Import Web Agent Bench leaderboard from agent-bench

.gitattributes

@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+results/raw/**/results.json filter=lfs diff=lfs merge=lfs -text

.gitignore

@@ -0,0 +1,4 @@
+hf_token
+__pycache__/
+*.pyc
+.venv/

Makefile

@@ -0,0 +1,13 @@
+.PHONY: style format
+
+
+style:
+	python -m black --line-length 119 .
+	python -m isort .
+	ruff check --fix .
+
+
+quality:
+	python -m black --check --line-length 119 .
+	python -m isort --check-only .
+	ruff check .

PUBLISH.md

@@ -0,0 +1,32 @@
+# Публикация на Hugging Face
+
+Скопируйте **всё содержимое** `liderboard/` в репозиторий Space [MERA-evaluation/WebAgentBench](https://huggingface.co/spaces/MERA-evaluation/WebAgentBench).
+
+## Перед публикацией
+
+Из корня `agent-bench`:
+
+```bash
+python liderboard/src/export_results.py
+```
+
+Это обновит `liderboard/results/` из последних прогонов в `tests/eval/` (та же логика discovery, что у `scripts/render_eval_aggregate.py`). Старые файлы в `results/` удаляются перед экспортом.
+
+## Push
+
+Синхронизировать в локальный клон Space только файлы из последнего коммита:
+
+```bash
+./scripts/sync_liderboard_to_webagentbench.sh /path/to/WebAgentBench
+```
+
+Затем в каталоге WebAgentBench:
+
+```bash
+cd /path/to/WebAgentBench
+git add -A
+git commit -m "Update Web Agent Bench leaderboard"
+git push -u origin main
+```
+
+Используйте [HF write token](https://huggingface.co/settings/tokens) вместо пароля.

README.md

@@ -1,13 +1,30 @@
 ---
 title: WebAgentBench
-emoji: 🐠
-colorFrom: pink
+emoji: 🌐
+colorFrom: blue
 colorTo: green
 sdk: gradio
-sdk_version: 6.19.0
-python_version: '3.13'
+sdk_version: 6.9.0
 app_file: app.py
 pinned: false
+license: mit
+tags:
+- leaderboard
+short_description: Web Agent Bench leaderboard
 ---
 
-Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
+# Web Agent Bench Leaderboard
+
+Источник для Hugging Face Space. Полная документация: [agent-bench/liderboard](https://github.com/ai-forever/agent-bench/tree/liderboard/liderboard).
+
+## Генерация
+
+```bash
+# из корня agent-bench — после прогона eval
+./scripts/export_leaderboard.sh
+cd liderboard && python app.py
+```
+
+## Метрики
+
+Основная — **DAB Success Rate** (`run.success_rate` из `results.json`). См. [EVAL_README.md](https://github.com/ai-forever/agent-bench/blob/open_bench/docs/EVAL_README.md).

app.py

@@ -0,0 +1,429 @@
+"""Web Agent Bench leaderboard (Gradio Space source)."""
+
+from __future__ import annotations
+
+import html
+import json
+from pathlib import Path
+
+import gradio as gr
+import pandas as pd
+
+from src.bench_config import load_bench_config, section_order, taxonomy_order
+from src.load_entries import entry_by_id, entry_choices, load_entries, load_raw_results
+from src.display import dataframe_to_html
+from src.tables import (
+    category_description_html,
+    section_leaderboard_dataframe,
+    taxonomy_leaderboard_dataframe,
+)
+from src.ui import build_board, build_nav_js, build_shell_top, load_css
+
+
+def _category_table(entries, kind: str, item_id: str):
+    if kind == "section":
+        desc, df = category_description_html("section", item_id), section_leaderboard_dataframe(entries, item_id)
+    else:
+        desc, df = category_description_html("taxonomy", item_id), taxonomy_leaderboard_dataframe(entries, item_id)
+    return desc, dataframe_to_html(df)
+
+
+def _raw_tasks_dataframe(raw: dict | None) -> pd.DataFrame:
+    if not raw:
+        return pd.DataFrame()
+    rows = []
+    config = load_bench_config()
+    section_labels = {section_id: meta["label"] for section_id, meta in config["sections"].items()}
+    for test in raw.get("tests") or []:
+        ui = test.get("ui_taxonomy") or {}
+        domain = ui.get("domain") or "—"
+        rows.append(
+            {
+                "Section": section_labels.get(domain, domain),
+                "Task": test.get("test_name") or "—",
+                "Success": "✓" if test.get("success") else "✗",
+                "Duration": test.get("duration_seconds"),
+                "Steps": test.get("agent_steps"),
+            }
+        )
+    return pd.DataFrame(rows)
+
+
+def _run_metrics_dataframe(entry, raw: dict | None) -> pd.DataFrame:
+    run = (raw or {}).get("run") or {}
+    rows = [
+        {"Metric": "Success %", "Value": entry.success_pct},
+        {"Metric": "Tasks passed", "Value": f"{entry.passed_tasks}/{entry.total_tasks}"},
+        {"Metric": "Section avg %", "Value": round(entry.section_avg_rate * 100, 1) if entry.section_avg_rate is not None else None},
+        {"Metric": "Harness", "Value": entry.harness},
+        {"Metric": "Provider", "Value": entry.provider or run.get("provider")},
+        {"Metric": "Mock", "Value": entry.mock or run.get("mock")},
+        {"Metric": "Finished", "Value": entry.finished},
+        {"Metric": "Submitted at", "Value": entry.submitted_at or run.get("finished_at") or run.get("started_at")},
+        {"Metric": "Avg duration (s)", "Value": entry.avg_duration_seconds or run.get("avg_duration_seconds")},
+        {"Metric": "Avg steps", "Value": entry.avg_agent_steps or run.get("avg_agent_steps")},
+        {"Metric": "Tokens / task", "Value": entry.avg_tokens_per_task or run.get("avg_tokens_per_task")},
+        {"Metric": "Total tokens", "Value": entry.total_tokens or run.get("total_tokens")},
+        {"Metric": "Total $", "Value": entry.total_cost_usd},
+        {"Metric": "Pass@k", "Value": entry.pass_at_k},
+        {"Metric": "Agent completion", "Value": entry.agent_completion_rate},
+        {"Metric": "Agent/DAB agreement", "Value": entry.agent_dab_agreement_rate},
+        {"Metric": "Source", "Value": entry.source_path},
+        {"Metric": "Raw results", "Value": entry.raw_results_path},
+    ]
+    return pd.DataFrame(rows)
+
+
+def _ui_taxonomy_dataframe(raw: dict | None) -> pd.DataFrame:
+    if not raw:
+        return pd.DataFrame()
+    stats = raw.get("ui_taxonomy_stats") or {}
+    by_primary = stats.get("by_primary") or {}
+    rows = []
+    for class_id, row in sorted(by_primary.items()):
+        if not isinstance(row, dict):
+            continue
+        rows.append(
+            {
+                "Class": class_id,
+                "Success %": round(float(row["success_rate"]) * 100, 1) if isinstance(row.get("success_rate"), (int, float)) else None,
+                "Passed": row.get("passed"),
+                "Failed": row.get("failed"),
+                "Total": row.get("total"),
+            }
+        )
+    return pd.DataFrame(rows)
+
+
+def _submission_summary(entry) -> str:
+    section_note = ""
+    if entry.section_avg_rate is not None and entry.success_rate is not None:
+        section_pct = round(entry.section_avg_rate * 100, 1)
+        success_pct = entry.success_pct
+        if abs(section_pct - (success_pct or 0)) >= 0.5:
+            section_note = (
+                f'<p class="wab-submission-note">Section-average score: <strong>{section_pct}%</strong>. '
+                f"Primary leaderboard metric is DAB task pass rate: "
+                f"<strong>{success_pct}%</strong> ({entry.passed_tasks}/{entry.total_tasks}).</p>"
+            )
+
+    badges = ", ".join(html.escape(badge) for badge in entry.ui_badges) if entry.ui_badges else "—"
+    return (
+        '<div class="wab-submission-summary">'
+        f'<h3 class="wab-submission-title">{html.escape(entry.model)}</h3>'
+        f'<p><strong>Harness:</strong> <span class="wab-inline-mono">{html.escape(entry.harness)}</span></p>'
+        f"<p><strong>Success:</strong> {entry.success_pct}% "
+        f"({entry.passed_tasks}/{entry.total_tasks} tasks)</p>"
+        f"<p><strong>UI badges:</strong> {badges}</p>"
+        f"{section_note}"
+        "</div>"
+    )
+
+
+def _submission_tables(entry_id: str, entries):
+    entry = entry_by_id(entries, entry_id)
+    if entry is None:
+        empty = dataframe_to_html(pd.DataFrame())
+        return "No submission selected.", empty, empty, empty
+
+    summary = _submission_summary(entry)
+    raw = load_raw_results(entry)
+    if raw is None:
+        return (
+            summary
+            + '<p class="wab-submission-note">Raw <span class="wab-inline-mono">results.json</span> not found. '
+            "Run <span class=\"wab-inline-mono\">./scripts/export_leaderboard.sh</span> after eval.</p>",
+            dataframe_to_html(_run_metrics_dataframe(entry, None)),
+            dataframe_to_html(pd.DataFrame(), empty_note="Raw results not available."),
+            dataframe_to_html(pd.DataFrame(), empty_note="Raw results not available."),
+        )
+
+    return (
+        summary,
+        dataframe_to_html(_run_metrics_dataframe(entry, raw)),
+        dataframe_to_html(_raw_tasks_dataframe(raw)),
+        dataframe_to_html(_ui_taxonomy_dataframe(raw)),
+    )
+
+
+def _raw_json_placeholder(entry_id: str, entries) -> str:
+    entry = entry_by_id(entries, entry_id)
+    if entry is None:
+        return "{}"
+    path = entry.raw_results_path or f"results/raw/{entry_id}/results.json"
+    return (
+        f"// Raw results for {entry.model} × {entry.harness}\n"
+        f"// File: {path}\n"
+        '// Click "Load full JSON" to fetch the exported results.json.'
+    )
+
+
+def _raw_json_full(entry_id: str, entries) -> str:
+    entry = entry_by_id(entries, entry_id)
+    if entry is None:
+        return "{}"
+    raw = load_raw_results(entry)
+    if raw is None:
+        return "{}"
+    return json.dumps(raw, ensure_ascii=False, indent=2)
+
+
+_DETAIL_TAB_IDS = ("submission", "sections", "taxonomy", "metrics", "about")
+
+
+def _parse_nav_payload(payload: str) -> tuple[str, str]:
+    tab, category = (payload.split(":", 1) + [""])[:2]
+    return tab, category
+
+
+def build_demo(entries=None):
+    config = load_bench_config()
+    entries = entries if entries is not None else load_entries()
+    view_choices = [(f'{view["emoji"]} {view["label"]}', view["id"]) for view in config["views"]]
+    submission_choices = entry_choices(entries)
+    default_submission = submission_choices[0][1] if submission_choices else None
+    section_choices = [(config["sections"][section_id]["label"], section_id) for section_id in section_order()]
+    taxonomy_choices = [(config["ui_classes"][class_id]["label"], class_id) for class_id in taxonomy_order()]
+    default_section = section_choices[0][1] if section_choices else None
+    default_taxonomy = taxonomy_choices[0][1] if taxonomy_choices else None
+
+    theme = gr.themes.Base(
+        primary_hue=gr.themes.colors.orange,
+        neutral_hue=gr.themes.colors.gray,
+    ).set(
+        body_background_fill="*neutral_950",
+        body_background_fill_dark="*neutral_950",
+        block_background_fill="*neutral_900",
+        block_background_fill_dark="*neutral_900",
+        block_border_color="*neutral_800",
+        block_border_color_dark="*neutral_800",
+        body_text_color="*neutral_50",
+        body_text_color_dark="*neutral_50",
+    )
+
+    with gr.Blocks(title="Web Agent Bench Leaderboard", elem_classes=["wab-root"]) as demo:
+        gr.HTML(build_shell_top(entries))
+        view = gr.Radio(
+            choices=view_choices,
+            value="success",
+            show_label=False,
+            elem_classes=["wab-view-nav-radio"],
+            container=False,
+        )
+        board = gr.HTML(build_board(entries, "success"))
+
+        def on_view_change(view_id: str):
+            return build_board(entries, view_id)
+
+        view.change(on_view_change, inputs=view, outputs=board)
+
+        gr.HTML('<div id="wab-detail-anchor" class="wab-detail-anchor"></div>')
+
+        with gr.Tabs(elem_id="wab-detail-tabs", elem_classes=["wab-tabs"]):
+            with gr.Tab("Submission", elem_classes=["wab-tab-panel"]):
+                gr.HTML(
+                    '<p class="wab-detail-note">Inspect the full exported '
+                    '<span class="wab-inline-mono">results.json</span> for a model × harness run.</p>'
+                )
+                submission = gr.Dropdown(
+                    choices=submission_choices,
+                    value=default_submission,
+                    label="Submission",
+                    interactive=True,
+                    filterable=True,
+                    elem_id="wab-submission-dropdown",
+                    elem_classes=["wab-submission-dropdown"],
+                )
+                initial_detail = _submission_tables(default_submission, entries) if default_submission else (
+                    "No submissions yet.",
+                    dataframe_to_html(pd.DataFrame()),
+                    dataframe_to_html(pd.DataFrame()),
+                    dataframe_to_html(pd.DataFrame()),
+                )
+                detail_summary = gr.HTML(value=initial_detail[0], elem_classes=["wab-submission-summary-wrap"])
+                with gr.Accordion("Run metrics", open=True, elem_classes=["wab-accordion"]):
+                    detail_metrics = gr.HTML(value=initial_detail[1], elem_classes=["wab-detail-html"])
+                with gr.Accordion("Tasks", open=False, elem_classes=["wab-accordion"]):
+                    detail_tasks = gr.HTML(value=initial_detail[2], elem_classes=["wab-detail-html"])
+                with gr.Accordion("UI taxonomy", open=False, elem_classes=["wab-accordion"]):
+                    detail_ui = gr.HTML(value=initial_detail[3], elem_classes=["wab-detail-html"])
+                with gr.Accordion("Raw JSON", open=False, elem_classes=["wab-accordion"]):
+                    gr.Markdown(
+                        '<p class="wab-detail-note">Large files load on demand so tables above stay responsive.</p>'
+                    )
+                    load_json_btn = gr.Button("Load full JSON", elem_classes=["wab-load-json-btn"])
+                    detail_json = gr.Code(
+                        value=_raw_json_placeholder(default_submission, entries) if default_submission else "{}",
+                        language="json",
+                        lines=20,
+                        elem_classes=["wab-json"],
+                    )
+
+                def on_submission_change(entry_id: str):
+                    return _submission_tables(entry_id, entries) + (_raw_json_placeholder(entry_id, entries),)
+
+                submission.change(
+                    on_submission_change,
+                    inputs=[submission],
+                    outputs=[detail_summary, detail_metrics, detail_tasks, detail_ui, detail_json],
+                )
+
+                def on_load_full_json(entry_id: str):
+                    return _raw_json_full(entry_id, entries)
+
+                load_json_btn.click(
+                    on_load_full_json,
+                    inputs=[submission],
+                    outputs=[detail_json],
+                )
+
+            with gr.Tab("Sections", elem_classes=["wab-tab-panel"]):
+                gr.HTML(
+                    '<p class="wab-detail-note">Leaderboard по доменам бенчмарка. '
+                    "Выберите категорию — описание появится под кнопками.</p>"
+                )
+                section_pick = gr.Radio(
+                    choices=section_choices,
+                    value=default_section,
+                    show_label=False,
+                    elem_id="wab-section-pick",
+                    elem_classes=["wab-category-radio"],
+                    container=False,
+                )
+                section_initial = _category_table(entries, "section", default_section) if default_section else ("", "")
+                section_desc = gr.HTML(value=section_initial[0], elem_classes=["wab-category-desc-wrap"])
+                section_table = gr.HTML(value=section_initial[1], elem_classes=["wab-detail-html"])
+
+                def on_section_change(section_id: str):
+                    return _category_table(entries, "section", section_id)
+
+                section_pick.change(
+                    on_section_change,
+                    inputs=[section_pick],
+                    outputs=[section_desc, section_table],
+                )
+
+            with gr.Tab("UI taxonomy", elem_classes=["wab-tab-panel"]):
+                gr.HTML(
+                    '<p class="wab-detail-note">Leaderboard по UI-классам (primary). '
+                    "Выберите класс — описание появится под кнопками.</p>"
+                )
+                taxonomy_pick = gr.Radio(
+                    choices=taxonomy_choices,
+                    value=default_taxonomy,
+                    show_label=False,
+                    elem_id="wab-taxonomy-pick",
+                    elem_classes=["wab-category-radio"],
+                    container=False,
+                )
+                taxonomy_initial = _category_table(entries, "taxonomy", default_taxonomy) if default_taxonomy else ("", "")
+                taxonomy_desc = gr.HTML(value=taxonomy_initial[0], elem_classes=["wab-category-desc-wrap"])
+                taxonomy_table = gr.HTML(value=taxonomy_initial[1], elem_classes=["wab-detail-html"])
+
+                def on_taxonomy_change(class_id: str):
+                    return _category_table(entries, "taxonomy", class_id)
+
+                taxonomy_pick.change(
+                    on_taxonomy_change,
+                    inputs=[taxonomy_pick],
+                    outputs=[taxonomy_desc, taxonomy_table],
+                )
+
+            with gr.Tab("Metrics", elem_classes=["wab-tab-panel"]):
+                metric_rows = [
+                    {
+                        "Model": entry.model,
+                        "Harness": entry.harness,
+                        "Success %": entry.success_pct,
+                        "Tasks": f"{entry.passed_tasks}/{entry.total_tasks}",
+                        "Avg duration": entry.avg_duration_seconds,
+                        "Avg steps": entry.avg_agent_steps,
+                        "Tokens/task": entry.avg_tokens_per_task,
+                        "Total $": entry.total_cost_usd,
+                        "Pass@k": entry.pass_at_k,
+                    }
+                    for entry in entries
+                ]
+                gr.HTML(
+                    value=dataframe_to_html(pd.DataFrame(metric_rows)),
+                    elem_classes=["wab-detail-html"],
+                )
+
+            with gr.Tab("About", elem_classes=["wab-tab-panel"]):
+                gr.Markdown((Path(__file__).parent / "docs" / "description.md").read_text(encoding="utf-8"))
+
+        nav_payload = gr.Textbox(
+            value="",
+            show_label=False,
+            visible="hidden",
+            elem_id="wab-nav-payload",
+            elem_classes=["wab-nav-trigger-hidden"],
+        )
+        nav_btn = gr.Button(
+            "Navigate",
+            visible="hidden",
+            elem_id="wab-nav-go",
+            elem_classes=["wab-nav-trigger-hidden"],
+        )
+
+        def on_badge_nav(payload: str = ""):
+            no_update = gr.update()
+            if not payload:
+                return (no_update,) * 6
+
+            tab, category = _parse_nav_payload(payload)
+            section_pick_up = section_desc_up = section_table_up = no_update
+            tax_pick_up = tax_desc_up = tax_table_up = no_update
+
+            if tab == "sections" and category:
+                section_pick_up = gr.update(value=category)
+                desc, table = _category_table(entries, "section", category)
+                section_desc_up = gr.update(value=desc)
+                section_table_up = gr.update(value=table)
+            elif tab == "taxonomy" and category:
+                tax_pick_up = gr.update(value=category)
+                desc, table = _category_table(entries, "taxonomy", category)
+                tax_desc_up = gr.update(value=desc)
+                tax_table_up = gr.update(value=table)
+
+            return (
+                section_pick_up,
+                section_desc_up,
+                section_table_up,
+                tax_pick_up,
+                tax_desc_up,
+                tax_table_up,
+            )
+
+        nav_btn.click(
+            on_badge_nav,
+            inputs=[nav_payload],
+            outputs=[
+                section_pick,
+                section_desc,
+                section_table,
+                taxonomy_pick,
+                taxonomy_desc,
+                taxonomy_table,
+            ],
+            js=(
+                "(payload) => {"
+                "const p = window.__wabPendingNav || payload || '';"
+                "const tab = (p.split(':')[0] || 'submission');"
+                "if (window.wabOpenDetailTab) window.wabOpenDetailTab(tab);"
+                "document.getElementById('wab-detail-anchor')"
+                "?.scrollIntoView({behavior:'smooth', block:'start'});"
+                "return p;"
+                "}"
+            ),
+        )
+
+        demo.load(lambda: None, None, None, js=build_nav_js())
+
+    return demo, theme, load_css()
+
+
+demo, theme, css = build_demo()
+
+if __name__ == "__main__":
+    demo.launch(theme=theme, css=css, share=False)

