tests/runtime/test_interactive_equivalence.py

"""Golden: the same action sequence produces an identical SessionTrace whether
driven through InteractiveSession (HTTP path) or SessionRunner + a scripted
HumanAgent (CLI path). This pins the two paths together so they cannot drift."""
from __future__ import annotations

import proteus.game.scenarios  # noqa: F401
from proteus.game.agents.human import HumanAgent
from proteus.game.engine.difficulty import Difficulty
from proteus.game.runtime.interactive import InteractiveSession
from proteus.game.runtime.session import SessionRunner

ACTIONS = ["up", "up", "left", "stay", "right", "up"]


def _scripted_human():
    feed = iter(ACTIONS)

    def input_fn(_prompt: str) -> str:
        return next(feed)

    def output_fn(_text: str) -> None:
        return None

    return HumanAgent(input_fn=input_fn, output_fn=output_fn)


def test_interactive_matches_session_runner():
    runner = SessionRunner(
        "template", _scripted_human(),
        difficulty=Difficulty.EASY, seed=42,
        play_turns=len(ACTIONS), use_probe=False,
    )
    cli_trace = runner.run()

    sess = InteractiveSession(
        "template", difficulty=Difficulty.EASY, seed=42,
        play_turns=len(ACTIONS), use_probe=False,
    )
    for a in ACTIONS:
        if sess.state()["outcome"] is not None:
            break
        sess.step(a)
    web_trace = sess.finish()

    # Both are human; everything must match field-for-field.
    assert web_trace.model == cli_trace.model == "human"
    assert web_trace.cut_frames == cli_trace.cut_frames
    assert web_trace.outcome == cli_trace.outcome
    assert web_trace.metrics == cli_trace.metrics
    assert len(web_trace.turns) == len(cli_trace.turns)
    for wt, ct in zip(web_trace.turns, cli_trace.turns):
        assert wt.model_dump() == ct.model_dump()
    # Full-trace equality is the strongest pin.
    assert web_trace.model_dump() == cli_trace.model_dump()