test_loopx_turn_executor.py

from __future__ import annotations

import json
import sys
import threading
from collections.abc import Mapping
from pathlib import Path

import pytest

from loopx.cli_commands import turn_cadence
from loopx.cli_commands.turn_cadence import ManagedCadenceStart, managed_cadence_start
from loopx.control_plane.effect_runtime import effect_runtime_result
from loopx.control_plane.goals import source_session_turn_effects
from loopx.control_plane.goals.first_party_host_admission import (
    FirstPartyHostGoalAdmission,
    FirstPartyHostRuntimeRejected,
)
from loopx.control_plane.goals.source_session_recreation import (
    RecreateGoalRequest,
    recreate_goal_instance,
)
from loopx.control_plane.goals.source_session_registry_state import (
    alias_digest,
    guard_path,
)
from loopx.control_plane.projects.registry_codec import (
    load_project_registry,
    source_session_registry_transaction,
)
from loopx.control_plane.turn_driver import executor as turn_executor
from loopx.control_plane.turn_driver import (
    LOOPX_TURN_RESULT_SCHEMA_VERSION,
    TurnRecoveryBlockedError,
    build_loopx_turn_plan,
    load_loopx_turn_plan_from_journal,
    run_loopx_turn_once,
    validate_loopx_turn_host_result,
)
from loopx.control_plane.turn_driver.journal_store import (
    find_loopx_turn_key_by_settlement_identity,
)
from loopx.control_plane.turn_driver.subagent_execution_topology import (
    OPAQUE_REF_PATTERN,
    child_execution_receipts_json_schema,
)
from loopx.control_plane.turn_driver.executor import (
    BuiltInHostError,
    LOOPX_TURN_JOURNAL_SCHEMA_VERSION,
    _task_validation_stage,
    turn_journal_path,
)
from loopx.control_plane.turn_driver.host_binding import managed_executor_binding
from loopx.control_plane.turn_driver.settlement import execute_turn_driver_settlement
from loopx.control_plane.turn_driver.transaction import TRANSACTION_PHASES
from loopx.file_lock import exclusive_cross_runtime_file_lock


INSTANCE_A = "ginst_aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa"
INSTANCE_B = "ginst_bbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbb"


def _write_source_registry(path: Path, instance_id: str) -> None:
    payload = {
        "schema_version": "0.2",
        "registry_role": "project-local",
        "profile_id": "source_session_v1",
        "common_runtime_root": str(path.parent),
        "projects": [],
        "goals": [
            {
                "id": "fixture-goal",
                "goal_instance_id": instance_id,
                "status": "active",
                "execution_authority": False,
            }
        ],
        "session_bindings": [],
        "session_receipts": [],
        "lifetime_receipts": [],
        "retired_goal_instances": [],
    }
    create = None if path.exists() else lambda: payload
    with source_session_registry_transaction(
        path,
        operation="turn_host_goal_instance_test",
        create=create,
    ) as transaction:
        current = transaction.payload_copy()
        current["goals"] = payload["goals"]
        transaction.commit(current)


def _replace_source_goal(path: Path, instance_id: str) -> None:
    with exclusive_cross_runtime_file_lock(
        guard_path(path, "fixture-goal"),
        operation="turn_host_goal_instance_test_recreate",
    ):
        _write_source_registry(path, instance_id)


def _plan() -> dict[str, object]:
    return build_loopx_turn_plan(
        {
            "ok": True,
            "schema_version": "loopx_turn_envelope_v0",
            "goal_id": "fixture-goal",
            "agent_id": "codex-fixture",
            "should_run": True,
            "effective_action": "normal_run",
            "action": {
                "must_attempt": True,
                "delivery_allowed": True,
                "quiet_noop_allowed": False,
                "selected_todo": {
                    "todo_id": "todo_fixture0001",
                    "text": "Advance one public fixture",
                },
            },
            "user": {
                "action_required": False,
                "open_count": 0,
                "notify": "DONT_NOTIFY",
            },
            "writeback": {"spend_after_validation": True},
            "scheduler": {"action": "run_now"},
            "action_signature": {
                "matches": True,
                "source_hash": "sha256:fixture",
                "envelope_hash": "sha256:fixture",
            },
            "compaction": {"within_budget": True},
        },
        host="generic-cli",
        execution_mode="isolated-headless",
    )


def _source_plan() -> dict[str, object]:
    plan = _plan()
    envelope = plan["turn_envelope"]
    assert isinstance(envelope, dict)
    return build_loopx_turn_plan(
        envelope,
        host="generic-cli",
        execution_mode="isolated-headless",
        goal_ref={
            "goal_id": "fixture-goal",
            "goal_instance_id": INSTANCE_A,
        },
    )


def _codex_plan() -> dict[str, object]:
    plan = _plan()
    envelope = plan["turn_envelope"]
    assert isinstance(envelope, dict)
    return build_loopx_turn_plan(
        envelope,
        host="codex-cli",
        execution_mode="isolated-headless",
    )


def _managed_plan(*, runtime_available: bool) -> dict[str, object]:
    """One dsh plan carrying the executor readback the command layer attaches."""

    plan = _plan()
    envelope = plan["turn_envelope"]
    assert isinstance(envelope, dict)
    managed = build_loopx_turn_plan(
        envelope,
        host="dsh",
        execution_mode="isolated-headless",
    )
    managed["managed_executor"] = managed_executor_binding(
        "dsh",
        environ={"DEEPSEEK_API_KEY": "fixture-operator-credential"},
        module_probe=lambda _module: runtime_available,
    )
    return managed


def test_turn_journal_resolves_only_from_exact_settlement_identity(
    tmp_path: Path,
) -> None:
    plan = _plan()
    transaction = plan["transaction"]
    assert isinstance(transaction, dict)
    turn_key = str(transaction["turn_key"])
    settlement = transaction["settlement_plan"]
    assert isinstance(settlement, dict)
    identity = settlement["identity"]
    assert isinstance(identity, dict)
    runtime_root = tmp_path / "runtime"
    path = turn_journal_path(runtime_root, goal_id="fixture-goal", turn_key=turn_key)
    path.parent.mkdir(parents=True)
    path.write_text(
        json.dumps(
            {
                "schema_version": LOOPX_TURN_JOURNAL_SCHEMA_VERSION,
                "goal_id": "fixture-goal",
                "turn_key": turn_key,
                "status": "in_progress",
                "completed_phases": [],
                "plan": plan,
            }
        ),
        encoding="utf-8",
    )

    assert (
        find_loopx_turn_key_by_settlement_identity(
            runtime_root,
            goal_id="fixture-goal",
            agent_id="codex-fixture",
            todo_id="todo_fixture0001",
            turn_instance_id=str(identity["turn_instance_id"]),
        )
        == turn_key
    )
    assert (
        find_loopx_turn_key_by_settlement_identity(
            runtime_root,
            goal_id="fixture-goal",
            agent_id="other-agent",
            todo_id="todo_fixture0001",
            turn_instance_id=str(identity["turn_instance_id"]),
        )
        is None
    )


def _adaptive_observation_plan(
    *,
    required_write_scopes: list[str] | None = None,
) -> dict[str, object]:
    plan = _plan()
    envelope = plan["turn_envelope"]
    assert isinstance(envelope, dict)
    envelope["task_orchestration_contract"] = {
        "schema_version": "task_orchestration_contract_v2",
        "mode": "adaptive",
        "coordinator_agent_id": "codex-fixture",
        "primary_todo_id": "todo_fixture0001",
        "child_brief_defaults": {
            "schema_version": "subagent_control_plane_handoff_v0",
            "parent_goal_id": "fixture-goal",
            "authority_artifact": "quota_should_run.goal_boundary",
            "latest_state_ref": "quota_should_run.action_signature.source_hash",
            "context_policy": {
                "selection_owner": "task_coordinator",
                "default": "fresh",
                "allowed": ["fresh"],
            },
            "expected_output": "public_safe_evidence",
            "execution_policy": {
                "timeout": "bounded_by_host_turn",
                "cancel": "task_coordinator_or_host_timeout",
            },
            "child_guard_policy": "prevention_first_v0",
            "validation_policy": "report validation commands and results",
            "acceptance": [
                "report completed scope and evidence",
                "do not write LoopX state or spend quota",
            ],
        },
        "eligible_child_lanes": [
            {
                "todo_id": "todo_child001",
                "task_domain": "validation",
                "execution_kind": "ephemeral_child",
                "child_brief": {
                    "todo_id": "todo_child001",
                    "objective": "Validate one independent fixture.",
                    "action_kind": "validate",
                    "task_domain": "validation",
                    "required_capabilities": [],
                    "task_repository": None,
                    "required_write_scopes": required_write_scopes or [],
                    "workspace_isolation": (
                        "independent_git_worktree"
                        if required_write_scopes
                        else "not_required"
                    ),
                },
            }
        ],
        "writeback_owner": "task_coordinator",
    }
    return build_loopx_turn_plan(
        envelope,
        host="codex-cli",
        execution_mode="isolated-headless",
    )


def _adaptive_plan() -> dict[str, object]:
    plan = _plan()
    envelope = plan["turn_envelope"]
    assert isinstance(envelope, dict)
    envelope["action"]["selected_todo"] = {"todo_id": "todo-stale"}
    envelope["task_orchestration_contract"] = {
        "schema_version": "task_orchestration_contract_v2",
        "mode": "adaptive",
        "primary_todo_id": "todo_fixture0001",
    }
    return build_loopx_turn_plan(
        envelope,
        host="generic-cli",
        execution_mode="isolated-headless",
    )


def _host_result(
    plan: dict[str, object], *, kind: str = "validated_progress"
) -> dict[str, object]:
    transaction = plan["transaction"]
    assert isinstance(transaction, dict)
    result: dict[str, object] = {
        "schema_version": LOOPX_TURN_RESULT_SCHEMA_VERSION,
        "turn_key": transaction["turn_key"],
        "result_kind": kind,
        "completed_phases": ["host_execute", "typed_result"],
    }
    if kind in {"validated_progress", "validated_completion"}:
        result.update(
            classification=(
                "fixture_progress"
                if kind == "validated_progress"
                else "fixture_completion"
            ),
            recommended_action=(
                "Continue the public fixture."
                if kind == "validated_progress"
                else "Refresh the active goal after this Todo completion."
            ),
            next_action=(
                "Run the next public fixture check."
                if kind == "validated_progress"
                else "Select the next Todo from a fresh decision."
            ),
            delivery_batch_scale="implementation",
            delivery_outcome="outcome_progress",
            vision_unchanged_reason="The fixture objective is unchanged after validated progress.",
            summary=(
                "One public fixture advanced."
                if kind == "validated_progress"
                else "One public fixture completed."
            ),
        )
    return result


def _child_execution_receipt(
    plan: dict[str, object],
    *,
    effect_classes: list[str] | None = None,
    evidence_refs: list[str] | None = None,
) -> dict[str, object]:
    topology = plan["subagent_execution_topology"]
    assert isinstance(topology, dict)
    lanes = topology["lanes"]
    assert isinstance(lanes, list)
    lane = lanes[0]
    assert isinstance(lane, dict)
    return {
        "schema_version": "subagent_host_execution_receipt_v0",
        "bundle_id": topology["bundle_id"],
        "lane_id": lane["lane_id"],
        "goal_id": topology["goal_id"],
        "todo_id": lane["todo_id"],
        "execution_kind": lane["execution_kind"],
        "runtime_id": "codex-cli",
        "worker_ref": "worker:fixture-child",
        "source_state_ref": topology["source_state_ref"],
        "task_packet_digest": lane["task_packet_digest"],
        "context_mode": lane["task_packet"]["context"]["mode"],
        "workspace_ref": None,
        "status": "completed",
        "effect_classes": (
            ["local_read"] if effect_classes is None else effect_classes
        ),
        "evidence_refs": (
            ["artifact:fixture-review"] if evidence_refs is None else evidence_refs
        ),
        "raw_transcript_copied": False,
    }


def test_task_validation_stage_reads_result_kind_through_effect_turn(
    tmp_path: Path,
) -> None:
    plan = _plan()
    result = _host_result(plan, kind="wait")
    journal = {
        "schema_version": LOOPX_TURN_JOURNAL_SCHEMA_VERSION,
        "goal_id": "fixture-goal",
        "turn_key": plan["transaction"]["turn_key"],
        "status": "in_progress",
        "completed_phases": list(TRANSACTION_PHASES[:2]),
        "plan": plan,
    }
    journal_path = tmp_path / "journal.json"
    turn_executor._write_journal(
        journal_path,
        {**journal, "completed_phases": []},
    )
    turn_executor._write_journal(journal_path, journal)

    completed, payload = _task_validation_stage(
        plan,
        result,
        task_validator=None,
        completed_phases=list(TRANSACTION_PHASES[:2]),
        journal=journal,
        persist_journal=lambda snapshot: turn_executor._write_journal(
            journal_path,
            snapshot,
        ),
        effects={},
    )

    assert completed == list(TRANSACTION_PHASES[:3])
    assert journal["status"] == "stopped"
    assert payload is not None
    assert payload["status"] == "stopped"


def test_typed_settlement_fails_closed_when_journal_receipt_payload_is_missing() -> (
    None
):
    transaction = _plan()["transaction"]
    assert isinstance(transaction, dict)
    calls = {"writeback": 0, "spend": 0, "checkpoint": 0}

    result = execute_turn_driver_settlement(
        transaction,
        transaction_phases=TRANSACTION_PHASES,
        completed_phases=TRANSACTION_PHASES[:4],
        writeback_payload=None,
        quota_spend_payload=None,
        writeback=lambda: (
            calls.__setitem__("writeback", calls["writeback"] + 1)
            or {"ok": True, "appended": True}
        ),
        spend=lambda: (
            calls.__setitem__("spend", calls["spend"] + 1)
            or {"ok": True, "appended": True}
        ),
        checkpoint=lambda _kind, _payload, _phases: calls.__setitem__(
            "checkpoint", calls["checkpoint"] + 1
        ),
    )

    assert result.failure is not None
    assert result.failure.kind.value == "receipt_missing"
    assert result.failure.step_kind.value == "durable_writeback"
    assert [receipt.step_kind.value for receipt in result.receipts] == ["validation"]
    assert calls == {"writeback": 0, "spend": 0, "checkpoint": 0}


def test_typed_settlement_fails_closed_when_plan_has_no_settlement_plan() -> None:
    transaction = _plan()["transaction"]
    assert isinstance(transaction, dict)
    legacy = {
        key: value for key, value in transaction.items() if key != "settlement_plan"
    }
    calls = {"writeback": 0, "spend": 0, "checkpoint": 0}

    result = execute_turn_driver_settlement(
        legacy,
        transaction_phases=TRANSACTION_PHASES,
        completed_phases=TRANSACTION_PHASES[:3],
        writeback_payload=None,
        quota_spend_payload=None,
        writeback=lambda: (
            calls.__setitem__("writeback", calls["writeback"] + 1)
            or {"ok": True, "appended": True}
        ),
        spend=lambda: (
            calls.__setitem__("spend", calls["spend"] + 1)
            or {"ok": True, "appended": True}
        ),
        checkpoint=lambda _kind, _payload, _phases: calls.__setitem__(
            "checkpoint", calls["checkpoint"] + 1
        ),
    )

    assert result.failure is not None
    assert result.failure.kind.value == "receipt_missing"
    assert result.failure.step_kind.value == "validation"
    assert "typed settlement plan" in result.failure.reason
    assert result.receipts == ()
    assert calls == {"writeback": 0, "spend": 0, "checkpoint": 0}


def _host_argv(result_path: Path, count_path: Path) -> list[str]:
    script = """
import json
import pathlib
import sys
request = json.load(sys.stdin)
result = json.loads(pathlib.Path(sys.argv[1]).read_text())
result["turn_key"] = request["turn_key"]
count = pathlib.Path(sys.argv[2])
count.write_text(str(int(count.read_text()) + 1 if count.exists() else 1))
json.dump(result, sys.stdout)
"""
    return [sys.executable, "-c", script, str(result_path), str(count_path)]


def _callbacks(calls: dict[str, int]):
    def writeback(_result: dict[str, object]) -> dict[str, object]:
        calls["writeback"] += 1
        return {"ok": True, "appended": True, "classification": "fixture_progress"}

    def spend() -> dict[str, object]:
        calls["spend"] += 1
        return {"ok": True, "appended": True, "slots": 1}

    def scheduler(_spend: dict[str, object]) -> dict[str, object]:
        calls["scheduler"] += 1
        return {"completed": True, "acknowledged": False, "disposition": "not_required"}

    return writeback, spend, scheduler


def _journal(runtime_root: Path) -> dict[str, object]:
    journal_paths = [
        path
        for path in (runtime_root / "goals" / "fixture-goal" / "turns").glob("*.json")
        if not path.name.endswith(".lock.holder.json")
    ]
    assert len(journal_paths) == 1
    return json.loads(journal_paths[0].read_text(encoding="utf-8"))


def _passing_validator(
    _plan: dict[str, object],
    _result: dict[str, object],
) -> dict[str, object]:
    return {
        "status": "passed",
        "validator_kind": "fixture",
        "summary": "independent fixture postconditions passed",
    }