assets/pinchbench.css

@@ -0,0 +1,864 @@
+:root {
+  --bg: #121212;
+  --bg-card: #171717;
+  --bg-elevated: #1f1f1f;
+  --border: #2a2a2a;
+  --text: #fafafa;
+  --text-muted: #9a9a9a;
+  --primary: #f97316;
+  --primary-soft: rgba(249, 115, 22, 0.14);
+  --success: #22c55e;
+  --warning: #f59e0b;
+  --danger: #ef4444;
+  --radius: 10px;
+  --mono: "SF Mono", "JetBrains Mono", "Fira Code", ui-monospace, monospace;
+  --sans: Inter, -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;
+}
+
+.gradio-container {
+  background: var(--bg) !important;
+  color: var(--text) !important;
+  font-family: var(--sans) !important;
+  max-width: 1200px !important;
+  width: 100% !important;
+  margin: 0 auto !important;
+}
+
+.gradio-container .wab-root,
+.gradio-container .wab-tabs,
+.gradio-container .wab-tab-panel,
+.gradio-container .wab-detail-html,
+.gradio-container .wab-submission-summary-wrap,
+.gradio-container .html-container {
+  width: 100% !important;
+  max-width: 100% !important;
+  box-sizing: border-box !important;
+}
+
+.gradio-container .markdown code,
+.gradio-container .prose code,
+.gradio-container code {
+  background: var(--bg-elevated) !important;
+  color: #fdba74 !important;
+  border: 1px solid var(--border) !important;
+  border-radius: 6px !important;
+  padding: 0.1rem 0.35rem !important;
+}
+
+.wab-inline-mono {
+  font-family: var(--mono);
+  font-size: 0.86rem;
+  color: #fdba74;
+  background: var(--bg-elevated);
+  border: 1px solid var(--border);
+  border-radius: 6px;
+  padding: 0.1rem 0.35rem;
+}
+
+.wab-submission-summary {
+  margin: 0.5rem 0 1rem;
+  padding: 1rem 1.1rem;
+  background: var(--bg-card);
+  border: 1px solid var(--border);
+  border-radius: var(--radius);
+}
+
+.wab-submission-title {
+  margin: 0 0 0.65rem;
+  font-size: 1.15rem;
+  font-weight: 700;
+  color: var(--text);
+}
+
+.wab-submission-summary p {
+  margin: 0.35rem 0;
+  color: var(--text);
+  line-height: 1.45;
+}
+
+.wab-submission-note {
+  margin: 0.75rem 0 0 !important;
+  padding: 0.75rem 0.85rem;
+  background: var(--bg-elevated);
+  border-left: 3px solid var(--primary);
+  border-radius: 0 8px 8px 0;
+  color: var(--text-muted) !important;
+  font-size: 0.88rem;
+}
+
+.wab-category-desc,
+.wab-category-desc-wrap p {
+  margin: 0 0 0.85rem !important;
+  color: var(--text-muted);
+  font-size: 0.9rem;
+  line-height: 1.45;
+}
+
+.wab-category-desc strong,
+.wab-category-desc-wrap strong {
+  color: var(--text);
+}
+
+.wab-shell {
+  color: var(--text);
+  padding: 0.25rem 0 2rem;
+}
+
+.wab-topbar {
+  display: flex;
+  align-items: center;
+  justify-content: space-between;
+  gap: 1rem;
+  margin-bottom: 1.5rem;
+  flex-wrap: wrap;
+}
+
+.wab-brand {
+  display: flex;
+  align-items: center;
+  gap: 0.75rem;
+}
+
+.wab-logo {
+  width: 42px;
+  height: 42px;
+  border-radius: 12px;
+  background: linear-gradient(135deg, #fb923c, #ea580c);
+  display: grid;
+  place-items: center;
+  font-size: 1.35rem;
+  box-shadow: 0 8px 24px rgba(249, 115, 22, 0.25);
+}
+
+.wab-brand h1 {
+  margin: 0;
+  font-size: 1.35rem;
+  font-weight: 700;
+  letter-spacing: -0.02em;
+}
+
+.wab-brand p {
+  margin: 0.15rem 0 0;
+  color: var(--text-muted);
+  font-size: 0.85rem;
+}
+
+.wab-stats {
+  display: flex;
+  gap: 0.75rem;
+  flex-wrap: wrap;
+}
+
+.wab-stat {
+  background: var(--bg-card);
+  border: 1px solid var(--border);
+  border-radius: 999px;
+  padding: 0.45rem 0.85rem;
+  font-size: 0.82rem;
+  color: var(--text-muted);
+}
+
+.wab-stat strong {
+  color: var(--text);
+}
+
+.wab-hero {
+  margin-bottom: 1.75rem;
+}
+
+.wab-hero h2 {
+  margin: 0 0 0.35rem;
+  font-size: clamp(1.6rem, 3vw, 2.2rem);
+  font-weight: 700;
+  letter-spacing: -0.03em;
+}
+
+.wab-hero p {
+  margin: 0;
+  color: var(--text-muted);
+  font-size: 1rem;
+}
+
+.wab-section-label {
+  color: var(--text-muted);
+  font-size: 0.78rem;
+  text-transform: uppercase;
+  letter-spacing: 0.08em;
+  margin: 0 0 0.75rem;
+}
+
+.wab-quick-grid {
+  display: grid;
+  grid-template-columns: repeat(auto-fit, minmax(220px, 1fr));
+  gap: 0.9rem;
+  margin-bottom: 2rem;
+}
+
+.wab-quick-card {
+  background: var(--bg-card);
+  border: 1px solid var(--border);
+  border-radius: var(--radius);
+  padding: 1rem 1rem 0.95rem;
+  transition: border-color 0.15s ease, transform 0.15s ease;
+}
+
+.wab-quick-card:hover {
+  border-color: rgba(249, 115, 22, 0.45);
+  transform: translateY(-1px);
+}
+
+.wab-quick-kicker {
+  color: var(--primary);
+  font-size: 0.76rem;
+  font-weight: 700;
+  text-transform: uppercase;
+  letter-spacing: 0.06em;
+  margin-bottom: 0.35rem;
+}
+
+.wab-quick-score {
+  font-size: 1.65rem;
+  font-weight: 700;
+  line-height: 1.1;
+  margin-bottom: 0.35rem;
+}
+
+.wab-quick-model {
+  font-family: var(--mono);
+  font-size: 0.82rem;
+  color: var(--text);
+  word-break: break-word;
+}
+
+.wab-quick-meta {
+  margin-top: 0.55rem;
+  color: var(--text-muted);
+  font-size: 0.78rem;
+  line-height: 1.35;
+}
+
+.wab-panel {
+  background: var(--bg-card);
+  border: 1px solid var(--border);
+  border-radius: calc(var(--radius) + 2px);
+  overflow: hidden;
+  margin-bottom: 1.5rem;
+}
+
+.wab-panel-head {
+  padding: 1.15rem 1.2rem 0.9rem;
+  border-bottom: 1px solid var(--border);
+}
+
+.wab-panel-head h3 {
+  margin: 0 0 0.35rem;
+  font-size: 1.2rem;
+  display: flex;
+  align-items: center;
+  gap: 0.45rem;
+}
+
+.wab-panel-head p {
+  margin: 0;
+  color: var(--text-muted);
+  font-size: 0.9rem;
+}
+
+.wab-table-wrap {
+  overflow-x: auto;
+}
+
+.wab-table {
+  width: 100%;
+  border-collapse: collapse;
+}
+
+.wab-board .wab-table {
+  min-width: 720px;
+}
+
+.wab-table th {
+  text-align: left;
+  font-size: 0.72rem;
+  text-transform: uppercase;
+  letter-spacing: 0.06em;
+  color: var(--text-muted);
+  padding: 0.85rem 1rem;
+  border-bottom: 1px solid var(--border);
+  background: var(--bg-elevated);
+  white-space: nowrap;
+}
+
+.wab-table td {
+  padding: 0.9rem 1rem;
+  border-bottom: 1px solid var(--border);
+  vertical-align: middle;
+}
+
+.wab-table tr:last-child td {
+  border-bottom: none;
+}
+
+.wab-table tbody tr:hover {
+  background: rgba(255, 255, 255, 0.02);
+}
+
+.wab-rank {
+  width: 2rem;
+  font-size: 1.1rem;
+}
+
+.wab-model-cell {
+  min-width: 220px;
+}
+
+.wab-model-name {
+  font-family: var(--mono);
+  font-size: 0.86rem;
+  font-weight: 600;
+}
+
+.wab-badges {
+  display: flex;
+  flex-wrap: wrap;
+  gap: 0.35rem;
+  margin-top: 0.4rem;
+  pointer-events: auto;
+}
+
+.wab-board .html-container,
+.wab-board .prose {
+  pointer-events: auto !important;
+}
+
+.wab-badge {
+  display: inline-flex;
+  align-items: center;
+  gap: 0.25rem;
+  border-radius: 999px;
+  padding: 0.15rem 0.55rem;
+  font-size: 0.68rem;
+  font-weight: 600;
+  background: var(--bg-elevated);
+  border: 1px solid var(--border);
+  color: var(--text-muted);
+  line-height: 1.25;
+}
+
+.wab-badge-link {
+  cursor: pointer;
+  transition: border-color 0.15s ease, background 0.15s ease, opacity 0.15s ease;
+}
+
+.wab-badge-link:hover {
+  opacity: 0.92;
+}
+
+.wab-badge-stat {
+  opacity: 0.72;
+  font-weight: 500;
+}
+
+.wab-badge.domain {
+  color: #fdba74;
+  border-color: rgba(249, 115, 22, 0.35);
+  background: var(--primary-soft);
+}
+
+.wab-badge.domain:hover {
+  border-color: rgba(249, 115, 22, 0.55);
+}
+
+.wab-badge.taxonomy {
+  color: #93c5fd;
+  border-color: rgba(59, 130, 246, 0.35);
+  background: rgba(59, 130, 246, 0.12);
+}
+
+.wab-badge.taxonomy:hover {
+  border-color: rgba(59, 130, 246, 0.55);
+}
+
+.wab-hero-links {
+  margin: 0.85rem 0 0;
+}
+
+.wab-jump-link {
+  color: var(--primary);
+  text-decoration: none;
+  font-size: 0.92rem;
+  font-weight: 600;
+}
+
+.wab-jump-link:hover {
+  text-decoration: underline;
+}
+
+.wab-detail-anchor {
+  scroll-margin-top: 1.25rem;
+  height: 0;
+}
+
+.wab-nav-trigger-hidden {
+  position: absolute !important;
+  width: 1px !important;
+  height: 1px !important;
+  padding: 0 !important;
+  margin: 0 !important;
+  overflow: hidden !important;
+  clip: rect(0, 0, 0, 0) !important;
+  white-space: nowrap !important;
+  border: 0 !important;
+}
+
+.wab-harness {
+  font-family: var(--mono);
+  font-size: 0.78rem;
+  color: var(--text-muted);
+}
+
+.wab-category-legend {
+  display: flex;
+  flex-wrap: wrap;
+  gap: 0.45rem;
+  margin: 0 0 0.85rem;
+}
+
+.wab-legend-item {
+  display: inline-flex;
+  align-items: center;
+  border-radius: 999px;
+  padding: 0.2rem 0.6rem;
+  font-size: 0.72rem;
+  color: var(--text-muted);
+  border: 1px dashed var(--border);
+  cursor: help;
+}
+
+.wab-category-radio {
+  margin: 0 0 0.85rem !important;
+  padding: 0 !important;
+  background: transparent !important;
+  border: none !important;
+  box-shadow: none !important;
+}
+
+.wab-category-radio fieldset,
+.wab-category-radio > .wrap {
+  display: flex !important;
+  gap: 0.5rem !important;
+  flex-wrap: wrap !important;
+  border: none !important;
+  padding: 0 !important;
+  margin: 0 !important;
+  background: transparent !important;
+}
+
+.wab-category-radio label {
+  display: inline-flex !important;
+  align-items: center !important;
+  border: 1px solid var(--border) !important;
+  border-radius: 999px !important;
+  padding: 0.4rem 0.85rem !important;
+  font-size: 0.82rem !important;
+  color: var(--text-muted) !important;
+  background: var(--bg-card) !important;
+  cursor: pointer !important;
+  margin: 0 !important;
+  box-shadow: none !important;
+}
+
+.wab-category-radio label.selected,
+.wab-category-radio input:checked + label,
+.wab-category-radio .selected {
+  color: var(--primary) !important;
+  border-color: rgba(249, 115, 22, 0.45) !important;
+  background: var(--primary-soft) !important;
+}
+
+.wab-category-radio input {
+  display: none !important;
+}
+
+.wab-provider {
+  font-size: 0.8rem;
+  color: var(--text-muted);
+  text-transform: lowercase;
+}
+
+.wab-score-cell {
+  min-width: 150px;
+}
+
+.wab-score-value {
+  font-weight: 700;
+  font-size: 0.95rem;
+  margin-bottom: 0.35rem;
+}
+
+.wab-bar {
+  height: 7px;
+  border-radius: 999px;
+  background: #2b2b2b;
+  overflow: hidden;
+}
+
+.wab-bar > span {
+  display: block;
+  height: 100%;
+  border-radius: 999px;
+}
+
+.wab-score-success { color: var(--success); }
+.wab-score-warning { color: var(--warning); }
+.wab-score-danger { color: var(--danger); }
+.wab-bar-success { background: linear-gradient(90deg, #16a34a, #22c55e); }
+.wab-bar-warning { background: linear-gradient(90deg, #d97706, #f59e0b); }
+.wab-bar-danger { background: linear-gradient(90deg, #dc2626, #ef4444); }
+
+.wab-footnote {
+  padding: 0.85rem 1.2rem 1rem;
+  color: var(--text-muted);
+  font-size: 0.82rem;
+  border-top: 1px solid var(--border);
+}
+
+.wab-links a {
+  color: var(--primary);
+  text-decoration: none;
+}
+
+.wab-links a:hover {
+  text-decoration: underline;
+}
+
+.wab-tabs .tab-nav button {
+  background: var(--bg-card) !important;
+  color: var(--text-muted) !important;
+  border: 1px solid var(--border) !important;
+  border-radius: 999px !important;
+  margin-right: 0.45rem !important;
+  pointer-events: auto !important;
+  cursor: pointer !important;
+}
+
+.wab-tabs .tab-nav {
+  pointer-events: auto !important;
+  position: relative;
+  z-index: 2;
+}
+
+.wab-tabs .tab-nav button.selected {
+  background: var(--primary-soft) !important;
+  color: var(--primary) !important;
+  border-color: rgba(249, 115, 22, 0.45) !important;
+}
+
+.wab-tabs .tabitem {
+  border: none !important;
+  background: transparent !important;
+  padding-top: 1rem !important;
+  overflow: visible !important;
+}
+
+.wab-tab-panel {
+  padding-top: 0.25rem !important;
+}
+
+.wab-accordion {
+  margin-bottom: 0.65rem !important;
+  border: 1px solid var(--border) !important;
+  border-radius: var(--radius) !important;
+  background: var(--bg-card) !important;
+}
+
+.wab-accordion > .label-wrap {
+  background: var(--bg-elevated) !important;
+  color: var(--text) !important;
+}
+
+.wab-accordion .accordion {
+  background: var(--bg-card) !important;
+}
+
+.wab-detail-table-wrap {
+  margin: 0;
+  width: 100%;
+}
+
+.wab-detail-table {
+  min-width: 0 !important;
+  width: 100% !important;
+  table-layout: auto;
+}
+
+.wab-detail-table td,
+.wab-detail-table th {
+  font-size: 0.84rem;
+  color: var(--text) !important;
+  background: var(--bg-card) !important;
+  border-color: var(--border) !important;
+}
+
+.wab-detail-table th {
+  color: var(--text-muted) !important;
+  background: var(--bg-elevated) !important;
+}
+
+.wab-metric-value {
+  font-weight: 700;
+  color: #86efac !important;
+}
+
+.wab-mono-cell {
+  font-family: var(--mono);
+  font-size: 0.8rem;
+  color: var(--text-muted) !important;
+}
+
+.wab-detail-html {
+  width: 100%;
+  overflow-x: auto;
+}
+
+.wab-detail-html .html-container {
+  padding: 0 !important;
+  overflow: visible !important;
+}
+
+.wab-load-json-btn {
+  margin-bottom: 0.65rem !important;
+}
+
+.wab-load-json-btn button {
+  background: var(--bg-elevated) !important;
+  color: var(--text) !important;
+  border: 1px solid var(--border) !important;
+  border-radius: 999px !important;
+  font-size: 0.82rem !important;
+}
+
+.wab-load-json-btn button:hover {
+  border-color: rgba(249, 115, 22, 0.45) !important;
+  color: var(--primary) !important;
+}
+
+.wab-accordion .tabitem,
+.wab-accordion > .wrap,
+.wab-accordion .form,
+.wab-accordion .html-container {
+  overflow: visible !important;
+  height: auto !important;
+  max-height: none !important;
+}
+
+.wab-submission-dropdown,
+#wab-submission-dropdown {
+  width: 100% !important;
+}
+
+.wab-submission-dropdown label,
+.wab-submission-dropdown .wrap,
+.wab-submission-dropdown input,
+.wab-submission-dropdown textarea,
+.wab-submission-dropdown .single-select,
+.wab-submission-dropdown .secondary-wrap,
+.wab-submission-dropdown [data-testid="textbox"] input,
+#wab-submission-dropdown input,
+#wab-submission-dropdown textarea {
+  color: var(--text) !important;
+  -webkit-text-fill-color: var(--text) !important;
+  background: var(--bg-card) !important;
+  border-color: var(--border) !important;
+}
+
+.wab-submission-dropdown input::placeholder,
+#wab-submission-dropdown input::placeholder {
+  color: var(--text-muted) !important;
+  -webkit-text-fill-color: var(--text-muted) !important;
+}
+
+.gradio-container ul.options,
+.gradio-container .options,
+.gradio-container [role="listbox"],
+.gradio-container .gr-dropdown-list,
+.gradio-container .dropdown-options {
+  background: var(--bg-card) !important;
+  border: 1px solid var(--border) !important;
+  color: var(--text) !important;
+}
+
+.gradio-container ul.options li,
+.gradio-container .options li,
+.gradio-container [role="listbox"] [role="option"],
+.gradio-container .dropdown-option,
+.gradio-container .item {
+  color: var(--text) !important;
+  background: var(--bg-card) !important;
+}
+
+.gradio-container ul.options li:hover,
+.gradio-container .options li:hover,
+.gradio-container [role="listbox"] [role="option"]:hover,
+.gradio-container .dropdown-option:hover,
+.gradio-container .item:hover {
+  color: var(--primary) !important;
+  background: var(--bg-elevated) !important;
+}
+
+.gradio-container ul.options li.selected,
+.gradio-container [role="listbox"] [role="option"][aria-selected="true"] {
+  color: var(--primary) !important;
+  background: var(--primary-soft) !important;
+}
+
+.wab-dataframe,
+.wab-dataframe .wrap,
+.wab-dataframe .dataframe-wrap,
+.wab-dataframe table,
+.wab-dataframe thead,
+.wab-dataframe tbody,
+.wab-dataframe tr,
+.wab-dataframe td,
+.wab-dataframe th,
+.wab-dataframe input,
+.wab-dataframe textarea,
+.wab-dataframe .cell-wrap,
+.wab-dataframe .svelte-drgcpz,
+.wab-dataframe [data-testid="dataframe"] table,
+.wab-dataframe [data-testid="dataframe"] td,
+.wab-dataframe [data-testid="dataframe"] th {
+  background: var(--bg-card) !important;
+  color: var(--text) !important;
+  border-color: var(--border) !important;
+}
+
+.wab-dataframe th {
+  color: var(--text-muted) !important;
+  background: var(--bg-elevated) !important;
+}
+
+.wab-dataframe .table-wrap {
+  border: 1px solid var(--border) !important;
+  border-radius: var(--radius) !important;
+  overflow: auto !important;
+}
+
+.wab-view-nav {
+  display: flex;
+  gap: 0.5rem;
+  flex-wrap: wrap;
+  margin-bottom: 1.25rem;
+}
+
+.wab-view-nav-radio {
+  margin: 0 0 1.25rem !important;
+  padding: 0 !important;
+  background: transparent !important;
+  border: none !important;
+  box-shadow: none !important;
+}
+
+.wab-view-nav-radio fieldset,
+.wab-view-nav-radio > .wrap {
+  display: flex !important;
+  gap: 0.5rem !important;
+  flex-wrap: wrap !important;
+  border: none !important;
+  padding: 0 !important;
+  margin: 0 !important;
+  background: transparent !important;
+}
+
+.wab-view-nav-radio label {
+  display: inline-flex !important;
+  align-items: center !important;
+  gap: 0.35rem !important;
+  border: 1px solid var(--border) !important;
+  border-radius: 999px !important;
+  padding: 0.4rem 0.85rem !important;
+  font-size: 0.82rem !important;
+  color: var(--text-muted) !important;
+  background: var(--bg-card) !important;
+  cursor: pointer !important;
+  margin: 0 !important;
+  box-shadow: none !important;
+}
+
+.wab-view-nav-radio label:hover {
+  border-color: rgba(249, 115, 22, 0.35) !important;
+  color: var(--text) !important;
+}
+
+.wab-view-nav-radio label.selected,
+.wab-view-nav-radio input:checked + label,
+.wab-view-nav-radio .selected {
+  color: var(--primary) !important;
+  border-color: rgba(249, 115, 22, 0.45) !important;
+  background: var(--primary-soft) !important;
+}
+
+.wab-view-nav-radio input {
+  display: none !important;
+}
+
+.wab-view-pill {
+  display: inline-flex;
+  align-items: center;
+  gap: 0.35rem;
+  border: 1px solid var(--border);
+  border-radius: 999px;
+  padding: 0.4rem 0.85rem;
+  font-size: 0.82rem;
+  color: var(--text-muted);
+  background: var(--bg-card);
+}
+
+.wab-view-pill.active {
+  color: var(--primary);
+  border-color: rgba(249, 115, 22, 0.45);
+  background: var(--primary-soft);
+}
+
+.wab-board {
+  color: var(--text);
+}
+
+.wab-detail-note {
+  color: var(--text-muted);
+  font-size: 0.88rem;
+  margin: 0 0 0.75rem;
+}
+
+.wab-json textarea,
+.wab-json pre {
+  font-family: var(--mono) !important;
+  font-size: 0.78rem !important;
+  background: var(--bg-card) !important;
+  color: var(--text) !important;
+}
+
+#arena_leaderboard_dataframe,
+#arena_leaderboard_dataframe table,
+[data-testid="dataframe"] table,
+[data-testid="dataframe"] td,
+[data-testid="dataframe"] th {
+  background: var(--bg-card) !important;
+  color: var(--text) !important;
+}
+
+#arena_leaderboard_dataframe th,
+[data-testid="dataframe"] th {
+  background: var(--bg-elevated) !important;
+  color: var(--text-muted) !important;
+  border-color: var(--border) !important;
+}
+
+#arena_leaderboard_dataframe td,
+[data-testid="dataframe"] td {
+  border-color: var(--border) !important;
+}
+
+footer {
+  display: none !important;
+}

