import gradio as gr
from content_moderation import build_cm_tab
from jailbreak import build_jailbreak_tab
from shared import GLOBAL_CSS, build_footer


with gr.Blocks() as demo:
    gr.HTML(GLOBAL_CSS)

    gr.Markdown("# BELLS-Operational: Supervision Systems Benchmark")

    with gr.Column(elem_classes=["intro-section"]):
        gr.Markdown(
            """
            **BELLS-O (Benchmark for the Evaluation of LLM Supervision Systems – Operational)** is a benchmark for evaluating the operational performance of LLM supervision systems.

            It compares specialized guardrails and generalist LLMs acting as supervisors on misuse detection tasks — including input moderation, output moderation, and jailbreak detection — using four deployment-relevant metrics: Detection Rate, False Positive Rate, Latency, and Cost.            """
        )

    with gr.Tabs():
        with gr.Tab("Content Moderation - Input"):
            build_cm_tab("input")

        with gr.Tab("Content Moderation - Output"):
            build_cm_tab("output")

        with gr.Tab("Jailbreak"):
            build_jailbreak_tab()

    build_footer()

if __name__ == "__main__":
    demo.launch()