app.py

import html
import json
import os
from typing import List

import gradio as gr
import pandas as pd

from topic_pipeline import (
    OUTPUT_DIR,
    parse_notebooklm_tccm_text,
    run_complete_pipeline,
    write_tccm_dual_validation,
)


os.makedirs(OUTPUT_DIR, exist_ok=True)


def _exists(name: str) -> bool:
    return os.path.exists(os.path.join(OUTPUT_DIR, name))


def _load_json(name: str):
    with open(os.path.join(OUTPUT_DIR, name), "r", encoding="utf-8") as f:
        return json.load(f)


def _download_files() -> List[str]:
    names = [
        "comparison.csv",
        "taxonomy_map.json",
        "topic_model_report.md",
        "narrative.txt",
        "cluster_optimization_log.csv",
        "llm_council_validation.csv",
        "tccm_validation.csv",
        "tccm_dual_validation.csv",
        "notebooklm_extraction.csv",
        "compliance_checklist.csv",
        "compliance_checklist.json",
        "run_metadata.json",
        "combined_labels.json",
    ]
    return [os.path.join(OUTPUT_DIR, name) for name in names if _exists(name)]


def _phase_html() -> str:
    phases = [
        ("Corpus", _exists("corpus_config.json")),
        ("Embeddings", _exists("combined_emb.npy")),
        ("Optimization", _exists("cluster_optimization_log.csv")),
        ("Clusters", _exists("combined_labels.json")),
        ("Council", _exists("llm_council_validation.csv")),
        ("TCCM", _exists("tccm_validation.csv")),
        ("Compliance", _exists("compliance_checklist.csv")),
        ("Report", _exists("topic_model_report.md")),
    ]
    chips = []
    for name, done in phases:
        bg = "#0f766e" if done else "#334155"
        mark = "OK" if done else "--"
        chips.append(
            f"<span style='display:inline-flex;gap:6px;align-items:center;"
            f"padding:6px 12px;border-radius:999px;background:{bg};"
            f"color:white;font-size:12px;font-weight:700'>{mark} {name}</span>"
        )
    return "<div style='display:flex;gap:8px;flex-wrap:wrap'>" + "".join(chips) + "</div>"


def _cluster_table():
    if not _exists("combined_labels.json"):
        return []
    rows = []
    for s in _load_json("combined_labels.json"):
        rows.append([
            s.get("cluster_id"),
            s.get("label"),
            s.get("category"),
            s.get("paper_count"),
            s.get("confidence"),
            s.get("agreement_score"),
            "; ".join(s.get("keywords", [])[:8]),
            " | ".join(s.get("top_titles", [])[:3]),
            s.get("reasoning", ""),
        ])
    return rows


def _council_table():
    path = os.path.join(OUTPUT_DIR, "llm_council_validation.csv")
    if not os.path.exists(path):
        return []
    return pd.read_csv(path).head(120)


def _council_viz_html() -> str:
    path = os.path.join(OUTPUT_DIR, "llm_council_validation.csv")
    if not os.path.exists(path):
        return (
            "<div class='council-empty'>Run the pipeline to activate the LLM Council "
            "validation board.</div>"
        )

    df = pd.read_csv(path)
    if df.empty:
        return "<div class='council-empty'>Council validation file is empty.</div>"

    grouped = list(df.groupby(["cluster_id", "final_label"], sort=False))[:6]
    rows = []
    avg_agreement = float(df["agreement_score"].mean()) if "agreement_score" in df else 0
    avg_confidence = float(df["confidence"].mean()) if "confidence" in df else 0
    llm_member_present = df["member"].astype(str).str.contains("LLM|Mistral", case=False, regex=True).any()
    llm_status = "Mistral LLM active" if llm_member_present else "Local semantic fallback active"

