Combat model · source snapshot f3b66b07f
Duris Power Atlas
Explore 192 race and class combinations, with 711 builds from level 31 and 56 Human and Orc multiclass builds, across 13 report levels and three gear tiers. Compare the model’s one-on-one matchups, then see what drives the differences.
- Combinations
- 192
- Multiclass builds
- 56
- Report levels
- 13
- Gear tiers
- 3
- Reported engine checks
- ~37,000
- Reported deviations >1%
- 0
Historical model snapshot · run 2026-09-16 06:41:20 (timezone unspecified). These results and reported engine checks belong to that source revision; they do not track current game balance.
Loading the model snapshot…
Findings
What the numbers say
- Class decides; race multiplies.Class explains 80–91% of the spread in one-on-one strength at every level from 11, race 11–27%. Within a class, race scales the result rather than reordering the classes.
- One-on-one, casters out-damage melee from level 11.At level 56 in end-game gear a warrior lands about 2 damage per second on a geared player. A sorcerer's best spell does 46, an ethermancer's 59. Melee is cut three ways: the global −75%, armour that keeps reducing damage down to −750, and about half of all swings parried, blocked or dodged.
- Big bodies win.Troll, Ogre and Firbolg run 2–6× their class average from level 11 on, almost entirely through hit points, which grow with the square of racial Constitution. Humans are the weakest race from level 36 and sit in the bottom two of most classes (0.61–0.77× from level 31).
- The gaps open at three levels.Level 11 brings the first real attack spells, and the spread between classes goes from 24× to 129×. Level 21 brings psionicist inflict pain and druid earthquake. At level 51 end-game gear arrives, and stacked globe and ward items block most circle 1–7 arcane spells.
- Pulse, shrug and innates matter less.Pulse moves a combination by about ±10%, mostly helping small-race fighters. Shrug matters mainly at end-game, where it lifts the shrugging races most. The only large innate is troll regeneration.
- The halfling warrior is weak, but not the weakest.It ranks 13th to 19th of 21 warriors. At level 56 its hit points hold it back (×0.66) and its size (×0.87), partly offset by faster pulse (×1.17). Two warrior races rank below it. The warrior class as a whole has the bigger problem in the late game.
- Specialisation can matter more than race.From level 31 every build is in the field and each race and class is scored with its best one. Ethermancer edges Shaman at level 56 end-game (7.3 against 6.9). Healer clerics, end-game conjurers who stay unspecialised and rogue–bard multiclasses are the weakest builds of all. 22 specialisations never strictly beat the alternatives for any race, level or gear, and 6 never even tie for best.
- Multiclassing pays in good gear, rarely at end-game.At level 46 in good gear 26 of the 56 Human and Orc pairs beat their race's best single class; Orc Bard/Conjurer leads at 5.2 against 2.6. At level 56 end-game only Orc Ethermancer/Shaman does (7.7 against 7.2). Most pairs beat their primary class alone (40 of 56 at level 56), and rogue pairings are the weakest builds in the game. Adding them barely moves the single-class order (rank correlation 0.995–1.000).
Read the index as an order, not a margin. Every duel is one continuous fight on full memorisation, with no fleeing and no re-memorising, so running out of spell slots decides many caster fights and the gaps (×100 and more) overstate real play. Ranking by plain share of duels won gives almost the same order (rank correlation 0.95–0.99). Openers such as backstab are left out of the duel, so rogues are understated.
The atlas
Every combination, by level and gear
Each cell is one race and class, scored with its best single-class build from level 31; multiclass builds have their own section below. The duel index is the geometric mean, over every opponent build (248 through report level 26, 767 from level 31, multiclass builds included), of how much longer they take to kill you than you take to kill them: 1 means even, 3 means a geometric mean time-to-kill ratio of 3, not a threefold chance of winning. Blue is stronger than even, red weaker. The last column is the race's strength relative to others of the same class; the bottom row is each class's average. Hover or focus a cell to preview its detail. Click, tap, or press Enter on a cell to open its full breakdown. Scroll sideways to see all classes; race labels stay in view.
Standing by level
How classes and races rise and fall from 1 to 56
These follow the usual gear for each level: basic to 35, good from 36 to 50, end-game from 51. Classes show their average duel index; races show their strength relative to the rest of each class they can take (1 = typical for the class).
Specialisations
How much the specialisation matters
Specialisations open at level 30. From level 31 every specialisation, and staying unspecialised, is its own build, scored against all 767 builds (711 single-class, 56 multiclass); the headline uses each race and class's best build. Each figure is the specialisation's average duel index over the races that can take it, at three points on the usual gear path. To compare specialisations race by race at any level, choose a class under Specialisation in the atlas above.