def test_late_host_result_cannot_enter_a_recreated_goal(
    tmp_path: Path,
) -> None:
    registry = tmp_path / "project" / ".loopx" / "registry.json"
    _write_source_registry(registry, INSTANCE_A)
    plan = _source_plan()
    admission = FirstPartyHostGoalAdmission.for_plan(
        registry_path=registry,
        goal_id="fixture-goal",
        planned_goal_ref=plan["goal_ref"],
    )
    calls = {"writeback": 0, "spend": 0, "scheduler": 0}
    writeback, spend, scheduler = _callbacks(calls)
    runtime_root = tmp_path / "runtime"

    def stale_host_result(_request: Mapping[str, object]) -> dict[str, object]:
        _replace_source_goal(registry, INSTANCE_B)
        return _host_result(plan)

    with pytest.raises(FirstPartyHostRuntimeRejected) as exc_info:
        run_loopx_turn_once(
            plan,
            host_runner=stale_host_result,
            project=tmp_path,
            runtime_root=runtime_root,
            goal_id="fixture-goal",
            timeout_seconds=5,
            execute=True,
            task_validator=_passing_validator,
            writeback=writeback,
            spend=spend,
            scheduler=scheduler,
            goal_admission=admission,
        )

    assert exc_info.value.code == "stale_goal_instance"
    assert calls == {"writeback": 0, "spend": 0, "scheduler": 0}
    journal = _journal(runtime_root)
    assert "host_result" not in journal
    assert journal["completed_phases"] == []


def test_stale_source_plan_is_rejected_before_host_start(
    tmp_path: Path,
) -> None:
    registry = tmp_path / "project" / ".loopx" / "registry.json"
    _write_source_registry(registry, INSTANCE_A)
    plan = _source_plan()
    admission = FirstPartyHostGoalAdmission.for_plan(
        registry_path=registry,
        goal_id="fixture-goal",
        planned_goal_ref=plan["goal_ref"],
    )
    _replace_source_goal(registry, INSTANCE_B)
    calls = {"host": 0, "writeback": 0, "spend": 0, "scheduler": 0}
    writeback, spend, scheduler = _callbacks(calls)
    runtime_root = tmp_path / "runtime"

    def forbidden_host(_request: Mapping[str, object]) -> dict[str, object]:
        calls["host"] += 1
        return _host_result(plan)

    with pytest.raises(FirstPartyHostRuntimeRejected) as exc_info:
        run_loopx_turn_once(
            plan,
            host_runner=forbidden_host,
            project=tmp_path,
            runtime_root=runtime_root,
            goal_id="fixture-goal",
            timeout_seconds=5,
            execute=True,
            task_validator=_passing_validator,
            writeback=writeback,
            spend=spend,
            scheduler=scheduler,
            goal_admission=admission,
        )

    assert exc_info.value.code == "stale_goal_instance"
    assert calls == {"host": 0, "writeback": 0, "spend": 0, "scheduler": 0}
    transaction = plan["transaction"]
    assert isinstance(transaction, dict)
    assert (
        turn_journal_path(
            runtime_root,
            goal_id="fixture-goal",
            turn_key=str(transaction["turn_key"]),
        ).exists()
        is False
    )


def test_cached_host_result_cannot_resume_after_goal_recreation(
    tmp_path: Path,
) -> None:
    registry = tmp_path / "project" / ".loopx" / "registry.json"
    _write_source_registry(registry, INSTANCE_A)
    plan = _source_plan()
    transaction = plan["transaction"]
    assert isinstance(transaction, dict)
    runtime_root = tmp_path / "runtime"
    path = turn_journal_path(
        runtime_root,
        goal_id="fixture-goal",
        turn_key=str(transaction["turn_key"]),
    )
    journal = {
        "schema_version": LOOPX_TURN_JOURNAL_SCHEMA_VERSION,
        "turn_key": transaction["turn_key"],
        "goal_id": "fixture-goal",
        "status": "in_progress",
        "host": {"kind": "generic-cli"},
        "completed_phases": [],
        "plan": plan,
    }
    admission = FirstPartyHostGoalAdmission.for_plan(
        registry_path=registry,
        goal_id="fixture-goal",
        planned_goal_ref=plan["goal_ref"],
    )

    def write_source_journal() -> None:
        with admission.source_journal_admission(
            runtime_root=runtime_root,
        ) as source_admission:
            assert source_admission is not None
            turn_executor._write_journal(
                path,
                journal,
                source_admission=source_admission,
            )

    write_source_journal()
    journal.update(
        completed_phases=list(TRANSACTION_PHASES[:2]),
        host_result=_host_result(plan),
        result_kind="validated_progress",
    )
    write_source_journal()
    _replace_source_goal(registry, INSTANCE_B)
    calls = {"writeback": 0, "spend": 0, "scheduler": 0}
    writeback, spend, scheduler = _callbacks(calls)

    with pytest.raises(FirstPartyHostRuntimeRejected) as exc_info:
        run_loopx_turn_once(
            plan,
            host_runner=lambda _request: pytest.fail(
                "cached host result must not relaunch the Host"
            ),
            project=tmp_path,
            runtime_root=runtime_root,
            goal_id="fixture-goal",
            timeout_seconds=5,
            execute=True,
            task_validator=_passing_validator,
            writeback=writeback,
            spend=spend,
            scheduler=scheduler,
            goal_admission=admission,
        )

    assert exc_info.value.code == "stale_goal_instance"
    assert calls == {"writeback": 0, "spend": 0, "scheduler": 0}


def test_host_result_requires_bounded_public_material_fields() -> None:
    plan = _plan()
    result = _host_result(plan)
    result["raw_trajectory"] = "not allowed"

    validation = validate_loopx_turn_host_result(plan, result)

    assert validation["ok"] is False
    assert "unsupported host result fields" in " ".join(validation["errors"])


def test_child_receipt_schema_excludes_registered_peer_authority() -> None:
    schema = child_execution_receipts_json_schema()
    item_schema = schema["items"]
    properties = item_schema["properties"]

    assert properties["execution_kind"]["enum"] == ["ephemeral_child"]
    assert {"agent_id", "session_ref", "task_lease_ref"}.isdisjoint(properties)
    assert {"agent_id", "session_ref", "task_lease_ref"}.isdisjoint(
        item_schema["required"]
    )
    assert properties["worker_ref"]["pattern"] == OPAQUE_REF_PATTERN
    assert properties["evidence_refs"]["items"]["pattern"] == OPAQUE_REF_PATTERN
    assert properties["evidence_refs"]["minItems"] == 1


def test_child_receipt_rejects_unknown_context_mode() -> None:
    plan = _adaptive_observation_plan()
    result = _host_result(plan)
    result["child_execution_receipts"] = [
        {
            **_child_execution_receipt(plan),
            "context_mode": "implicit_parent_history",
        }
    ]

    validation = validate_loopx_turn_host_result(plan, result)

    assert validation["ok"] is False
    assert "context_mode is unsupported" in " ".join(validation["errors"])


def test_host_result_reconciles_aligned_child_receipt() -> None:
    plan = _adaptive_observation_plan()
    result = _host_result(plan)
    result["child_execution_receipts"] = [_child_execution_receipt(plan)]

    validation = validate_loopx_turn_host_result(plan, result)

    assert validation["ok"] is True
    reconciliation = validation["result"]["subagent_reconciliation"]
    assert reconciliation["status"] == "reconciled"
    assert reconciliation["observation_only"] is True
    assert reconciliation["settlement_enforced"] is False
    assert reconciliation["child_guard"] == {
        "schema_version": "child_execution_guard_v0",
        "enforced_boundaries": ["pre_spawn_task_packet"],
        "validated_observations": [
            "receipt_task_packet_binding",
            "context_mode_binding",
            "workspace_boundary",
            "effect_boundary",
        ],
        "projected_dispositions": [
            "stop_child",
            "quarantine_evidence",
            "continue_parent",
            "fallback_actions",
        ],
        "unsupported_boundaries": [
            "evidence_acceptance_enforcement",
            "live_host_tool_interception",
            "automatic_host_child_termination",
        ],
        "parent_acceptance_required": True,
    }
    assert reconciliation["parent_blocked"] is False
    assert reconciliation["parent_continuation"] == "continue"
    assert reconciliation["counts"] == {
        "planned": 1,
        "pre_spawn_rejected": 0,
        "observed": 1,
        "aligned": 1,
        "incomplete": 0,
        "rejected": 0,
        "cancelled": 0,
        "drifted": 0,
        "orphaned": 0,
    }
    lane = reconciliation["lanes"][0]
    assert lane["status"] == "aligned"
    assert lane["evidence_disposition"] == "candidate_for_parent_acceptance"
    assert lane["recommended_child_action"] == "return_to_parent"
    assert lane["parent_blocked"] is False
    assert lane["candidate_evidence_refs"] == ["artifact:fixture-review"]
    assert "quarantined_evidence_refs" not in lane


def test_pre_spawn_rejection_remains_visible_without_blocking_parent() -> None:
    plan = _adaptive_observation_plan()
    topology = plan["subagent_execution_topology"]
    topology["pre_spawn_rejections"] = [
        {
            "schema_version": "child_execution_rejection_v0",
            "todo_id": "todo_child002",
            "stage": "pre_spawn",
            "reason_codes": ["child_task_packet_incomplete"],
            "launch_allowed": False,
            "recommended_child_action": "do_not_launch",
            "parent_blocked": False,
            "parent_continuation": "continue",
            "fallback_actions": [
                "retry_fresh",
                "replace_child",
                "serial_takeover",
                "ignore_optional_result",
            ],
        }
    ]
    result = _host_result(plan)
    result["child_execution_receipts"] = [_child_execution_receipt(plan)]

    validation = validate_loopx_turn_host_result(plan, result)

    assert validation["ok"] is True
    reconciliation = validation["result"]["subagent_reconciliation"]
    assert reconciliation["status"] == "guarded"
    assert reconciliation["counts"]["pre_spawn_rejected"] == 1
    assert reconciliation["parent_blocked"] is False
    assert reconciliation["pre_spawn_rejections"] == topology["pre_spawn_rejections"]


def test_host_result_observes_missing_and_drifted_child_receipts() -> None:
    plan = _adaptive_observation_plan()
    missing = validate_loopx_turn_host_result(plan, _host_result(plan))

    assert missing["ok"] is True
    assert missing["result"]["subagent_reconciliation"]["status"] == "incomplete"
    assert missing["result"]["subagent_reconciliation"]["lanes"][0]["reason_codes"] == [
        "worker_receipt_missing"
    ]

    drifted_result = _host_result(plan)
    drifted_result["child_execution_receipts"] = [
        _child_execution_receipt(
            plan,
            effect_classes=["external_write"],
        )
    ]
    drifted = validate_loopx_turn_host_result(plan, drifted_result)

    assert drifted["ok"] is True
    reconciliation = drifted["result"]["subagent_reconciliation"]
    assert reconciliation["status"] == "drifted"
    assert reconciliation["lanes"][0]["reason_codes"] == [
        "side_effect_boundary_exceeded"
    ]
    assert reconciliation["lanes"][0]["evidence_disposition"] == "quarantined"
    assert reconciliation["lanes"][0]["recommended_child_action"] == "stop_child"
    assert reconciliation["lanes"][0]["parent_blocked"] is False
    assert reconciliation["lanes"][0]["quarantined_evidence_refs"] == [
        "artifact:fixture-review"
    ]
    assert "candidate_evidence_refs" not in reconciliation["lanes"][0]
    assert reconciliation["lanes"][0]["fallback_actions"] == [
        "retry_fresh",
        "replace_child",
        "serial_takeover",
        "ignore_optional_result",
    ]

    packet_mismatch_result = _host_result(plan)
    packet_mismatch_result["child_execution_receipts"] = [
        {
            **_child_execution_receipt(plan),
            "task_packet_digest": "sha256:" + "0" * 64,
        }
    ]
    packet_mismatch = validate_loopx_turn_host_result(plan, packet_mismatch_result)
    assert packet_mismatch["ok"] is True
    assert packet_mismatch["result"]["subagent_reconciliation"]["lanes"][0][
        "reason_codes"
    ] == ["task_packet_mismatch"]

    context_mismatch_result = _host_result(plan)
    context_mismatch_result["child_execution_receipts"] = [
        {
            **_child_execution_receipt(plan),
            "context_mode": "forked_snapshot",
        }
    ]
    context_mismatch = validate_loopx_turn_host_result(plan, context_mismatch_result)
    assert context_mismatch["ok"] is True
    context_lane = context_mismatch["result"]["subagent_reconciliation"]["lanes"][0]
    assert context_lane["reason_codes"] == ["context_mode_mismatch"]
    assert context_lane["evidence_disposition"] == "quarantined"

    no_evidence_result = _host_result(plan)
    no_evidence_result["child_execution_receipts"] = [
        _child_execution_receipt(plan, evidence_refs=[])
    ]
    no_evidence = validate_loopx_turn_host_result(plan, no_evidence_result)
    assert no_evidence["ok"] is True
    assert no_evidence["result"]["subagent_reconciliation"]["lanes"][0][
        "reason_codes"
    ] == ["aggregate_settlement_without_lane_evidence"]


@pytest.mark.parametrize(
    ("receipt_status", "lane_status"),
    [
        ("failed", "rejected"),
        ("rejected", "rejected"),
        ("cancelled", "cancelled"),
    ],
)
def test_host_result_guards_terminal_unsuccessful_child_receipts(
    receipt_status: str,
    lane_status: str,
) -> None:
    plan = _adaptive_observation_plan()
    result = _host_result(plan)
    result["child_execution_receipts"] = [
        {
            **_child_execution_receipt(plan),
            "status": receipt_status,
        }
    ]

    validation = validate_loopx_turn_host_result(plan, result)

    assert validation["ok"] is True
    reconciliation = validation["result"]["subagent_reconciliation"]
    assert reconciliation["status"] == "guarded"
    assert reconciliation["parent_blocked"] is False
    lane = reconciliation["lanes"][0]
    assert lane["status"] == lane_status
    assert lane["evidence_disposition"] == "quarantined"
    assert lane["recommended_child_action"] == "stop_child"


def test_terminal_unsuccessful_child_takes_priority_over_incomplete_sibling() -> None:
    plan = _adaptive_observation_plan()
    topology = plan["subagent_execution_topology"]
    lanes = topology["lanes"]
    assert isinstance(lanes, list)
    sibling = {
        **lanes[0],
        "lane_id": "lane_sibling",
        "todo_id": "todo_sibling",
    }
    lanes.append(sibling)
    result = _host_result(plan)
    result["child_execution_receipts"] = [
        {
            **_child_execution_receipt(plan),
            "status": "failed",
        }
    ]

    validation = validate_loopx_turn_host_result(plan, result)

    assert validation["ok"] is True
    reconciliation = validation["result"]["subagent_reconciliation"]
    assert reconciliation["status"] == "guarded"
    assert reconciliation["counts"]["rejected"] == 1
    assert reconciliation["counts"]["incomplete"] == 1


def test_host_result_requires_planned_workspace_for_writing_child() -> None:
    plan = _adaptive_observation_plan(required_write_scopes=["src/**"])
    topology = plan["subagent_execution_topology"]
    lane = topology["lanes"][0]
    receipt = _child_execution_receipt(
        plan,
        effect_classes=["local_read", "held_workspace_write"],
    )
    result = _host_result(plan)
    result["child_execution_receipts"] = [receipt]

    missing_workspace = validate_loopx_turn_host_result(plan, result)
    assert missing_workspace["ok"] is True
    assert missing_workspace["result"]["subagent_reconciliation"]["lanes"][0][
        "reason_codes"
    ] == ["workspace_mismatch"]

    receipt["workspace_ref"] = lane["workspace_ref"]
    aligned = validate_loopx_turn_host_result(plan, result)
    assert aligned["ok"] is True
    assert aligned["result"]["subagent_reconciliation"]["status"] == "reconciled"


def test_disabled_host_result_rejects_unadmitted_child_receipt() -> None:
    plan = _plan()
    receipt = {
        "schema_version": "subagent_host_execution_receipt_v0",
        "bundle_id": "bundle_unadmitted",
        "lane_id": "lane_unadmitted",
        "goal_id": "fixture-goal",
        "todo_id": "todo_unadmitted",
        "execution_kind": "ephemeral_child",
        "runtime_id": "generic-cli",
        "worker_ref": "worker:unadmitted",
        "source_state_ref": "sha256:fixture",
        "task_packet_digest": "sha256:" + "f" * 64,
        "context_mode": "fresh",
        "workspace_ref": None,
        "status": "completed",
        "effect_classes": ["local_read"],
        "evidence_refs": ["artifact:unadmitted"],
        "raw_transcript_copied": False,
    }
    result = _host_result(plan)
    result["child_execution_receipts"] = [receipt]

    rejected = validate_loopx_turn_host_result(plan, result)
    assert rejected["ok"] is False
    assert "unsupported host result fields: child_execution_receipts" in (
        " ".join(rejected["errors"])
    )
    assert "subagent_reconciliation" not in rejected["result"]


def test_enabled_host_result_rejects_receipt_local_path() -> None:
    plan = _adaptive_observation_plan()
    result = _host_result(plan)
    receipt = _child_execution_receipt(plan)
    receipt["workspace_ref"] = "/Users/example/raw-worker-path"
    result["child_execution_receipts"] = [receipt]

    rejected = validate_loopx_turn_host_result(plan, result)
    assert rejected["ok"] is False
    assert "absolute local path" in " ".join(rejected["errors"])


