Defence reduction: the one stat that improves as you stack
✍️ Alice・愛麗沙
Defence reduction is not an ordinary percentage stat. It is one of the very few things in this game whose marginal return goes up. Enemy defence is mitigated along a hyperbola: damage multiplier = 1 ÷ (1 + enemy defence × (1 − reduction rate) ÷ 14828). Because reduction acts on the denominator, every slice you take makes the remaining denominator smaller and the next slice sharper — on the same boss, the first 10% is worth roughly +7.2% and the fourth 10% is worth +9.1%. The inverse holds just as hard: against soft trash it is nearly worthless, and against a high-defence boss it is the most expensive stat on your character. It is the one damage stat whose value is decided by the content you are running, so carry it per fight rather than parking it on a permanent priority list.
What the curve looks like
The mitigation expression the Korean community measured is a single line:
I = 1 ÷ (1 + enemy defence × (1 − reduction rate) ÷ 14828)
Three properties, each of them the opposite of the intuition players arrive with.
There is no flat subtraction. This is not damage-minus-defence, so you will never be pushed down to chip damage by a defence value that outgrows your attack.
It never reaches zero. The denominator is 1 plus a positive number, so the multiplier is always greater than zero. No amount of defence kills your output outright; it only compresses it.
14828 is a constant, and two independently written community calculators use the same figure.
| Bucket | Content | Stacking | Confidence |
|---|---|---|---|
| A 攻擊力 | 武器基礎攻擊、角色攻擊、刻印、寵物、時裝套裝、固定值附魔全部相加成一個總攻擊；徽章的攻擊%只乘武器那一段 | Adds within its bucket | Confident |
| B 攻擊力% | 所有「攻擊力增加%」的來源，包含裝備、技能、附魔、神器，全部在這一格相加 | Adds within its bucket | Confident |
| C 傷害增加 | 技能威力、增傷詞條、共鳴、神器增傷等相加後，再乘上一個「敵方受傷提升」的子乘區（防具破壞在這裡） | Two nested sub-multipliers | Confident |
| D 強化桶 | 連擊、強擊、廣域、連段、絕招五種強化——彼此相加、各自內部相乘；連擊與強擊共用同一池，只看總和 | Adds within its bucket | Confident |
| E 寶石標籤 | 寶石加成的技能標籤傷害，自成一個乘區，跟傷害增加分開 | Its own multiplier | Fairly confident |
| F 暴擊 | 期望值寫法：1＋暴擊率×(暴擊倍率−1)；沒有外加%時可化簡成 1＋暴擊／10000 | Its own roll-based multiplier | Confident |
| G 無防備 | 只在敵人被破防、進入無防備視窗時才生效；破防極限狀態再乘 1.5 倍 | Its own roll-based multiplier | Moderate |
| H 技能係數強化 | 各職業專屬的技能傷害增益，加法或乘法依來源而定——作者明講要各職自行測試 | Depends on the source | Unconfirmed |
| I 防禦減免 | 1÷(1＋敵防×(1−減防率)／14828)，雙曲線、永不歸零，沒有固定扣減 | Its own multiplier | Confident |
| J 反擊 | 反擊命中時固定 ×1.1，其餘為 1 | Its own roll-based multiplier | Moderate |
| K 追擊 | 期望值寫法；一次追擊打出該擊的 100%，單發模擬時視為 1 | Its own roll-based multiplier | Confident |
| L 最終傷害增加 | 最外層乘區：最終傷害%、飾品超越、魔導抗性差額都掛在這裡 | Its own multiplier | Confident |
The damage formula the KR community maintains — the operator has never published any combat equation. Its skeleton: values in the same bucket add, different buckets multiply, and the judgement types (crit / defenceless / pursuit) multiply against each other. Taiwan Season 0 runs the same structure. Every constant comes from long-running KR measurement, and two independent implementations agree on the numbers.