bench_config.json

@@ -0,0 +1,232 @@
+{
+  "benchmark": {
+    "name": "Web Agent Bench",
+    "mock": "bench",
+    "task_count": 51,
+    "full_bench_min_tasks": 6,
+    "primary_metric": "success_rate",
+    "primary_metric_label": "DAB Success Rate",
+    "primary_metric_help": "Доля задач, где все conditions из dab_check прошли (all_passed). Не путать с Agent Completion.",
+    "docs_url": "https://github.com/ai-forever/agent-bench/blob/open_bench/docs/EVAL_README.md"
+  },
+  "views": [
+    {
+      "id": "success",
+      "label": "Success Rate",
+      "emoji": "🦀",
+      "sort_key": "success_rate",
+      "ascending": false
+    },
+    {
+      "id": "speed",
+      "label": "Speed",
+      "emoji": "⚡",
+      "sort_key": "avg_duration_seconds",
+      "ascending": true
+    },
+    {
+      "id": "cost",
+      "label": "Cost",
+      "emoji": "💰",
+      "sort_key": "total_cost_usd",
+      "ascending": true
+    }
+  ],
+  "quick_picks": {
+    "success": [
+      {
+        "id": "best_overall",
+        "label": "Best Overall",
+        "metric": "success_rate",
+        "subtitle": "Highest DAB success rate on full bench (>5 tasks)."
+      },
+      {
+        "id": "best_full_bench",
+        "label": "Best Full Bench",
+        "metric": "full_bench_success_rate",
+        "subtitle": "Best success on runs with all 51 tasks."
+      }
+    ],
+    "speed": [
+      {
+        "id": "fastest",
+        "label": "Fastest",
+        "metric": "avg_duration_seconds",
+        "subtitle": "Lowest average task duration (seconds)."
+      }
+    ],
+    "cost": [
+      {
+        "id": "best_budget",
+        "label": "Best Budget",
+        "metric": "total_cost_usd",
+        "subtitle": "Lowest total LLM cost for the run."
+      },
+      {
+        "id": "best_value",
+        "label": "Best Value",
+        "metric": "value_score",
+        "subtitle": "Highest success rate per dollar spent."
+      }
+    ]
+  },
+  "ui_classes": {
+    "NAV": {
+      "label": "NAV",
+      "description": "Навигация: переходы между разделами, меню, вкладки хаба, карточки каталога."
+    },
+    "BASKET": {
+      "label": "BASKET",
+      "description": "Корзина: добавление, удаление и изменение количества товаров."
+    },
+    "FAV": {
+      "label": "FAV",
+      "description": "Избранное: добавление и удаление товаров из списка желаний."
+    },
+    "GOV": {
+      "label": "GOV",
+      "description": "Портал услуг: заказ, просмотр и скачивание документов."
+    },
+    "SEARCH": {
+      "label": "SEARCH",
+      "description": "Поиск: ввод запроса, подсказки и отправка формы поиска."
+    },
+    "DATE": {
+      "label": "DATE",
+      "description": "Выбор даты и периода в формах поиска и бронирования."
+    },
+    "SELECT_AC": {
+      "label": "SELECT",
+      "description": "Выбор из списка: станции, города, подсказки autocomplete."
+    },
+    "CARD": {
+      "label": "CARD",
+      "description": "Карточки в каталоге: выбор позиции из сетки или списка."
+    },
+    "PAY": {
+      "label": "PAY",
+      "description": "Оплата: выбор способа оплаты и подтверждение платежа."
+    },
+    "AUTH": {
+      "label": "AUTH",
+      "description": "Авторизация: логин, телефон и подтверждение кода."
+    },
+    "FILTER": {
+      "label": "FILTER",
+      "description": "Фильтры и сортировка в каталогах и списках."
+    },
+    "BTN": {
+      "label": "BTN",
+      "description": "Кнопки и ссылки: CTA, «В корзину», «Найти»."
+    }
+  },
+  "sections": {
+    "shop": {
+      "label": "Маркет",
+      "description": "Каталог маркетплейса: корзина, избранное, поиск и фильтрация.",
+      "core": true,
+      "domain": "shop",
+      "tasks": [
+        "ecommerce_basket_and_favorites",
+        "ecommerce_basket_any_product",
+        "ecommerce_basket_multiple",
+        "ecommerce_basket_named_product",
+        "ecommerce_basket_named_product_02",
+        "ecommerce_basket_named_product_03",
+        "ecommerce_basket_price_10k",
+        "ecommerce_basket_price_1k",
+        "ecommerce_basket_price_constraint",
+        "ecommerce_basket_typo",
+        "ecommerce_favorites",
+        "ecommerce_favorites_02",
+        "ecommerce_favorites_any_product"
+      ]
+    },
+    "books": {
+      "label": "Книги",
+      "description": "Каталог цифровых книг: поиск, корзина и избранное.",
+      "core": false,
+      "domain": "books",
+      "tasks": [
+        "digital_books_audio_basket",
+        "digital_books_author_cheapest_basket",
+        "digital_books_author_cheapest_favorites",
+        "digital_books_author_expensive_basket",
+        "digital_books_named_product_basket",
+        "digital_books_named_product_basket_02",
+        "digital_books_named_product_favorites"
+      ]
+    },
+    "grocery": {
+      "label": "Продукты",
+      "description": "Каталог продуктов: навигация по разделам и корзина.",
+      "core": false,
+      "domain": "grocery",
+      "tasks": [
+        "bench_grocery_navigation",
+        "grocery_basket_any_product",
+        "grocery_basket_named_product",
+        "grocery_basket_named_product_02"
+      ]
+    },
+    "rail": {
+      "label": "Поезда",
+      "description": "Бронирование поездок: маршрут, даты, места, пассажиры и оплата.",
+      "core": true,
+      "domain": "rail",
+      "tasks": [
+        "rail_atomic_select_city_from",
+        "rail_atomic_select_city_to",
+        "rail_atomic_select_date",
+        "rail_atomic_submit_search",
+        "rail_book_and_pay",
+        "rail_book_and_pay_2_passengers",
+        "rail_book_to_cart",
+        "rail_book_to_cart_2_passengers",
+        "rail_book_to_cart_business"
+      ]
+    },
+    "hotels": {
+      "label": "Отели",
+      "description": "Бронирование размещения: город, даты, гости и выбор объекта.",
+      "core": false,
+      "domain": "hotels",
+      "tasks": [
+        "hotel_atomic_select_city",
+        "hotel_atomic_select_end_date",
+        "hotel_atomic_select_guests",
+        "hotel_atomic_select_start_date",
+        "hotel_search_scenario",
+        "hotel_search_scenario_miami",
+        "hotel_search_scenario_typo",
+        "hotel_search_scenario_with_hotel"
+      ]
+    },
+    "gov": {
+      "label": "Услуги",
+      "description": "Портал услуг: заказ, просмотр и скачивание документов.",
+      "core": false,
+      "domain": "gov",
+      "tasks": [
+        "government_download_certificate",
+        "government_download_certificate_techno",
+        "government_order_certificate",
+        "government_view_certificate",
+        "government_view_certificate_2025",
+        "government_view_certificate_casual"
+      ]
+    },
+    "hub": {
+      "label": "Главная",
+      "description": "Стартовый хаб: переходы между разделами обезличенного бенчмарка.",
+      "core": true,
+      "domain": "hub",
+      "tasks": [
+        "bench_hub_navigation_books",
+        "bench_hub_navigation_gov",
+        "bench_hub_navigation_rail",
+        "bench_hub_navigation_shop"
+      ]
+    }
+  }
+}