    for (cluster_id, final_label), group in grouped:
        votes = []
        for _, row in group.iterrows():
            member = html.escape(str(row.get("member", "")))
            label = html.escape(str(row.get("member_label", "")))
            method = html.escape(str(row.get("method", "")))
            votes.append(
                "<div class='council-vote'>"
                "<div class='vote-dot'></div>"
                f"<div><strong>{member}</strong><span>{label}</span><small>{method}</small></div>"
                "</div>"
            )
        confidence = int(float(group["confidence"].iloc[0]) * 100)
        agreement = int(float(group["agreement_score"].iloc[0]) * 100)
        rows.append(
            "<div class='council-cluster'>"
            "<div class='cluster-head'>"
            f"<span>Cluster {html.escape(str(cluster_id))}</span>"
            f"<strong>{html.escape(str(final_label))}</strong>"
            "</div>"
            "<div class='council-flow'>"
            + "".join(votes) +
            "<div class='final-label'>"
            "<small>Accepted label</small>"
            f"<strong>{html.escape(str(final_label))}</strong>"
            f"<span>{confidence}% confidence | {agreement}% agreement</span>"
            "</div>"
            "</div>"
            "</div>"
        )

    return (
        "<div class='council-board'>"
        "<div class='council-top'>"
        "<div><h3>LLM Council Validation Running In-App</h3>"
        "<p>Three independent validators inspect each cluster label, compare votes, "
        "and write the accepted label plus agreement score into the export file.</p></div>"
        "<div class='council-metrics'>"
        f"<div><strong>{len(df['cluster_id'].unique())}</strong><span>clusters checked</span></div>"
        f"<div><strong>{int(avg_agreement * 100)}%</strong><span>avg agreement</span></div>"
        f"<div><strong>{int(avg_confidence * 100)}%</strong><span>avg confidence</span></div>"
        f"<div><strong>{html.escape(llm_status)}</strong><span>council mode</span></div>"
        "</div></div>"
        "<div class='council-lane'>"
        "<div class='pulse-node'>1<br><span>Keyword Extractor</span></div>"
        "<div class='pulse-line'></div>"
        "<div class='pulse-node'>2<br><span>PAJAIS Mapper</span></div>"
        "<div class='pulse-line'></div>"
        "<div class='pulse-node'>3<br><span>LLM / Semantic Judge</span></div>"
        "<div class='pulse-line'></div>"
        "<div class='pulse-node final'>OK<br><span>Validated Label</span></div>"
        "</div>"
        + "".join(rows) +
        "</div>"
    )


def _optimizer_table():
    path = os.path.join(OUTPUT_DIR, "cluster_optimization_log.csv")
    if not os.path.exists(path):
        return []
    df = pd.read_csv(path)
    cols = [
        c for c in [
            "algorithm",
            "umap_n_neighbors",
            "umap_n_components",
            "hdbscan_min_cluster_size",
            "hdbscan_min_samples",
            "n_clusters",
            "noise_ratio",
            "min_size",
            "max_size",
            "too_small",
            "too_large",
            "silhouette_cosine",
            "score",
            "optimizer_recommendation",
        ] if c in df.columns
    ]
    return df[cols].head(80)


def _tccm_table():
    path = os.path.join(OUTPUT_DIR, "tccm_validation.csv")
    if not os.path.exists(path):
        return []
    return pd.read_csv(path).head(100)


def _tccm_dual_table():
    path = os.path.join(OUTPUT_DIR, "tccm_dual_validation.csv")
    if not os.path.exists(path):
        return []
    return pd.read_csv(path).head(100)


def _compliance_table():
    path = os.path.join(OUTPUT_DIR, "compliance_checklist.csv")
    if not os.path.exists(path):
        return []
    return pd.read_csv(path)