@pytest.mark.parametrize(
    ("field", "value", "expected_error"),
    [
        # A drive-qualified path is now recognized as a local path, so the
        # shared public-safety rule reports it before the opaque-shape check.
        # Both rules reject the value; only the diagnostic differs.
        (
            "worker_ref",
            "C:/workspace/private/worker.json",
            "contains an absolute local path",
        ),
        (
            "evidence_refs",
            ["file:/tmp/private-result.json"],
            "contains a local path",
        ),
    ],
)
def test_enabled_host_result_rejects_path_shaped_opaque_refs(
    field: str,
    value: object,
    expected_error: str,
) -> None:
    plan = _adaptive_observation_plan()
    result = _host_result(plan)
    receipt = _child_execution_receipt(plan)
    receipt[field] = value
    result["child_execution_receipts"] = [receipt]

    rejected = validate_loopx_turn_host_result(plan, result)

    assert rejected["ok"] is False
    assert expected_error in " ".join(rejected["errors"])
    assert "child_execution_receipts" not in rejected["result"]
    rejected_value = value[0] if isinstance(value, list) else value
    assert rejected_value not in json.dumps(
        rejected["result"],
        ensure_ascii=False,
    )
    reconciliation = rejected["result"]["subagent_reconciliation"]
    assert reconciliation["counts"]["observed"] == 0
    assert reconciliation["lanes"][0]["receipt_present"] is False


def test_observation_only_reconciliation_does_not_change_settlement(
    tmp_path: Path,
) -> None:
    plan = _adaptive_observation_plan()
    calls = {"writeback": 0, "spend": 0, "scheduler": 0}
    writeback, spend, scheduler = _callbacks(calls)

    committed = run_loopx_turn_once(
        plan,
        host_runner=lambda _request: _host_result(plan),
        project=tmp_path,
        runtime_root=tmp_path / "runtime",
        goal_id="fixture-goal",
        timeout_seconds=5,
        execute=True,
        task_validator=_passing_validator,
        writeback=writeback,
        spend=spend,
        scheduler=scheduler,
    )

    assert committed["ok"] is True
    assert committed["status"] == "committed"
    assert committed["subagent_reconciliation"]["status"] == "incomplete"
    assert committed["subagent_reconciliation"]["settlement_enforced"] is False
    assert committed["subagent_reconciliation"]["parent_blocked"] is False
    assert (
        _journal(tmp_path / "runtime")["host_result"]["subagent_reconciliation"]
        == committed["subagent_reconciliation"]
    )
    assert calls == {"writeback": 1, "spend": 1, "scheduler": 1}


def test_run_once_preview_has_no_host_or_journal_effects(tmp_path: Path) -> None:
    plan = _plan()

    payload = run_loopx_turn_once(
        plan,
        host_argv=[sys.executable, "-c", "raise SystemExit(9)"],
        project=tmp_path,
        runtime_root=tmp_path / "runtime",
        goal_id="fixture-goal",
        timeout_seconds=5,
        execute=False,
    )

    assert payload["ok"] is True
    assert payload["status"] == "preview"
    assert payload["effects"] == {
        "host_invoked": False,
        "state_written": False,
        "quota_spent": False,
        "scheduler_acknowledged": False,
    }
    assert not (tmp_path / "runtime").exists()


def test_managed_start_waits_before_host_and_replay_needs_no_new_admission(
    tmp_path: Path,
) -> None:
    plan = _plan()
    calls = {"host": 0, "admit": 0, "writeback": 0, "spend": 0, "scheduler": 0}
    writeback, spend, scheduler = _callbacks(calls)

    def host(_request: object) -> dict[str, object]:
        calls["host"] += 1
        return _host_result(plan)

    def wait(_identity: object) -> dict[str, object]:
        calls["admit"] += 1
        return {
            "admitted": False,
            "reason": "minimum_interval_wait",
            "next_eligible_at_ms": 9000,
        }

    common = dict(
        host_runner=host,
        project=tmp_path,
        runtime_root=tmp_path / "runtime",
        goal_id="fixture-goal",
        timeout_seconds=5,
        execute=True,
        task_validator=_passing_validator,
        writeback=writeback,
        spend=spend,
        scheduler=scheduler,
    )
    denied = run_loopx_turn_once(plan, admit_start=wait, **common)
    assert denied["status"] == "interval_wait"
    assert denied["admission"]["next_eligible_at_ms"] == 9000
    assert denied["effects"]["host_invoked"] is False
    assert calls == {"host": 0, "admit": 1, "writeback": 0, "spend": 0, "scheduler": 0}
    assert not list(
        (tmp_path / "runtime" / "goals" / "fixture-goal" / "turns").glob("*.json")
    )

    def allow(_identity: object) -> dict[str, object]:
        calls["admit"] += 1
        return {"admitted": True, "reserved": True, "reason": "admitted"}

    committed = run_loopx_turn_once(plan, admit_start=allow, **common)
    assert committed["status"] == "committed"
    assert calls["host"] == 1
    replay = run_loopx_turn_once(plan, admit_start=wait, **common)
    assert replay["replayed"] is True
    assert calls["admit"] == 2
    assert calls["host"] == 1


def _configure_managed_floor(runtime_root: Path, *, minutes_value: int) -> None:
    """Write the owner floor through the shipped TypeScript cadence store."""

    configured = effect_runtime_result(
        "quota.automation_cadence.manage",
        {
            "runtime_root": str(runtime_root),
            "goal_id": "fixture-goal",
            "agent_id": None,
            "automation_id": None,
            "operation": "configure",
            "expected_revision": 0,
            "min_interval_minutes": minutes_value,
            "owner_reference": "fixture-owner",
            "execute": True,
        },
    )
    assert configured["min_interval_minutes"] == minutes_value, configured


def _cadence_starts(runtime_root: Path) -> list[dict[str, object]]:
    """Read the real TypeScript cadence store that admits managed starts."""

    stores = []
    for path in sorted(runtime_root.rglob("*.json")):
        if path.name.endswith(".lock.holder.json"):
            continue
        payload = json.loads(path.read_text(encoding="utf-8"))
        if (
            isinstance(payload, dict)
            and str(payload.get("schema_version", "")).startswith(
                "automation_cadence_store"
            )
            and payload.get("goal_id") == "fixture-goal"
        ):
            stores.append(payload)
    assert len(stores) == 1, stores
    return list(stores[0]["starts"])


class _FrozenTurnClock:
    """Freeze the managed-start clock so a test can cross the owner floor."""

    def __init__(self, now_ms: int) -> None:
        self._now_ns = now_ms * 1_000_000

    def time_ns(self) -> int:
        return self._now_ns


def _managed_cadence(runtime_root: Path) -> ManagedCadenceStart:
    return managed_cadence_start(
        runtime_root=runtime_root,
        goal_id="fixture-goal",
        agent_id="codex-fixture",
        automation_id=None,
        manual_reason=None,
    )


def _managed_start_fixture(
    tmp_path: Path,
) -> tuple[dict[str, object], Path, dict[str, int], dict[str, object]]:
    plan = _plan()
    runtime_root = tmp_path / "runtime"
    calls = {"host": 0, "writeback": 0, "spend": 0, "scheduler": 0}
    writeback, spend, scheduler = _callbacks(calls)

    def host(_request: object) -> dict[str, object]:
        calls["host"] += 1
        return _host_result(plan)

    _configure_managed_floor(runtime_root, minutes_value=1)
    return (
        plan,
        runtime_root,
        calls,
        {
            "host_runner": host,
            "project": tmp_path,
            "runtime_root": runtime_root,
            "goal_id": "fixture-goal",
            "timeout_seconds": 5,
            "execute": True,
            "task_validator": _passing_validator,
            "writeback": writeback,
            "spend": spend,
            "scheduler": scheduler,
        },
    )


def test_reserved_managed_start_recovers_after_death_before_the_first_journal_write(
    tmp_path: Path,
    monkeypatch: pytest.MonkeyPatch,
) -> None:
    """A crash between the cadence reservation and the journal must not strand a Turn.

    The reservation is written by the real TypeScript cadence store and the
    process is interrupted before any Turn journal write, so the restart asks
    with the same request identity.
    """

    plan, runtime_root, calls, common = _managed_start_fixture(tmp_path)
    turn_key = str(plan["transaction"]["turn_key"])  # type: ignore[index]

    def die_after_reservation(identity: Mapping[str, object]) -> dict[str, object]:
        reservation = _managed_cadence(runtime_root).admit(identity)
        assert reservation["admitted"] is True and reservation["reserved"] is True
        raise RuntimeError("simulated process death after the cadence reservation")

    with pytest.raises(RuntimeError, match="simulated process death"):
        run_loopx_turn_once(plan, admit_start=die_after_reservation, **common)

    assert [
        (row["state"], row["request_id"]) for row in _cadence_starts(runtime_root)
    ] == [("reserved", f"{turn_key}:1")]
    assert not list((runtime_root / "goals" / "fixture-goal" / "turns").glob("*.json"))
    assert calls == {"host": 0, "writeback": 0, "spend": 0, "scheduler": 0}

    started_at_ms = int(_cadence_starts(runtime_root)[0]["started_at_ms"])
    monkeypatch.setattr(turn_cadence, "time", _FrozenTurnClock(started_at_ms + 120_000))
    restart = _managed_cadence(runtime_root)
    recovered = run_loopx_turn_once(
        plan, admit_start=restart.admit, confirm_start=restart.confirm, **common
    )

    assert recovered["status"] == "committed", recovered
    assert recovered["admission"]["resumed"] is True
    assert calls == {"host": 1, "writeback": 1, "spend": 1, "scheduler": 1}
    assert [
        (row["state"], row["request_id"]) for row in _cadence_starts(runtime_root)
    ] == [("started", f"{turn_key}:1")]


def test_reserved_managed_start_recovers_after_death_before_the_attempt_record(
    tmp_path: Path,
    monkeypatch: pytest.MonkeyPatch,
) -> None:
    """The same recovery holds when the journal exists without an attempt."""

    plan, runtime_root, calls, common = _managed_start_fixture(tmp_path)
    turn_key = str(plan["transaction"]["turn_key"])  # type: ignore[index]
    cadence = _managed_cadence(runtime_root)
    journal_writes = turn_executor._write_journal

    def die_before_attempt_record(path: Path, journal: dict[str, object]) -> None:
        if "host_attempt_count" in journal:
            raise RuntimeError("simulated process death before the attempt record")
        journal_writes(path, journal)

    monkeypatch.setattr(turn_executor, "_write_journal", die_before_attempt_record)
    with pytest.raises(RuntimeError, match="simulated process death"):
        run_loopx_turn_once(
            plan, admit_start=cadence.admit, confirm_start=cadence.confirm, **common
        )
    monkeypatch.setattr(turn_executor, "_write_journal", journal_writes)

    journal = _journal(runtime_root)
    assert "host_attempt_count" not in journal
    assert journal["admission"]["reserved"] is True
    assert [
        (row["state"], row["request_id"]) for row in _cadence_starts(runtime_root)
    ] == [("reserved", f"{turn_key}:1")]
    assert calls == {"host": 0, "writeback": 0, "spend": 0, "scheduler": 0}

    started_at_ms = int(_cadence_starts(runtime_root)[0]["started_at_ms"])
    monkeypatch.setattr(turn_cadence, "time", _FrozenTurnClock(started_at_ms + 120_000))
    restart = _managed_cadence(runtime_root)
    recovered = run_loopx_turn_once(
        plan, admit_start=restart.admit, confirm_start=restart.confirm, **common
    )

    assert recovered["status"] == "committed", recovered
    assert calls == {"host": 1, "writeback": 1, "spend": 1, "scheduler": 1}
    assert [
        (row["state"], row["request_id"]) for row in _cadence_starts(runtime_root)
    ] == [("started", f"{turn_key}:1")]


def test_run_once_rejects_oversized_built_in_host_result(tmp_path: Path) -> None:
    plan = _plan()
    calls = {"writeback": 0, "spend": 0, "scheduler": 0}
    writeback, spend, scheduler = _callbacks(calls)
    oversized = _host_result(plan)
    oversized["summary"] = "x" * 13_000

    payload = run_loopx_turn_once(
        plan,
        host_runner=lambda _request: oversized,
        project=tmp_path,
        runtime_root=tmp_path / "runtime",
        goal_id="fixture-goal",
        timeout_seconds=5,
        execute=True,
        writeback=writeback,
        spend=spend,
        scheduler=scheduler,
    )

    assert payload["ok"] is False
    assert payload["reason"] == "built-in host result exceeded the result budget"
    assert calls == {"writeback": 0, "spend": 0, "scheduler": 0}


def test_run_once_explicitly_retries_failed_host_without_duplicate_effects(
    tmp_path: Path,
) -> None:
    plan = _plan()
    calls = {"host": 0, "admit": 0, "writeback": 0, "spend": 0, "scheduler": 0}
    writeback, spend, scheduler = _callbacks(calls)

    def host(_request: dict[str, object]) -> dict[str, object]:
        calls["host"] += 1
        if calls["host"] == 1:
            raise BuiltInHostError("codex_cli_model_requires_newer_codex")
        return _host_result(plan)

    def admit(identity: dict[str, object]) -> dict[str, object]:
        calls["admit"] += 1
        assert identity["attempt"] == (1 if calls["admit"] == 1 else 2)
        return {"admitted": calls["admit"] != 2, "reason": "minimum_interval_wait"}

    kwargs = {
        "host_runner": host,
        "project": tmp_path,
        "runtime_root": tmp_path / "runtime",
        "goal_id": "fixture-goal",
        "timeout_seconds": 5,
        "execute": True,
        "task_validator": _passing_validator,
        "writeback": writeback,
        "spend": spend,
        "scheduler": scheduler,
        "admit_start": admit,
    }
    failed = run_loopx_turn_once(plan, **kwargs)
    replayed = run_loopx_turn_once(plan, **kwargs)
    waiting = run_loopx_turn_once(plan, retry_failed=True, **kwargs)
    recovered = run_loopx_turn_once(plan, retry_failed=True, **kwargs)

    assert failed["reason"] == "codex_cli_model_requires_newer_codex"
    assert failed["result_kind"] == "host_failure"
    assert failed["receipt"]["result_kind"] == "host_failure"
    assert failed["receipt"]["failed_phase"] == "host_execute"
    assert replayed["replayed"] is True
    assert waiting["status"] == "interval_wait"
    assert waiting["effects"]["host_invoked"] is False
    assert recovered["status"] == "committed"
    assert calls == {"host": 2, "admit": 3, "writeback": 1, "spend": 1, "scheduler": 1}


def test_invalid_host_result_reinvocation_reenters_admission(tmp_path: Path) -> None:
    plan = _plan()
    calls = {"host": 0, "admit": 0, "writeback": 0, "spend": 0, "scheduler": 0}
    writeback, spend, scheduler = _callbacks(calls)

    def host(_request: object) -> dict[str, object]:
        calls["host"] += 1
        return {"invalid": True} if calls["host"] == 1 else _host_result(plan)

    def admit(identity: dict[str, object]) -> dict[str, object]:
        calls["admit"] += 1
        assert identity["attempt"] == (1 if calls["admit"] == 1 else 2)
        return {"admitted": calls["admit"] != 2, "reason": "minimum_interval_wait"}

    common = dict(
        host_runner=host,
        admit_start=admit,
        project=tmp_path,
        runtime_root=tmp_path / "runtime",
        goal_id="fixture-goal",
        timeout_seconds=5,
        execute=True,
        task_validator=_passing_validator,
        writeback=writeback,
        spend=spend,
        scheduler=scheduler,
    )
    first = run_loopx_turn_once(plan, **common)
    waiting = run_loopx_turn_once(plan, retry_failed=True, **common)
    recovered = run_loopx_turn_once(plan, retry_failed=True, **common)

    assert first["result_kind"] == "validation_failed"
    assert (
        waiting["status"] == "interval_wait"
        and waiting["effects"]["host_invoked"] is False
    )
    assert recovered["status"] == "committed"
    assert calls == {"host": 2, "admit": 3, "writeback": 1, "spend": 1, "scheduler": 1}


def test_run_once_bounds_provider_capacity_retries_without_spending_quota(
    tmp_path: Path,
) -> None:
    plan = _plan()
    calls = {"host": 0, "writeback": 0, "spend": 0, "scheduler": 0}
    writeback, spend, scheduler = _callbacks(calls)

    def host(_request: dict[str, object]) -> dict[str, object]:
        calls["host"] += 1
        raise BuiltInHostError(
            "codex_cli_provider_capacity",
            failure_kind="provider_capacity",
        )

    common = {
        "host_runner": host,
        "project": tmp_path,
        "runtime_root": tmp_path / "runtime",
        "goal_id": "fixture-goal",
        "timeout_seconds": 5,
        "execute": True,
        "writeback": writeback,
        "spend": spend,
        "scheduler": scheduler,
    }

    first = run_loopx_turn_once(plan, **common)
    second = run_loopx_turn_once(plan, retry_failed=True, **common)
    third = run_loopx_turn_once(plan, retry_failed=True, **common)

    assert first["host_failure"]["attempt"] == 1
    assert first["host_failure"]["retry"]["backoff_seconds"] == 30
    assert second["host_failure"]["attempt"] == 2
    assert second["host_failure"]["retry"]["backoff_seconds"] == 60
    assert third["host_failure"]["attempt"] == 3
    assert third["host_failure"]["retry"]["max_attempts"] == 3
    with pytest.raises(TurnRecoveryBlockedError) as exc_info:
        run_loopx_turn_once(plan, retry_failed=True, **common)
    assert exc_info.value.decision["reason"] == "host_retry_budget_exhausted"
    assert calls == {"host": 3, "writeback": 0, "spend": 0, "scheduler": 0}


def test_run_once_never_blindly_retries_output_budget_exhaustion(
    tmp_path: Path,
) -> None:
    plan = _plan()
    calls = {"host": 0, "writeback": 0, "spend": 0, "scheduler": 0}
    writeback, spend, scheduler = _callbacks(calls)

    def host(_request: dict[str, object]) -> dict[str, object]:
        calls["host"] += 1
        raise BuiltInHostError(
            "dsh_output_budget_exhausted_no_final",
            failure_kind="output_budget_exhausted",
        )

    common = {
        "host_runner": host,
        "project": tmp_path,
        "runtime_root": tmp_path / "runtime",
        "goal_id": "fixture-goal",
        "timeout_seconds": 5,
        "execute": True,
        "writeback": writeback,
        "spend": spend,
        "scheduler": scheduler,
    }

    failed = run_loopx_turn_once(plan, **common)

    assert failed["reason"] == "dsh_output_budget_exhausted_no_final"
    assert failed["host_failure"] == {
        "schema_version": "loopx_turn_host_failure_v0",
        "kind": "output_budget_exhausted",
        "attempt": 1,
        "retryable": False,
    }
    with pytest.raises(TurnRecoveryBlockedError) as exc_info:
        run_loopx_turn_once(plan, retry_failed=True, **common)
    assert exc_info.value.decision["reason"] == "host_retry_not_available"
    assert calls == {"host": 1, "writeback": 0, "spend": 0, "scheduler": 0}


