Combat lab: 2,900 measured duels, deaths against three enemies cut by a quarter
A reproducible test bench makes all nine classes (levels 5, 10, 20 and 40) fight real monsters summoned at the exact level on an isolated test server. It confirmed that same-level duels are won 99% of the time and guided new tactics against several attackers: against three enemies, deaths drop from 34% to 26%, against two from 4% to 2%. Changes are ready on the development branch, not yet deployed.
- Commit
- 1a13566
- Build
- Build success (7 min) · Target worldserver
The test bench
On a separate test server (copies of the databases, never the public server's), each character in the plan is raised to the target level (class trainer and class quest spells, talents, quest reward gear, hunter pet), isolated in its own phase on a Barrens plain, then fights real normal monsters summoned at the exact level. The brain fights exactly as in the live game. Eight cases: same level, +1, +2, +3, two and three monsters, ambush at same level and +2; ten fights per case, the same monster sequence before and after each change.
What it showed
- Same-level duel: 99% wins for every class, +2: 99%, two monsters: 95%. Deaths on the public server therefore do not come from ordinary duels but mostly from multi-monster fights (66% of deaths) and monsters three levels higher.
- Three against one: only 51% wins; low-level warlocks, priests and mages almost always died.
- Defects found along the way: Frost Nova and War Stomp cast out of range, Entangling Roots broken at once by Moonfire then recast in a loop, reactive abilities (Overpower, Victory Rush) refused by the server at every decision.
The changes
- Crowd control on the second attacker (Polymorph, Fear, Entangling Roots, Blind, Gouge, Hammer of Justice) while finishing the first, never waking a controlled monster with an area spell; it is finished afterwards.
- The Voidwalker engages and takes aggro before the first spell; the hunter's or warlock's pet goes after the monster hitting its master.
- Area defensive spells only when an attacker is within their radius; a broken crowd control is not recast in a loop.
- Casters open at near maximum range.
- Rest up to 98% health and 95% mana; rest from 90% health (75% mana for casters). Measured: against two enemies, 92% wins starting at 85% / 50%, 94% at 90% / 75%, 96% when full.
- Target choice: a spot whose neighbours are three levels higher is avoided, quests taken up to one level above.
Results (before → after)
| Case | Wins | Deaths | |---|---|---| | Same level | 99% → 99% | 0% → 0% | | +2 | 99% → 100% | 0% → 0% | | +3 | 94% → 97% | 5% → 2% | | Two monsters | 95% → 96% | 4% → 2% | | Three monsters | 51% → 59% | 34% → 26% |
Warlocks against three enemies: 15% → 62% wins; priests: 35% → 65%. Overall, 165 deaths → 115 over about 2,900 fights.
Next
The changes await deployment on the public server.