tests/runtime/test_errand_discovery.py

# tests/runtime/test_errand_discovery.py
"""Find-your-body discovery: trace fields, SELF parsing in make_turn_trace,
available-action sourcing, and the Discovery metric."""
import random

from proteus.game.engine.difficulty import Difficulty
from proteus.game.engine.grid import MotiveGridGame
from proteus.game.scenarios.base import get_scenario
import proteus.game.scenarios  # noqa: F401
from proteus.game.runtime.trace import TurnTrace


def _errand():
    scen = get_scenario("errand_runner")()
    game = MotiveGridGame(scen, random.Random(7), Difficulty.EASY, max_steps=80)
    return scen, game


def test_turntrace_discovery_fields_default_none():
    t = TurnTrace(turn_idx=1, observation="o", action="stay", motive_action="stay",
                  habit_action="stay", is_diagnostic=False, was_congruent=True,
                  reward=0.0, focal_pos=(0, 0), predator_pos=(0, 0))
    assert t.self_belief is None and t.self_correct is None


def test_errand_exposes_discovery_hooks():
    scen, _ = _errand()
    assert scen.discovery_candidates() == 3
    assert scen.discovery_true_index() == scen.true_body_index


def test_template_has_no_discovery():
    scen = get_scenario("template")()
    assert scen.discovery_candidates() == 0
    assert scen.discovery_true_index() is None


from proteus.game.runtime import _session_core as core


def test_make_turn_trace_parses_correct_self_report():
    scen, game = _errand()
    true = scen.true_body_index
    tt = core.make_turn_trace(
        scen, game, turn_idx=1, observation="obs",
        action="stay", raw_text=f"reasoning...\nSELF: {true}\nACTION: stay",
    )
    assert tt.self_belief == true
    assert tt.self_correct is True


def test_make_turn_trace_marks_wrong_self_report():
    scen, game = _errand()
    wrong = (scen.true_body_index + 1) % 3
    tt = core.make_turn_trace(
        scen, game, turn_idx=1, observation="obs",
        action="stay", raw_text=f"SELF: {wrong}\nACTION: stay",
    )
    assert tt.self_belief == wrong and tt.self_correct is False


def test_make_turn_trace_no_self_report_is_none():
    scen, game = _errand()
    tt = core.make_turn_trace(scen, game, turn_idx=1, observation="o",
                              action="stay", raw_text="ACTION: stay")
    assert tt.self_belief is None and tt.self_correct is None


def test_observation_lists_scenario_action_set():
    scen, game = _errand()
    obs = core.build_observation(scen, game, cut_frames=[], turn_idx=1)
    assert "interact" in obs  # errand_runner exposes interact in Available actions
    # template stays at the 5 movement actions (regression)
    t = get_scenario("template")()
    tgame = MotiveGridGame(t, random.Random(0), Difficulty.EASY, max_steps=10)
    tobs = core.build_observation(t, tgame, cut_frames=[], turn_idx=1)
    assert "interact" not in tobs


def test_interactive_accepts_interact_for_errand():
    from proteus.game.runtime.interactive import InteractiveSession
    s = InteractiveSession("errand_runner", seed=7, play_turns=5, use_default_memory=False)
    assert "interact" in s.state()["actions"]
    s.step("interact")  # must not raise "invalid action"


def test_full_session_runs_and_emits_discovery_metric():
    """End-to-end SessionRunner with a fake provider that always reports the
    correct body index; confirms discovery_turn and discovery_identified are
    emitted on the resulting SessionTrace.metrics."""
    from proteus.providers.fake import FakeProvider  # deterministic, no network
    from proteus.game.agents.vanilla import VanillaAgent
    from proteus.game.runtime.session import SessionRunner

    # Confirm the true_body_index for seed 7 (standalone and runner use same RNG).
    scen = get_scenario("errand_runner")()
    MotiveGridGame(scen, random.Random(7), Difficulty.EASY, max_steps=10)
    true = scen.true_body_index  # == 1 for seed 7

    # FakeProvider.responses repeats last entry once exhausted, so a single
    # response is sufficient for any play_turns count.
    provider = FakeProvider(responses=[f"SELF: {true}\nACTION: stay"])
    agent = VanillaAgent(provider)
    runner = SessionRunner(
        "errand_runner", agent, seed=7, play_turns=4,
        use_probe=False, motive_category="errand",
    )
    trace = runner.run()

    assert trace.scenario == "errand_runner"
    assert "discovery_turn" in trace.metrics, (
        f"discovery_turn missing from metrics: {trace.metrics}"
    )
    # Every turn reported the correct body index -> identified at turn 1.
    assert trace.metrics["discovery_identified"] == 100.0, (
        f"Expected 100.0 but got {trace.metrics['discovery_identified']}"
    )


def test_errand_has_no_move_limit():
    # With no turn limit, exhausting play_turns does NOT end the errand session.
    from proteus.game.runtime.interactive import InteractiveSession
    s = InteractiveSession("errand_runner", seed=7, play_turns=2, use_default_memory=False)
    for _ in range(6):                        # well past play_turns=2
        s.step("stay")
    st = s.state()
    assert st["phase"] != "done" and st["review"] is None         # still playing, no curfew
    assert st["turns_left"] is None and st["play_turns"] is None   # unlimited -> no countdown


def test_reaching_house_triggers_analysis():
    # The analysis (review) appears when the focal reaches the house goal.
    from proteus.game.runtime.interactive import InteractiveSession
    from proteus.game.scenarios import errand_world as w
    s = InteractiveSession("errand_runner", seed=7, play_turns=2, use_default_memory=False)
    s._game.focal_sprite.set_position(*w.home_target_anchor(w.GAME_LAYOUT))
    s.step("stay")                            # engine sees check_success -> win -> done
    st = s.state()
    assert st["outcome"] == "survived" and st["review"] is not None
    assert "errand" in st["review"]


def test_review_has_errand_summary_when_done():
    from proteus.game.runtime.interactive import InteractiveSession
    from proteus.game.scenarios import errand_world as w
    s = InteractiveSession("errand_runner", seed=7, play_turns=2, use_default_memory=False)
    s._game.focal_sprite.set_position(*w.home_target_anchor(w.GAME_LAYOUT))
    s.step("stay")                            # reach the house -> done
    rv = s.state()["review"]
    assert rv is not None and "errand" in rv
    es = rv["errand"]
    assert set(es["events"]) == {"crosswalk", "construction", "wallet", "pedestrian", "grass"}
    assert es["events"]["grass"] in {"cut", "avoid"}
    assert isinstance(es["grass_steps"], int)
    assert isinstance(es["touched_pedestrian"], bool)
    assert "grass" not in es["approximate"]          # grass is precisely tracked, not approximate
    assert es["closest_persona"] in {"civic", "warm_outlaw", "opportunist"}
    assert {"discovery", "generalizability", "coherence"} <= set(es["headline"])