app.py

import html
from pathlib import Path
from datetime import datetime

import pandas as pd
import gradio as gr

CSV_PATH = Path("leaderboard.csv")

# Full breakdown columns (Appendix Table 6)
COLS = [
    "Model",
    "Score",
    "Completeness",
    "Grounding",
    "Success Rate",
    "Recovery Rate",
    "Flexibility",
    "Order",
    "Info Diversity",
    "Format",
    "Tradeoff",
    "Tool Calls",
    "# Turns",
    "Progress Tracking",
    "Goal Decomposition",
]

PERCENT_COLS = {
    "Success Rate",
    "Recovery Rate",
    "Flexibility",
    "Order",
    "Info Diversity",
    "Format",
    "Tradeoff",
}

LABEL_MAP = {
    "Info Diversity": "Info Div.",
    "# Turns": "Turns",
    "Progress Tracking": "Progress",
    "Goal Decomposition": "Goal Decomp.",
}

# Light, Arena-like, high-contrast style (prevents "light bg + light text" issues)
ARENA_CSS = r"""
:root { color-scheme: light; }
html, body { background: #f6f7fb !important; }

/* Gradio theme tokens */
.gradio-container{
  max-width: 1200px !important;
  margin: 0 auto !important;
  padding: 18px 16px 24px !important;

  --body-background-fill: #f6f7fb !important;
  --body-background-fill-hover: #f6f7fb !important;

  --body-text-color: #0f172a !important;
  --body-text-color-subdued: #334155 !important;

  --block-background-fill: #ffffff !important;
  --block-background-fill-hover: #ffffff !important;
  --block-border-color: #e5e7eb !important;

  --input-background-fill: #ffffff !important;
  --input-background-fill-hover: #ffffff !important;
  --input-border-color: #e2e8f0 !important;
  --input-border-color-focus: #93c5fd !important;
  --input-text-color: #0f172a !important;
  --input-placeholder-color: #94a3b8 !important;

  --button-secondary-background-fill: #ffffff !important;
  --button-secondary-background-fill-hover: #f8fafc !important;
  --button-secondary-border-color: #e2e8f0 !important;
  --button-secondary-text-color: #0f172a !important;

  --button-primary-background-fill: #2563eb !important;
  --button-primary-background-fill-hover: #1d4ed8 !important;
  --button-primary-border-color: #2563eb !important;
  --button-primary-text-color: #ffffff !important;

  --link-text-color: #2563eb !important;
  --link-text-color-active: #1d4ed8 !important;
}

/* Arena table card */
.arena-card{
  background: #ffffff;
  border: 1px solid #e5e7eb;
  border-radius: 14px;
  box-shadow: 0 1px 2px rgba(16,24,40,0.06);
  overflow: hidden;
}
.arena-wrap{ width: 100%; overflow-x: auto; }

table.arena-table{
  width: 100%;
  min-width: 1300px; /* wide table, scrolls horizontally */
  border-collapse: separate;
  border-spacing: 0;
  font-size: 13px;
  color: #0f172a;
}

/* IMPORTANT: override any global "prose table" borders */
table.arena-table th, table.arena-table td{
  border: none !important;
  overflow: visible !important;
  text-overflow: clip !important;
}

table.arena-table thead th{
  position: sticky;
  top: 0;
  z-index: 2;
  background: #f8fafc;
  color: #334155 !important;
  font-weight: 650;
  text-align: left;
  padding: 10px 12px;
  border-bottom: 1px solid #e2e8f0 !important;
  white-space: nowrap;
}

table.arena-table tbody td{
  padding: 10px 12px;
  border-bottom: 1px solid #eef2f7 !important;
  white-space: nowrap;
  color: #0f172a !important;
}

table.arena-table tbody tr:nth-child(even){ background: #fbfdff; }
table.arena-table tbody tr:hover{ background: #f1f5f9; }

table.arena-table th.num, table.arena-table td.num{
  text-align: right;
  font-variant-numeric: tabular-nums;
}

table.arena-table td.model{ font-weight: 650; }
table.arena-table td.rank{ width: 52px; color: #64748b !important; }