def _compliance_html() -> str:
    path = os.path.join(OUTPUT_DIR, "compliance_checklist.csv")
    if not os.path.exists(path):
        return (
            "<div class='compliance-empty'>Run the pipeline to generate the professor-requirement "
            "compliance checklist.</div>"
        )
    df = pd.read_csv(path)
    color_map = {
        "PASS": "#0f766e",
        "FAIL": "#b91c1c",
        "CONFIG_REQUIRED": "#b45309",
        "ENV_FALLBACK": "#b45309",
        "INPUT_REQUIRED": "#b45309",
        "PARTIAL": "#7c3aed",
        "MANUAL_REQUIRED": "#475569",
        "REVIEW": "#7c3aed",
    }
    rows = []
    for _, row in df.iterrows():
        status = str(row.get("Status", "REVIEW"))
        color = color_map.get(status, "#475569")
        rows.append(
            "<div class='compliance-row'>"
            f"<span style='background:{color}'>{html.escape(status)}</span>"
            f"<strong>{html.escape(str(row.get('Requirement', '')))}</strong>"
            f"<p>{html.escape(str(row.get('Evidence', '')))}</p>"
            f"<small>{html.escape(str(row.get('File', '')))}</small>"
            "</div>"
        )
    return (
        "<div class='compliance-board'>"
        "<h3>Professor Requirement Compliance Checklist</h3>"
        "<p>This separates completed app evidence from items that still need API secrets, "
        "NotebookLM/full-text inputs, or mentor approval.</p>"
        "<div class='compliance-grid'>" + "".join(rows) + "</div></div>"
    )


def _tccm_dual_status_html() -> str:
    path = os.path.join(OUTPUT_DIR, "tccm_dual_validation.csv")
    if not os.path.exists(path):
        return (
            "<div class='compliance-empty'>Upload NotebookLM and second-LLM extraction CSVs "
            "to generate TCCM dual validation.</div>"
        )
    df = pd.read_csv(path)
    status_col = "Final_TCCM_Compliance_Status"
    if status_col not in df.columns:
        return "<div class='compliance-empty'>TCCM dual validation is pending source uploads.</div>"
    counts = df[status_col].value_counts().to_dict()
    cards = []
    for status, count in counts.items():
        ok = "COMPLIANT" in str(status)
        color = "#0f766e" if ok else "#b45309"
        cards.append(
            f"<div class='tccm-card'><strong style='color:{color}'>{count}</strong>"
            f"<span>{html.escape(str(status))}</span></div>"
        )
    return (
        "<div class='tccm-status'><h3>TCCM Dual Validation Status</h3>"
        "<p>Required by email: NotebookLM extraction plus another LLM/extraction method. "
        "This screen reconciles those files with regex/semantic extraction.</p>"
        "<div>" + "".join(cards) + "</div></div>"
    )


def _on_tccm_dual_validate(notebook_file, second_file):
    notebook_path = notebook_file if isinstance(notebook_file, str) else getattr(notebook_file, "name", "")
    second_path = second_file if isinstance(second_file, str) else getattr(second_file, "name", "")
    write_tccm_dual_validation(notebook_path, second_path)
    return _tccm_dual_status_html(), _tccm_dual_table(), _download_files()


def _on_notebooklm_paste(notebook_text):
    if not str(notebook_text or "").strip():
        return (
            "<div class='compliance-empty'>Paste the NotebookLM table text first.</div>",
            _tccm_dual_table(),
            _download_files(),
        )
    notebook_path = parse_notebooklm_tccm_text(notebook_text)
    write_tccm_dual_validation(notebook_path, "")
    count = len(pd.read_csv(notebook_path)) if os.path.exists(notebook_path) else 0
    status = (
        f"<div class='tccm-status'><h3>NotebookLM Paste Imported</h3>"
        f"<p>Parsed {count} NotebookLM rows into <code>outputs/notebooklm_extraction.csv</code>. "
        "Merged with the independent regex/semantic extractor in "
        "<code>outputs/tccm_dual_validation.csv</code>. Upload a second-LLM CSV as well "
        "for full NotebookLM + second LLM compliance.</p></div>"
        + _tccm_dual_status_html()
    )
    return status, _tccm_dual_table(), _download_files()


def _chart_iframe(name: str) -> str:
    path = os.path.join(OUTPUT_DIR, "combined_charts", name)
    if not os.path.exists(path):
        return (
            "<div style='height:320px;display:grid;place-items:center;"
            "background:#0f172a;color:#94a3b8;border-radius:8px'>"
            "Run the pipeline to generate this chart.</div>"
        )
    with open(path, "r", encoding="utf-8") as f:
        srcdoc = f.read().replace("&", "&amp;").replace('"', "&quot;")
    return (
        f"<iframe srcdoc=\"{srcdoc}\" width='100%' height='500' "
        "style='border:0;border-radius:8px;background:#0f172a'></iframe>"
    )