@pytest.mark.parametrize("native_children", [False, True])
def test_run_once_resumes_session_observed_by_recoverable_failed_turn(
    tmp_path: Path, native_children: bool,
) -> None:
    plan = _codex_plan()
    if native_children:
        plan["turn_envelope"]["agent_context"] = {"contributions": [
            {"capability_id": "multi_subagent", "facts": {"max_children": 3}}]}
    calls = {"host": 0, "writeback": 0, "spend": 0, "scheduler": 0}
    session_actions: list[str] = []
    writeback, spend, scheduler = _callbacks(calls)

    def host(request: dict[str, object]) -> dict[str, object]:
        calls["host"] += 1
        if native_children:
            assert request["host_attempt"] == calls["host"]
            assert request["host_attempt"] == _journal(tmp_path / "runtime")["host_attempt_count"]
        else:
            assert "host_attempt" not in request
        session = request["session"]
        assert isinstance(session, dict)
        session_actions.append(str(session["action"]))
        if calls["host"] == 1:
            raise BuiltInHostError(
                "codex_cli_timeout",
                recovery_kind="resume_session",
            )
        return _host_result(plan)

    def session_binding(
        _turn_envelope: Mapping[str, object],
    ) -> dict[str, object]:
        return {
            "schema_version": "loopx_turn_session_binding_v0",
            "goal_id": "fixture-goal",
            "agent_id": "codex-fixture",
            "todo_id": "todo_fixture0001",
        }

    common = {
        "host_runner": host,
        "session_binding_resolver": session_binding,
        "project": tmp_path,
        "runtime_root": tmp_path / "runtime",
        "goal_id": "fixture-goal",
        "timeout_seconds": 5,
        "execute": True,
        "task_validator": _passing_validator,
        "writeback": writeback,
        "spend": spend,
        "scheduler": scheduler,
    }

    failed = run_loopx_turn_once(plan, **common)
    transaction = plan["transaction"]
    assert isinstance(transaction, dict)
    inspected = turn_executor.inspect_loopx_turn_journal(
        tmp_path / "runtime",
        goal_id="fixture-goal",
        agent_id="codex-fixture",
        turn_key=str(transaction["turn_key"]),
        retry_failed=True,
        session_binding_resolver=session_binding,
    )
    recovered = run_loopx_turn_once(plan, retry_failed=True, **common)

    assert failed["reason"] == "codex_cli_timeout"
    assert inspected["recovery_decision"]["can_continue"] is True
    assert inspected["recovery_decision"]["resume_from"] == "host_execute"
    assert inspected["recovery_decision"]["reinvoke_host"] is True
    assert recovered["recovery"]["planned"] == inspected["recovery_decision"]
    assert recovered["status"] == "committed"
    assert session_actions == ["start_new", "resume"]
    replay = run_loopx_turn_once(plan, **common)
    assert replay["replayed"] is True
    assert _journal(tmp_path / "runtime")["host_attempt_count"] == 2
    assert calls == {"host": 2, "writeback": 1, "spend": 1, "scheduler": 1}


@pytest.mark.parametrize("host_recovery", ["corrupted", {}])
def test_run_once_rejects_corrupted_failed_turn_recovery_before_host_retry(
    tmp_path: Path,
    host_recovery: object,
) -> None:
    plan = _codex_plan()
    calls = {"host": 0, "writeback": 0, "spend": 0, "scheduler": 0}
    writeback, spend, scheduler = _callbacks(calls)

    def host(request: dict[str, object]) -> dict[str, object]:
        calls["host"] += 1
        if calls["host"] == 1:
            raise BuiltInHostError(
                "codex_cli_timeout",
                recovery_kind="resume_session",
            )
        return _host_result(plan)

    common = {
        "host_runner": host,
        "session_binding_resolver": lambda _turn_envelope: {
            "schema_version": "loopx_turn_session_binding_v0",
            "goal_id": "fixture-goal",
            "agent_id": "codex-fixture",
            "todo_id": "todo_fixture0001",
        },
        "project": tmp_path,
        "runtime_root": tmp_path / "runtime",
        "goal_id": "fixture-goal",
        "timeout_seconds": 5,
        "execute": True,
        "task_validator": _passing_validator,
        "writeback": writeback,
        "spend": spend,
        "scheduler": scheduler,
    }

    failed = run_loopx_turn_once(plan, **common)
    journal_paths = [
        path
        for path in (tmp_path / "runtime" / "goals" / "fixture-goal" / "turns").glob(
            "*.json"
        )
        if not path.name.endswith(".lock.holder.json")
    ]
    assert len(journal_paths) == 1
    journal = json.loads(journal_paths[0].read_text(encoding="utf-8"))
    journal["host_recovery"] = host_recovery
    journal_paths[0].write_text(json.dumps(journal), encoding="utf-8")

    with pytest.raises(ValueError, match="host recovery"):
        run_loopx_turn_once(plan, retry_failed=True, **common)

    assert failed["reason"] == "codex_cli_timeout"
    assert calls == {"host": 1, "writeback": 0, "spend": 0, "scheduler": 0}


def test_run_once_recoverable_failed_turn_rejects_session_identity_drift(
    tmp_path: Path,
) -> None:
    plan = _codex_plan()
    calls = {"host": 0, "writeback": 0, "spend": 0, "scheduler": 0}
    writeback, spend, scheduler = _callbacks(calls)

    def host(_request: dict[str, object]) -> dict[str, object]:
        calls["host"] += 1
        raise BuiltInHostError(
            "codex_cli_timeout",
            recovery_kind="resume_session",
        )

    common = {
        "host_runner": host,
        "session_binding_resolver": lambda _turn_envelope: {
            "schema_version": "loopx_turn_session_binding_v0",
            "goal_id": "fixture-goal",
            "agent_id": "codex-from-another-agent",
            "todo_id": "todo_fixture0001",
        },
        "project": tmp_path,
        "runtime_root": tmp_path / "runtime",
        "goal_id": "fixture-goal",
        "timeout_seconds": 5,
        "execute": True,
        "task_validator": _passing_validator,
        "writeback": writeback,
        "spend": spend,
        "scheduler": scheduler,
    }

    failed = run_loopx_turn_once(plan, **common)
    transaction = plan["transaction"]
    assert isinstance(transaction, dict)
    inspected = turn_executor.inspect_loopx_turn_journal(
        tmp_path / "runtime",
        goal_id="fixture-goal",
        agent_id="codex-fixture",
        turn_key=str(transaction["turn_key"]),
        retry_failed=True,
        session_binding_resolver=common["session_binding_resolver"],
    )
    assert inspected["recovery_decision"]["action"] == "blocked"
    assert (
        inspected["recovery_decision"]["reason"] == "session_binding_identity_mismatch"
    )
    with pytest.raises(ValueError, match="session binding does not match") as exc_info:
        run_loopx_turn_once(plan, retry_failed=True, **common)
    assert exc_info.value.decision == inspected["recovery_decision"]

    assert failed["reason"] == "codex_cli_timeout"
    assert calls == {"host": 1, "writeback": 0, "spend": 0, "scheduler": 0}


def test_run_once_commits_once_and_replays_without_duplicate_effects(
    tmp_path: Path,
) -> None:
    plan = _plan()
    result_path = tmp_path / "result.json"
    result_path.write_text(json.dumps(_host_result(plan)), encoding="utf-8")
    count_path = tmp_path / "host-count"
    calls = {"writeback": 0, "spend": 0, "scheduler": 0}
    writeback, spend, scheduler = _callbacks(calls)
    kwargs = {
        "host_argv": _host_argv(result_path, count_path),
        "project": tmp_path,
        "runtime_root": tmp_path / "runtime",
        "goal_id": "fixture-goal",
        "timeout_seconds": 5,
        "execute": True,
        "task_validator": _passing_validator,
        "writeback": writeback,
        "spend": spend,
        "scheduler": scheduler,
    }

    first = run_loopx_turn_once(plan, **kwargs)
    replay = run_loopx_turn_once(plan, **kwargs)

    assert first["ok"] is True
    assert first["status"] == "committed"
    assert first["receipt"]["status"] == "committed"
    assert [
        receipt["step_kind"] for receipt in first["settlement_result"]["receipts"]
    ] == ["validation", "durable_writeback", "quota_spend"]
    assert first["effects"]["host_invoked"] is True
    assert first["effects"]["state_written"] is True
    assert first["effects"]["quota_spent"] is True
    assert replay["replayed"] is True
    assert not any(replay["effects"].values())
    assert count_path.read_text(encoding="utf-8") == "1"
    assert calls == {"writeback": 1, "spend": 1, "scheduler": 1}

    # The route is a persisted compatibility surface, not just in-process state.
    # Exercise the actual TypeScript-backed journal writer and Python resume reader.
    transaction = plan["transaction"]
    assert isinstance(transaction, dict)
    turn_key = str(transaction["turn_key"])
    stored = json.loads(
        turn_journal_path(
            tmp_path / "runtime",
            goal_id="fixture-goal",
            turn_key=turn_key,
        ).read_text(encoding="utf-8")
    )
    assert stored["plan"]["route"]["kind"] == "ready_for_host"
    resumed = load_loopx_turn_plan_from_journal(
        tmp_path / "runtime",
        goal_id="fixture-goal",
        turn_key=turn_key,
    )
    assert resumed["route"] == stored["plan"]["route"]


def test_provider_can_commit_before_its_journal_checkpoint(
    tmp_path: Path,
    monkeypatch: pytest.MonkeyPatch,
) -> None:
    plan = _plan()
    calls = {"writeback": 0, "spend": 0, "scheduler": 0}
    _writeback, spend, scheduler = _callbacks(calls)
    provider_records: dict[str, dict[str, object]] = {}

    def writeback(_result: dict[str, object], effect_ref: str) -> dict[str, object]:
        calls["writeback"] += 1
        payload = {"ok": True, "appended": True, "effect_ref": effect_ref}
        provider_records[effect_ref] = payload
        return payload

    write_journal = turn_executor._write_journal

    def fail_before_writeback_checkpoint(
        path: Path,
        journal: Mapping[str, object],
    ) -> None:
        if "writeback" in journal and "quota_spend" not in journal:
            raise RuntimeError("injected crash before writeback checkpoint")
        write_journal(path, journal)

    monkeypatch.setattr(
        turn_executor,
        "_write_journal",
        fail_before_writeback_checkpoint,
    )

    with pytest.raises(
        RuntimeError,
        match="injected crash before writeback checkpoint",
    ):
        run_loopx_turn_once(
            plan,
            host_runner=lambda _request: _host_result(plan),
            project=tmp_path,
            runtime_root=tmp_path / "runtime",
            goal_id="fixture-goal",
            timeout_seconds=5,
            execute=True,
            task_validator=_passing_validator,
            writeback=writeback,
            spend=spend,
            scheduler=scheduler,
        )

    journal = _journal(tmp_path / "runtime")
    assert calls == {"writeback": 1, "spend": 0, "scheduler": 0}
    assert journal["completed_phases"] == list(TRANSACTION_PHASES[:3])
    assert "writeback" not in journal
    prepared = journal["effect_attempts"]["durable_writeback"]
    assert prepared["status"] == "prepared"
    assert prepared["effect_ref"] in provider_records

    monkeypatch.setattr(turn_executor, "_write_journal", write_journal)
    resumed = run_loopx_turn_once(
        plan,
        host_runner=lambda _request: pytest.fail("host must not run during recovery"),
        project=tmp_path,
        runtime_root=tmp_path / "runtime",
        goal_id="fixture-goal",
        timeout_seconds=5,
        execute=True,
        task_validator=_passing_validator,
        writeback=writeback,
        writeback_resolver=lambda effect_ref: {
            "kind": "committed",
            "payload": provider_records[effect_ref],
        },
        spend=spend,
        scheduler=scheduler,
    )

    assert resumed["status"] == "committed"
    assert calls == {"writeback": 1, "spend": 1, "scheduler": 1}
    assert "effect_attempts" not in _journal(tmp_path / "runtime")


def test_recreation_waits_for_admitted_provider_checkpoint(
    tmp_path: Path,
) -> None:
    registry = tmp_path / "project" / ".loopx" / "registry.json"
    runtime_root = registry.parent
    _write_source_registry(registry, INSTANCE_A)
    plan = _source_plan()
    admission = FirstPartyHostGoalAdmission.for_plan(
        registry_path=registry,
        goal_id="fixture-goal",
        planned_goal_ref=plan["goal_ref"],
    )
    provider_committed = threading.Event()
    allow_provider_return = threading.Event()
    turn_finished = threading.Event()
    turn_errors: list[BaseException] = []
    calls = {"writeback": 0, "spend": 0, "scheduler": 0}

    def writeback(
        _result: dict[str, object],
        effect_ref: str,
    ) -> dict[str, object]:
        calls["writeback"] += 1
        provider_committed.set()
        assert allow_provider_return.wait(timeout=5)
        return {"ok": True, "appended": True, "effect_ref": effect_ref}

    def run_turn() -> None:
        try:
            run_loopx_turn_once(
                plan,
                host_runner=lambda _request: _host_result(plan),
                project=tmp_path,
                runtime_root=runtime_root,
                goal_id="fixture-goal",
                timeout_seconds=5,
                execute=True,
                task_validator=_passing_validator,
                writeback=writeback,
                spend=lambda: (
                    calls.__setitem__("spend", calls["spend"] + 1)
                    or {"ok": True, "appended": True}
                ),
                scheduler=lambda _spend: (
                    calls.__setitem__("scheduler", calls["scheduler"] + 1)
                    or {"completed": True}
                ),
                goal_admission=admission,
            )
        except BaseException as exc:
            turn_errors.append(exc)
        finally:
            turn_finished.set()

    turn_thread = threading.Thread(target=run_turn)
    turn_thread.start()
    assert provider_committed.wait(timeout=5), turn_errors

    request = RecreateGoalRequest(
        registry_path=registry,
        goal_id="fixture-goal",
        goal_instance_id=INSTANCE_A,
        operation_id="recreate-after-provider-commit",
    )
    blocked = recreate_goal_instance(request)

    assert blocked["ok"] is False
    assert blocked["status"] == "drain_required"
    assert load_project_registry(registry)["goals"][0]["goal_instance_id"] == INSTANCE_A
    gate_path = (
        registry.parent
        / ".loopx"
        / "lifecycle"
        / "goal-instance"
        / "turn-settlement"
        / alias_digest("fixture-goal")
        / "gate.json"
    )
    gate = json.loads(gate_path.read_text(encoding="utf-8"))
    assert gate["state"] == "closing"

    allow_provider_return.set()
    assert turn_finished.wait(timeout=5)
    turn_thread.join(timeout=5)

    assert len(turn_errors) == 1
    assert isinstance(turn_errors[0], FirstPartyHostRuntimeRejected)
    assert turn_errors[0].code == "goal_retirement_in_progress"
    journal = _journal(runtime_root)
    assert journal["writeback"]["appended"] is True
    assert "effect_attempts" not in journal
    assert calls == {"writeback": 1, "spend": 0, "scheduler": 0}

    recreated = recreate_goal_instance(request)
    assert recreated["ok"] is True
    assert recreated["replayed"] is False
    assert recreated["goal_ref"]["goal_instance_id"] != INSTANCE_A
    assert (
        load_project_registry(registry)["goals"][0]["goal_instance_id"]
        == (recreated["goal_ref"]["goal_instance_id"])
    )


