Build efficiency: stacking, value per point and travel
Checked against patch 0.5.5 · confidence: verified · topic id build-efficiency
This topic is reasoning from the formulas in the other topics, not a set of game rules. Its inputs are verified there, and the arithmetic below was recomputed on 2026-11-16. Sections 5 and 7 are advice, not rules.
1. The marginal value of "increased" falls as you stack it
All increased modifiers that apply to the same value are summed. If you already have S% increased, another x% is worth:
relative gain = x / (100 + S)
| Already have | +12% increased is worth |
|---|---|
| 0% | 12.0% more |
| 100% | 6.0% more |
| 200% | 4.0% more |
| 300% | 3.0% more |
So the tenth damage node is worth far less than the first. Remember that gear and gems also add to S.
2. Different multipliers multiply
Total damage is a product of separate terms, roughly:
base x (1 + increased) x more x (1 + crit chance x crit bonus) x speed x (1 - enemy mitigation)
Points are best spent on the term where they buy the largest relative increase. A build with 250% increased damage and a 5% crit chance gains more from crit; a build with heavy crit and no speed gains more from speed. The same holds for defence: life pool x 1/(1 - layer 1) x 1/(1 - layer 2).
3. Capped stats are worth nothing past the cap
Resistances (75%, or up to 90% with maximum-resistance modifiers), block chance (50% base maximum), chance to evade (95%), chance to deflect (95%), physical damage reduction (90%), critical hit chance (100%). Check the cap before recommending more.
4. Value per point, including travel
A notable three nodes away costs four points, not one. Compare options by total value of every node on the path divided by points spent.
- Travel through attribute nodes has low but non-zero value (requirements, 2 life or mana per point).
- Travel through small nodes that help the build is nearly free.
- A cluster is efficient when its small nodes are also useful. A notable behind two useless small nodes is expensive.
suggest_passivescomputes this: it returns value per point with the travel included.search_passivesreturns the distance in points.
5. When travel is worth it
Long travel is worth it for things you cannot get nearby:
- a keystone the build depends on,
- a "more" multiplier or a new term in the product (for example the only source of a mechanic),
- a jewel socket (when jewels are part of the plan),
- a cluster with several strong notables, so the travel cost is shared.
It is not worth it for one more "increased" notable of a kind you already stack.
6. Shared paths
Required nodes should share as much path as possible. Finding the cheapest tree that contains a set of nodes is a Steiner tree problem; never do it by hand. plan_tree solves it and reports whether the result is proven optimal. evaluate_build reports when the same notables and keystones can be had for fewer points.
7. Offence and defence
A dead character deals no damage. There is no rule for how many points go to defence: it depends on gear and on the content the player runs. State the assumption you made instead of guessing.
Worked example
Two options for 6 points, current total 180% increased lightning damage, 20% crit chance, 100% crit bonus:
A: 6 nodes, 72% increased Lightning Damage -> 72 / 280 = +25.7%
B: 6 nodes, 90% increased Critical Hit Chance (base 9%, currently 122% increased)
chance 20.0% -> 28.1%; (1 + 0.281) / (1 + 0.200) = +6.7%
A is better here. With 400% increased damage already, A would be 72 / 500 = +14.4%, and with a 160% crit bonus B would be (1 + 0.281 x 1.6) / (1 + 0.20 x 1.6) = +9.8%: the gap narrows. Compute it; do not assume.
Workflow for an AI
- Ask or infer the build's damage type, delivery (attack or spell), defence type and level.
search_passivesfor keystones and notables that define it; read them withget_passives.plan_treewith those asrequired, weightedgoalsfor the rest.evaluate_build, thensuggest_passivesto spend or reallocate remaining points.- Explain trade-offs with the formulas above and cite the topic.
Common mistakes
- Recommending "more damage nodes" without checking how much increased the build already has.
- Ignoring travel cost.
- Doing path arithmetic by hand instead of calling the tools.
Sources
- Derived from the formulas and caps in modifier-types, hit-damage-order, critical-hits, armour, block, accuracy-evasion-deflection and resistances, which carry their own sources
- GGG passive tree export 0.5.5: node counts and stat magnitudes
- Planner on this site: src/engine/planner.ts (Steiner tree search and value-per-point fill)