def _cards_html() -> str:
    if not _exists("combined_labels.json"):
        return (
            "<div style='padding:28px;color:#64748b;background:#f8fafc;"
            "border:1px solid #e2e8f0;border-radius:8px'>"
            "Clusters will appear here after a complete run.</div>"
        )
    cards = []
    for s in _load_json("combined_labels.json"):
        evidence = html.escape(" | ".join(s.get("top_titles", [])[:3]))
        label = html.escape(s.get("label", "Cluster"))
        category = html.escape(s.get("category", "Unmapped"))
        keywords = html.escape(", ".join(s.get("keywords", [])[:8]))
        conf = int(float(s.get("confidence", 0)) * 100)
        cards.append(
            "<div style='border:1px solid #d8dee9;border-left:4px solid #0f766e;"
            "border-radius:8px;padding:14px;background:white;min-height:170px'>"
            f"<div style='font-size:15px;font-weight:800;color:#0f172a'>{label}</div>"
            f"<div style='font-size:12px;color:#475569;margin-top:4px'>{category}</div>"
            f"<div style='margin-top:10px;font-size:12px;color:#334155'>"
            f"{s.get('paper_count', 0)} papers | confidence {conf}% | agreement {s.get('agreement_score', 0)}</div>"
            f"<div style='margin-top:8px;font-size:12px;color:#0f766e'>{keywords}</div>"
            f"<div style='margin-top:10px;font-size:12px;color:#64748b;line-height:1.45'>{evidence}</div>"
            "</div>"
        )
    return (
        "<div style='display:grid;grid-template-columns:repeat(auto-fit,minmax(280px,1fr));"
        "gap:12px'>" + "".join(cards) + "</div>"
    )


def _summary_markdown(result=None) -> str:
    if result is None and not _exists("run_metadata.json"):
        return (
            "Upload the Scopus CSV and click **Run Complete Pipeline**. "
            "The app will generate paper-level Title+Abstract+DOI embeddings, optimize "
            "UMAP/HDBSCAN clustering, label 15-25 clusters through an in-app council, "
            "map them to PAJAIS, and export TCCM validation files."
        )
    meta = result or {}
    if not meta:
        meta = {
            "parameters": _load_json("run_metadata.json").get("selected_parameters", {}),
            "embedding": _load_json("run_metadata.json").get("embedding", {}),
            "clusters": _load_json("combined_labels.json") if _exists("combined_labels.json") else [],
            "taxonomy": _load_json("taxonomy_map.json") if _exists("taxonomy_map.json") else {},
            "config": _load_json("corpus_config.json") if _exists("corpus_config.json") else {},
        }
    params = meta.get("parameters", {})
    emb = meta.get("embedding", {})
    tax = meta.get("taxonomy", {}).get("coverage_stats", {})
    cfg = meta.get("config", {})
    return (
        f"**Run complete.** Analysed {cfg.get('rows', 'N/A')} papers from "
        f"{cfg.get('journal', 'the corpus')} ({cfg.get('year_min')} to {cfg.get('year_max')}).\n\n"
        f"Selected clustering: `{params.get('algorithm')}` with "
        f"`{params.get('n_clusters')}` clusters, min size `{params.get('min_size')}`, "
        f"max size `{params.get('max_size')}`, noise ratio `{params.get('noise_ratio')}`.\n\n"
        f"Embedding: `{emb.get('embedding_model')}`. PAJAIS mapped: "
        f"`{tax.get('mapped', 0)}`; novel: `{tax.get('novel', 0)}`. "
        "Download the optimizer log and council validation for the final submission appendix."
    )