/* optional: keep Model column visible while horizontal scrolling */
table.arena-table thead th:first-child,
table.arena-table tbody td:first-child{
  position: sticky;
  left: 0;
  z-index: 3;
  background: #f8fafc;
}
table.arena-table tbody td:first-child{
  background: #ffffff;
}
"""

def _to_float(x):
    """Parse numeric and percent strings into float for sorting."""
    if x is None:
        return float("nan")
    if isinstance(x, (int, float)) and not pd.isna(x):
        return float(x)
    s = str(x).strip()
    if not s:
        return float("nan")
    if s.endswith("%"):
        s = s[:-1].strip()
    s = s.replace(",", "")
    try:
        return float(s)
    except Exception:
        return float("nan")

def load_df() -> pd.DataFrame:
    if not CSV_PATH.exists():
        return pd.DataFrame(columns=COLS)
    df = pd.read_csv(CSV_PATH)
    for c in COLS:
        if c not in df.columns:
            df[c] = ""
    return df[COLS]

def format_cell(col: str, val) -> str:
    if val is None or (isinstance(val, float) and pd.isna(val)):
        return ""
    s = str(val).strip()
    if col in PERCENT_COLS:
        return s
    f = _to_float(val)
    if pd.isna(f):
        return s
    return f"{f:.2f}"

def prepare_df(query: str, sort_by: str, descending: bool) -> pd.DataFrame:
    df = load_df()
    if query:
        q = query.lower().strip()
        df = df[df["Model"].astype(str).str.lower().str.contains(q, na=False)]

    if sort_by in df.columns:
        df = df.assign(_s=df[sort_by].map(_to_float))
        df = df.sort_values("_s", ascending=not descending, na_position="last").drop(columns=["_s"])

    df = df.reset_index(drop=True)
    df.insert(0, "Rank", range(1, len(df) + 1))
    return df

def render_table(df: pd.DataFrame) -> str:
    if df.empty:
        return "<div class='arena-card' style='padding:14px;'>No entries found.</div>"

    cols = list(df.columns)

    ths = []
    for c in cols:
        label = LABEL_MAP.get(c, c)
        cls = "num" if c not in ("Model",) else ""
        if c == "Rank":
            cls = "rank num"
        ths.append(f"<th class='{cls}'>{html.escape(label)}</th>")

    rows = []
    for _, row in df.iterrows():
        tds = []
        for c in cols:
            if c == "Rank":
                cls = "rank num"
                val = row[c]
            elif c == "Model":
                cls = "model"
                val = row[c]
            else:
                cls = "num"
                val = format_cell(c, row[c])
            tds.append(f"<td class='{cls}'>{html.escape(str(val))}</td>")
        rows.append("<tr>" + "".join(tds) + "</tr>")

    return f"""
    <div class="arena-card">
      <div class="arena-wrap">
        <table class="arena-table">
          <thead><tr>{''.join(ths)}</tr></thead>
          <tbody>{''.join(rows)}</tbody>
        </table>
      </div>
    </div>
    """

def update(q: str, s: str, d: bool) -> str:
    return render_table(prepare_df(q, s, d))

SORT_CHOICES = [c for c in COLS if c != "Model"]

with gr.Blocks(title="ToolGym Leaderboard", css=ARENA_CSS) as demo:
    gr.Markdown("# 🏆 ToolGym Leaderboard")
    gr.Markdown("Full leaderboard breakdown. Update by editing `leaderboard.csv` via PR.")

    with gr.Row():
        query = gr.Textbox(label="Search", placeholder="e.g., deepseek, gemini, qwen ...")
        sort_by = gr.Dropdown(label="Sort by", choices=SORT_CHOICES, value="Score")
        descending = gr.Checkbox(value=True, label="Descending")

    table = gr.HTML()

    query.change(update, inputs=[query, sort_by, descending], outputs=table)
    sort_by.change(update, inputs=[query, sort_by, descending], outputs=table)
    descending.change(update, inputs=[query, sort_by, descending], outputs=table)
    demo.load(update, inputs=[query, sort_by, descending], outputs=table)

    ts = ""
    if CSV_PATH.exists():
        ts = datetime.utcfromtimestamp(CSV_PATH.stat().st_mtime).strftime("%Y-%m-%d %H:%M UTC")
    gr.Markdown(f"<small>Source: <code>leaderboard.csv</code>{(' · Last updated: ' + ts) if ts else ''}</small>")

    with gr.Accordion("Submit / Update", open=False):
        gr.Markdown(
            "- Open a PR editing `leaderboard.csv`.\n"
            "- Include: model name, evaluation setting/commit hash, and the metrics.\n"
        )

demo.launch()