docs/description.md

@@ -0,0 +1,75 @@
+# Web Agent Bench — лидерборд
+
+Платформа оценки веб-агентов на **обезличенном** бенчмарке (**51** задача, **7** разделов хаба). Все mock-сайты и формулировки в UI не привязаны к реальным брендам. Подробности: [EVAL_README.md](https://github.com/ai-forever/agent-bench/blob/open_bench/docs/EVAL_README.md).
+
+## Бенчмарк
+
+| Раздел | Ключ | Задач |
+|--------|------|------:|
+| Маркет | `shop` | 13 |
+| Книги | `books` | 7 |
+| Продукты | `grocery` | 4 |
+| Поезда | `rail` | 9 |
+| Отели | `hotels` | 8 |
+| Услуги | `gov` | 6 |
+| Главная | `hub` | 4 |
+
+Агент работает через harness (`browser-use`, `ouroboros-cut`, `ouroboros-full-*`, `hermes-ouroboros`) с выбранной LLM на mock-сайтах DAB.
+
+## Метрики лидерборда
+
+### Основная: DAB Success Rate
+
+| Поле | Источник | Описание |
+|------|----------|----------|
+| `success_rate` | `run.success_rate` | `passed_tasks / total_tasks` |
+| Успех задачи | `tests[].success` | Все `conditions` прошли (`dab_check.all_passed`) |
+
+Не путать с **Agent Completion** — агент мог вызвать `done`, но условия DAB не выполнены.
+
+### Скорость и эффективность
+
+| Поле | Описание |
+|------|----------|
+| `avg_duration_seconds` | Среднее время задачи (create_track + agent + check) |
+| `avg_agent_steps` | Среднее число шагов browser-use |
+| `avg_tokens_per_task` | Средние токены на задачу |
+| `total_cost_usd` | Оценка стоимости LLM (`bench_eval.llm_cost`) |
+
+### Надёжность
+
+| Поле | Описание |
+|------|----------|
+| `pass_at_k` | Pass@k по историческим прогонам (extended_metrics) |
+| `agent_completion_rate` | Доля задач с `agent_is_done` |
+| `agent_dab_agreement_rate` | Согласие completion и DAB |
+
+### UI-таксономия
+
+Агрегаты `ui_taxonomy_stats` в `results.json`:
+
+- `by_checked_class` — классы из `conditions` (BASKET, DATE, NAV, …)
+- `by_primary` — главный UI-класс задачи
+- `by_ui_pattern` — паттерны виджетов (`date:popup_grid`, …)
+
+На лидерборде показываются топ UI-бейджи из `by_primary`.
+
+## Виды лидерборда (как PinchBench)
+
+1. **Success Rate** — DAB success, основной рейтинг
+2. **Speed** — `avg_duration_seconds` (меньше — лучше)
+3. **Cost** — стоимость и value (success / $)
+
+## Экспорт результатов
+
+```bash
+python liderboard/src/export_results.py
+```
+
+Берёт последний full-bench прогон (>5 задач) для каждой пары model × harness из `tests/eval/`.
+
+## Ссылки
+
+- [Репозиторий](https://github.com/ai-forever/agent-bench)
+- [Задачи бенчмарка](https://github.com/ai-forever/agent-bench/tree/open_bench/tests/bench)
+- [Таксономия](https://github.com/ai-forever/agent-bench/blob/open_bench/docs/TAXONOMY.md)

pyproject.toml

@@ -0,0 +1,12 @@
+[tool.ruff]
+select = ["E", "F"]
+ignore = ["E501"]
+line-length = 119
+fixable = ["A", "B", "C", "D", "E", "F", "G", "I", "N", "Q", "S", "T", "W", "ANN", "ARG", "BLE", "COM", "DJ", "DTZ", "EM", "ERA", "EXE", "FBT", "ICN", "INP", "ISC", "NPY", "PD", "PGH", "PIE", "PL", "PT", "PTH", "PYI", "RET", "RSE", "RUF", "SIM", "SLF", "TCH", "TID", "TRY", "UP", "YTT"]
+
+[tool.isort]
+profile = "black"
+line_length = 119
+
+[tool.black]
+line-length = 119

requirements.txt

@@ -0,0 +1,4 @@
+gradio==6.9.0
+jinja2>=3.1.5
+numpy
+pandas

results/deepseek_deepseek-v4-flash__browser-use.json

@@ -0,0 +1,205 @@
+{
+  "model": "deepseek/deepseek-v4-flash",
+  "harness": "browser-use",
+  "provider": "openrouter",
+  "mock": "bench",
+  "finished": true,
+  "source_path": "/home/jovyan/emelyanov/agent-bench/tests/eval/deepseek_deepseek-v4-flash/browser-use/2026-06-23_21-48-53__ebfee145/results.json",
+  "submitted_at": "2026-06-24T00:59:05.569517+00:00",
+  "metrics": {
+    "success_rate": 0.45098039215686275,
+    "passed_tasks": 23,
+    "failed_tasks": 28,
+    "total_tasks": 51,
+    "avg_duration_seconds": 430.82946465441995,
+    "avg_agent_steps": 12.07843137254902,
+    "avg_tokens_per_task": 136423.29411764705,
+    "total_tokens": 6957588,
+    "total_cost_usd": 0.681866178,
+    "avg_cost_per_task_usd": 0.01336992505882353,
+    "pass_at_k": null,
+    "agent_completion_rate": 0.8235294117647058,
+    "agent_dab_agreement_rate": 0.6274509803921569,
+    "section_avg_rate": 0.5166143380429095
+  },
+  "sections": {
+    "shop": {
+      "label": "Маркет",
+      "success_rate": 0.46153846153846156,
+      "passed": 6,
+      "total": 13,
+      "tasks": {
+        "ecommerce_basket_and_favorites": false,
+        "ecommerce_basket_any_product": false,
+        "ecommerce_basket_multiple": false,
+        "ecommerce_basket_named_product": true,
+        "ecommerce_basket_named_product_02": true,
+        "ecommerce_basket_named_product_03": false,
+        "ecommerce_basket_price_10k": false,
+        "ecommerce_basket_price_1k": false,
+        "ecommerce_basket_price_constraint": true,
+        "ecommerce_basket_typo": false,
+        "ecommerce_favorites": true,
+        "ecommerce_favorites_02": true,
+        "ecommerce_favorites_any_product": true
+      }
+    },
+    "books": {
+      "label": "Книги",
+      "success_rate": 0.5714285714285714,
+      "passed": 4,
+      "total": 7,
+      "tasks": {
+        "digital_books_audio_basket": true,
+        "digital_books_author_cheapest_basket": true,
+        "digital_books_author_cheapest_favorites": false,
+        "digital_books_author_expensive_basket": false,
+        "digital_books_named_product_basket": true,
+        "digital_books_named_product_basket_02": false,
+        "digital_books_named_product_favorites": true
+      }
+    },
+    "grocery": {
+      "label": "Продукты",
+      "success_rate": 0.75,
+      "passed": 3,
+      "total": 4,
+      "tasks": {
+        "bench_grocery_navigation": true,
+        "grocery_basket_any_product": true,
+        "grocery_basket_named_product": true,
+        "grocery_basket_named_product_02": false
+      }
+    },
+    "rail": {
+      "label": "Поезда",
+      "success_rate": 0.0,
+      "passed": 0,
+      "total": 9,
+      "tasks": {
+        "rail_atomic_select_city_from": false,
+        "rail_atomic_select_city_to": false,
+        "rail_atomic_select_date": false,
+        "rail_atomic_submit_search": false,
+        "rail_book_and_pay": false,
+        "rail_book_and_pay_2_passengers": false,
+        "rail_book_to_cart": false,
+        "rail_book_to_cart_2_passengers": false,
+        "rail_book_to_cart_business": false
+      }
+    },
+    "hotels": {
+      "label": "Отели",
+      "success_rate": 0.5,
+      "passed": 4,
+      "total": 8,
+      "tasks": {
+        "hotel_atomic_select_city": true,
+        "hotel_atomic_select_end_date": false,
+        "hotel_atomic_select_guests": true,
+        "hotel_atomic_select_start_date": true,
+        "hotel_search_scenario": false,
+        "hotel_search_scenario_miami": true,
+        "hotel_search_scenario_typo": false,
+        "hotel_search_scenario_with_hotel": false
+      }
+    },
+    "gov": {
+      "label": "Услуги",
+      "success_rate": 0.3333333333333333,
+      "passed": 2,
+      "total": 6,
+      "tasks": {
+        "government_download_certificate": false,
+        "government_download_certificate_techno": true,
+        "government_order_certificate": true,
+        "government_view_certificate": false,
+        "government_view_certificate_2025": false,
+        "government_view_certificate_casual": false
+      }
+    },
+    "hub": {
+      "label": "Главная",
+      "success_rate": 1.0,
+      "passed": 4,
+      "total": 4,
+      "tasks": {
+        "bench_hub_navigation_books": true,
+        "bench_hub_navigation_gov": true,
+        "bench_hub_navigation_rail": true,
+        "bench_hub_navigation_shop": true
+      }
+    }
+  },
+  "ui_badges": [
+    "NAV",
+    "FAV",
+    "COUNTER",
+    "SELECT_AC"
+  ],
+  "ui_classes": {
+    "BASKET": {
+      "success_rate": 0.4090909090909091,
+      "passed": 9,
+      "total": 22,
+      "failed": 13
+    },
+    "NAV": {
+      "success_rate": 1.0,
+      "passed": 4,
+      "total": 4,
+      "failed": 0
+    },
+    "FAV": {
+      "success_rate": 0.8,
+      "passed": 4,
+      "total": 5,
+      "failed": 1
+    },
+    "GOV": {
+      "success_rate": 0.3333333333333333,
+      "passed": 2,
+      "total": 6,
+      "failed": 4
+    },
+    "SELECT_AC": {
+      "success_rate": 0.5,
+      "passed": 1,
+      "total": 2,
+      "failed": 1
+    },
+    "DATE": {
+      "success_rate": 0.25,
+      "passed": 1,
+      "total": 4,
+      "failed": 3
+    },
+    "COUNTER": {
+      "success_rate": 0.6666666666666666,
+      "passed": 2,
+      "total": 3,
+      "failed": 1
+    },
+    "CARD": {
+      "success_rate": 0.0,
+      "passed": 0,
+      "total": 1,
+      "failed": 1
+    },
+    "SELECT_LIST": {
+      "success_rate": 0.0,
+      "passed": 0,
+      "total": 2,
+      "failed": 2
+    },
+    "PAY": {
+      "success_rate": 0.0,
+      "passed": 0,
+      "total": 2,
+      "failed": 2
+    }
+  },
+  "section_avg_rate": 0.5166143380429095,
+  "raw_results_path": "results/raw/deepseek_deepseek-v4-flash__browser-use/results.json",
+  "entry_id": "deepseek_deepseek-v4-flash__browser-use"
+}

results/deepseek_deepseek-v4-flash__openhands.json

@@ -0,0 +1,205 @@
+{
+  "model": "deepseek/deepseek-v4-flash",
+  "harness": "openhands",
+  "provider": "openrouter",
+  "mock": "bench",
+  "finished": true,
+  "source_path": "/home/jovyan/emelyanov/agent-bench/tests/eval/deepseek_deepseek-v4-flash/openhands/2026-06-23_21-49-06__1bd3084d/results.json",
+  "submitted_at": "2026-06-23T23:25:08.701701+00:00",
+  "metrics": {
+    "success_rate": 0.3333333333333333,
+    "passed_tasks": 17,
+    "failed_tasks": 34,
+    "total_tasks": 51,
+    "avg_duration_seconds": 214.01237942468302,
+    "avg_agent_steps": 18.96078431372549,
+    "avg_tokens_per_task": 515313.09803921566,
+    "total_tokens": 26280968,
+    "total_cost_usd": 3.7022001800000006,
+    "avg_cost_per_task_usd": 0.07259216039215688,
+    "pass_at_k": null,
+    "agent_completion_rate": 0.49019607843137253,
+    "agent_dab_agreement_rate": 0.7647058823529411,
+    "section_avg_rate": 0.3830455259026687
+  },
+  "sections": {
+    "shop": {
+      "label": "Маркет",
+      "success_rate": 0.5384615384615384,
+      "passed": 7,
+      "total": 13,
+      "tasks": {
+        "ecommerce_basket_and_favorites": false,
+        "ecommerce_basket_any_product": false,
+        "ecommerce_basket_multiple": true,
+        "ecommerce_basket_named_product": true,
+        "ecommerce_basket_named_product_02": true,
+        "ecommerce_basket_named_product_03": false,
+        "ecommerce_basket_price_10k": false,
+        "ecommerce_basket_price_1k": false,
+        "ecommerce_basket_price_constraint": true,
+        "ecommerce_basket_typo": false,
+        "ecommerce_favorites": true,
+        "ecommerce_favorites_02": true,
+        "ecommerce_favorites_any_product": true
+      }
+    },
+    "books": {
+      "label": "Книги",
+      "success_rate": 0.14285714285714285,
+      "passed": 1,
+      "total": 7,
+      "tasks": {
+        "digital_books_audio_basket": false,
+        "digital_books_author_cheapest_basket": false,
+        "digital_books_author_cheapest_favorites": false,
+        "digital_books_author_expensive_basket": false,
+        "digital_books_named_product_basket": false,
+        "digital_books_named_product_basket_02": false,
+        "digital_books_named_product_favorites": true
+      }
+    },
+    "grocery": {
+      "label": "Продукты",
+      "success_rate": 0.75,
+      "passed": 3,
+      "total": 4,
+      "tasks": {
+        "bench_grocery_navigation": true,
+        "grocery_basket_any_product": true,
+        "grocery_basket_named_product": true,
+        "grocery_basket_named_product_02": false
+      }
+    },
+    "rail": {
+      "label": "Поезда",
+      "success_rate": 0.0,
+      "passed": 0,
+      "total": 9,
+      "tasks": {
+        "rail_atomic_select_city_from": false,
+        "rail_atomic_select_city_to": false,
+        "rail_atomic_select_date": false,
+        "rail_atomic_submit_search": false,
+        "rail_book_and_pay": false,
+        "rail_book_and_pay_2_passengers": false,
+        "rail_book_to_cart": false,
+        "rail_book_to_cart_2_passengers": false,
+        "rail_book_to_cart_business": false
+      }
+    },
+    "hotels": {
+      "label": "Отели",
+      "success_rate": 0.25,
+      "passed": 2,
+      "total": 8,
+      "tasks": {
+        "hotel_atomic_select_city": true,
+        "hotel_atomic_select_end_date": false,
+        "hotel_atomic_select_guests": true,
+        "hotel_atomic_select_start_date": false,
+        "hotel_search_scenario": false,
+        "hotel_search_scenario_miami": false,
+        "hotel_search_scenario_typo": false,
+        "hotel_search_scenario_with_hotel": false
+      }
+    },
+    "gov": {
+      "label": "Услуги",
+      "success_rate": 0.0,
+      "passed": 0,
+      "total": 6,
+      "tasks": {
+        "government_download_certificate": false,
+        "government_download_certificate_techno": false,
+        "government_order_certificate": false,
+        "government_view_certificate": false,
+        "government_view_certificate_2025": false,
+        "government_view_certificate_casual": false
+      }
+    },
+    "hub": {
+      "label": "Главная",
+      "success_rate": 1.0,
+      "passed": 4,
+      "total": 4,
+      "tasks": {
+        "bench_hub_navigation_books": true,
+        "bench_hub_navigation_gov": true,
+        "bench_hub_navigation_rail": true,
+        "bench_hub_navigation_shop": true
+      }
+    }
+  },
+  "ui_badges": [
+    "NAV",
+    "FAV",
+    "SELECT_AC",
+    "COUNTER"
+  ],
+  "ui_classes": {
+    "BASKET": {
+      "success_rate": 0.3181818181818182,
+      "passed": 7,
+      "total": 22,
+      "failed": 15
+    },
+    "NAV": {
+      "success_rate": 1.0,
+      "passed": 4,
+      "total": 4,
+      "failed": 0
+    },
+    "FAV": {
+      "success_rate": 0.8,
+      "passed": 4,
+      "total": 5,
+      "failed": 1
+    },
+    "GOV": {
+      "success_rate": 0.0,
+      "passed": 0,
+      "total": 6,
+      "failed": 6
+    },
+    "SELECT_AC": {
+      "success_rate": 0.5,
+      "passed": 1,
+      "total": 2,
+      "failed": 1
+    },
+    "DATE": {
+      "success_rate": 0.0,
+      "passed": 0,
+      "total": 4,
+      "failed": 4
+    },
+    "COUNTER": {
+      "success_rate": 0.3333333333333333,
+      "passed": 1,
+      "total": 3,
+      "failed": 2
+    },
+    "CARD": {
+      "success_rate": 0.0,
+      "passed": 0,
+      "total": 1,
+      "failed": 1
+    },
+    "SELECT_LIST": {
+      "success_rate": 0.0,
+      "passed": 0,
+      "total": 2,
+      "failed": 2
+    },
+    "PAY": {
+      "success_rate": 0.0,
+      "passed": 0,
+      "total": 2,
+      "failed": 2
+    }
+  },
+  "section_avg_rate": 0.3830455259026687,
+  "raw_results_path": "results/raw/deepseek_deepseek-v4-flash__openhands/results.json",
+  "entry_id": "deepseek_deepseek-v4-flash__openhands"
+}

