tests/test_option_controller.py

import unittest

from terrarium.large_world import generate_large_world
from terrarium.neural_schema import NEURAL_ACTION_SCHEMA_VERSION, NEURAL_OPTIONS
from terrarium.option_controller import (
    OPTION_RESOLUTION_SCHEMA_VERSION,
    apply_resolution_step,
    option_scores_from_context,
    resolve_option,
)


class OptionControllerTest(unittest.TestCase):
    def test_all_neural_options_resolve_without_direct_coordinates(self):
        world = generate_large_world(seed=515)
        position = tuple(world["creature_spawn"])
        for option in NEURAL_OPTIONS:
            resolution = resolve_option(
                option,
                world,
                position,
                user_position=tuple(world["user_spawn"]),
                resources={"hunger": 0.45, "energy": 0.7, "fear": 0.2},
                temperament={"careful": 0.5, "playful": 0.4, "social": 0.4, "curious": 0.6},
                slow_steering={
                    "directive": {"mode": "inspect", "strength": 0.6, "urgency": 0.3},
                    "option_priors": {"inspect_novelty": 0.2},
                },
            )
            self.assertEqual(resolution["schema_version"], OPTION_RESOLUTION_SCHEMA_VERSION)
            self.assertEqual(resolution["action_schema_version"], NEURAL_ACTION_SCHEMA_VERSION)
            self.assertEqual(resolution["option"], option)
            self.assertNotIn("coordinates", resolution)
            self.assertNotIn("resource_mutation", resolution)
            self.assertNotIn("direct_logits", resolution)

    def test_unsafe_user_approach_is_refused_by_controller(self):
        world = generate_large_world(seed=616)
        creature = tuple(world["creature_spawn"])
        user = tuple(world["user_spawn"])
        x0, y0 = creature
        x1, y1 = user
        for x in range(min(x0, x1), max(x0, x1) + 1):
            world["terrain"][y0][x] = "hazard"
        for y in range(min(y0, y1), max(y0, y1) + 1):
            world["terrain"][y][x1] = "hazard"
        world["terrain"][y0][x0] = "open"
        world["terrain"][y1][x1] = "open"

        resolution = resolve_option(
            "approach_user",
            world,
            creature,
            user_position=user,
            resources={"hunger": 0.2, "energy": 0.8, "fear": 0.1},
            temperament={"careful": 0.9, "social": 0.7},
        )

        self.assertEqual(resolution["primitive_action"], "wait")
        self.assertEqual(resolution["arbiter_override"], "unsafe_user_approach")

    def test_resolution_step_moves_only_for_passable_move_actions(self):
        world = generate_large_world(seed=717)
        position = tuple(world["creature_spawn"])
        resolution = resolve_option("explore_frontier", world, position, visited={position})
        next_position = apply_resolution_step(world, position, resolution)
        if resolution["primitive_action"].startswith("move_"):
            self.assertNotEqual(next_position, position)
        else:
            self.assertEqual(next_position, position)
        x, y = next_position
        self.assertNotEqual(world["terrain"][y][x], "wall")

    def test_option_scores_reflect_temperament_memory_and_slow_steering(self):
        world = generate_large_world(seed=818)
        position = tuple(world["creature_spawn"])
        careful = option_scores_from_context(
            world,
            position,
            user_position=tuple(world["user_spawn"]),
            resources={"hunger": 0.3, "energy": 0.8, "fear": 0.2},
            temperament={"careful": 0.9, "playful": 0.1, "social": 0.2, "curious": 0.4},
            memory={"negative_pull": 0.7},
        )
        playful = option_scores_from_context(
            world,
            position,
            user_position=tuple(world["user_spawn"]),
            resources={"hunger": 0.3, "energy": 0.8, "fear": 0.2},
            temperament={"careful": 0.1, "playful": 0.9, "social": 0.8, "curious": 0.4},
            memory={"negative_pull": 0.0},
            slow_steering={"directive": {"mode": "approach_user", "strength": 0.6}},
        )

        self.assertGreater(careful["avoid_or_retreat"], playful["avoid_or_retreat"])
        self.assertGreater(playful["approach_user"], careful["approach_user"])
        self.assertGreater(playful["follow_slow_directive"], careful["follow_slow_directive"])


if __name__ == "__main__":
    unittest.main()