def test_source_provider_cannot_change_admitted_effect_ref(
    tmp_path: Path,
) -> None:
    registry = tmp_path / "project" / ".loopx" / "registry.json"
    runtime_root = registry.parent
    _write_source_registry(registry, INSTANCE_A)
    plan = _source_plan()
    admission = FirstPartyHostGoalAdmission.for_plan(
        registry_path=registry,
        goal_id="fixture-goal",
        planned_goal_ref=plan["goal_ref"],
    )

    failed = run_loopx_turn_once(
        plan,
        host_runner=lambda _request: _host_result(plan),
        project=tmp_path,
        runtime_root=runtime_root,
        goal_id="fixture-goal",
        timeout_seconds=5,
        execute=True,
        task_validator=_passing_validator,
        writeback=lambda _result, _effect_ref: {
            "ok": True,
            "appended": True,
            "effect_ref": "different-effect",
        },
        spend=lambda: pytest.fail("quota must not run after identity drift"),
        scheduler=lambda _spend: pytest.fail(
            "scheduler must not run after identity drift"
        ),
        goal_admission=admission,
    )
    assert failed["status"] == "failed"
    assert failed["settlement_result"]["failure"]["kind"] == "identity_mismatch"

    blocked = recreate_goal_instance(
        RecreateGoalRequest(
            registry_path=registry,
            goal_id="fixture-goal",
            goal_instance_id=INSTANCE_A,
            operation_id="recreate-after-provider-identity-drift",
        )
    )
    assert blocked["status"] == "drain_required"
    assert blocked["pending_effects"][0]["reason"] == "provider_readback_required"


def test_recreation_waits_for_scheduler_and_post_settlement(
    tmp_path: Path,
) -> None:
    registry = tmp_path / "project" / ".loopx" / "registry.json"
    runtime_root = registry.parent
    _write_source_registry(registry, INSTANCE_A)
    plan = _source_plan()
    admission = FirstPartyHostGoalAdmission.for_plan(
        registry_path=registry,
        goal_id="fixture-goal",
        planned_goal_ref=plan["goal_ref"],
    )
    request = RecreateGoalRequest(
        registry_path=registry,
        goal_id="fixture-goal",
        goal_instance_id=INSTANCE_A,
        operation_id="recreate-during-settlement-tail",
    )
    blocked: list[dict[str, object]] = []
    observed_instances: list[str] = []

    def attempt_recreation() -> None:
        blocked.append(recreate_goal_instance(request))
        observed_instances.append(
            str(load_project_registry(registry)["goals"][0]["goal_instance_id"])
        )

    def scheduler(_spend: dict[str, object]) -> dict[str, object]:
        attempt_recreation()
        return {"completed": True, "acknowledged": True}

    def post_settlement(
        _plan: Mapping[str, object],
        _result: Mapping[str, object],
        _evidence: Mapping[str, object],
    ) -> dict[str, object]:
        attempt_recreation()
        return {"status": "recorded"}

    settled = run_loopx_turn_once(
        plan,
        host_runner=lambda _request: _host_result(plan),
        project=tmp_path,
        runtime_root=runtime_root,
        goal_id="fixture-goal",
        timeout_seconds=5,
        execute=True,
        task_validator=_passing_validator,
        writeback=lambda _result: {"ok": True, "appended": True},
        spend=lambda: {"ok": True, "appended": True},
        scheduler=scheduler,
        post_settlement=post_settlement,
        goal_admission=admission,
    )

    assert settled["status"] == "committed"
    assert [result["status"] for result in blocked] == [
        "drain_required",
        "drain_required",
    ]
    assert observed_instances == [INSTANCE_A, INSTANCE_A]

    recreated = recreate_goal_instance(request)
    assert recreated["ok"] is True
    assert recreated["goal_ref"]["goal_instance_id"] != INSTANCE_A


def test_recreation_cannot_publish_before_tail_hold_checkpoint(
    tmp_path: Path,
    monkeypatch: pytest.MonkeyPatch,
) -> None:
    registry = tmp_path / "project" / ".loopx" / "registry.json"
    runtime_root = registry.parent
    _write_source_registry(registry, INSTANCE_A)
    plan = _source_plan()
    admission = FirstPartyHostGoalAdmission.for_plan(
        registry_path=registry,
        goal_id="fixture-goal",
        planned_goal_ref=plan["goal_ref"],
    )
    hold_started = threading.Event()
    allow_hold = threading.Event()
    turn_finished = threading.Event()
    turn_errors: list[BaseException] = []
    original_hold = type(
        admission.turn_effect_admission(
            turn_key=str(plan["transaction"]["turn_key"]),
            journal_path=turn_executor.turn_journal_path(
                runtime_root,
                goal_id="fixture-goal",
                turn_key=str(plan["transaction"]["turn_key"]),
            ),
        )
    ).hold

    def blocked_hold(
        self: object,
        step_kind: object,
        effect_ref: str,
        persist_journal: object,
    ) -> None:
        hold_started.set()
        assert allow_hold.wait(timeout=5)
        original_hold(self, step_kind, effect_ref, persist_journal)

    monkeypatch.setattr(
        "loopx.control_plane.goals.first_party_host_admission."
        "FirstPartyHostTurnEffectAdmission.hold",
        blocked_hold,
    )

    def run_turn() -> None:
        try:
            run_loopx_turn_once(
                plan,
                host_runner=lambda _request: _host_result(plan),
                project=tmp_path,
                runtime_root=runtime_root,
                goal_id="fixture-goal",
                timeout_seconds=5,
                execute=True,
                task_validator=_passing_validator,
                writeback=lambda _result: {"ok": True, "appended": True},
                spend=lambda: {"ok": True, "appended": True},
                scheduler=lambda _spend: {"completed": True},
                goal_admission=admission,
            )
        except BaseException as exc:
            turn_errors.append(exc)
        finally:
            turn_finished.set()

    turn_thread = threading.Thread(target=run_turn)
    turn_thread.start()
    assert hold_started.wait(timeout=5), turn_errors
    request = RecreateGoalRequest(
        registry_path=registry,
        goal_id="fixture-goal",
        goal_instance_id=INSTANCE_A,
        operation_id="recreate-before-tail-hold",
    )

    blocked = recreate_goal_instance(request)

    assert blocked["status"] == "drain_required"
    assert load_project_registry(registry)["goals"][0]["goal_instance_id"] == INSTANCE_A
    allow_hold.set()
    assert turn_finished.wait(timeout=5)
    turn_thread.join(timeout=5)
    assert turn_errors == []
    assert recreate_goal_instance(request)["ok"] is True


@pytest.mark.parametrize(
    ("result_kind", "tail_step"),
    [
        ("validated_progress", "quota_spend"),
        ("validated_completion", "terminal_closeout"),
    ],
)
def test_final_provider_checkpoint_persists_tail_hold_before_confirmation(
    tmp_path: Path,
    monkeypatch: pytest.MonkeyPatch,
    result_kind: str,
    tail_step: str,
) -> None:
    registry = tmp_path / "project" / ".loopx" / "registry.json"
    runtime_root = registry.parent
    _write_source_registry(registry, INSTANCE_A)
    plan = _source_plan()
    admission = FirstPartyHostGoalAdmission.for_plan(
        registry_path=registry,
        goal_id="fixture-goal",
        planned_goal_ref=plan["goal_ref"],
    )

    def crash_before_tail_confirmation(
        _self: object,
        _step_kind: object,
        _effect_ref: str,
    ) -> None:
        raise RuntimeError("injected crash after final provider checkpoint")

    monkeypatch.setattr(
        turn_executor.TurnSettlementJournalAdapter,
        "hold_tail",
        crash_before_tail_confirmation,
    )
    completion_callbacks = (
        {
            "completion_writeback": lambda _result: pytest.fail(
                "terminal completion must not use the pre-spend lifecycle callback"
            ),
            "completion_intent": lambda _result: {
                "todo_id": "todo_fixture0001",
                "continuation": "no_followup",
            },
            "terminal_closeout": lambda _result: {
                "ok": True,
                "appended": True,
                "completion": {
                    "todo_id": "todo_fixture0001",
                    "continuation": "no_followup",
                },
            },
        }
        if result_kind == "validated_completion"
        else {}
    )

    with pytest.raises(
        RuntimeError,
        match="injected crash after final provider checkpoint",
    ):
        run_loopx_turn_once(
            plan,
            host_runner=lambda _request: _host_result(plan, kind=result_kind),
            project=tmp_path,
            runtime_root=runtime_root,
            goal_id="fixture-goal",
            timeout_seconds=5,
            execute=True,
            task_validator=_passing_validator,
            writeback=lambda _result: {"ok": True, "appended": True},
            spend=lambda: {"ok": True, "appended": True},
            scheduler=lambda _spend: pytest.fail(
                "scheduler must not run before tail confirmation"
            ),
            goal_admission=admission,
            **completion_callbacks,
        )

    journal = _journal(runtime_root)
    hold = journal["source_effect_hold"]
    assert hold["status"] == "held"
    assert hold["step_kind"] == tail_step
    assert "effect_attempts" not in journal

    blocked = recreate_goal_instance(
        RecreateGoalRequest(
            registry_path=registry,
            goal_id="fixture-goal",
            goal_instance_id=INSTANCE_A,
            operation_id=f"recreate-after-{tail_step}-checkpoint",
        )
    )
    assert blocked["status"] == "drain_required"
    assert blocked["pending_effects"] == [
        {
            "turn_key": str(plan["transaction"]["turn_key"]),
            "step_kind": tail_step,
            "reason": "turn_tail_recovery_required",
            "recovery_action": (
                "Resume this Turn to finish its settlement tail, "
                "then retry recreate-goal."
            ),
        }
    ]
    assert load_project_registry(registry)["goals"][0]["goal_instance_id"] == INSTANCE_A

    journal_path = turn_journal_path(
        runtime_root,
        goal_id="fixture-goal",
        turn_key=str(plan["transaction"]["turn_key"]),
    )
    hold["schema_version"] = "loopx_source_turn_effect_hold_invalid"
    journal_path.write_text(json.dumps(journal), encoding="utf-8")

    conflicted = recreate_goal_instance(
        RecreateGoalRequest(
            registry_path=registry,
            goal_id="fixture-goal",
            goal_instance_id=INSTANCE_A,
            operation_id=f"recreate-after-{tail_step}-checkpoint",
        )
    )
    assert conflicted["changed"] is False
    assert conflicted["replayed"] is True
    assert conflicted["pending_effects"][0]["reason"] == "turn_tail_conflict"


@pytest.mark.parametrize("first_scheduler_result", ["exception", "incomplete"])
def test_recreation_waits_for_tail_recovery(
    tmp_path: Path,
    first_scheduler_result: str,
) -> None:
    registry = tmp_path / "project" / ".loopx" / "registry.json"
    runtime_root = registry.parent
    _write_source_registry(registry, INSTANCE_A)
    plan = _source_plan()
    admission = FirstPartyHostGoalAdmission.for_plan(
        registry_path=registry,
        goal_id="fixture-goal",
        planned_goal_ref=plan["goal_ref"],
    )
    calls = {"host": 0, "writeback": 0, "spend": 0, "scheduler": 0}

    def host(_request: Mapping[str, object]) -> dict[str, object]:
        calls["host"] += 1
        return _host_result(plan)

    def writeback(_result: Mapping[str, object]) -> dict[str, object]:
        calls["writeback"] += 1
        return {"ok": True, "appended": True}

    def spend() -> dict[str, object]:
        calls["spend"] += 1
        return {"ok": True, "appended": True}

    def first_scheduler(_spend: Mapping[str, object]) -> dict[str, object]:
        calls["scheduler"] += 1
        if first_scheduler_result == "exception":
            raise KeyboardInterrupt
        return {"completed": False, "acknowledged": False}

    common = {
        "host_runner": host,
        "project": tmp_path,
        "runtime_root": runtime_root,
        "goal_id": "fixture-goal",
        "timeout_seconds": 5,
        "execute": True,
        "task_validator": _passing_validator,
        "writeback": writeback,
        "spend": spend,
        "goal_admission": admission,
    }
    if first_scheduler_result == "exception":
        with pytest.raises(KeyboardInterrupt):
            run_loopx_turn_once(plan, scheduler=first_scheduler, **common)
    else:
        first = run_loopx_turn_once(plan, scheduler=first_scheduler, **common)
        assert first["status"] == "scheduler_action_required"

    request = RecreateGoalRequest(
        registry_path=registry,
        goal_id="fixture-goal",
        goal_instance_id=INSTANCE_A,
        operation_id=f"recreate-after-tail-{first_scheduler_result}",
    )
    blocked = recreate_goal_instance(request)
    assert blocked["status"] == "drain_required"
    assert blocked["pending_effects"][0]["reason"] == ("turn_tail_recovery_required")

    def healthy_scheduler(_spend: Mapping[str, object]) -> dict[str, object]:
        calls["scheduler"] += 1
        return {"completed": True, "acknowledged": True}

    recovered = run_loopx_turn_once(plan, scheduler=healthy_scheduler, **common)
    assert recovered["status"] == "committed"
    assert calls == {"host": 1, "writeback": 1, "spend": 1, "scheduler": 2}
    assert recreate_goal_instance(request)["ok"] is True


@pytest.mark.parametrize(
    ("readback_kind", "recreation_completes"),
    [
        ("committed", True),
        ("absent", True),
        ("unknown", False),
    ],
)
def test_recreation_reconciles_crashed_source_provider_without_reexecution(
    tmp_path: Path,
    monkeypatch: pytest.MonkeyPatch,
    readback_kind: str,
    recreation_completes: bool,
) -> None:
    registry = tmp_path / "project" / ".loopx" / "registry.json"
    runtime_root = registry.parent
    _write_source_registry(registry, INSTANCE_A)
    plan = _source_plan()
    admission = FirstPartyHostGoalAdmission.for_plan(
        registry_path=registry,
        goal_id="fixture-goal",
        planned_goal_ref=plan["goal_ref"],
    )
    calls = {"writeback": 0, "spend": 0, "scheduler": 0}
    provider_records: dict[str, dict[str, object]] = {}

    def writeback(
        _result: dict[str, object],
        effect_ref: str,
    ) -> dict[str, object]:
        calls["writeback"] += 1
        payload = {"ok": True, "appended": True, "effect_ref": effect_ref}
        provider_records[effect_ref] = payload
        return payload

    write_journal = turn_executor._write_journal

    def fail_before_writeback_checkpoint(
        path: Path,
        journal: Mapping[str, object],
        **kwargs: object,
    ) -> None:
        if "writeback" in journal and "quota_spend" not in journal:
            raise RuntimeError("injected source checkpoint crash")
        write_journal(path, journal, **kwargs)

    monkeypatch.setattr(
        turn_executor,
        "_write_journal",
        fail_before_writeback_checkpoint,
    )
    with pytest.raises(RuntimeError, match="injected source checkpoint crash"):
        run_loopx_turn_once(
            plan,
            host_runner=lambda _request: _host_result(plan),
            project=tmp_path,
            runtime_root=runtime_root,
            goal_id="fixture-goal",
            timeout_seconds=5,
            execute=True,
            task_validator=_passing_validator,
            writeback=writeback,
            spend=lambda: pytest.fail("quota must not run before writeback checkpoint"),
            scheduler=lambda _spend: pytest.fail(
                "scheduler must not run before writeback checkpoint"
            ),
            goal_admission=admission,
        )
    monkeypatch.setattr(turn_executor, "_write_journal", write_journal)

    request = RecreateGoalRequest(
        registry_path=registry,
        goal_id="fixture-goal",
        goal_instance_id=INSTANCE_A,
        operation_id=f"recreate-after-{readback_kind}-readback",
    )
    blocked = recreate_goal_instance(request)
    assert blocked["status"] == "drain_required"

    effect_ref = next(iter(provider_records))

    def resolver(_effect_ref: str) -> dict[str, object]:
        assert _effect_ref == effect_ref
        if readback_kind == "committed":
            return {"kind": "committed", "payload": provider_records[effect_ref]}
        if readback_kind == "absent":
            return {"kind": "absent"}
        return {"kind": "unknown", "reason": "provider unavailable"}

    def forbidden_writeback(
        _result: dict[str, object],
        _effect_ref: str,
    ) -> dict[str, object]:
        pytest.fail("retirement recovery must not re-execute the provider")

    try:
        resumed = run_loopx_turn_once(
            plan,
            host_runner=lambda _request: pytest.fail(
                "settlement recovery must not relaunch the Host"
            ),
            project=tmp_path,
            runtime_root=runtime_root,
            goal_id="fixture-goal",
            timeout_seconds=5,
            execute=True,
            task_validator=_passing_validator,
            writeback=forbidden_writeback,
            writeback_resolver=resolver,
            spend=lambda: (
                calls.__setitem__("spend", calls["spend"] + 1)
                or {"ok": True, "appended": True}
            ),
            scheduler=lambda _spend: (
                calls.__setitem__("scheduler", calls["scheduler"] + 1)
                or {"completed": True}
            ),
            goal_admission=admission,
        )
    except FirstPartyHostRuntimeRejected as exc:
        assert readback_kind == "committed"
        assert exc.code == "goal_retirement_in_progress"
    else:
        assert resumed["status"] == "failed"

    assert calls == {"writeback": 1, "spend": 0, "scheduler": 0}
    retried = recreate_goal_instance(request)
    assert retried["ok"] is recreation_completes
    if recreation_completes:
        assert retried["goal_ref"]["goal_instance_id"] != INSTANCE_A
    else:
        assert retried["status"] == "drain_required"
        assert load_project_registry(registry)["goals"][0]["goal_instance_id"] == (
            INSTANCE_A
        )