results/deepseek_deepseek-v4-flash__openmanus.json

@@ -0,0 +1,205 @@
+{
+  "model": "deepseek/deepseek-v4-flash",
+  "harness": "openmanus",
+  "provider": "openrouter",
+  "mock": "bench",
+  "finished": true,
+  "source_path": "/home/jovyan/emelyanov/agent-bench/tests/eval/deepseek_deepseek-v4-flash/openmanus/2026-06-23_21-49-14__b1d56b67/results.json",
+  "submitted_at": "2026-06-24T00:24:03.691109+00:00",
+  "metrics": {
+    "success_rate": 0.5490196078431373,
+    "passed_tasks": 28,
+    "failed_tasks": 23,
+    "total_tasks": 51,
+    "avg_duration_seconds": 355.42385389261386,
+    "avg_agent_steps": 11.803921568627452,
+    "avg_tokens_per_task": 135782.68,
+    "total_tokens": 6789134,
+    "total_cost_usd": 1.0452048600000003,
+    "avg_cost_per_task_usd": 0.02049421294117648,
+    "pass_at_k": null,
+    "agent_completion_rate": 0.82,
+    "agent_dab_agreement_rate": 0.74,
+    "section_avg_rate": 0.5972004186289901
+  },
+  "sections": {
+    "shop": {
+      "label": "Маркет",
+      "success_rate": 0.6923076923076923,
+      "passed": 9,
+      "total": 13,
+      "tasks": {
+        "ecommerce_basket_and_favorites": false,
+        "ecommerce_basket_any_product": true,
+        "ecommerce_basket_multiple": true,
+        "ecommerce_basket_named_product": true,
+        "ecommerce_basket_named_product_02": true,
+        "ecommerce_basket_named_product_03": false,
+        "ecommerce_basket_price_10k": false,
+        "ecommerce_basket_price_1k": false,
+        "ecommerce_basket_price_constraint": true,
+        "ecommerce_basket_typo": true,
+        "ecommerce_favorites": true,
+        "ecommerce_favorites_02": true,
+        "ecommerce_favorites_any_product": true
+      }
+    },
+    "books": {
+      "label": "Книги",
+      "success_rate": 0.5714285714285714,
+      "passed": 4,
+      "total": 7,
+      "tasks": {
+        "digital_books_audio_basket": true,
+        "digital_books_author_cheapest_basket": false,
+        "digital_books_author_cheapest_favorites": false,
+        "digital_books_author_expensive_basket": false,
+        "digital_books_named_product_basket": true,
+        "digital_books_named_product_basket_02": true,
+        "digital_books_named_product_favorites": true
+      }
+    },
+    "grocery": {
+      "label": "Продукты",
+      "success_rate": 0.75,
+      "passed": 3,
+      "total": 4,
+      "tasks": {
+        "bench_grocery_navigation": true,
+        "grocery_basket_any_product": true,
+        "grocery_basket_named_product": true,
+        "grocery_basket_named_product_02": false
+      }
+    },
+    "rail": {
+      "label": "Поезда",
+      "success_rate": 0.0,
+      "passed": 0,
+      "total": 9,
+      "tasks": {
+        "rail_atomic_select_city_from": false,
+        "rail_atomic_select_city_to": false,
+        "rail_atomic_select_date": false,
+        "rail_atomic_submit_search": false,
+        "rail_book_and_pay": false,
+        "rail_book_and_pay_2_passengers": false,
+        "rail_book_to_cart": false,
+        "rail_book_to_cart_2_passengers": false,
+        "rail_book_to_cart_business": false
+      }
+    },
+    "hotels": {
+      "label": "Отели",
+      "success_rate": 0.5,
+      "passed": 4,
+      "total": 8,
+      "tasks": {
+        "hotel_atomic_select_city": true,
+        "hotel_atomic_select_end_date": true,
+        "hotel_atomic_select_guests": true,
+        "hotel_atomic_select_start_date": true,
+        "hotel_search_scenario": false,
+        "hotel_search_scenario_miami": false,
+        "hotel_search_scenario_typo": false,
+        "hotel_search_scenario_with_hotel": false
+      }
+    },
+    "gov": {
+      "label": "Услуги",
+      "success_rate": 0.6666666666666666,
+      "passed": 4,
+      "total": 6,
+      "tasks": {
+        "government_download_certificate": true,
+        "government_download_certificate_techno": true,
+        "government_order_certificate": true,
+        "government_view_certificate": false,
+        "government_view_certificate_2025": false,
+        "government_view_certificate_casual": true
+      }
+    },
+    "hub": {
+      "label": "Главная",
+      "success_rate": 1.0,
+      "passed": 4,
+      "total": 4,
+      "tasks": {
+        "bench_hub_navigation_books": true,
+        "bench_hub_navigation_gov": true,
+        "bench_hub_navigation_rail": true,
+        "bench_hub_navigation_shop": true
+      }
+    }
+  },
+  "ui_badges": [
+    "NAV",
+    "FAV",
+    "GOV",
+    "BASKET"
+  ],
+  "ui_classes": {
+    "BASKET": {
+      "success_rate": 0.5454545454545454,
+      "passed": 12,
+      "total": 22,
+      "failed": 10
+    },
+    "NAV": {
+      "success_rate": 1.0,
+      "passed": 4,
+      "total": 4,
+      "failed": 0
+    },
+    "FAV": {
+      "success_rate": 0.8,
+      "passed": 4,
+      "total": 5,
+      "failed": 1
+    },
+    "GOV": {
+      "success_rate": 0.6666666666666666,
+      "passed": 4,
+      "total": 6,
+      "failed": 2
+    },
+    "SELECT_AC": {
+      "success_rate": 0.5,
+      "passed": 1,
+      "total": 2,
+      "failed": 1
+    },
+    "DATE": {
+      "success_rate": 0.5,
+      "passed": 2,
+      "total": 4,
+      "failed": 2
+    },
+    "COUNTER": {
+      "success_rate": 0.3333333333333333,
+      "passed": 1,
+      "total": 3,
+      "failed": 2
+    },
+    "CARD": {
+      "success_rate": 0.0,
+      "passed": 0,
+      "total": 1,
+      "failed": 1
+    },
+    "SELECT_LIST": {
+      "success_rate": 0.0,
+      "passed": 0,
+      "total": 2,
+      "failed": 2
+    },
+    "PAY": {
+      "success_rate": 0.0,
+      "passed": 0,
+      "total": 2,
+      "failed": 2
+    }
+  },
+  "section_avg_rate": 0.5972004186289901,
+  "raw_results_path": "results/raw/deepseek_deepseek-v4-flash__openmanus/results.json",
+  "entry_id": "deepseek_deepseek-v4-flash__openmanus"
+}

results/deepseek_deepseek-v4-flash__ouroboros-cut.json

@@ -0,0 +1,205 @@
+{
+  "model": "deepseek/deepseek-v4-flash",
+  "harness": "ouroboros-cut",
+  "provider": "openrouter",
+  "mock": "bench",
+  "finished": true,
+  "source_path": "/home/jovyan/emelyanov/agent-bench/tests/eval/deepseek_deepseek-v4-flash/ouroboros-cut/2026-06-23_22-23-19__07dd12ff/results.json",
+  "submitted_at": "2026-06-24T01:55:26.128559+00:00",
+  "metrics": {
+    "success_rate": 0.49019607843137253,
+    "passed_tasks": 25,
+    "failed_tasks": 26,
+    "total_tasks": 51,
+    "avg_duration_seconds": 490.06015126830806,
+    "avg_agent_steps": 11.607843137254902,
+    "avg_tokens_per_task": 124507.70588235294,
+    "total_tokens": 6349893,
+    "total_cost_usd": 0.5917200799999999,
+    "avg_cost_per_task_usd": 0.011602354509803918,
+    "pass_at_k": null,
+    "agent_completion_rate": 0.8235294117647058,
+    "agent_dab_agreement_rate": 0.6666666666666666,
+    "section_avg_rate": 0.5385923600209315
+  },
+  "sections": {
+    "shop": {
+      "label": "Маркет",
+      "success_rate": 0.6153846153846154,
+      "passed": 8,
+      "total": 13,
+      "tasks": {
+        "ecommerce_basket_and_favorites": false,
+        "ecommerce_basket_any_product": true,
+        "ecommerce_basket_multiple": true,
+        "ecommerce_basket_named_product": false,
+        "ecommerce_basket_named_product_02": true,
+        "ecommerce_basket_named_product_03": false,
+        "ecommerce_basket_price_10k": false,
+        "ecommerce_basket_price_1k": false,
+        "ecommerce_basket_price_constraint": true,
+        "ecommerce_basket_typo": true,
+        "ecommerce_favorites": true,
+        "ecommerce_favorites_02": true,
+        "ecommerce_favorites_any_product": true
+      }
+    },
+    "books": {
+      "label": "Книги",
+      "success_rate": 0.5714285714285714,
+      "passed": 4,
+      "total": 7,
+      "tasks": {
+        "digital_books_audio_basket": false,
+        "digital_books_author_cheapest_basket": true,
+        "digital_books_author_cheapest_favorites": false,
+        "digital_books_author_expensive_basket": false,
+        "digital_books_named_product_basket": true,
+        "digital_books_named_product_basket_02": true,
+        "digital_books_named_product_favorites": true
+      }
+    },
+    "grocery": {
+      "label": "Продукты",
+      "success_rate": 0.75,
+      "passed": 3,
+      "total": 4,
+      "tasks": {
+        "bench_grocery_navigation": true,
+        "grocery_basket_any_product": true,
+        "grocery_basket_named_product": true,
+        "grocery_basket_named_product_02": false
+      }
+    },
+    "rail": {
+      "label": "Поезда",
+      "success_rate": 0.0,
+      "passed": 0,
+      "total": 9,
+      "tasks": {
+        "rail_atomic_select_city_from": false,
+        "rail_atomic_select_city_to": false,
+        "rail_atomic_select_date": false,
+        "rail_atomic_submit_search": false,
+        "rail_book_and_pay": false,
+        "rail_book_and_pay_2_passengers": false,
+        "rail_book_to_cart": false,
+        "rail_book_to_cart_2_passengers": false,
+        "rail_book_to_cart_business": false
+      }
+    },
+    "hotels": {
+      "label": "Отели",
+      "success_rate": 0.5,
+      "passed": 4,
+      "total": 8,
+      "tasks": {
+        "hotel_atomic_select_city": true,
+        "hotel_atomic_select_end_date": true,
+        "hotel_atomic_select_guests": true,
+        "hotel_atomic_select_start_date": true,
+        "hotel_search_scenario": false,
+        "hotel_search_scenario_miami": false,
+        "hotel_search_scenario_typo": false,
+        "hotel_search_scenario_with_hotel": false
+      }
+    },
+    "gov": {
+      "label": "Услуги",
+      "success_rate": 0.3333333333333333,
+      "passed": 2,
+      "total": 6,
+      "tasks": {
+        "government_download_certificate": false,
+        "government_download_certificate_techno": true,
+        "government_order_certificate": true,
+        "government_view_certificate": false,
+        "government_view_certificate_2025": false,
+        "government_view_certificate_casual": false
+      }
+    },
+    "hub": {
+      "label": "Главная",
+      "success_rate": 1.0,
+      "passed": 4,
+      "total": 4,
+      "tasks": {
+        "bench_hub_navigation_books": true,
+        "bench_hub_navigation_gov": true,
+        "bench_hub_navigation_rail": true,
+        "bench_hub_navigation_shop": true
+      }
+    }
+  },
+  "ui_badges": [
+    "NAV",
+    "FAV",
+    "SELECT_AC",
+    "DATE"
+  ],
+  "ui_classes": {
+    "BASKET": {
+      "success_rate": 0.5,
+      "passed": 11,
+      "total": 22,
+      "failed": 11
+    },
+    "NAV": {
+      "success_rate": 1.0,
+      "passed": 4,
+      "total": 4,
+      "failed": 0
+    },
+    "FAV": {
+      "success_rate": 0.8,
+      "passed": 4,
+      "total": 5,
+      "failed": 1
+    },
+    "GOV": {
+      "success_rate": 0.3333333333333333,
+      "passed": 2,
+      "total": 6,
+      "failed": 4
+    },
+    "SELECT_AC": {
+      "success_rate": 0.5,
+      "passed": 1,
+      "total": 2,
+      "failed": 1
+    },
+    "DATE": {
+      "success_rate": 0.5,
+      "passed": 2,
+      "total": 4,
+      "failed": 2
+    },
+    "COUNTER": {
+      "success_rate": 0.3333333333333333,
+      "passed": 1,
+      "total": 3,
+      "failed": 2
+    },
+    "CARD": {
+      "success_rate": 0.0,
+      "passed": 0,
+      "total": 1,
+      "failed": 1
+    },
+    "SELECT_LIST": {
+      "success_rate": 0.0,
+      "passed": 0,
+      "total": 2,
+      "failed": 2
+    },
+    "PAY": {
+      "success_rate": 0.0,
+      "passed": 0,
+      "total": 2,
+      "failed": 2
+    }
+  },
+  "section_avg_rate": 0.5385923600209315,
+  "raw_results_path": "results/raw/deepseek_deepseek-v4-flash__ouroboros-cut/results.json",
+  "entry_id": "deepseek_deepseek-v4-flash__ouroboros-cut"
+}

results/deepseek_deepseek-v4-flash__ouroboros-full-evolving.json

@@ -0,0 +1,205 @@
+{
+  "model": "deepseek/deepseek-v4-flash",
+  "harness": "ouroboros-full-evolving",
+  "provider": "openrouter",
+  "mock": "bench",
+  "finished": true,
+  "source_path": "/home/jovyan/emelyanov/agent-bench/tests/eval/deepseek_deepseek-v4-flash/ouroboros-full-evolving/2026-06-23_22-23-32__lm-mpi-job-90bbb689-8f98-4242-a478-9ed4d96171f3-mpimaster-0-1168-18362-1782253409/results.json",
+  "submitted_at": "2026-06-24T12:17:31.606961+00:00",
+  "metrics": {
+    "success_rate": 0.49019607843137253,
+    "passed_tasks": 25,
+    "failed_tasks": 26,
+    "total_tasks": 51,
+    "avg_duration_seconds": 980.6634050323902,
+    "avg_agent_steps": 22.725490196078432,
+    "avg_tokens_per_task": 1672382.5227272727,
+    "total_tokens": 73584831,
+    "total_cost_usd": 3.50092,
+    "avg_cost_per_task_usd": 0.06864549019607842,
+    "pass_at_k": null,
+    "agent_completion_rate": 0.7058823529411765,
+    "agent_dab_agreement_rate": 0.7843137254901961,
+    "section_avg_rate": 0.5300889586603873
+  },
+  "sections": {
+    "shop": {
+      "label": "Маркет",
+      "success_rate": 0.6153846153846154,
+      "passed": 8,
+      "total": 13,
+      "tasks": {
+        "ecommerce_basket_and_favorites": false,
+        "ecommerce_basket_any_product": true,
+        "ecommerce_basket_multiple": true,
+        "ecommerce_basket_named_product": true,
+        "ecommerce_basket_named_product_02": true,
+        "ecommerce_basket_named_product_03": false,
+        "ecommerce_basket_price_10k": false,
+        "ecommerce_basket_price_1k": false,
+        "ecommerce_basket_price_constraint": true,
+        "ecommerce_basket_typo": true,
+        "ecommerce_favorites": true,
+        "ecommerce_favorites_02": true,
+        "ecommerce_favorites_any_product": false
+      }
+    },
+    "books": {
+      "label": "Книги",
+      "success_rate": 0.42857142857142855,
+      "passed": 3,
+      "total": 7,
+      "tasks": {
+        "digital_books_audio_basket": true,
+        "digital_books_author_cheapest_basket": true,
+        "digital_books_author_cheapest_favorites": false,
+        "digital_books_author_expensive_basket": false,
+        "digital_books_named_product_basket": false,
+        "digital_books_named_product_basket_02": true,
+        "digital_books_named_product_favorites": false
+      }
+    },
+    "grocery": {
+      "label": "Продукты",
+      "success_rate": 0.5,
+      "passed": 2,
+      "total": 4,
+      "tasks": {
+        "bench_grocery_navigation": true,
+        "grocery_basket_any_product": true,
+        "grocery_basket_named_product": false,
+        "grocery_basket_named_product_02": false
+      }
+    },
+    "rail": {
+      "label": "Поезда",
+      "success_rate": 0.0,
+      "passed": 0,
+      "total": 9,
+      "tasks": {
+        "rail_atomic_select_city_from": false,
+        "rail_atomic_select_city_to": false,
+        "rail_atomic_select_date": false,
+        "rail_atomic_submit_search": false,
+        "rail_book_and_pay": false,
+        "rail_book_and_pay_2_passengers": false,
+        "rail_book_to_cart": false,
+        "rail_book_to_cart_2_passengers": false,
+        "rail_book_to_cart_business": false
+      }
+    },
+    "hotels": {
+      "label": "Отели",
+      "success_rate": 0.5,
+      "passed": 4,
+      "total": 8,
+      "tasks": {
+        "hotel_atomic_select_city": true,
+        "hotel_atomic_select_end_date": false,
+        "hotel_atomic_select_guests": true,
+        "hotel_atomic_select_start_date": true,
+        "hotel_search_scenario": false,
+        "hotel_search_scenario_miami": false,
+        "hotel_search_scenario_typo": true,
+        "hotel_search_scenario_with_hotel": false
+      }
+    },
+    "gov": {
+      "label": "Услуги",
+      "success_rate": 0.6666666666666666,
+      "passed": 4,
+      "total": 6,
+      "tasks": {
+        "government_download_certificate": true,
+        "government_download_certificate_techno": false,
+        "government_order_certificate": true,
+        "government_view_certificate": false,
+        "government_view_certificate_2025": true,
+        "government_view_certificate_casual": true
+      }
+    },
+    "hub": {
+      "label": "Главная",
+      "success_rate": 1.0,
+      "passed": 4,
+      "total": 4,
+      "tasks": {
+        "bench_hub_navigation_books": true,
+        "bench_hub_navigation_gov": true,
+        "bench_hub_navigation_rail": true,
+        "bench_hub_navigation_shop": true
+      }
+    }
+  },
+  "ui_badges": [
+    "SELECT_AC",
+    "NAV",
+    "GOV",
+    "BASKET"
+  ],
+  "ui_classes": {
+    "BASKET": {
+      "success_rate": 0.5,
+      "passed": 11,
+      "total": 22,
+      "failed": 11
+    },
+    "NAV": {
+      "success_rate": 1.0,
+      "passed": 4,
+      "total": 4,
+      "failed": 0
+    },
+    "FAV": {
+      "success_rate": 0.4,
+      "passed": 2,
+      "total": 5,
+      "failed": 3
+    },
+    "GOV": {
+      "success_rate": 0.6666666666666666,
+      "passed": 4,
+      "total": 6,
+      "failed": 2
+    },
+    "SELECT_AC": {
+      "success_rate": 1.0,
+      "passed": 2,
+      "total": 2,
+      "failed": 0
+    },
+    "DATE": {
+      "success_rate": 0.25,
+      "passed": 1,
+      "total": 4,
+      "failed": 3
+    },
+    "COUNTER": {
+      "success_rate": 0.3333333333333333,
+      "passed": 1,
+      "total": 3,
+      "failed": 2
+    },
+    "CARD": {
+      "success_rate": 0.0,
+      "passed": 0,
+      "total": 1,
+      "failed": 1
+    },
+    "SELECT_LIST": {
+      "success_rate": 0.0,
+      "passed": 0,
+      "total": 2,
+      "failed": 2
+    },
+    "PAY": {
+      "success_rate": 0.0,
+      "passed": 0,
+      "total": 2,
+      "failed": 2
+    }
+  },
+  "section_avg_rate": 0.5300889586603873,
+  "raw_results_path": "results/raw/deepseek_deepseek-v4-flash__ouroboros-full-evolving/results.json",
+  "entry_id": "deepseek_deepseek-v4-flash__ouroboros-full-evolving"
+}

