from pathlib import Path
import json
import subprocess
import sys


ROOT = Path(__file__).resolve().parents[1]


SCENARIOS = [
    {
        "id": "B01",
        "name": "single_agnet_failure_isolation",
        "claim": "single Agnet failure does not prevent result convergence",
        "why": "Robustness is a core swarm property: one failed individual should not collapse the group result.",
        "test": "tests.test_swarm_behavior_academic.SwarmBehaviorAcademicTest.test_single_agnet_failure_isolated_by_redundant_convergence",
    },
    {
        "id": "B02",
        "name": "emergent_consensus",
        "claim": "group consensus can beat any single weak local signal",
        "why": "Emergence means global behavior appears from local evidence and shared environment updates.",
        "test": "tests.test_swarm_behavior_academic.SwarmBehaviorAcademicTest.test_emergent_consensus_accumulates_local_evidence",
    },
    {
        "id": "B03",
        "name": "pheromone_stigmergy",
        "claim": "pheromone state biases work selection and records positive feedback",
        "why": "Stigmergy is the indirect coordination mechanism that distinguishes a swarm from a plain chain.",
        "test": "tests.test_swarm_behavior_academic.SwarmBehaviorAcademicTest.test_pheromone_biases_claim_order_and_records_positive_feedback",
    },
    {
        "id": "B04",
        "name": "handoff_continuity",
        "claim": "handoff preserves target agent, active-agent state, and context payload",
        "why": "LangGraph Swarm centers on dynamic control handoff between named specialized agents.",
        "test": "tests.test_swarm_behavior_academic.SwarmBehaviorAcademicTest.test_langgraph_style_handoff_preserves_active_agent_and_payload",
    },
]


def main() -> None:
    results = []
    for scenario in SCENARIOS:
        command = [sys.executable, "-B", "-m", "unittest", scenario["test"]]
        completed = subprocess.run(
            command,
            cwd=ROOT,
            text=True,
            capture_output=True,
            timeout=120,
        )
        results.append(
            {
                "id": scenario["id"],
                "name": scenario["name"],
                "claim": scenario["claim"],
                "why": scenario["why"],
                "command": " ".join(command),
                "passed": completed.returncode == 0,
                "evidence": summarize_output(completed.stdout, completed.stderr),
            }
        )
        if completed.returncode != 0:
            break

    report = {
        "standard": "swarm-behavior-academic-v1",
        "status": "PASS" if len(results) == len(SCENARIOS) and all(item["passed"] for item in results) else "FAIL",
        "basis": [
            "Swarm claims: decentralized/self-organizing behavior, stigmergy, robustness, emergence, convergence.",
            "LangGraph Swarm claims: named agents, active-agent routing, and create_handoff_tool-style transfer.",
        ],
        "scenarios": results,
    }
    print(json.dumps(report, ensure_ascii=False, indent=2))
    if report["status"] != "PASS":
        raise SystemExit(1)


def summarize_output(stdout: str, stderr: str) -> dict[str, str]:
    combined = "\n".join(part.strip() for part in [stdout, stderr] if part.strip())
    return {"tail": combined[-1000:] if combined else "<no output>"}


if __name__ == "__main__":
    main()