is not an arbitrary window. It is derived by fixed-point iteration. The target's health divided by damage per second gives a duration, but damage itself depends on duration because mana runs out, so we iterate until convergence. This calculation determines whether the pool lasts, and the Barbarian's Fury uptime is resolved within the same fixed point.

Proving the simulator tells the truth

A simulator never checked against the server is a second implementation of the game, with its own bugs. We needed a test comparing the two.

This fidelity test runs the server's real combat code against a purpose-built player and creature, then compares the result with the simulator. Tolerance: 2%, with a minimum allowance of one and a half damage points, because the server rounds at every step while the simulator rounds once.

The test found three errors on the day it was written:

  1. The Cleric's healing passives stack multiplicatively (+43.75%), not additively.
  2. An anti-boss passive was counted as a permanent bonus even though it has a per-turn lockout.
  3. The equipment selector ignored class armor rules and dressed Barbarians in plate and Bards in mail, which the server forbids.

Three errors that would have produced plausible, internally consistent, incorrect guides.

What the tool revealed about the game

This is where the detour paid off. The simulator was written to publish guides; its first job was finding gaps.

Four out of six armor slots did nothing

Leggings, boots, gloves and cloaks at the darksteel, silver, mithril and dragon tiers all came out with zero armor and no bonuses. Every piece above level 7 was therefore strictly worse than its steel equivalent, and an optimized level 30 character kept level 7 pieces to the end.

Nobody had noticed. While playing, you pick up a higher-tier piece, equip it and carry on. The simulator searches for the best available equipment: it immediately returned high-level builds wearing steel boots.

The cause was in the generator, not the data. It calculated armor for chest pieces and helmets, and zero for the other four slots. The rule now lives in one table where secondary slots are expressed as fractions of the same archetype's chest piece, so they inherit tier progression without adding a single constant.

Shields protected against nothing

From level 7 to level 20, no shield counted. The “is this a shield?” test checks an armor-category column, and tiered shields had been inserted without it. Their armor values, from 7 to 13, were purely decorative. Only the four level 1 shields worked.

The level cap made no sense

With a simulator, we can ask questions we could not ask before. Three measurements were enough to rule out the level 200 cap: level 40 represented 0.45% of the journey to the cap, going from 199 to 200 alone cost four times the entire journey from level 1 to 40, and a capped character reached 5,044 health against a bestiary whose toughest creature has 2,983.

The cap dropped to 100, the value the server's constants were calibrated for. The code already said so in a comment. The eleven equipment tiers were spread across the entire curve: each now lasts eight to ten levels instead of two to five. Silver and mithril had become outdated in two levels despite costing a boss heart, the worst cost-to-lifespan ratio in the whole progression.

The chain, end to end

A build begins with a plain-language description, “two-handed melee Barbarian, prioritize Strength,” containing not a single number. A selector equips it with the best gear actually available in game at its level, the simulator makes it fight, and the six metrics land on the page. After a formula or spell change, the guides regenerate together. None is updated by hand, so none can fall behind the others.

The lesson

We wrote a tool to fill a webpage, and it became a balance audit. Hardly surprising, really: a simulator is the first player who actually tries to play optimally, without tiring and without assuming that a high-tier piece must be better just because it exists.

The guides are in the guide's Builds tab. The assumptions appear at the top of every build. Start there.