Multiclass
What multiclassing buys
Tier levels
Where the gaps open
Spell circles open at levels 1, 6, 11 … 56 (circle N at 5N − 4). The steps below compare each report level with the one before, on the usual gear. A combination is flagged when it moves 15 percentile points or ×1.42 in index.
| Step | What changes |
|---|---|
| 1 → 6 | First attack spells. Minotaur sorcerer ×3.5 (burning hands, chill touch); druids fall. |
| 6 → 11 | Casters overtake melee. Shamans get pythonsting and scalding blast (ogre and firbolg shamans ×12); summoners get shocking grasp. The class spread goes from 24× to 129×. |
| 16 → 21 | Psionicist inflict pain (gith psionicists ×35–39); druid earthquake (×7–9). |
| 21 → 26 | Earthen maul for druids (×3.5); necromancer cone of cold, enervation and minor globe (×2.3). |
| 26 → 31 | Specialisations at 30: shock reaver (earth reaver for duergar) lifts reavers ×4.5. |
| 31 → 36 | Basic to good gear. Melee gains (drow rogues ×3.0, duergar rogues ×2.0). |
| 50 → 51 | End-game gear, the largest shift after level 16. Stacked globe and spirit-ward items block most circle 1–7 arcane spells: summoners ×0.22, while conjurers hold level (×1.1) by switching from water to fire specialisation; ethermancer 2.6 → 8.4, blighter 2.2 → 6.8, druid 2.6 → 5.8. |
| 51 → 56 | Conjurers hold level (×1.0). Paladins (×1.36) and reavers (×1.35) gain most; berserkers (×0.74) and clerics (×0.77) lose most. |
Racial factors
What makes a race strong
Each racial term was switched off in turn, with gear held fixed, to see how far it moves a combination's index. Hit points are the dominant racial factor. Pulse, shrug and innates are real but smaller, apart from troll regeneration.
Each cell is the multiplier the race's own term applies to its index, averaged over its classes. Above 1 the term helps the race; below 1 it holds it back.
Case study
The halfling warrior
The halfling warrior is below average among warriors at most levels, and the larger problem is the warrior class itself, which trails most classes in the late game. On good gear it climbs to 13th of 21 as a Guardian; at level 56 its best build is Swordsman. The table follows its best build at each level.
End-game
End-game turns on globe gear
Every end-game kit the model builds stacks permanent globe and ward items, which zero the matching spells (fight.c:4562–4584), almost every circle 1–7 arcane attack. Removing just those four gear effects, with the same items and specialisations, moves every caster's numbers but barely reorders the classes (rank correlation 0.95–0.96; only the top two swap). Globe items still decide which spells land at end-game, so how easy they are to get is a major end-game balance lever.
Reality check
Against real PvP picks
The public PvP log records 2,210 distinct level-50+ characters since 2017. Pick rates are normalised so 1 means a race is picked as often as its side and number of legal classes would predict. The model agrees with players most on races and weakly on classes: players over-pick psionicists and necromancers and under-pick shamans and ethermancers, which the model rates highest. Pick rates also reflect group roles, history and fashion, so this is a sanity check, not proof.
Method and limits
How this was built
- Source. Races, classes, skills, spells, melee and spell formulas were transcribed from master f3b66b07f, which includes the revised racial pulse (#333). The original report references file and line citations in separate working files; those files are not included in this snapshot.
- Reported engine check. The original report states that a harness compiled all 432 server files, loaded the live properties and enumerated every random outcome. Against it, the model's round length, cast time, HP, stats, to-hit and melee damage showed 0 deviations over 1% in about 37,000 checks. Spell resists, shrug and saves are exact. 16 of 1,334 spell-damage cases fell outside sampling tolerance, each explained.
- Gear. Three kits were built per combination from the 10,424 wearable items in booted zones, within the engine's caps: gear HP ≤ 3 × level, hitroll ≤ 75, damroll ≤ 50 × the race multiplier, and stats ≤ 100 plus racial-maximum gear. Basic uses typical level 20–35 items, good uses upper-quartile level 40–50 items, and end-game uses the best level 55+ items. Class and race restrictions were respected.
- Duel. Time-to-kill both ways from melee, spells, pets, control and healing, with skills at the maximum for each level, full spell slots. Below level 30 the field is the 192 race and class combinations; from level 31 it is all 711 builds (every specialisation plus unspecialised), each counted once, and each race and class is scored with its best build. The 56 multiclass builds join the field at every level (248 builds through level 26, 767 from level 31). The index is capped at 1/100–100 per matchup.
- Multiclass. Only Humans and Orcs can multiclass, after reaching level 50; the engine then restarts the character at level 1, so the 56 allowed primary and secondary pairs are scored at every report level. A multiclass build takes the better hit points, damage and to-hit of its two classes and the faster combat speed, cannot specialise, and has halved spell slots unless both classes are pure casters. Skills only the secondary class has arrive five levels later at 95% of the normal cap, and secondary spells the primary lacks count one circle higher. Each pair wears its primary or secondary class's kit, whichever scores better at that tier. A second harness mode built each pair the way the engine does: hit points, round length and spell slots matched in all 2,184 multiclass records.
- Not modelled. Openers (backstab and the like appear only as a separate burst figure), bard songs, reaver weapon procs, spell disables inside the duel, groups, fleeing and re-memorising, and randomly generated gear.
- Engine findings along the way. Normal melee and all spell damage are multiplied by 0.25. Sanctuary only works against melee for clerics and paladins. The spell-penetration cap is 60, because its properties key is misspelt. Thri-Kreen and Barbarian levelling factors fall back to 1.0 for the same kind of reason. Minotaur charge can't hit Medium or Small races. Duergar battlerage haste lasts longer than its cooldown.