@pytest.mark.parametrize(
    ("failure_point", "provider_calls"),
    [
        ("prepared_journal", 0),
        ("admission_release", 1),
    ],
)
def test_recreation_repairs_partial_source_effect_admission(
    tmp_path: Path,
    monkeypatch: pytest.MonkeyPatch,
    failure_point: str,
    provider_calls: int,
) -> None:
    registry = tmp_path / "project" / ".loopx" / "registry.json"
    runtime_root = registry.parent
    _write_source_registry(registry, INSTANCE_A)
    plan = _source_plan()
    admission = FirstPartyHostGoalAdmission.for_plan(
        registry_path=registry,
        goal_id="fixture-goal",
        planned_goal_ref=plan["goal_ref"],
    )
    calls = {"writeback": 0}
    write_journal = turn_executor._write_journal
    remove_admission = source_session_turn_effects._remove_admission

    if failure_point == "prepared_journal":

        def fail_prepared_journal(
            path: Path,
            journal: Mapping[str, object],
            **kwargs: object,
        ) -> None:
            if "effect_attempts" in journal:
                raise RuntimeError("injected prepared journal failure")
            write_journal(path, journal, **kwargs)

        monkeypatch.setattr(
            turn_executor,
            "_write_journal",
            fail_prepared_journal,
        )
        expected_error = "injected prepared journal failure"
    else:

        def fail_admission_release(_path: Path) -> None:
            raise RuntimeError("injected admission release failure")

        monkeypatch.setattr(
            source_session_turn_effects,
            "_remove_admission",
            fail_admission_release,
        )
        expected_error = "injected admission release failure"

    def writeback(
        _result: dict[str, object],
        effect_ref: str,
    ) -> dict[str, object]:
        calls["writeback"] += 1
        return {"ok": True, "appended": True, "effect_ref": effect_ref}

    with pytest.raises(RuntimeError, match=expected_error):
        run_loopx_turn_once(
            plan,
            host_runner=lambda _request: _host_result(plan),
            project=tmp_path,
            runtime_root=runtime_root,
            goal_id="fixture-goal",
            timeout_seconds=5,
            execute=True,
            task_validator=_passing_validator,
            writeback=writeback,
            spend=lambda: pytest.fail("quota must not run after the injected failure"),
            scheduler=lambda _spend: pytest.fail(
                "scheduler must not run after the injected failure"
            ),
            goal_admission=admission,
        )

    monkeypatch.setattr(turn_executor, "_write_journal", write_journal)
    monkeypatch.setattr(
        source_session_turn_effects,
        "_remove_admission",
        remove_admission,
    )
    recreated = recreate_goal_instance(
        RecreateGoalRequest(
            registry_path=registry,
            goal_id="fixture-goal",
            goal_instance_id=INSTANCE_A,
            operation_id=f"recreate-after-{failure_point}",
        )
    )

    assert calls["writeback"] == provider_calls
    assert recreated["ok"] is True
    assert recreated["goal_ref"]["goal_instance_id"] != INSTANCE_A


def test_recreation_replay_repairs_gate_after_registry_publication(
    tmp_path: Path,
    monkeypatch: pytest.MonkeyPatch,
) -> None:
    registry = tmp_path / "project" / ".loopx" / "registry.json"
    _write_source_registry(registry, INSTANCE_A)
    request = RecreateGoalRequest(
        registry_path=registry,
        goal_id="fixture-goal",
        goal_instance_id=INSTANCE_A,
        operation_id="recreate-before-gate-reopen",
    )
    write_sidecar = source_session_turn_effects.write_journal
    failed_once = False

    def fail_gate_reopen(path: Path, payload: dict[str, object]) -> None:
        nonlocal failed_once
        if payload.get("state") == "open" and not failed_once:
            failed_once = True
            raise OSError("injected gate reopen failure")
        write_sidecar(path, payload)

    monkeypatch.setattr(
        source_session_turn_effects,
        "write_journal",
        fail_gate_reopen,
    )
    with pytest.raises(OSError, match="injected gate reopen failure"):
        recreate_goal_instance(request)
    instance_b = load_project_registry(registry)["goals"][0]["goal_instance_id"]
    assert instance_b != INSTANCE_A

    monkeypatch.setattr(
        source_session_turn_effects,
        "write_journal",
        write_sidecar,
    )
    replayed = recreate_goal_instance(request)

    assert replayed["ok"] is True
    assert replayed["replayed"] is True
    assert replayed["goal_ref"]["goal_instance_id"] == instance_b


def test_historical_recreation_replay_preserves_later_goal_gate(
    tmp_path: Path,
) -> None:
    registry = tmp_path / "project" / ".loopx" / "registry.json"
    _write_source_registry(registry, INSTANCE_A)
    first_request = RecreateGoalRequest(
        registry_path=registry,
        goal_id="fixture-goal",
        goal_instance_id=INSTANCE_A,
        operation_id="recreate-a-to-b",
    )
    first = recreate_goal_instance(first_request)
    instance_b = str(first["goal_ref"]["goal_instance_id"])
    second = recreate_goal_instance(
        RecreateGoalRequest(
            registry_path=registry,
            goal_id="fixture-goal",
            goal_instance_id=instance_b,
            operation_id="recreate-b-to-c",
        )
    )
    instance_c = str(second["goal_ref"]["goal_instance_id"])

    replayed = recreate_goal_instance(first_request)

    assert replayed["ok"] is True
    assert replayed["replayed"] is True
    assert replayed["goal_ref"]["goal_instance_id"] == instance_b
    assert load_project_registry(registry)["goals"][0]["goal_instance_id"] == instance_c
    gate = json.loads(
        source_session_turn_effects.source_turn_effect_gate_path(
            registry,
            "fixture-goal",
        ).read_text(encoding="utf-8")
    )
    assert gate["state"] == "open"
    assert gate["goal_ref"]["goal_instance_id"] == instance_c


def test_run_once_legacy_plan_without_settlement_plan_is_upgraded(
    tmp_path: Path,
) -> None:
    plan = _plan()
    transaction = plan["transaction"]
    assert isinstance(transaction, dict)
    transaction.pop("settlement_plan", None)
    result_path = tmp_path / "result.json"
    result_path.write_text(json.dumps(_host_result(plan)), encoding="utf-8")
    count_path = tmp_path / "host-count"
    calls = {"writeback": 0, "spend": 0, "scheduler": 0}
    writeback, spend, scheduler = _callbacks(calls)

    committed = run_loopx_turn_once(
        plan,
        host_argv=_host_argv(result_path, count_path),
        project=tmp_path,
        runtime_root=tmp_path / "runtime",
        goal_id="fixture-goal",
        timeout_seconds=5,
        execute=True,
        task_validator=_passing_validator,
        writeback=writeback,
        spend=spend,
        scheduler=scheduler,
    )

    assert committed["ok"] is True
    assert committed["status"] == "committed"
    assert committed["receipt"]["status"] == "committed"
    assert committed["settlement_result"]["ok"] is True
    assert [
        receipt["step_kind"] for receipt in committed["settlement_result"]["receipts"]
    ] == ["validation", "durable_writeback", "quota_spend"]
    assert calls == {"writeback": 1, "spend": 1, "scheduler": 1}


def test_adaptive_completion_upgrades_legacy_plan_with_primary_todo_identity(
    tmp_path: Path,
) -> None:
    plan = _adaptive_plan()
    transaction = plan["transaction"]
    assert isinstance(transaction, dict)
    transaction.pop("settlement_plan", None)
    calls = {"completion": 0, "spend": 0, "scheduler": 0}

    def completion_writeback(_result: dict[str, object]) -> dict[str, object]:
        calls["completion"] += 1
        return {
            "ok": True,
            "appended": True,
            "completion": {
                "todo_id": "todo_fixture0001",
                "continuation": "active_goal",
            },
        }

    committed = run_loopx_turn_once(
        plan,
        host_runner=lambda _request: _host_result(
            plan,
            kind="validated_completion",
        ),
        project=tmp_path,
        runtime_root=tmp_path / "runtime",
        goal_id="fixture-goal",
        timeout_seconds=5,
        execute=True,
        task_validator=_passing_validator,
        writeback=lambda _result: {"ok": True, "appended": True},
        completion_writeback=completion_writeback,
        completion_intent=lambda _result: {
            "todo_id": "todo_fixture0001",
            "continuation": "active_goal",
        },
        terminal_closeout=lambda _result: {
            "ok": True,
            "appended": True,
            "completion": {
                "todo_id": "todo_fixture0001",
                "continuation": "no_followup",
            },
        },
        spend=lambda: (
            calls.__setitem__("spend", calls["spend"] + 1)
            or {"ok": True, "appended": True}
        ),
        scheduler=lambda _spend: (
            calls.__setitem__("scheduler", calls["scheduler"] + 1)
            or {"completed": True, "acknowledged": True}
        ),
    )

    assert committed["status"] == "committed"
    assert committed["receipt"]["lineage"]["todo_id"] == "todo_fixture0001"
    assert calls == {"completion": 1, "spend": 1, "scheduler": 1}


def test_validated_completion_requires_explicit_lifecycle_writeback(
    tmp_path: Path,
) -> None:
    plan = _plan()
    result_path = tmp_path / "result.json"
    result_path.write_text(
        json.dumps(_host_result(plan, kind="validated_completion")),
        encoding="utf-8",
    )
    calls = {"writeback": 0, "completion": 0, "spend": 0, "scheduler": 0}

    def writeback(_result: dict[str, object]) -> dict[str, object]:
        calls["writeback"] += 1
        return {"ok": True, "appended": True}

    def spend() -> dict[str, object]:
        calls["spend"] += 1
        return {"ok": True, "appended": True}

    def scheduler(_spend: dict[str, object]) -> dict[str, object]:
        calls["scheduler"] += 1
        return {"completed": True, "acknowledged": True}

    payload = run_loopx_turn_once(
        plan,
        host_argv=_host_argv(result_path, tmp_path / "host-count"),
        project=tmp_path,
        runtime_root=tmp_path / "runtime",
        goal_id="fixture-goal",
        timeout_seconds=5,
        execute=True,
        task_validator=_passing_validator,
        writeback=writeback,
        spend=spend,
        scheduler=scheduler,
    )

    assert payload["result_kind"] == "validation_failed"
    assert payload["receipt"]["failed_phase"] == "validation"
    assert calls == {"writeback": 0, "completion": 0, "spend": 0, "scheduler": 0}


def test_validated_completion_commits_once_with_lifecycle_outcome(
    tmp_path: Path,
) -> None:
    plan = _plan()
    result_path = tmp_path / "result.json"
    result_path.write_text(
        json.dumps(_host_result(plan, kind="validated_completion")),
        encoding="utf-8",
    )
    calls = {"writeback": 0, "completion": 0, "spend": 0, "scheduler": 0}
    writeback, spend, scheduler = _callbacks(calls)

    def completion_writeback(_result: dict[str, object]) -> dict[str, object]:
        calls["completion"] += 1
        return {
            "ok": True,
            "appended": True,
            "completion": {
                "todo_id": "todo_fixture0001",
                "continuation": "active_goal",
            },
        }

    kwargs = {
        "host_argv": _host_argv(result_path, tmp_path / "host-count"),
        "project": tmp_path,
        "runtime_root": tmp_path / "runtime",
        "goal_id": "fixture-goal",
        "timeout_seconds": 5,
        "execute": True,
        "task_validator": _passing_validator,
        "writeback": writeback,
        "completion_writeback": completion_writeback,
        "completion_intent": lambda _result: {
            "todo_id": "todo_fixture0001",
            "continuation": "active_goal",
        },
        "terminal_closeout": lambda _result: {
            "ok": True,
            "appended": True,
            "completion": {
                "todo_id": "todo_fixture0001",
                "continuation": "no_followup",
            },
        },
        "spend": spend,
        "scheduler": scheduler,
    }
    first = run_loopx_turn_once(plan, **kwargs)
    replay = run_loopx_turn_once(plan, **kwargs)

    assert first["status"] == "committed"
    assert first["effects"]["state_written"] is True
    assert first["effects"]["quota_spent"] is True
    assert replay["replayed"] is True
    assert not any(replay["effects"].values())
    assert calls == {"writeback": 0, "completion": 1, "spend": 1, "scheduler": 1}
    assert _journal(tmp_path / "runtime")["writeback"]["completion"] == {
        "todo_id": "todo_fixture0001",
        "continuation": "active_goal",
    }


def test_invalid_completion_outcome_fails_closed_without_spending(
    tmp_path: Path,
) -> None:
    plan = _plan()
    calls = {"completion": 0, "spend": 0, "scheduler": 0}
    writeback, _spend, _scheduler = _callbacks(calls)

    def completion_writeback(_result: dict[str, object]) -> dict[str, object]:
        calls["completion"] += 1
        return {
            "ok": True,
            "appended": True,
            "completion": {"todo_id": "todo_other", "continuation": "active_goal"},
        }

    payload = run_loopx_turn_once(
        plan,
        host_runner=lambda _request: _host_result(plan, kind="validated_completion"),
        project=tmp_path,
        runtime_root=tmp_path / "runtime",
        goal_id="fixture-goal",
        timeout_seconds=5,
        execute=True,
        task_validator=_passing_validator,
        writeback=writeback,
        completion_writeback=completion_writeback,
        completion_intent=lambda _result: {
            "todo_id": "todo_other",
            "continuation": "active_goal",
        },
        terminal_closeout=lambda _result: {
            "ok": True,
            "appended": True,
            "completion": {
                "todo_id": "todo_fixture0001",
                "continuation": "no_followup",
            },
        },
        spend=lambda: (
            calls.__setitem__("spend", calls["spend"] + 1)
            or {
                "ok": True,
                "appended": True,
            }
        ),
        scheduler=lambda _spend: (
            calls.__setitem__("scheduler", calls["scheduler"] + 1)
            or {"completed": True, "acknowledged": True}
        ),
    )

    assert payload["result_kind"] == "writeback_failed"
    assert payload["receipt"]["result_kind"] == "writeback_failed"
    assert payload["receipt"]["failed_phase"] == "durable_writeback"
    assert payload["settlement_result"]["failure"]["kind"] == "writeback_rejected"
    assert calls == {"completion": 0, "spend": 0, "scheduler": 0}


def test_terminal_closeout_runs_only_after_matching_spend_receipt(
    tmp_path: Path,
) -> None:
    plan = _plan()
    events: list[str] = []

    payload = run_loopx_turn_once(
        plan,
        host_runner=lambda _request: _host_result(
            plan,
            kind="validated_completion",
        ),
        project=tmp_path,
        runtime_root=tmp_path / "runtime",
        goal_id="fixture-goal",
        timeout_seconds=5,
        execute=True,
        task_validator=_passing_validator,
        writeback=lambda _result: (
            events.append("writeback") or {"ok": True, "appended": True}
        ),
        completion_writeback=lambda _result: pytest.fail(
            "terminal completion must not use the pre-spend lifecycle callback"
        ),
        completion_intent=lambda _result: {
            "todo_id": "todo_fixture0001",
            "continuation": "no_followup",
        },
        spend=lambda: events.append("spend") or {"ok": True, "appended": True},
        terminal_closeout=lambda _result: (
            events.append("terminal_closeout")
            or {
                "ok": True,
                "appended": True,
                "completion": {
                    "todo_id": "todo_fixture0001",
                    "continuation": "no_followup",
                },
            }
        ),
        scheduler=lambda _spend: (
            events.append("scheduler") or {"completed": True, "acknowledged": True}
        ),
    )

    assert payload["status"] == "committed"
    assert events == ["writeback", "spend", "terminal_closeout", "scheduler"]
    assert [
        receipt["step_kind"] for receipt in payload["settlement_result"]["receipts"]
    ] == ["validation", "durable_writeback", "quota_spend", "terminal_closeout"]
    assert payload["todo_completion"] == {
        "todo_id": "todo_fixture0001",
        "continuation": "no_followup",
    }


def test_terminal_closeout_lost_receipt_retries_without_repeating_effects(
    tmp_path: Path,
) -> None:
    plan = _plan()
    calls = {
        "writeback": 0,
        "spend": 0,
        "terminal_attempt": 0,
        "terminal_mutation": 0,
        "scheduler": 0,
    }
    terminal_committed = False

    def terminal_closeout(_result: dict[str, object]) -> dict[str, object]:
        nonlocal terminal_committed
        calls["terminal_attempt"] += 1
        if not terminal_committed:
            terminal_committed = True
            calls["terminal_mutation"] += 1
            return {
                "ok": False,
                "appended": False,
                "reason": "terminal closeout receipt was interrupted",
            }
        return {
            "ok": True,
            "appended": True,
            "completion": {
                "todo_id": "todo_fixture0001",
                "continuation": "no_followup",
            },
        }

