UtilityAI & Scoring System

Utility AI replaces sprawling if-else chains and hand-tuned behavior trees with something a designer can actually reason about: every option scores itself from 0 to 1, and the best score wins. Add a consideration, watch the numbers move, ship.

WHY SCORING INSTEAD OF A TREE

A behavior tree makes you encode priority as structure, so every new rule means restructuring the graph. Scoring keeps priority as data. "Prefer close targets, but never shoot through a wall, and don't flip-flop between two equally good ones" is three considerations, not a re-plumbed tree.

WORKS WITH NO SCRIPTS AT ALL

Drop the Utility Brain component on an object, pick a rule asset, pick where candidates come from - a list, a tag, a radius, or this object's children - and press Play. Designers tune weights and curves in the Inspector while the game runs. Programmers get the same engine as a strongly typed C# API for scoring their own types, not just GameObjects.

YOU CAN SEE WHY IT DECIDED

The Rule Editor breaks every candidate's score down consideration by consideration while the game is running, so a wrong decision is a number you can look at instead of a guess. Rule validation and candidate diagnostics name the object at fault, and every error the plugin prints says what it was doing at the time.

FIVE DEMO SCENES, FOUR OF THEM RETUNABLE

Village Defense is a playable micro-game where bad rules genuinely lose the tower. Target Selection, Job Board and Consideration Lab each isolate one idea and let you drag the weights while they run. A hub scene navigates between them in one click.

BUILT TO SIT BESIDE YOUR OTHER TOOLS

Works on its own, and works better together: if you own PostanteGames StatSystem, Utility AI reads real runtime stats as decision inputs; if you own TaskSystem, the decisions it makes become real queued tasks. Both bridges activate automatically and neither is required - there is no dependency in either direction.

Full C# source, no DLLs, no third-party runtime dependency. English and Turkish documentation. 479 EditMode tests, all green.