def _run(file_obj):
    if file_obj is None:
        return (
            "Upload a CSV first.",
            _phase_html(),
            _cluster_table(),
            _cards_html(),
            _optimizer_table(),
            _compliance_html(),
            _compliance_table(),
            _council_viz_html(),
            _council_table(),
            _tccm_table(),
            _tccm_dual_status_html(),
            _tccm_dual_table(),
            _chart_iframe("intertopic_map.html"),
            _chart_iframe("bar_chart.html"),
            _chart_iframe("treemap.html"),
            _download_files(),
        )
    filepath = file_obj if isinstance(file_obj, str) else file_obj.name
    result = run_complete_pipeline(filepath)
    return (
        _summary_markdown(result),
        _phase_html(),
        _cluster_table(),
        _cards_html(),
        _optimizer_table(),
        _compliance_html(),
        _compliance_table(),
        _council_viz_html(),
        _council_table(),
        _tccm_table(),
        _tccm_dual_status_html(),
        _tccm_dual_table(),
        _chart_iframe("intertopic_map.html"),
        _chart_iframe("bar_chart.html"),
        _chart_iframe("treemap.html"),
        result["deliverables"],
    )


def _refresh():
    return (
        _summary_markdown(),
        _phase_html(),
        _cluster_table(),
        _cards_html(),
        _optimizer_table(),
        _compliance_html(),
        _compliance_table(),
        _council_viz_html(),
        _council_table(),
        _tccm_table(),
        _tccm_dual_status_html(),
        _tccm_dual_table(),
        _chart_iframe("intertopic_map.html"),
        _chart_iframe("bar_chart.html"),
        _chart_iframe("treemap.html"),
        _download_files(),
    )