    common = {
        "host_runner": lambda _request: _host_result(
            plan,
            kind="validated_completion",
        ),
        "project": tmp_path,
        "runtime_root": tmp_path / "runtime",
        "goal_id": "fixture-goal",
        "timeout_seconds": 5,
        "execute": True,
        "task_validator": _passing_validator,
        "writeback": lambda _result: (
            calls.__setitem__("writeback", calls["writeback"] + 1)
            or {"ok": True, "appended": True}
        ),
        "completion_writeback": lambda _result: pytest.fail(
            "terminal completion must not use the pre-spend lifecycle callback"
        ),
        "completion_intent": lambda _result: {
            "todo_id": "todo_fixture0001",
            "continuation": "no_followup",
        },
        "spend": lambda: (
            calls.__setitem__("spend", calls["spend"] + 1)
            or {"ok": True, "appended": True}
        ),
        "terminal_closeout": terminal_closeout,
        "scheduler": lambda _spend: (
            calls.__setitem__("scheduler", calls["scheduler"] + 1)
            or {"completed": True, "acknowledged": True}
        ),
    }

    failed = run_loopx_turn_once(plan, **common)
    recovered = run_loopx_turn_once(plan, retry_failed=True, **common)

    assert failed["result_kind"] == "terminal_closeout_failed"
    assert failed["receipt"]["failed_phase"] == "terminal_closeout"
    assert failed["effects"]["quota_spent"] is True
    assert recovered["status"] == "committed"
    assert calls == {
        "writeback": 1,
        "spend": 1,
        "terminal_attempt": 2,
        "terminal_mutation": 1,
        "scheduler": 1,
    }


def test_validated_completion_recovers_after_writeback_without_repeating_completion(
    tmp_path: Path,
) -> None:
    plan = _plan()
    calls = {"completion": 0, "spend": 0, "scheduler": 0}

    def completion_writeback(_result: dict[str, object]) -> dict[str, object]:
        calls["completion"] += 1
        return {
            "ok": True,
            "appended": True,
            "completion": {
                "todo_id": "todo_fixture0001",
                "continuation": "active_goal",
            },
        }

    def interrupted_spend() -> dict[str, object]:
        calls["spend"] += 1
        raise SystemExit(8)

    def scheduler(_spend: dict[str, object]) -> dict[str, object]:
        calls["scheduler"] += 1
        return {"completed": True, "acknowledged": True}

    def healthy_spend() -> dict[str, object]:
        calls["spend"] += 1
        return {"ok": True, "appended": True}

    common = {
        "host_runner": lambda _request: _host_result(
            plan,
            kind="validated_completion",
        ),
        "project": tmp_path,
        "runtime_root": tmp_path / "runtime",
        "goal_id": "fixture-goal",
        "timeout_seconds": 5,
        "execute": True,
        "task_validator": _passing_validator,
        "writeback": lambda _result: {"ok": True, "appended": True},
        "completion_writeback": completion_writeback,
        "completion_intent": lambda _result: {
            "todo_id": "todo_fixture0001",
            "continuation": "active_goal",
        },
        "terminal_closeout": lambda _result: {
            "ok": True,
            "appended": True,
            "completion": {
                "todo_id": "todo_fixture0001",
                "continuation": "no_followup",
            },
        },
        "scheduler": scheduler,
    }
    with pytest.raises(SystemExit):
        run_loopx_turn_once(plan, spend=interrupted_spend, **common)

    recovered = run_loopx_turn_once(
        plan,
        spend=healthy_spend,
        spend_resolver=lambda _effect_ref: {"kind": "absent"},
        **common,
    )

    assert recovered["status"] == "committed"
    assert calls == {"completion": 1, "spend": 2, "scheduler": 1}


def test_run_once_recovers_after_process_exit_before_writeback(tmp_path: Path) -> None:
    plan = _plan()
    result_path = tmp_path / "result.json"
    result_path.write_text(json.dumps(_host_result(plan)), encoding="utf-8")
    count_path = tmp_path / "host-count"
    calls = {"writeback": 0, "spend": 0, "scheduler": 0}
    healthy_writeback, spend, scheduler = _callbacks(calls)

    def interrupted_writeback(_result: dict[str, object]) -> dict[str, object]:
        raise SystemExit(7)

    common = {
        "host_argv": _host_argv(result_path, count_path),
        "project": tmp_path,
        "runtime_root": tmp_path / "runtime",
        "goal_id": "fixture-goal",
        "timeout_seconds": 5,
        "execute": True,
        "task_validator": _passing_validator,
        "spend": spend,
        "scheduler": scheduler,
    }
    with pytest.raises(SystemExit):
        run_loopx_turn_once(plan, writeback=interrupted_writeback, **common)

    interrupted = _journal(tmp_path / "runtime")
    assert interrupted["completed_phases"] == [
        "host_execute",
        "typed_result",
        "validation",
    ]
    assert interrupted["task_validation"]["status"] == "passed"
    assert "writeback" not in interrupted

    recovered = run_loopx_turn_once(
        plan,
        writeback=healthy_writeback,
        writeback_resolver=lambda _effect_ref: {"kind": "absent"},
        **common,
    )

    assert recovered["status"] == "committed"
    assert count_path.read_text(encoding="utf-8") == "1"
    assert calls == {"writeback": 1, "spend": 1, "scheduler": 1}


@pytest.mark.parametrize(
    ("drift_kind", "expected_violations"),
    (
        (
            "goal_and_agent",
            ["goal_mismatch", "owner_mismatch", "journal_not_terminal"],
        ),
        (
            "todo_binding",
            ["settlement_binding_mismatch", "journal_not_terminal"],
        ),
    ),
)
def test_run_once_blocks_drifted_settlement_identity_before_any_provider(
    tmp_path: Path,
    drift_kind: str,
    expected_violations: list[str],
) -> None:
    plan = _plan()
    transaction = plan["transaction"]
    assert isinstance(transaction, dict)
    settlement_plan = transaction["settlement_plan"]
    assert isinstance(settlement_plan, dict)
    identity = settlement_plan["identity"]
    assert isinstance(identity, dict)
    if drift_kind == "goal_and_agent":
        identity["goal_id"] = "other-goal"
        identity["agent_id"] = "other-agent"
    else:
        identity["todo_id"] = "todo_other0002"
    identity["effect_id"] = ":".join(
        (
            str(identity["goal_id"]),
            str(identity["agent_id"]),
            str(identity["todo_id"]),
            str(identity["turn_instance_id"]),
        )
    )
    turn_key = str(transaction["turn_key"])
    runtime_root = tmp_path / "runtime"
    journal_path = turn_executor.turn_journal_path(
        runtime_root,
        goal_id="fixture-goal",
        turn_key=turn_key,
    )
    journal_path.parent.mkdir(parents=True)
    journal_path.write_text(
        json.dumps(
            {
                "schema_version": LOOPX_TURN_JOURNAL_SCHEMA_VERSION,
                "goal_id": "fixture-goal",
                "turn_key": turn_key,
                "status": "in_progress",
                "completed_phases": list(TRANSACTION_PHASES[:3]),
                "plan": plan,
                "host_result": _host_result(plan),
                "task_validation": _passing_validator(plan, _host_result(plan)),
            }
        ),
        encoding="utf-8",
    )
    loaded_plan = load_loopx_turn_plan_from_journal(
        runtime_root,
        goal_id="fixture-goal",
        turn_key=turn_key,
    )
    inspected = turn_executor.inspect_loopx_turn_journal(
        runtime_root,
        goal_id="fixture-goal",
        agent_id="codex-fixture",
        turn_key=turn_key,
    )
    assert inspected["violations"] == expected_violations
    assert inspected["journal_consistent"] is False
    assert inspected["recovery_decision"]["action"] == "blocked"

    calls = {
        "host": 0,
        "validation": 0,
        "writeback": 0,
        "writeback_readback": 0,
        "spend": 0,
        "spend_readback": 0,
        "scheduler": 0,
        "terminal_closeout": 0,
        "terminal_closeout_readback": 0,
    }

    def unexpected_provider(name: str):
        def invoke(*_args: object, **_kwargs: object) -> dict[str, object]:
            calls[name] += 1
            raise AssertionError(f"{name} provider must not run")

        return invoke

    with pytest.raises(TurnRecoveryBlockedError) as exc_info:
        run_loopx_turn_once(
            loaded_plan,
            host_runner=unexpected_provider("host"),
            project=tmp_path,
            runtime_root=runtime_root,
            goal_id="fixture-goal",
            timeout_seconds=5,
            execute=True,
            task_validator=unexpected_provider("validation"),
            writeback=unexpected_provider("writeback"),
            writeback_resolver=unexpected_provider("writeback_readback"),
            spend=unexpected_provider("spend"),
            spend_resolver=unexpected_provider("spend_readback"),
            scheduler=unexpected_provider("scheduler"),
            terminal_closeout=unexpected_provider("terminal_closeout"),
            terminal_closeout_resolver=unexpected_provider(
                "terminal_closeout_readback"
            ),
        )

    assert exc_info.value.decision == inspected["recovery_decision"]
    assert all(count == 0 for count in calls.values())


def test_run_once_recovers_saved_host_result_from_validation_without_host_retry(
    tmp_path: Path,
) -> None:
    plan = _plan()
    calls = {"host": 0, "validation": 0, "writeback": 0, "spend": 0, "scheduler": 0}
    writeback, spend, scheduler = _callbacks(calls)

    def host(_request: dict[str, object]) -> dict[str, object]:
        calls["host"] += 1
        return _host_result(plan)

    def interrupted_validation(
        _plan: dict[str, object],
        _result: dict[str, object],
    ) -> dict[str, object]:
        calls["validation"] += 1
        raise SystemExit(9)

    common = {
        "host_runner": host,
        "project": tmp_path,
        "runtime_root": tmp_path / "runtime",
        "goal_id": "fixture-goal",
        "timeout_seconds": 5,
        "execute": True,
        "writeback": writeback,
        "spend": spend,
        "scheduler": scheduler,
    }
    with pytest.raises(SystemExit):
        run_loopx_turn_once(
            plan,
            task_validator=interrupted_validation,
            **common,
        )

    transaction = plan["transaction"]
    assert isinstance(transaction, dict)
    inspected = turn_executor.inspect_loopx_turn_journal(
        tmp_path / "runtime",
        goal_id="fixture-goal",
        agent_id="codex-fixture",
        turn_key=str(transaction["turn_key"]),
    )
    assert inspected["replay_legal"] is False
    assert inspected["recovery_decision"]["can_continue"] is True
    assert inspected["recovery_decision"]["resume_from"] == "validation"
    assert inspected["recovery_decision"]["reinvoke_host"] is False

    recovered = run_loopx_turn_once(
        plan,
        task_validator=_passing_validator,
        **common,
    )

    assert recovered["status"] == "committed"
    assert recovered["recovery"]["planned"] == inspected["recovery_decision"]
    assert recovered["recovery"]["actual"] == {
        "status": "finished",
        "journal_status": "committed",
        "completed_phases": list(TRANSACTION_PHASES),
        "host_invoked": False,
    }
    audited = turn_executor.inspect_loopx_turn_journal(
        tmp_path / "runtime",
        goal_id="fixture-goal",
        agent_id="codex-fixture",
        turn_key=str(transaction["turn_key"]),
    )
    assert audited["last_recovery"] == recovered["recovery"]
    assert calls == {
        "host": 1,
        "validation": 1,
        "writeback": 1,
        "spend": 1,
        "scheduler": 1,
    }


def test_run_once_resumes_after_writeback_without_duplicate_effects(
    tmp_path: Path,
) -> None:
    plan = _plan()
    result_path = tmp_path / "result.json"
    result_path.write_text(json.dumps(_host_result(plan)), encoding="utf-8")
    count_path = tmp_path / "host-count"
    calls = {"writeback": 0, "spend": 0, "scheduler": 0}
    writeback, healthy_spend, scheduler = _callbacks(calls)

    def interrupted_spend() -> dict[str, object]:
        raise SystemExit(8)

    common = {
        "host_argv": _host_argv(result_path, count_path),
        "project": tmp_path,
        "runtime_root": tmp_path / "runtime",
        "goal_id": "fixture-goal",
        "timeout_seconds": 5,
        "execute": True,
        "task_validator": _passing_validator,
        "writeback": writeback,
        "scheduler": scheduler,
    }
    with pytest.raises(SystemExit):
        run_loopx_turn_once(plan, spend=interrupted_spend, **common)

    interrupted = _journal(tmp_path / "runtime")
    assert interrupted["completed_phases"] == [
        "host_execute",
        "typed_result",
        "validation",
        "durable_writeback",
    ]
    assert interrupted["writeback"]["appended"] is True
    assert "quota_spend" not in interrupted

    transaction = plan["transaction"]
    assert isinstance(transaction, dict)
    resumed_plan = load_loopx_turn_plan_from_journal(
        tmp_path / "runtime",
        goal_id="fixture-goal",
        turn_key=str(transaction["turn_key"]),
    )
    recovered = run_loopx_turn_once(
        resumed_plan,
        spend=healthy_spend,
        spend_resolver=lambda _effect_ref: {"kind": "absent"},
        **common,
    )

    assert recovered["status"] == "committed"
    assert count_path.read_text(encoding="utf-8") == "1"
    assert calls == {"writeback": 1, "spend": 1, "scheduler": 1}


def test_cancellation_before_writeback_preserves_prefix_and_resumes(
    tmp_path: Path,
) -> None:
    plan = _plan()
    calls = {"host": 0, "writeback": 0, "spend": 0}

    def host(_request: dict[str, object]) -> dict[str, object]:
        calls["host"] += 1
        return _host_result(plan)

    def cancelled_writeback(_result: dict[str, object]) -> dict[str, object]:
        calls["writeback"] += 1
        raise KeyboardInterrupt

    common = {
        "host_runner": host,
        "project": tmp_path,
        "runtime_root": tmp_path / "runtime",
        "goal_id": "fixture-goal",
        "timeout_seconds": 5,
        "execute": True,
        "task_validator": _passing_validator,
        "spend": lambda: (
            calls.__setitem__("spend", calls["spend"] + 1)
            or {"ok": True, "appended": True}
        ),
        "scheduler": lambda _spend: {"completed": True, "acknowledged": True},
    }
    with pytest.raises(KeyboardInterrupt):
        run_loopx_turn_once(plan, writeback=cancelled_writeback, **common)

    assert _journal(tmp_path / "runtime")["completed_phases"] == [
        "host_execute",
        "typed_result",
        "validation",
    ]
    recovered = run_loopx_turn_once(
        plan,
        writeback=lambda _result: (
            calls.__setitem__("writeback", calls["writeback"] + 1)
            or {"ok": True, "appended": True}
        ),
        writeback_resolver=lambda _effect_ref: {"kind": "absent"},
        **common,
    )

    assert recovered["status"] == "committed"
    assert calls == {"host": 1, "writeback": 2, "spend": 1}


def test_cancellation_during_scheduler_preserves_settlement_and_resumes(
    tmp_path: Path,
) -> None:
    plan = _plan()
    calls = {"host": 0, "writeback": 0, "spend": 0, "scheduler": 0}

    def host(_request: dict[str, object]) -> dict[str, object]:
        calls["host"] += 1
        return _host_result(plan)

    def writeback(_result: dict[str, object]) -> dict[str, object]:
        calls["writeback"] += 1
        return {"ok": True, "appended": True}

    def spend() -> dict[str, object]:
        calls["spend"] += 1
        return {"ok": True, "appended": True}

    def cancelled_scheduler(_spend: dict[str, object]) -> dict[str, object]:
        calls["scheduler"] += 1
        raise KeyboardInterrupt

    common = {
        "host_runner": host,
        "project": tmp_path,
        "runtime_root": tmp_path / "runtime",
        "goal_id": "fixture-goal",
        "timeout_seconds": 5,
        "execute": True,
        "task_validator": _passing_validator,
        "writeback": writeback,
        "spend": spend,
    }
    with pytest.raises(KeyboardInterrupt):
        run_loopx_turn_once(plan, scheduler=cancelled_scheduler, **common)

    interrupted_journal = _journal(tmp_path / "runtime")
    assert interrupted_journal["completed_phases"] == [
        "host_execute",
        "typed_result",
        "validation",
        "durable_writeback",
        "quota_spend",
    ]
    assert [
        receipt["step_kind"]
        for receipt in interrupted_journal["settlement_result"]["receipts"]
    ] == ["validation", "durable_writeback", "quota_spend"]

    def healthy_scheduler(_spend: dict[str, object]) -> dict[str, object]:
        calls["scheduler"] += 1
        return {"completed": True, "acknowledged": True}

    recovered = run_loopx_turn_once(plan, scheduler=healthy_scheduler, **common)

    assert recovered["status"] == "committed"
    assert calls == {"host": 1, "writeback": 1, "spend": 1, "scheduler": 2}


def test_permission_denial_from_host_is_typed_and_explicitly_retried(
    tmp_path: Path,
) -> None:
    plan = _plan()
    calls = {"host": 0, "writeback": 0, "spend": 0}

    def host(_request: dict[str, object]) -> dict[str, object]:
        calls["host"] += 1
        if calls["host"] == 1:
            raise PermissionError("host denied")
        return _host_result(plan)