results/deepseek_deepseek-v4-flash__ouroboros-full-isolated.json

@@ -0,0 +1,205 @@
+{
+  "model": "deepseek/deepseek-v4-flash",
+  "harness": "ouroboros-full-isolated",
+  "provider": "openrouter",
+  "mock": "bench",
+  "finished": true,
+  "source_path": "/home/jovyan/emelyanov/agent-bench/tests/eval/deepseek_deepseek-v4-flash/ouroboros-full-isolated/2026-06-24_11-36-03__lm-mpi-job-5f3144e2-b474-4212-aa8e-67a910368cec-mpimaster-0-485-8835-1782300960/results.json",
+  "submitted_at": "2026-06-24T15:22:04.805657+00:00",
+  "metrics": {
+    "success_rate": 0.5882352941176471,
+    "passed_tasks": 30,
+    "failed_tasks": 21,
+    "total_tasks": 51,
+    "avg_duration_seconds": 529.9419007587667,
+    "avg_agent_steps": 16.980392156862745,
+    "avg_tokens_per_task": 971737.2,
+    "total_tokens": 34010802,
+    "total_cost_usd": 1.6644430000000001,
+    "avg_cost_per_task_usd": 0.03263613725490196,
+    "pass_at_k": null,
+    "agent_completion_rate": 0.6862745098039216,
+    "agent_dab_agreement_rate": 0.7450980392156863,
+    "section_avg_rate": 0.64141810570382
+  },
+  "sections": {
+    "shop": {
+      "label": "Маркет",
+      "success_rate": 0.6923076923076923,
+      "passed": 9,
+      "total": 13,
+      "tasks": {
+        "ecommerce_basket_and_favorites": false,
+        "ecommerce_basket_any_product": true,
+        "ecommerce_basket_multiple": true,
+        "ecommerce_basket_named_product": true,
+        "ecommerce_basket_named_product_02": true,
+        "ecommerce_basket_named_product_03": false,
+        "ecommerce_basket_price_10k": false,
+        "ecommerce_basket_price_1k": false,
+        "ecommerce_basket_price_constraint": true,
+        "ecommerce_basket_typo": true,
+        "ecommerce_favorites": true,
+        "ecommerce_favorites_02": true,
+        "ecommerce_favorites_any_product": true
+      }
+    },
+    "books": {
+      "label": "Книги",
+      "success_rate": 0.7142857142857143,
+      "passed": 5,
+      "total": 7,
+      "tasks": {
+        "digital_books_audio_basket": true,
+        "digital_books_author_cheapest_basket": true,
+        "digital_books_author_cheapest_favorites": false,
+        "digital_books_author_expensive_basket": false,
+        "digital_books_named_product_basket": true,
+        "digital_books_named_product_basket_02": true,
+        "digital_books_named_product_favorites": true
+      }
+    },
+    "grocery": {
+      "label": "Продукты",
+      "success_rate": 0.75,
+      "passed": 3,
+      "total": 4,
+      "tasks": {
+        "bench_grocery_navigation": true,
+        "grocery_basket_any_product": true,
+        "grocery_basket_named_product": true,
+        "grocery_basket_named_product_02": false
+      }
+    },
+    "rail": {
+      "label": "Поезда",
+      "success_rate": 0.0,
+      "passed": 0,
+      "total": 9,
+      "tasks": {
+        "rail_atomic_select_city_from": false,
+        "rail_atomic_select_city_to": false,
+        "rail_atomic_select_date": false,
+        "rail_atomic_submit_search": false,
+        "rail_book_and_pay": false,
+        "rail_book_and_pay_2_passengers": false,
+        "rail_book_to_cart": false,
+        "rail_book_to_cart_2_passengers": false,
+        "rail_book_to_cart_business": false
+      }
+    },
+    "hotels": {
+      "label": "Отели",
+      "success_rate": 0.5,
+      "passed": 4,
+      "total": 8,
+      "tasks": {
+        "hotel_atomic_select_city": true,
+        "hotel_atomic_select_end_date": true,
+        "hotel_atomic_select_guests": true,
+        "hotel_atomic_select_start_date": true,
+        "hotel_search_scenario": false,
+        "hotel_search_scenario_miami": false,
+        "hotel_search_scenario_typo": false,
+        "hotel_search_scenario_with_hotel": false
+      }
+    },
+    "gov": {
+      "label": "Услуги",
+      "success_rate": 0.8333333333333334,
+      "passed": 5,
+      "total": 6,
+      "tasks": {
+        "government_download_certificate": true,
+        "government_download_certificate_techno": true,
+        "government_order_certificate": true,
+        "government_view_certificate": false,
+        "government_view_certificate_2025": true,
+        "government_view_certificate_casual": true
+      }
+    },
+    "hub": {
+      "label": "Главная",
+      "success_rate": 1.0,
+      "passed": 4,
+      "total": 4,
+      "tasks": {
+        "bench_hub_navigation_books": true,
+        "bench_hub_navigation_gov": true,
+        "bench_hub_navigation_rail": true,
+        "bench_hub_navigation_shop": true
+      }
+    }
+  },
+  "ui_badges": [
+    "NAV",
+    "GOV",
+    "FAV",
+    "BASKET"
+  ],
+  "ui_classes": {
+    "BASKET": {
+      "success_rate": 0.5909090909090909,
+      "passed": 13,
+      "total": 22,
+      "failed": 9
+    },
+    "NAV": {
+      "success_rate": 1.0,
+      "passed": 4,
+      "total": 4,
+      "failed": 0
+    },
+    "FAV": {
+      "success_rate": 0.8,
+      "passed": 4,
+      "total": 5,
+      "failed": 1
+    },
+    "GOV": {
+      "success_rate": 0.8333333333333334,
+      "passed": 5,
+      "total": 6,
+      "failed": 1
+    },
+    "SELECT_AC": {
+      "success_rate": 0.5,
+      "passed": 1,
+      "total": 2,
+      "failed": 1
+    },
+    "DATE": {
+      "success_rate": 0.5,
+      "passed": 2,
+      "total": 4,
+      "failed": 2
+    },
+    "COUNTER": {
+      "success_rate": 0.3333333333333333,
+      "passed": 1,
+      "total": 3,
+      "failed": 2
+    },
+    "CARD": {
+      "success_rate": 0.0,
+      "passed": 0,
+      "total": 1,
+      "failed": 1
+    },
+    "SELECT_LIST": {
+      "success_rate": 0.0,
+      "passed": 0,
+      "total": 2,
+      "failed": 2
+    },
+    "PAY": {
+      "success_rate": 0.0,
+      "passed": 0,
+      "total": 2,
+      "failed": 2
+    }
+  },
+  "section_avg_rate": 0.64141810570382,
+  "raw_results_path": "results/raw/deepseek_deepseek-v4-flash__ouroboros-full-isolated/results.json",
+  "entry_id": "deepseek_deepseek-v4-flash__ouroboros-full-isolated"
+}

results/google_gemini-2.5-flash__browser-use.json

@@ -0,0 +1,205 @@
+{
+  "model": "google/gemini-2.5-flash",
+  "harness": "browser-use",
+  "provider": "openrouter",
+  "mock": "bench",
+  "finished": true,
+  "source_path": "/home/jovyan/emelyanov/agent-bench/tests/eval/google_gemini-2.5-flash/browser-use/2026-06-23_21-48-40__30b21cb0/results.json",
+  "submitted_at": "2026-06-23T23:26:25.408974+00:00",
+  "metrics": {
+    "success_rate": 0.4117647058823529,
+    "passed_tasks": 21,
+    "failed_tasks": 30,
+    "total_tasks": 51,
+    "avg_duration_seconds": 216.41764206044815,
+    "avg_agent_steps": 14.176470588235293,
+    "avg_tokens_per_task": 174671.92156862744,
+    "total_tokens": 8908268,
+    "total_cost_usd": 3.3214496000000002,
+    "avg_cost_per_task_usd": 0.06512646274509805,
+    "pass_at_k": null,
+    "agent_completion_rate": 0.8235294117647058,
+    "agent_dab_agreement_rate": 0.5882352941176471,
+    "section_avg_rate": 0.45866038723181585
+  },
+  "sections": {
+    "shop": {
+      "label": "Маркет",
+      "success_rate": 0.6153846153846154,
+      "passed": 8,
+      "total": 13,
+      "tasks": {
+        "ecommerce_basket_and_favorites": false,
+        "ecommerce_basket_any_product": false,
+        "ecommerce_basket_multiple": false,
+        "ecommerce_basket_named_product": true,
+        "ecommerce_basket_named_product_02": true,
+        "ecommerce_basket_named_product_03": true,
+        "ecommerce_basket_price_10k": false,
+        "ecommerce_basket_price_1k": false,
+        "ecommerce_basket_price_constraint": true,
+        "ecommerce_basket_typo": true,
+        "ecommerce_favorites": true,
+        "ecommerce_favorites_02": true,
+        "ecommerce_favorites_any_product": true
+      }
+    },
+    "books": {
+      "label": "Книги",
+      "success_rate": 0.42857142857142855,
+      "passed": 3,
+      "total": 7,
+      "tasks": {
+        "digital_books_audio_basket": true,
+        "digital_books_author_cheapest_basket": true,
+        "digital_books_author_cheapest_favorites": false,
+        "digital_books_author_expensive_basket": false,
+        "digital_books_named_product_basket": false,
+        "digital_books_named_product_basket_02": true,
+        "digital_books_named_product_favorites": false
+      }
+    },
+    "grocery": {
+      "label": "Продукты",
+      "success_rate": 0.75,
+      "passed": 3,
+      "total": 4,
+      "tasks": {
+        "bench_grocery_navigation": true,
+        "grocery_basket_any_product": true,
+        "grocery_basket_named_product": true,
+        "grocery_basket_named_product_02": false
+      }
+    },
+    "rail": {
+      "label": "Поезда",
+      "success_rate": 0.0,
+      "passed": 0,
+      "total": 9,
+      "tasks": {
+        "rail_atomic_select_city_from": false,
+        "rail_atomic_select_city_to": false,
+        "rail_atomic_select_date": false,
+        "rail_atomic_submit_search": false,
+        "rail_book_and_pay": false,
+        "rail_book_and_pay_2_passengers": false,
+        "rail_book_to_cart": false,
+        "rail_book_to_cart_2_passengers": false,
+        "rail_book_to_cart_business": false
+      }
+    },
+    "hotels": {
+      "label": "Отели",
+      "success_rate": 0.25,
+      "passed": 2,
+      "total": 8,
+      "tasks": {
+        "hotel_atomic_select_city": true,
+        "hotel_atomic_select_end_date": false,
+        "hotel_atomic_select_guests": true,
+        "hotel_atomic_select_start_date": false,
+        "hotel_search_scenario": false,
+        "hotel_search_scenario_miami": false,
+        "hotel_search_scenario_typo": false,
+        "hotel_search_scenario_with_hotel": false
+      }
+    },
+    "gov": {
+      "label": "Услуги",
+      "success_rate": 0.16666666666666666,
+      "passed": 1,
+      "total": 6,
+      "tasks": {
+        "government_download_certificate": false,
+        "government_download_certificate_techno": true,
+        "government_order_certificate": false,
+        "government_view_certificate": false,
+        "government_view_certificate_2025": false,
+        "government_view_certificate_casual": false
+      }
+    },
+    "hub": {
+      "label": "Главная",
+      "success_rate": 1.0,
+      "passed": 4,
+      "total": 4,
+      "tasks": {
+        "bench_hub_navigation_books": true,
+        "bench_hub_navigation_gov": true,
+        "bench_hub_navigation_rail": true,
+        "bench_hub_navigation_shop": true
+      }
+    }
+  },
+  "ui_badges": [
+    "NAV",
+    "FAV",
+    "SELECT_AC",
+    "BASKET"
+  ],
+  "ui_classes": {
+    "BASKET": {
+      "success_rate": 0.5,
+      "passed": 11,
+      "total": 22,
+      "failed": 11
+    },
+    "NAV": {
+      "success_rate": 1.0,
+      "passed": 4,
+      "total": 4,
+      "failed": 0
+    },
+    "FAV": {
+      "success_rate": 0.6,
+      "passed": 3,
+      "total": 5,
+      "failed": 2
+    },
+    "GOV": {
+      "success_rate": 0.16666666666666666,
+      "passed": 1,
+      "total": 6,
+      "failed": 5
+    },
+    "SELECT_AC": {
+      "success_rate": 0.5,
+      "passed": 1,
+      "total": 2,
+      "failed": 1
+    },
+    "DATE": {
+      "success_rate": 0.0,
+      "passed": 0,
+      "total": 4,
+      "failed": 4
+    },
+    "COUNTER": {
+      "success_rate": 0.3333333333333333,
+      "passed": 1,
+      "total": 3,
+      "failed": 2
+    },
+    "CARD": {
+      "success_rate": 0.0,
+      "passed": 0,
+      "total": 1,
+      "failed": 1
+    },
+    "SELECT_LIST": {
+      "success_rate": 0.0,
+      "passed": 0,
+      "total": 2,
+      "failed": 2
+    },
+    "PAY": {
+      "success_rate": 0.0,
+      "passed": 0,
+      "total": 2,
+      "failed": 2
+    }
+  },
+  "section_avg_rate": 0.45866038723181585,
+  "raw_results_path": "results/raw/google_gemini-2.5-flash__browser-use/results.json",
+  "entry_id": "google_gemini-2.5-flash__browser-use"
+}