Work it once and watch it climb
To compute anything you first have to assume an enemy defence value. This is the weakest link in the article, so it goes up front: no boss defence value has ever been published — not by an operator, and not as a per-boss table anywhere in the Korean community. The worked example below uses 30,000 as an illustrative figure. The point is the shape of the curve, not that number; substitute any positive value and the conclusion is identical.
Enemy defence 30,000:
| Reduction | Denominator | Damage multiplier | vs no reduction | Added by this band |
|---|---|---|---|---|
| 0% | 3.0233 | 0.3308 | — | — |
| 10% | 2.8209 | 0.3545 | +7.2% | +7.2% |
| 20% | 2.6186 | 0.3819 | +15.4% | +7.7% |
| 30% | 2.4162 | 0.4139 | +25.1% | +8.4% |
| 40% | 2.2139 | 0.4517 | +36.5% | +9.1% |
The right-hand column is the whole article: 7.2 → 7.7 → 8.4 → 9.1.
Each additional 10% is worth more than the 10% before it.
That makes it an outlier here. The enhancement bucket, crit, attack percentage — every one of them has diminishing returns, where the more you own the less the next point buys. Defence reduction runs the other way, because it is the only one operating on the denominator.
The harder the target, the more it pays
Swap in something softer. Defence 5,000:
- no reduction: 1 ÷ (1 + 5000/14828) = 0.7478
- 20% reduction: 1 ÷ (1 + 4000/14828) = 0.7875 → +5.3%
The same 20% is worth +15.4% against the 30,000-defence boss and +5.3% against the 5,000-defence trash mob.
A threefold difference, from one identical line of stat.
Which is why “should I stack defence reduction” has no universal answer. It is not really a stat. It is a bet on a specific piece of content.
Four things with confusingly similar names
This is the other half of the article’s reason to exist. The game has at least four things that all sound like “makes the enemy easier to hurt”:
- Defence reduction (defence reduction rate) — enters the denominator above, and is live for the entire fight.
- Armour break / unguarded — pushes the enemy into a timed window, and only while that window is open does a different multiplicative bucket wake up. It has nothing to do with the enemy’s defence value.
- Armour-destruction damage — sits in the second sub-multiplier of the damage-increase bucket (damage taken by the enemy). Also live for the whole fight, but in a completely different position from defence reduction.
- Damage reduction — this is damage you take. It is a defensive stat with no bearing on your output, and it sits next to armour break in the extra-effect list, which is exactly why people pick the wrong row.
Two and three are a full article’s worth of difference on their own. Here, only one thing needs to stick: defence reduction is inside the denominator, and the other three are all outside it.
Worth adding: the source model keeps its own explicit list of things it has not measured — cross, weak point, the enraged state, the counter state — with an instruction from its author not to fabricate values for them. So there is no number here for elemental matchups or weak-point multipliers. There being none is the correct outcome.
What to do with this
- Raise its priority for a group that runs high-defence bosses. Its realised value there is above its printed value, and it climbs as you stack it.
- Push it to the bottom of the list for trash clearing and dailies. Against soft targets it is among the worst stats you can own.
- Ask once whether your party is stacking it redundantly. If several sources multiply into the same (1 − r), two people each carrying 20% is not 40%. But the formula we have does not state how multiple sources combine, so no combination rule is offered here — only the fact that it has never been measured.
- Do not weigh defence reduction against armour break on the same scale. They live in different multiplicative buckets, which makes them multiplicative with each other, not an either-or.
What could overturn this. The constant 14828 and the hyperbolic shape come from a long-maintained community formula whose line is marked high confidence; no operator, in any region, has ever published a combat formula. The 30,000 in the table is an illustrative assumption and not any real boss’s number — if someone measures a genuine boss defence value, every percentage in that table has to be recomputed (the conclusion that the curve accelerates does not change; that is a property of the expression, not of the number). How multiple players’ reduction combines has no source at all, and a controlled test would be welcome.