    common = {
        "host_runner": host,
        "project": tmp_path,
        "runtime_root": tmp_path / "runtime",
        "goal_id": "fixture-goal",
        "timeout_seconds": 5,
        "execute": True,
        "task_validator": _passing_validator,
        "writeback": lambda _result: (
            calls.__setitem__("writeback", calls["writeback"] + 1)
            or {"ok": True, "appended": True}
        ),
        "spend": lambda: (
            calls.__setitem__("spend", calls["spend"] + 1)
            or {"ok": True, "appended": True}
        ),
        "scheduler": lambda _spend: {"completed": True, "acknowledged": True},
    }
    failed = run_loopx_turn_once(plan, **common)
    replayed = run_loopx_turn_once(plan, **common)
    recovered = run_loopx_turn_once(plan, retry_failed=True, **common)

    assert failed["result_kind"] == "host_failure"
    assert failed["receipt"]["failed_phase"] == "host_execute"
    assert failed["reason"] == "PermissionError"
    assert replayed["replayed"] is True
    assert recovered["status"] == "committed"
    assert calls == {"host": 2, "writeback": 1, "spend": 1}


def test_permission_denial_during_spend_preserves_writeback_and_resumes(
    tmp_path: Path,
) -> None:
    plan = _plan()
    calls = {"host": 0, "writeback": 0, "spend": 0}

    def host(_request: dict[str, object]) -> dict[str, object]:
        calls["host"] += 1
        return _host_result(plan)

    def writeback(_result: dict[str, object]) -> dict[str, object]:
        calls["writeback"] += 1
        return {"ok": True, "appended": True}

    def denied_spend() -> dict[str, object]:
        calls["spend"] += 1
        raise PermissionError("quota ledger denied")

    common = {
        "host_runner": host,
        "project": tmp_path,
        "runtime_root": tmp_path / "runtime",
        "goal_id": "fixture-goal",
        "timeout_seconds": 5,
        "execute": True,
        "task_validator": _passing_validator,
        "writeback": writeback,
        "scheduler": lambda _spend: {"completed": True, "acknowledged": True},
    }
    with pytest.raises(PermissionError):
        run_loopx_turn_once(plan, spend=denied_spend, **common)

    assert _journal(tmp_path / "runtime")["completed_phases"] == [
        "host_execute",
        "typed_result",
        "validation",
        "durable_writeback",
    ]
    recovered = run_loopx_turn_once(
        plan,
        spend=lambda: (
            calls.__setitem__("spend", calls["spend"] + 1)
            or {"ok": True, "appended": True}
        ),
        spend_resolver=lambda _effect_ref: {"kind": "absent"},
        **common,
    )

    assert recovered["status"] == "committed"
    assert calls == {"host": 1, "writeback": 1, "spend": 2}


def test_budget_rejection_is_typed_and_retry_does_not_repeat_writeback(
    tmp_path: Path,
) -> None:
    plan = _plan()
    calls = {"host": 0, "writeback": 0, "spend": 0}

    def host(_request: dict[str, object]) -> dict[str, object]:
        calls["host"] += 1
        return _host_result(plan)

    def writeback(_result: dict[str, object]) -> dict[str, object]:
        calls["writeback"] += 1
        return {"ok": True, "appended": True}

    def reject_budget() -> dict[str, object]:
        calls["spend"] += 1
        return {
            "ok": False,
            "appended": False,
            "reason": "quota budget rejected",
        }

    common = {
        "host_runner": host,
        "project": tmp_path,
        "runtime_root": tmp_path / "runtime",
        "goal_id": "fixture-goal",
        "timeout_seconds": 5,
        "execute": True,
        "task_validator": _passing_validator,
        "writeback": writeback,
        "scheduler": lambda _spend: {"completed": True, "acknowledged": True},
    }
    failed = run_loopx_turn_once(plan, spend=reject_budget, **common)

    assert failed["result_kind"] == "quota_spend_failed"
    assert failed["receipt"]["result_kind"] == "quota_spend_failed"
    assert failed["receipt"]["failed_phase"] == "quota_spend"
    assert failed["settlement_result"]["failure"]["kind"] == "budget_rejected"
    assert [
        receipt["step_kind"] for receipt in failed["settlement_result"]["receipts"]
    ] == ["validation", "durable_writeback"]
    assert failed["effects"]["state_written"] is True
    assert failed["effects"]["quota_spent"] is False

    recovered = run_loopx_turn_once(
        plan,
        spend=lambda: (
            calls.__setitem__("spend", calls["spend"] + 1)
            or {"ok": True, "appended": True}
        ),
        retry_failed=True,
        **common,
    )

    assert recovered["status"] == "committed"
    assert calls == {"host": 1, "writeback": 1, "spend": 2}


def test_run_once_fails_closed_without_independent_task_validator(
    tmp_path: Path,
) -> None:
    plan = _plan()
    calls = {"writeback": 0, "spend": 0, "scheduler": 0}
    writeback, spend, scheduler = _callbacks(calls)

    payload = run_loopx_turn_once(
        plan,
        host_runner=lambda _request: _host_result(plan),
        project=tmp_path,
        runtime_root=tmp_path / "runtime",
        goal_id="fixture-goal",
        timeout_seconds=5,
        execute=True,
        writeback=writeback,
        spend=spend,
        scheduler=scheduler,
    )

    assert payload["ok"] is False
    assert payload["status"] == "failed"
    assert payload["result_kind"] == "validation_failed"
    assert payload["validation"]["status"] == "unavailable"
    assert payload["validation"]["recovery_kind"] == "repair_required"
    assert payload["receipt"]["failed_phase"] == "validation"
    assert payload["receipt"]["completed_phases"] == ["host_execute", "typed_result"]
    assert calls == {"writeback": 0, "spend": 0, "scheduler": 0}


def test_run_once_retries_task_validation_without_reinvoking_host(
    tmp_path: Path,
) -> None:
    plan = _plan()
    calls = {"host": 0, "writeback": 0, "spend": 0, "scheduler": 0}
    writeback, spend, scheduler = _callbacks(calls)

    def host(_request: dict[str, object]) -> dict[str, object]:
        calls["host"] += 1
        return _host_result(plan)

    def reject(
        _plan: dict[str, object],
        _result: dict[str, object],
    ) -> dict[str, object]:
        return {
            "status": "failed",
            "validator_kind": "fixture",
            "summary": "independent fixture postcondition is absent",
            "recovery_kind": "replan_required",
        }

    common = {
        "host_runner": host,
        "project": tmp_path,
        "runtime_root": tmp_path / "runtime",
        "goal_id": "fixture-goal",
        "timeout_seconds": 5,
        "execute": True,
        "writeback": writeback,
        "spend": spend,
        "scheduler": scheduler,
    }
    failed = run_loopx_turn_once(plan, task_validator=reject, **common)
    recovered = run_loopx_turn_once(
        plan,
        task_validator=_passing_validator,
        retry_failed=True,
        **common,
    )

    assert failed["result_kind"] == "validation_failed"
    assert failed["validation"]["recovery_kind"] == "replan_required"
    assert recovered["status"] == "committed"
    assert recovered["effects"]["host_invoked"] is False
    assert calls == {"host": 1, "writeback": 1, "spend": 1, "scheduler": 1}


def test_material_result_cannot_use_not_required_validation_receipt(
    tmp_path: Path,
) -> None:
    plan = _plan()
    calls = {"writeback": 0, "spend": 0, "scheduler": 0}
    writeback, spend, scheduler = _callbacks(calls)

    payload = run_loopx_turn_once(
        plan,
        host_runner=lambda _request: _host_result(plan),
        project=tmp_path,
        runtime_root=tmp_path / "runtime",
        goal_id="fixture-goal",
        timeout_seconds=5,
        execute=True,
        task_validator=lambda _plan, _result: {
            "status": "not_required",
            "validator_kind": "fixture",
            "summary": "skip validation",
        },
        writeback=writeback,
        spend=spend,
        scheduler=scheduler,
    )

    assert payload["result_kind"] == "validation_failed"
    assert payload["validation"]["status"] == "inconclusive"
    assert "cannot skip" in payload["validation"]["summary"]
    assert calls == {"writeback": 0, "spend": 0, "scheduler": 0}


@pytest.mark.parametrize("result_kind", ["wait", "iteration_failed"])
def test_run_once_stops_without_writeback_or_spend(
    tmp_path: Path, result_kind: str
) -> None:
    plan = _plan()
    result_path = tmp_path / "result.json"
    result_path.write_text(
        json.dumps(_host_result(plan, kind=result_kind)), encoding="utf-8"
    )
    calls = {"writeback": 0, "spend": 0, "scheduler": 0}
    writeback, spend, scheduler = _callbacks(calls)

    payload = run_loopx_turn_once(
        plan,
        host_argv=_host_argv(result_path, tmp_path / "host-count"),
        project=tmp_path,
        runtime_root=tmp_path / "runtime",
        goal_id="fixture-goal",
        timeout_seconds=5,
        execute=True,
        writeback=writeback,
        spend=spend,
        scheduler=scheduler,
    )

    assert payload["ok"] is True
    assert payload["status"] == "stopped"
    assert payload["receipt"]["status"] == "stopped"
    assert calls == {"writeback": 0, "spend": 0, "scheduler": 0}


def test_run_once_projects_scheduler_action_without_false_ack(tmp_path: Path) -> None:
    plan = _plan()
    result_path = tmp_path / "result.json"
    result_path.write_text(json.dumps(_host_result(plan)), encoding="utf-8")
    calls = {"writeback": 0, "spend": 0, "scheduler": 0}
    writeback, spend, _scheduler = _callbacks(calls)

    def scheduler(_spend: dict[str, object]) -> dict[str, object]:
        calls["scheduler"] += 1
        return {
            "completed": False,
            "apply_needed": True,
            "disposition": "host_action_required",
        }

    payload = run_loopx_turn_once(
        plan,
        host_argv=_host_argv(result_path, tmp_path / "host-count"),
        project=tmp_path,
        runtime_root=tmp_path / "runtime",
        goal_id="fixture-goal",
        timeout_seconds=5,
        execute=True,
        task_validator=_passing_validator,
        writeback=writeback,
        spend=spend,
        scheduler=scheduler,
    )

    assert payload["ok"] is True
    assert payload["status"] == "scheduler_action_required"
    assert payload["receipt"]["next_phase"] == "scheduler_apply"
    assert payload["effects"]["scheduler_acknowledged"] is False


def test_run_once_resumes_scheduler_without_repeating_committed_effects(
    tmp_path: Path,
) -> None:
    plan = _plan()
    result_path = tmp_path / "result.json"
    result_path.write_text(json.dumps(_host_result(plan)), encoding="utf-8")
    count_path = tmp_path / "host-count"
    calls = {"writeback": 0, "spend": 0, "scheduler": 0}
    writeback, spend, _scheduler = _callbacks(calls)

    def scheduler(_spend: dict[str, object]) -> dict[str, object]:
        calls["scheduler"] += 1
        if calls["scheduler"] == 1:
            return {
                "completed": False,
                "apply_needed": True,
                "disposition": "host_action_required",
            }
        return {
            "completed": True,
            "acknowledged": True,
            "disposition": "applied_and_acknowledged",
        }

    kwargs = {
        "host_argv": _host_argv(result_path, count_path),
        "project": tmp_path,
        "runtime_root": tmp_path / "runtime",
        "goal_id": "fixture-goal",
        "timeout_seconds": 5,
        "execute": True,
        "task_validator": _passing_validator,
        "writeback": writeback,
        "spend": spend,
        "scheduler": scheduler,
    }
    first = run_loopx_turn_once(plan, **kwargs)
    transaction = plan["transaction"]
    assert isinstance(transaction, dict)
    inspected = turn_executor.inspect_loopx_turn_journal(
        tmp_path / "runtime",
        goal_id="fixture-goal",
        agent_id="codex-fixture",
        turn_key=str(transaction["turn_key"]),
    )
    resumed = run_loopx_turn_once(plan, **kwargs)

    assert first["status"] == "scheduler_action_required"
    assert inspected["replay_legal"] is False
    assert inspected["recovery_decision"]["resume_from"] == "scheduler_apply"
    assert inspected["recovery_decision"]["reinvoke_host"] is False
    assert resumed["recovery"]["planned"] == inspected["recovery_decision"]
    assert resumed["status"] == "committed"
    assert resumed["effects"]["scheduler_acknowledged"] is True
    assert count_path.read_text(encoding="utf-8") == "1"
    assert calls == {"writeback": 1, "spend": 1, "scheduler": 2}
    audited = turn_executor.inspect_loopx_turn_journal(
        tmp_path / "runtime",
        goal_id="fixture-goal",
        agent_id="codex-fixture",
        turn_key=str(transaction["turn_key"]),
    )
    assert audited["last_recovery"] == resumed["recovery"]


def test_run_once_fails_closed_when_the_managed_executor_cannot_launch(tmp_path):
    plan = _managed_plan(runtime_available=False)
    transaction = plan["transaction"]
    assert isinstance(transaction, dict)
    runtime_root = tmp_path / "runtime"
    journal = turn_journal_path(
        runtime_root,
        goal_id="fixture-goal",
        turn_key=str(transaction["turn_key"]),
    )

    payload = run_loopx_turn_once(
        plan,
        host_runner=lambda _request: pytest.fail(
            "an unavailable executor must not run"
        ),
        project=tmp_path,
        runtime_root=runtime_root,
        goal_id="fixture-goal",
        timeout_seconds=5,
        execute=True,
    )

    assert payload["ok"] is False
    assert payload["status"] == "unavailable"
    assert payload["reason"] == "dsh_runtime_unavailable"
    assert payload["effects"] == {
        "host_invoked": False,
        "state_written": False,
        "quota_spent": False,
        "scheduler_acknowledged": False,
    }
    assert payload["quota_slot_spend_count"] == 0
    assert payload["managed_executor"] == plan["managed_executor"]
    assert journal.exists() is False


def test_run_once_preview_reports_the_managed_executor_without_refusing(tmp_path):
    plan = _managed_plan(runtime_available=False)

    payload = run_loopx_turn_once(
        plan,
        host_runner=lambda _request: pytest.fail("preview must not run the host"),
        project=tmp_path,
        runtime_root=tmp_path / "runtime",
        goal_id="fixture-goal",
        timeout_seconds=5,
        execute=False,
    )

    assert payload["ok"] is True
    assert payload["status"] == "preview"
    assert payload["managed_executor"]["available"] is False
    assert (
        payload["managed_executor"]["unavailable_reason"] == "dsh_runtime_unavailable"
    )


def test_run_once_does_not_refuse_a_launchable_managed_executor(tmp_path):
    plan = _managed_plan(runtime_available=True)

    # The refusal is the only guard under test here: without writeback, spend,
    # and scheduler callbacks the executor stops at its own contract instead.
    with pytest.raises(ValueError, match="requires writeback, spend, and scheduler"):
        run_loopx_turn_once(
            plan,
            host_runner=lambda _request: pytest.fail(
                "host must not run without callbacks"
            ),
            project=tmp_path,
            runtime_root=tmp_path / "runtime",
            goal_id="fixture-goal",
            timeout_seconds=5,
            execute=True,
        )


def test_real_host_duration_is_observed_but_settlement_replay_is_not(
    tmp_path, monkeypatch
):
    """Production Turn entrypoint, actual Host subprocess and detached TS state."""
    import time
    from loopx import usage_ping

    machine = tmp_path / "machine"
    monkeypatch.setattr(usage_ping, "select_default_runtime_root", lambda: machine)
    for name in ("CI", "DO_NOT_TRACK", "LOOPX_USAGE_PING", "LOOPX_USAGE_POLICY"):
        monkeypatch.delenv(name, raising=False)
    monkeypatch.setenv("LOOPX_USAGE_PING_ENDPOINT", "http://127.0.0.1:1/v1/ping")
    usage_ping.control("enable")
    plan = _plan()
    host_file = tmp_path / "host.py"
    host_file.write_text(
        "import json, sys, time\njson.load(sys.stdin)\ntime.sleep(0.08)\nprint("
        + repr(json.dumps(_host_result(plan)))
        + ")\n"
    )
    calls = {"writeback": 0, "spend": 0, "scheduler": 0}
    writeback, spend, scheduler = _callbacks(calls)
    options = dict(
        host_argv=[sys.executable, str(host_file)],
        project=tmp_path,
        runtime_root=tmp_path / "runtime",
        goal_id="fixture-goal",
        timeout_seconds=5,
        execute=True,
        task_validator=_passing_validator,
        writeback=writeback,
        spend=spend,
        scheduler=scheduler,
    )
    result = run_loopx_turn_once(plan, **options)
    assert result["ok"], result
    local = machine / "usage-ping.json.goals"
    deadline = time.monotonic() + 5
    while not local.exists() and time.monotonic() < deadline:
        time.sleep(0.02)
    before = json.loads(local.read_text())
    intervals = before["goals"][0]["intervals"]
    assert sum(end - start for start, end in intervals) >= 80
    assert "fixture-goal" not in local.read_text()
    assert run_loopx_turn_once(plan, **options)["ok"]
    time.sleep(0.15)
    assert json.loads(local.read_text()) == before
    assert calls == {"writeback": 1, "spend": 1, "scheduler": 1}