results/google_gemini-2.5-flash__openhands.json

@@ -0,0 +1,205 @@
+{
+  "model": "google/gemini-2.5-flash",
+  "harness": "openhands",
+  "provider": "openrouter",
+  "mock": "bench",
+  "finished": true,
+  "source_path": "/home/jovyan/emelyanov/agent-bench/tests/eval/google_gemini-2.5-flash/openhands/2026-06-24_11-35-27__55edfa0a/results.json",
+  "submitted_at": "2026-06-24T14:41:37.265464+00:00",
+  "metrics": {
+    "success_rate": 0.13725490196078433,
+    "passed_tasks": 7,
+    "failed_tasks": 44,
+    "total_tasks": 51,
+    "avg_duration_seconds": 218.3109522406103,
+    "avg_agent_steps": 10.254901960784315,
+    "avg_tokens_per_task": 197287.19607843139,
+    "total_tokens": 10061647,
+    "total_cost_usd": 3.126107099999999,
+    "avg_cost_per_task_usd": 0.06129621764705881,
+    "pass_at_k": null,
+    "agent_completion_rate": 0.9607843137254902,
+    "agent_dab_agreement_rate": 0.13725490196078433,
+    "section_avg_rate": 0.2099686028257457
+  },
+  "sections": {
+    "shop": {
+      "label": "Маркет",
+      "success_rate": 0.07692307692307693,
+      "passed": 1,
+      "total": 13,
+      "tasks": {
+        "ecommerce_basket_and_favorites": false,
+        "ecommerce_basket_any_product": false,
+        "ecommerce_basket_multiple": false,
+        "ecommerce_basket_named_product": false,
+        "ecommerce_basket_named_product_02": true,
+        "ecommerce_basket_named_product_03": false,
+        "ecommerce_basket_price_10k": false,
+        "ecommerce_basket_price_1k": false,
+        "ecommerce_basket_price_constraint": false,
+        "ecommerce_basket_typo": false,
+        "ecommerce_favorites": false,
+        "ecommerce_favorites_02": false,
+        "ecommerce_favorites_any_product": false
+      }
+    },
+    "books": {
+      "label": "Книги",
+      "success_rate": 0.14285714285714285,
+      "passed": 1,
+      "total": 7,
+      "tasks": {
+        "digital_books_audio_basket": false,
+        "digital_books_author_cheapest_basket": true,
+        "digital_books_author_cheapest_favorites": false,
+        "digital_books_author_expensive_basket": false,
+        "digital_books_named_product_basket": false,
+        "digital_books_named_product_basket_02": false,
+        "digital_books_named_product_favorites": false
+      }
+    },
+    "grocery": {
+      "label": "Продукты",
+      "success_rate": 0.25,
+      "passed": 1,
+      "total": 4,
+      "tasks": {
+        "bench_grocery_navigation": true,
+        "grocery_basket_any_product": false,
+        "grocery_basket_named_product": false,
+        "grocery_basket_named_product_02": false
+      }
+    },
+    "rail": {
+      "label": "Поезда",
+      "success_rate": 0.0,
+      "passed": 0,
+      "total": 9,
+      "tasks": {
+        "rail_atomic_select_city_from": false,
+        "rail_atomic_select_city_to": false,
+        "rail_atomic_select_date": false,
+        "rail_atomic_submit_search": false,
+        "rail_book_and_pay": false,
+        "rail_book_and_pay_2_passengers": false,
+        "rail_book_to_cart": false,
+        "rail_book_to_cart_2_passengers": false,
+        "rail_book_to_cart_business": false
+      }
+    },
+    "hotels": {
+      "label": "Отели",
+      "success_rate": 0.0,
+      "passed": 0,
+      "total": 8,
+      "tasks": {
+        "hotel_atomic_select_city": false,
+        "hotel_atomic_select_end_date": false,
+        "hotel_atomic_select_guests": false,
+        "hotel_atomic_select_start_date": false,
+        "hotel_search_scenario": false,
+        "hotel_search_scenario_miami": false,
+        "hotel_search_scenario_typo": false,
+        "hotel_search_scenario_with_hotel": false
+      }
+    },
+    "gov": {
+      "label": "Услуги",
+      "success_rate": 0.0,
+      "passed": 0,
+      "total": 6,
+      "tasks": {
+        "government_download_certificate": false,
+        "government_download_certificate_techno": false,
+        "government_order_certificate": false,
+        "government_view_certificate": false,
+        "government_view_certificate_2025": false,
+        "government_view_certificate_casual": false
+      }
+    },
+    "hub": {
+      "label": "Главная",
+      "success_rate": 1.0,
+      "passed": 4,
+      "total": 4,
+      "tasks": {
+        "bench_hub_navigation_books": true,
+        "bench_hub_navigation_gov": true,
+        "bench_hub_navigation_rail": true,
+        "bench_hub_navigation_shop": true
+      }
+    }
+  },
+  "ui_badges": [
+    "NAV",
+    "BASKET",
+    "SELECT_LIST",
+    "SELECT_AC"
+  ],
+  "ui_classes": {
+    "BASKET": {
+      "success_rate": 0.13636363636363635,
+      "passed": 3,
+      "total": 22,
+      "failed": 19
+    },
+    "NAV": {
+      "success_rate": 1.0,
+      "passed": 4,
+      "total": 4,
+      "failed": 0
+    },
+    "FAV": {
+      "success_rate": 0.0,
+      "passed": 0,
+      "total": 5,
+      "failed": 5
+    },
+    "GOV": {
+      "success_rate": 0.0,
+      "passed": 0,
+      "total": 6,
+      "failed": 6
+    },
+    "SELECT_AC": {
+      "success_rate": 0.0,
+      "passed": 0,
+      "total": 2,
+      "failed": 2
+    },
+    "DATE": {
+      "success_rate": 0.0,
+      "passed": 0,
+      "total": 4,
+      "failed": 4
+    },
+    "COUNTER": {
+      "success_rate": 0.0,
+      "passed": 0,
+      "total": 3,
+      "failed": 3
+    },
+    "CARD": {
+      "success_rate": 0.0,
+      "passed": 0,
+      "total": 1,
+      "failed": 1
+    },
+    "SELECT_LIST": {
+      "success_rate": 0.0,
+      "passed": 0,
+      "total": 2,
+      "failed": 2
+    },
+    "PAY": {
+      "success_rate": 0.0,
+      "passed": 0,
+      "total": 2,
+      "failed": 2
+    }
+  },
+  "section_avg_rate": 0.2099686028257457,
+  "raw_results_path": "results/raw/google_gemini-2.5-flash__openhands/results.json",
+  "entry_id": "google_gemini-2.5-flash__openhands"
+}

results/google_gemini-2.5-flash__openmanus.json

@@ -0,0 +1,205 @@
+{
+  "model": "google/gemini-2.5-flash",
+  "harness": "openmanus",
+  "provider": "openrouter",
+  "mock": "bench",
+  "finished": true,
+  "source_path": "/home/jovyan/emelyanov/agent-bench/tests/eval/google_gemini-2.5-flash/openmanus/2026-06-23_23-34-21__60885ccc/results.json",
+  "submitted_at": "2026-06-24T01:10:34.661145+00:00",
+  "metrics": {
+    "success_rate": 0.47058823529411764,
+    "passed_tasks": 24,
+    "failed_tasks": 27,
+    "total_tasks": 51,
+    "avg_duration_seconds": 212.98681739418237,
+    "avg_agent_steps": 12.92156862745098,
+    "avg_tokens_per_task": 163701.80392156861,
+    "total_tokens": 8348792,
+    "total_cost_usd": 3.0831540000000004,
+    "avg_cost_per_task_usd": 0.06045400000000001,
+    "pass_at_k": null,
+    "agent_completion_rate": 0.8235294117647058,
+    "agent_dab_agreement_rate": 0.6470588235294118,
+    "section_avg_rate": 0.5173338566195709
+  },
+  "sections": {
+    "shop": {
+      "label": "Маркет",
+      "success_rate": 0.6153846153846154,
+      "passed": 8,
+      "total": 13,
+      "tasks": {
+        "ecommerce_basket_and_favorites": false,
+        "ecommerce_basket_any_product": false,
+        "ecommerce_basket_multiple": true,
+        "ecommerce_basket_named_product": true,
+        "ecommerce_basket_named_product_02": true,
+        "ecommerce_basket_named_product_03": true,
+        "ecommerce_basket_price_10k": false,
+        "ecommerce_basket_price_1k": false,
+        "ecommerce_basket_price_constraint": true,
+        "ecommerce_basket_typo": true,
+        "ecommerce_favorites": true,
+        "ecommerce_favorites_02": false,
+        "ecommerce_favorites_any_product": true
+      }
+    },
+    "books": {
+      "label": "Книги",
+      "success_rate": 0.7142857142857143,
+      "passed": 5,
+      "total": 7,
+      "tasks": {
+        "digital_books_audio_basket": true,
+        "digital_books_author_cheapest_basket": true,
+        "digital_books_author_cheapest_favorites": false,
+        "digital_books_author_expensive_basket": false,
+        "digital_books_named_product_basket": true,
+        "digital_books_named_product_basket_02": true,
+        "digital_books_named_product_favorites": true
+      }
+    },
+    "grocery": {
+      "label": "Продукты",
+      "success_rate": 0.75,
+      "passed": 3,
+      "total": 4,
+      "tasks": {
+        "bench_grocery_navigation": true,
+        "grocery_basket_any_product": true,
+        "grocery_basket_named_product": true,
+        "grocery_basket_named_product_02": false
+      }
+    },
+    "rail": {
+      "label": "Поезда",
+      "success_rate": 0.0,
+      "passed": 0,
+      "total": 9,
+      "tasks": {
+        "rail_atomic_select_city_from": false,
+        "rail_atomic_select_city_to": false,
+        "rail_atomic_select_date": false,
+        "rail_atomic_submit_search": false,
+        "rail_book_and_pay": false,
+        "rail_book_and_pay_2_passengers": false,
+        "rail_book_to_cart": false,
+        "rail_book_to_cart_2_passengers": false,
+        "rail_book_to_cart_business": false
+      }
+    },
+    "hotels": {
+      "label": "Отели",
+      "success_rate": 0.375,
+      "passed": 3,
+      "total": 8,
+      "tasks": {
+        "hotel_atomic_select_city": true,
+        "hotel_atomic_select_end_date": false,
+        "hotel_atomic_select_guests": true,
+        "hotel_atomic_select_start_date": false,
+        "hotel_search_scenario": false,
+        "hotel_search_scenario_miami": false,
+        "hotel_search_scenario_typo": true,
+        "hotel_search_scenario_with_hotel": false
+      }
+    },
+    "gov": {
+      "label": "Услуги",
+      "success_rate": 0.16666666666666666,
+      "passed": 1,
+      "total": 6,
+      "tasks": {
+        "government_download_certificate": false,
+        "government_download_certificate_techno": true,
+        "government_order_certificate": false,
+        "government_view_certificate": false,
+        "government_view_certificate_2025": false,
+        "government_view_certificate_casual": false
+      }
+    },
+    "hub": {
+      "label": "Главная",
+      "success_rate": 1.0,
+      "passed": 4,
+      "total": 4,
+      "tasks": {
+        "bench_hub_navigation_books": true,
+        "bench_hub_navigation_gov": true,
+        "bench_hub_navigation_rail": true,
+        "bench_hub_navigation_shop": true
+      }
+    }
+  },
+  "ui_badges": [
+    "SELECT_AC",
+    "NAV",
+    "FAV",
+    "BASKET"
+  ],
+  "ui_classes": {
+    "BASKET": {
+      "success_rate": 0.5909090909090909,
+      "passed": 13,
+      "total": 22,
+      "failed": 9
+    },
+    "NAV": {
+      "success_rate": 1.0,
+      "passed": 4,
+      "total": 4,
+      "failed": 0
+    },
+    "FAV": {
+      "success_rate": 0.6,
+      "passed": 3,
+      "total": 5,
+      "failed": 2
+    },
+    "GOV": {
+      "success_rate": 0.16666666666666666,
+      "passed": 1,
+      "total": 6,
+      "failed": 5
+    },
+    "SELECT_AC": {
+      "success_rate": 1.0,
+      "passed": 2,
+      "total": 2,
+      "failed": 0
+    },
+    "DATE": {
+      "success_rate": 0.0,
+      "passed": 0,
+      "total": 4,
+      "failed": 4
+    },
+    "COUNTER": {
+      "success_rate": 0.3333333333333333,
+      "passed": 1,
+      "total": 3,
+      "failed": 2
+    },
+    "CARD": {
+      "success_rate": 0.0,
+      "passed": 0,
+      "total": 1,
+      "failed": 1
+    },
+    "SELECT_LIST": {
+      "success_rate": 0.0,
+      "passed": 0,
+      "total": 2,
+      "failed": 2
+    },
+    "PAY": {
+      "success_rate": 0.0,
+      "passed": 0,
+      "total": 2,
+      "failed": 2
+    }
+  },
+  "section_avg_rate": 0.5173338566195709,
+  "raw_results_path": "results/raw/google_gemini-2.5-flash__openmanus/results.json",
+  "entry_id": "google_gemini-2.5-flash__openmanus"
+}

results/google_gemini-2.5-flash__ouroboros-cut.json

@@ -0,0 +1,205 @@
+{
+  "model": "google/gemini-2.5-flash",
+  "harness": "ouroboros-cut",
+  "provider": "openrouter",
+  "mock": "bench",
+  "finished": true,
+  "source_path": "/home/jovyan/emelyanov/agent-bench/tests/eval/google_gemini-2.5-flash/ouroboros-cut/2026-06-23_22-23-17__5359c150/results.json",
+  "submitted_at": "2026-06-23T23:54:44.480931+00:00",
+  "metrics": {
+    "success_rate": 0.43137254901960786,
+    "passed_tasks": 22,
+    "failed_tasks": 29,
+    "total_tasks": 51,
+    "avg_duration_seconds": 206.01449627326983,
+    "avg_agent_steps": 14.235294117647058,
+    "avg_tokens_per_task": 183224.0,
+    "total_tokens": 9344424,
+    "total_cost_usd": 3.4392526,
+    "avg_cost_per_task_usd": 0.06743632549019608,
+    "pass_at_k": null,
+    "agent_completion_rate": 0.8235294117647058,
+    "agent_dab_agreement_rate": 0.6078431372549019,
+    "section_avg_rate": 0.4696493982208268
+  },
+  "sections": {
+    "shop": {
+      "label": "Маркет",
+      "success_rate": 0.6923076923076923,
+      "passed": 9,
+      "total": 13,
+      "tasks": {
+        "ecommerce_basket_and_favorites": false,
+        "ecommerce_basket_any_product": true,
+        "ecommerce_basket_multiple": false,
+        "ecommerce_basket_named_product": true,
+        "ecommerce_basket_named_product_02": true,
+        "ecommerce_basket_named_product_03": true,
+        "ecommerce_basket_price_10k": false,
+        "ecommerce_basket_price_1k": false,
+        "ecommerce_basket_price_constraint": true,
+        "ecommerce_basket_typo": true,
+        "ecommerce_favorites": true,
+        "ecommerce_favorites_02": true,
+        "ecommerce_favorites_any_product": true
+      }
+    },
+    "books": {
+      "label": "Книги",
+      "success_rate": 0.42857142857142855,
+      "passed": 3,
+      "total": 7,
+      "tasks": {
+        "digital_books_audio_basket": true,
+        "digital_books_author_cheapest_basket": true,
+        "digital_books_author_cheapest_favorites": false,
+        "digital_books_author_expensive_basket": false,
+        "digital_books_named_product_basket": false,
+        "digital_books_named_product_basket_02": false,
+        "digital_books_named_product_favorites": true
+      }
+    },
+    "grocery": {
+      "label": "Продукты",
+      "success_rate": 0.75,
+      "passed": 3,
+      "total": 4,
+      "tasks": {
+        "bench_grocery_navigation": true,
+        "grocery_basket_any_product": true,
+        "grocery_basket_named_product": true,
+        "grocery_basket_named_product_02": false
+      }
+    },
+    "rail": {
+      "label": "Поезда",
+      "success_rate": 0.0,
+      "passed": 0,
+      "total": 9,
+      "tasks": {
+        "rail_atomic_select_city_from": false,
+        "rail_atomic_select_city_to": false,
+        "rail_atomic_select_date": false,
+        "rail_atomic_submit_search": false,
+        "rail_book_and_pay": false,
+        "rail_book_and_pay_2_passengers": false,
+        "rail_book_to_cart": false,
+        "rail_book_to_cart_2_passengers": false,
+        "rail_book_to_cart_business": false
+      }
+    },
+    "hotels": {
+      "label": "Отели",
+      "success_rate": 0.25,
+      "passed": 2,
+      "total": 8,
+      "tasks": {
+        "hotel_atomic_select_city": true,
+        "hotel_atomic_select_end_date": false,
+        "hotel_atomic_select_guests": true,
+        "hotel_atomic_select_start_date": false,
+        "hotel_search_scenario": false,
+        "hotel_search_scenario_miami": false,
+        "hotel_search_scenario_typo": false,
+        "hotel_search_scenario_with_hotel": false
+      }
+    },
+    "gov": {
+      "label": "Услуги",
+      "success_rate": 0.16666666666666666,
+      "passed": 1,
+      "total": 6,
+      "tasks": {
+        "government_download_certificate": false,
+        "government_download_certificate_techno": false,
+        "government_order_certificate": false,
+        "government_view_certificate": true,
+        "government_view_certificate_2025": false,
+        "government_view_certificate_casual": false
+      }
+    },
+    "hub": {
+      "label": "Главная",
+      "success_rate": 1.0,
+      "passed": 4,
+      "total": 4,
+      "tasks": {
+        "bench_hub_navigation_books": true,
+        "bench_hub_navigation_gov": true,
+        "bench_hub_navigation_rail": true,
+        "bench_hub_navigation_shop": true
+      }
+    }
+  },
+  "ui_badges": [
+    "NAV",
+    "FAV",
+    "SELECT_AC",
+    "BASKET"
+  ],
+  "ui_classes": {
+    "BASKET": {
+      "success_rate": 0.5,
+      "passed": 11,
+      "total": 22,
+      "failed": 11
+    },
+    "NAV": {
+      "success_rate": 1.0,
+      "passed": 4,
+      "total": 4,
+      "failed": 0
+    },
+    "FAV": {
+      "success_rate": 0.8,
+      "passed": 4,
+      "total": 5,
+      "failed": 1
+    },
+    "GOV": {
+      "success_rate": 0.16666666666666666,
+      "passed": 1,
+      "total": 6,
+      "failed": 5
+    },
+    "SELECT_AC": {
+      "success_rate": 0.5,
+      "passed": 1,
+      "total": 2,
+      "failed": 1
+    },
+    "DATE": {
+      "success_rate": 0.0,
+      "passed": 0,
+      "total": 4,
+      "failed": 4
+    },
+    "COUNTER": {
+      "success_rate": 0.3333333333333333,
+      "passed": 1,
+      "total": 3,
+      "failed": 2
+    },
+    "CARD": {
+      "success_rate": 0.0,
+      "passed": 0,
+      "total": 1,
+      "failed": 1
+    },
+    "SELECT_LIST": {
+      "success_rate": 0.0,
+      "passed": 0,
+      "total": 2,
+      "failed": 2
+    },
+    "PAY": {
+      "success_rate": 0.0,
+      "passed": 0,
+      "total": 2,
+      "failed": 2
+    }
+  },
+  "section_avg_rate": 0.4696493982208268,
+  "raw_results_path": "results/raw/google_gemini-2.5-flash__ouroboros-cut/results.json",
+  "entry_id": "google_gemini-2.5-flash__ouroboros-cut"
+}