CSS = """
.gradio-container { max-width: 1360px !important; }
.app-title { padding: 18px 0 8px; }
.app-title h1 { margin: 0; font-size: 30px; color: #0f172a; letter-spacing: 0; }
.app-title p { color: #475569; margin: 6px 0 0; }
.compliance-empty { padding: 24px; border: 1px dashed #94a3b8; border-radius: 8px; background: #f8fafc; color: #475569; }
.compliance-board, .tccm-status { background: #ffffff; border: 1px solid #d8dee9; border-radius: 10px; padding: 16px; }
.compliance-board h3, .tccm-status h3 { margin: 0; color: #0f172a; font-size: 20px; }
.compliance-board p, .tccm-status p { color: #475569; margin: 6px 0 14px; line-height: 1.45; }
.compliance-grid { display: grid; grid-template-columns: repeat(auto-fit, minmax(300px, 1fr)); gap: 10px; }
.compliance-row { border: 1px solid #e2e8f0; border-radius: 8px; padding: 12px; background: #f8fafc; }
.compliance-row span { display: inline-block; color: white; font-size: 11px; font-weight: 800; padding: 3px 8px; border-radius: 999px; margin-bottom: 8px; }
.compliance-row strong { display: block; color: #0f172a; font-size: 14px; }
.compliance-row p { font-size: 12px; margin: 6px 0; color: #475569; }
.compliance-row small { color: #64748b; font-size: 11px; }
.tccm-status > div { display: grid; grid-template-columns: repeat(auto-fit, minmax(220px, 1fr)); gap: 10px; }
.tccm-card { border: 1px solid #e2e8f0; border-radius: 8px; padding: 12px; background: #f8fafc; }
.tccm-card strong { display: block; font-size: 24px; }
.tccm-card span { color: #475569; font-size: 12px; font-weight: 700; }
.council-empty { padding: 28px; border: 1px dashed #94a3b8; color: #475569; border-radius: 8px; background: #f8fafc; }
.council-board { background: #08111f; color: #e5edf7; border-radius: 10px; padding: 18px; border: 1px solid #1e3a5f; }
.council-top { display: grid; grid-template-columns: minmax(280px, 1.2fr) minmax(320px, 1fr); gap: 16px; align-items: start; }
.council-top h3 { margin: 0; font-size: 20px; letter-spacing: 0; }
.council-top p { margin: 6px 0 0; color: #9fb3c8; line-height: 1.45; }
.council-metrics { display: grid; grid-template-columns: repeat(2, minmax(140px, 1fr)); gap: 8px; }
.council-metrics div { background: #10243a; border: 1px solid #1f4568; border-radius: 8px; padding: 10px; }
.council-metrics strong { display: block; color: #5eead4; font-size: 18px; }
.council-metrics span { color: #a8bed4; font-size: 12px; }
.council-lane { display: grid; grid-template-columns: 1fr 60px 1fr 60px 1fr 60px 1fr; gap: 8px; align-items: center; margin: 20px 0; }
.pulse-node { min-height: 62px; display: grid; place-items: center; text-align: center; border: 1px solid #2563eb; background: #0f2546; border-radius: 8px; color: #bfdbfe; font-weight: 800; animation: councilGlow 1.8s ease-in-out infinite; }
.pulse-node span { display: block; font-size: 11px; font-weight: 600; color: #dbeafe; margin-top: 3px; }
.pulse-node.final { border-color: #14b8a6; background: #0f3b39; color: #99f6e4; }
.pulse-line { height: 3px; border-radius: 999px; background: linear-gradient(90deg, #2563eb, #14b8a6, #2563eb); background-size: 220% 100%; animation: councilFlow 1.1s linear infinite; }
.council-cluster { margin-top: 10px; padding: 12px; border: 1px solid #1f4568; border-radius: 8px; background: #0c1b2c; }
.cluster-head { display: flex; justify-content: space-between; gap: 12px; color: #cbd5e1; margin-bottom: 10px; }
.cluster-head span { color: #7dd3fc; font-weight: 800; }
.cluster-head strong { color: #f8fafc; }
.council-flow { display: grid; grid-template-columns: repeat(4, minmax(160px, 1fr)); gap: 8px; }
.council-vote, .final-label { border-radius: 8px; padding: 10px; background: #10243a; border: 1px solid #1f4568; min-height: 82px; }
.council-vote { display: flex; gap: 8px; align-items: flex-start; }
.vote-dot { width: 10px; height: 10px; margin-top: 4px; border-radius: 50%; background: #5eead4; box-shadow: 0 0 14px #5eead4; animation: councilBlink 1.2s ease-in-out infinite; flex: 0 0 auto; }
.council-vote strong, .final-label strong { display: block; color: #e2e8f0; font-size: 13px; }
.council-vote span, .final-label span { display: block; color: #5eead4; font-size: 12px; margin-top: 3px; }
.council-vote small, .final-label small { display: block; color: #94a3b8; font-size: 11px; margin-top: 4px; line-height: 1.25; }
.final-label { border-color: #14b8a6; background: #0d302f; }
@keyframes councilFlow { from { background-position: 0% 0; } to { background-position: 220% 0; } }
@keyframes councilGlow { 0%, 100% { box-shadow: 0 0 0 rgba(37,99,235,0.2); } 50% { box-shadow: 0 0 22px rgba(20,184,166,0.45); } }
@keyframes councilBlink { 0%, 100% { opacity: .35; transform: scale(.75); } 50% { opacity: 1; transform: scale(1.15); } }
@media (max-width: 900px) {
  .council-top, .council-flow { grid-template-columns: 1fr; }
  .council-lane { grid-template-columns: 1fr; }
  .pulse-line { height: 18px; width: 3px; justify-self: center; }
}
"""


with gr.Blocks(title="SPJIMR Topic Modelling Submission App", css=CSS, theme=gr.themes.Soft()) as demo:
    gr.HTML(
        "<div class='app-title'><h1>SPJIMR Topic Modelling Submission App</h1>"
        "<p>Paper-level Title + Abstract + DOI vectors, UMAP/HDBSCAN optimization, "
        "LLM council validation, PAJAIS mapping, and TCCM extraction appendix.</p></div>"
    )
    phase = gr.HTML(value=_phase_html())

    with gr.Row():
        csv_file = gr.File(label="Upload Scopus CSV", file_types=[".csv"], scale=3)
        with gr.Column(scale=1):
            run_btn = gr.Button("Run Complete Pipeline", variant="primary")
            refresh_btn = gr.Button("Refresh Outputs")

    summary = gr.Markdown(value=_summary_markdown())

    with gr.Tabs():
        with gr.Tab("Clusters"):
            cluster_table = gr.Dataframe(
                headers=[
                    "Cluster ID", "Label", "PAJAIS Category", "Papers", "Confidence",
                    "Agreement", "Keywords", "Top 3 Titles", "Reasoning",
                ],
                value=_cluster_table(),
                wrap=True,
                interactive=False,
            )
            cluster_cards = gr.HTML(value=_cards_html())
        with gr.Tab("Optimization"):
            optimizer_table = gr.Dataframe(value=_optimizer_table(), wrap=True, interactive=False)
        with gr.Tab("Compliance"):
            compliance_panel = gr.HTML(value=_compliance_html())
            compliance_table = gr.Dataframe(value=_compliance_table(), wrap=True, interactive=False)
        with gr.Tab("Council Validation"):
            council_viz = gr.HTML(value=_council_viz_html())
            council_table = gr.Dataframe(value=_council_table(), wrap=True, interactive=False)
        with gr.Tab("TCCM Validation"):
            tccm_table = gr.Dataframe(value=_tccm_table(), wrap=True, interactive=False)
        with gr.Tab("TCCM Dual Validation"):
            gr.Markdown(
                "Upload the **NotebookLM extraction CSV** and the **second LLM extraction CSV** "
                "from full-text PDFs. Expected columns can include Title, DOI, Theory, Context, "
                "Variables/Constructs, Method, and Computational Techniques."
            )
            notebook_paste = gr.Textbox(
                label="Paste NotebookLM table output",
                lines=10,
                placeholder="Paste the copied NotebookLM table here, including Paper ID / Paper Citation / Study Type / DV / IV / Evidence lines...",
            )
            notebook_paste_btn = gr.Button("Import NotebookLM Paste", variant="secondary")
            with gr.Row():
                notebook_file = gr.File(label="NotebookLM extraction CSV", file_types=[".csv"])
                second_llm_file = gr.File(label="Second LLM extraction CSV", file_types=[".csv"])
            tccm_dual_btn = gr.Button("Validate NotebookLM + Second LLM", variant="primary")
            tccm_dual_status = gr.HTML(value=_tccm_dual_status_html())
            tccm_dual_table = gr.Dataframe(value=_tccm_dual_table(), wrap=True, interactive=False)
        with gr.Tab("Charts"):
            chart_map = gr.HTML(value=_chart_iframe("intertopic_map.html"))
            chart_bar = gr.HTML(value=_chart_iframe("bar_chart.html"))
            chart_tree = gr.HTML(value=_chart_iframe("treemap.html"))
        with gr.Tab("Downloads"):
            downloads = gr.File(value=_download_files(), label="Generated deliverables", file_count="multiple")

    outputs = [
        summary,
        phase,
        cluster_table,
        cluster_cards,
        optimizer_table,
        compliance_panel,
        compliance_table,
        council_viz,
        council_table,
        tccm_table,
        tccm_dual_status,
        tccm_dual_table,
        chart_map,
        chart_bar,
        chart_tree,
        downloads,
    ]
    run_btn.click(fn=_run, inputs=[csv_file], outputs=outputs, show_api=False, api_name=False)
    refresh_btn.click(fn=_refresh, inputs=None, outputs=outputs, show_api=False, api_name=False)
    tccm_dual_btn.click(
        fn=_on_tccm_dual_validate,
        inputs=[notebook_file, second_llm_file],
        outputs=[tccm_dual_status, tccm_dual_table, downloads],
        show_api=False,
        api_name=False,
    )
    notebook_paste_btn.click(
        fn=_on_notebooklm_paste,
        inputs=[notebook_paste],
        outputs=[tccm_dual_status, tccm_dual_table, downloads],
        show_api=False,
        api_name=False,
    )


if __name__ == "__main__":
    demo.launch(server_name="0.0.0.0", server_port=7860, ssr_mode=False)