results/google_gemini-2.5-flash__ouroboros-full-evolving.json

@@ -0,0 +1,205 @@
+{
+  "model": "google/gemini-2.5-flash",
+  "harness": "ouroboros-full-evolving",
+  "provider": "openrouter",
+  "mock": "bench",
+  "finished": true,
+  "source_path": "/home/jovyan/emelyanov/agent-bench/tests/eval/google_gemini-2.5-flash/ouroboros-full-evolving/2026-06-23_22-23-55__lm-mpi-job-c08a12ec-da72-4952-ade5-c65163a4ee8e-mpimaster-0-1169-31430-1782253432/results.json",
+  "submitted_at": "2026-06-24T00:53:05.022768+00:00",
+  "metrics": {
+    "success_rate": 0.13725490196078433,
+    "passed_tasks": 7,
+    "failed_tasks": 44,
+    "total_tasks": 51,
+    "avg_duration_seconds": 175.00107505621716,
+    "avg_agent_steps": 11.352941176470589,
+    "avg_tokens_per_task": 727298.4090909091,
+    "total_tokens": 32001130,
+    "total_cost_usd": 5.1799219999999995,
+    "avg_cost_per_task_usd": 0.10156709803921568,
+    "pass_at_k": null,
+    "agent_completion_rate": 0.6862745098039216,
+    "agent_dab_agreement_rate": 0.45098039215686275,
+    "section_avg_rate": 0.18524332810047098
+  },
+  "sections": {
+    "shop": {
+      "label": "Маркет",
+      "success_rate": 0.15384615384615385,
+      "passed": 2,
+      "total": 13,
+      "tasks": {
+        "ecommerce_basket_and_favorites": false,
+        "ecommerce_basket_any_product": false,
+        "ecommerce_basket_multiple": false,
+        "ecommerce_basket_named_product": false,
+        "ecommerce_basket_named_product_02": true,
+        "ecommerce_basket_named_product_03": false,
+        "ecommerce_basket_price_10k": false,
+        "ecommerce_basket_price_1k": false,
+        "ecommerce_basket_price_constraint": false,
+        "ecommerce_basket_typo": true,
+        "ecommerce_favorites": false,
+        "ecommerce_favorites_02": false,
+        "ecommerce_favorites_any_product": false
+      }
+    },
+    "books": {
+      "label": "Книги",
+      "success_rate": 0.14285714285714285,
+      "passed": 1,
+      "total": 7,
+      "tasks": {
+        "digital_books_audio_basket": false,
+        "digital_books_author_cheapest_basket": false,
+        "digital_books_author_cheapest_favorites": false,
+        "digital_books_author_expensive_basket": false,
+        "digital_books_named_product_basket": true,
+        "digital_books_named_product_basket_02": false,
+        "digital_books_named_product_favorites": false
+      }
+    },
+    "grocery": {
+      "label": "Продукты",
+      "success_rate": 0.0,
+      "passed": 0,
+      "total": 4,
+      "tasks": {
+        "bench_grocery_navigation": false,
+        "grocery_basket_any_product": false,
+        "grocery_basket_named_product": false,
+        "grocery_basket_named_product_02": false
+      }
+    },
+    "rail": {
+      "label": "Поезда",
+      "success_rate": 0.0,
+      "passed": 0,
+      "total": 9,
+      "tasks": {
+        "rail_atomic_select_city_from": false,
+        "rail_atomic_select_city_to": false,
+        "rail_atomic_select_date": false,
+        "rail_atomic_submit_search": false,
+        "rail_book_and_pay": false,
+        "rail_book_and_pay_2_passengers": false,
+        "rail_book_to_cart": false,
+        "rail_book_to_cart_2_passengers": false,
+        "rail_book_to_cart_business": false
+      }
+    },
+    "hotels": {
+      "label": "Отели",
+      "success_rate": 0.0,
+      "passed": 0,
+      "total": 8,
+      "tasks": {
+        "hotel_atomic_select_city": false,
+        "hotel_atomic_select_end_date": false,
+        "hotel_atomic_select_guests": false,
+        "hotel_atomic_select_start_date": false,
+        "hotel_search_scenario": false,
+        "hotel_search_scenario_miami": false,
+        "hotel_search_scenario_typo": false,
+        "hotel_search_scenario_with_hotel": false
+      }
+    },
+    "gov": {
+      "label": "Услуги",
+      "success_rate": 0.0,
+      "passed": 0,
+      "total": 6,
+      "tasks": {
+        "government_download_certificate": false,
+        "government_download_certificate_techno": false,
+        "government_order_certificate": false,
+        "government_view_certificate": false,
+        "government_view_certificate_2025": false,
+        "government_view_certificate_casual": false
+      }
+    },
+    "hub": {
+      "label": "Главная",
+      "success_rate": 1.0,
+      "passed": 4,
+      "total": 4,
+      "tasks": {
+        "bench_hub_navigation_books": true,
+        "bench_hub_navigation_gov": true,
+        "bench_hub_navigation_rail": true,
+        "bench_hub_navigation_shop": true
+      }
+    }
+  },
+  "ui_badges": [
+    "NAV",
+    "BASKET",
+    "SELECT_LIST",
+    "SELECT_AC"
+  ],
+  "ui_classes": {
+    "BASKET": {
+      "success_rate": 0.13636363636363635,
+      "passed": 3,
+      "total": 22,
+      "failed": 19
+    },
+    "NAV": {
+      "success_rate": 1.0,
+      "passed": 4,
+      "total": 4,
+      "failed": 0
+    },
+    "FAV": {
+      "success_rate": 0.0,
+      "passed": 0,
+      "total": 5,
+      "failed": 5
+    },
+    "GOV": {
+      "success_rate": 0.0,
+      "passed": 0,
+      "total": 6,
+      "failed": 6
+    },
+    "SELECT_AC": {
+      "success_rate": 0.0,
+      "passed": 0,
+      "total": 2,
+      "failed": 2
+    },
+    "DATE": {
+      "success_rate": 0.0,
+      "passed": 0,
+      "total": 4,
+      "failed": 4
+    },
+    "COUNTER": {
+      "success_rate": 0.0,
+      "passed": 0,
+      "total": 3,
+      "failed": 3
+    },
+    "CARD": {
+      "success_rate": 0.0,
+      "passed": 0,
+      "total": 1,
+      "failed": 1
+    },
+    "SELECT_LIST": {
+      "success_rate": 0.0,
+      "passed": 0,
+      "total": 2,
+      "failed": 2
+    },
+    "PAY": {
+      "success_rate": 0.0,
+      "passed": 0,
+      "total": 2,
+      "failed": 2
+    }
+  },
+  "section_avg_rate": 0.18524332810047098,
+  "raw_results_path": "results/raw/google_gemini-2.5-flash__ouroboros-full-evolving/results.json",
+  "entry_id": "google_gemini-2.5-flash__ouroboros-full-evolving"
+}

results/google_gemini-2.5-flash__ouroboros-full-isolated.json

@@ -0,0 +1,205 @@
+{
+  "model": "google/gemini-2.5-flash",
+  "harness": "ouroboros-full-isolated",
+  "provider": "openrouter",
+  "mock": "bench",
+  "finished": true,
+  "source_path": "/home/jovyan/emelyanov/agent-bench/tests/eval/google_gemini-2.5-flash/ouroboros-full-isolated/2026-06-24_11-35-56__lm-mpi-job-b9be367c-3bfc-46dc-b65d-c9e762e3781f-mpimaster-0-462-24504-1782300952/results.json",
+  "submitted_at": "2026-06-24T16:02:38.468021+00:00",
+  "metrics": {
+    "success_rate": 0.09803921568627451,
+    "passed_tasks": 5,
+    "failed_tasks": 46,
+    "total_tasks": 51,
+    "avg_duration_seconds": 607.7635685334019,
+    "avg_agent_steps": 13.313725490196079,
+    "avg_tokens_per_task": 717052.7,
+    "total_tokens": 21511581,
+    "total_cost_usd": 3.039380000000001,
+    "avg_cost_per_task_usd": 0.05959568627450982,
+    "pass_at_k": null,
+    "agent_completion_rate": 0.5882352941176471,
+    "agent_dab_agreement_rate": 0.47058823529411764,
+    "section_avg_rate": 0.17857142857142858
+  },
+  "sections": {
+    "shop": {
+      "label": "Маркет",
+      "success_rate": 0.0,
+      "passed": 0,
+      "total": 13,
+      "tasks": {
+        "ecommerce_basket_and_favorites": false,
+        "ecommerce_basket_any_product": false,
+        "ecommerce_basket_multiple": false,
+        "ecommerce_basket_named_product": false,
+        "ecommerce_basket_named_product_02": false,
+        "ecommerce_basket_named_product_03": false,
+        "ecommerce_basket_price_10k": false,
+        "ecommerce_basket_price_1k": false,
+        "ecommerce_basket_price_constraint": false,
+        "ecommerce_basket_typo": false,
+        "ecommerce_favorites": false,
+        "ecommerce_favorites_02": false,
+        "ecommerce_favorites_any_product": false
+      }
+    },
+    "books": {
+      "label": "Книги",
+      "success_rate": 0.0,
+      "passed": 0,
+      "total": 7,
+      "tasks": {
+        "digital_books_audio_basket": false,
+        "digital_books_author_cheapest_basket": false,
+        "digital_books_author_cheapest_favorites": false,
+        "digital_books_author_expensive_basket": false,
+        "digital_books_named_product_basket": false,
+        "digital_books_named_product_basket_02": false,
+        "digital_books_named_product_favorites": false
+      }
+    },
+    "grocery": {
+      "label": "Продукты",
+      "success_rate": 0.25,
+      "passed": 1,
+      "total": 4,
+      "tasks": {
+        "bench_grocery_navigation": false,
+        "grocery_basket_any_product": true,
+        "grocery_basket_named_product": false,
+        "grocery_basket_named_product_02": false
+      }
+    },
+    "rail": {
+      "label": "Поезда",
+      "success_rate": 0.0,
+      "passed": 0,
+      "total": 9,
+      "tasks": {
+        "rail_atomic_select_city_from": false,
+        "rail_atomic_select_city_to": false,
+        "rail_atomic_select_date": false,
+        "rail_atomic_submit_search": false,
+        "rail_book_and_pay": false,
+        "rail_book_and_pay_2_passengers": false,
+        "rail_book_to_cart": false,
+        "rail_book_to_cart_2_passengers": false,
+        "rail_book_to_cart_business": false
+      }
+    },
+    "hotels": {
+      "label": "Отели",
+      "success_rate": 0.0,
+      "passed": 0,
+      "total": 8,
+      "tasks": {
+        "hotel_atomic_select_city": false,
+        "hotel_atomic_select_end_date": false,
+        "hotel_atomic_select_guests": false,
+        "hotel_atomic_select_start_date": false,
+        "hotel_search_scenario": false,
+        "hotel_search_scenario_miami": false,
+        "hotel_search_scenario_typo": false,
+        "hotel_search_scenario_with_hotel": false
+      }
+    },
+    "gov": {
+      "label": "Услуги",
+      "success_rate": 0.0,
+      "passed": 0,
+      "total": 6,
+      "tasks": {
+        "government_download_certificate": false,
+        "government_download_certificate_techno": false,
+        "government_order_certificate": false,
+        "government_view_certificate": false,
+        "government_view_certificate_2025": false,
+        "government_view_certificate_casual": false
+      }
+    },
+    "hub": {
+      "label": "Главная",
+      "success_rate": 1.0,
+      "passed": 4,
+      "total": 4,
+      "tasks": {
+        "bench_hub_navigation_books": true,
+        "bench_hub_navigation_gov": true,
+        "bench_hub_navigation_rail": true,
+        "bench_hub_navigation_shop": true
+      }
+    }
+  },
+  "ui_badges": [
+    "NAV",
+    "BASKET",
+    "SELECT_LIST",
+    "SELECT_AC"
+  ],
+  "ui_classes": {
+    "BASKET": {
+      "success_rate": 0.045454545454545456,
+      "passed": 1,
+      "total": 22,
+      "failed": 21
+    },
+    "NAV": {
+      "success_rate": 1.0,
+      "passed": 4,
+      "total": 4,
+      "failed": 0
+    },
+    "FAV": {
+      "success_rate": 0.0,
+      "passed": 0,
+      "total": 5,
+      "failed": 5
+    },
+    "GOV": {
+      "success_rate": 0.0,
+      "passed": 0,
+      "total": 6,
+      "failed": 6
+    },
+    "SELECT_AC": {
+      "success_rate": 0.0,
+      "passed": 0,
+      "total": 2,
+      "failed": 2
+    },
+    "DATE": {
+      "success_rate": 0.0,
+      "passed": 0,
+      "total": 4,
+      "failed": 4
+    },
+    "COUNTER": {
+      "success_rate": 0.0,
+      "passed": 0,
+      "total": 3,
+      "failed": 3
+    },
+    "CARD": {
+      "success_rate": 0.0,
+      "passed": 0,
+      "total": 1,
+      "failed": 1
+    },
+    "SELECT_LIST": {
+      "success_rate": 0.0,
+      "passed": 0,
+      "total": 2,
+      "failed": 2
+    },
+    "PAY": {
+      "success_rate": 0.0,
+      "passed": 0,
+      "total": 2,
+      "failed": 2
+    }
+  },
+  "section_avg_rate": 0.17857142857142858,
+  "raw_results_path": "results/raw/google_gemini-2.5-flash__ouroboros-full-isolated/results.json",
+  "entry_id": "google_gemini-2.5-flash__ouroboros-full-isolated"
+}

results/raw/deepseek_deepseek-v4-flash__browser-use/results.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ad648b18592c0967f0d20be1c458588472f52f9cbf1be7b6c13c8de692b62931
+size 15125158

results/raw/deepseek_deepseek-v4-flash__openhands/results.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:97926f7ef63e7d5243a25ea85954ac905d56a45a3757a3fea73bfb686b5ceee1
+size 1039593

results/raw/deepseek_deepseek-v4-flash__openmanus/results.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:db173307fac9c66172d7975ca2aabbf9a68029cec87e5a3ce77ad6931ba37722
+size 14604809

results/raw/deepseek_deepseek-v4-flash__ouroboros-cut/results.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a4926b0169e027f5326c0df10c0eb7ceebe8a6256e577b7bcb8454403846c4af
+size 14613326

results/raw/deepseek_deepseek-v4-flash__ouroboros-full-evolving/results.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:739ea1997acc50bd12448ec22acf220cded58afd8a7c3f3b21546ce6050ec95f
+size 810416

results/raw/deepseek_deepseek-v4-flash__ouroboros-full-isolated/results.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d6d0fcc67278d52cfa73704b18a96e38bfe82c1410d56e6fbae107315983276a
+size 1070384

results/raw/google_gemini-2.5-flash__browser-use/results.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2b65d8973d0217faf5f988f4ebbf789fa83b2ac4ff50f64549b58f11d8fa7adb
+size 15555946

results/raw/google_gemini-2.5-flash__openhands/results.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:556cc1d2a5eb86867aaa1ed5c76169a6308e3f9a69f93460f3d8a84dea55d709
+size 829166

results/raw/google_gemini-2.5-flash__openmanus/results.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:821e31b5c78216797d8e821e974c19bca72f18c11d432f28b1d79190dfe82c9a
+size 14381161

results/raw/google_gemini-2.5-flash__ouroboros-cut/results.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e5cb7fb93f9992dc4ed79226d707627bb12e9704ddbe7355f72c3c98f8e8faaa
+size 16236181

results/raw/google_gemini-2.5-flash__ouroboros-full-evolving/results.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0bfed04d3567844b17e98cb2872bf1f97385cbd38dfd7feab991bf1266408610
+size 684024

results/raw/google_gemini-2.5-flash__ouroboros-full-isolated/results.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0d4e1b519827d61dc194cbad7e250d49e5cc08bd259ca6076eaa2f8b7b33b618
+size 869193

src/bench_config.py

@@ -0,0 +1,53 @@
+"""Load Web Agent Bench leaderboard configuration."""
+
+from __future__ import annotations
+
+import json
+from functools import lru_cache
+from pathlib import Path
+from typing import Any
+
+
+def config_path() -> Path:
+    return Path(__file__).resolve().parent.parent / "bench_config.json"
+
+
+@lru_cache(maxsize=1)
+def load_bench_config() -> dict[str, Any]:
+    return json.loads(config_path().read_text(encoding="utf-8"))
+
+
+def section_order() -> list[str]:
+    config = load_bench_config()
+    core = [key for key, meta in config["sections"].items() if meta.get("core")]
+    rest = [key for key in config["sections"] if key not in core]
+    return core + rest
+
+
+def section_label(section_id: str) -> str:
+    return load_bench_config()["sections"][section_id]["label"]
+
+
+def section_description(section_id: str) -> str:
+    return load_bench_config()["sections"][section_id].get("description") or ""
+
+
+def taxonomy_order() -> list[str]:
+    return list(load_bench_config().get("ui_classes", {}).keys())
+
+
+def taxonomy_label(class_id: str) -> str:
+    meta = load_bench_config().get("ui_classes", {}).get(class_id) or {}
+    return str(meta.get("label") or class_id)
+
+
+def taxonomy_description(class_id: str) -> str:
+    meta = load_bench_config().get("ui_classes", {}).get(class_id) or {}
+    return str(meta.get("description") or "")
+
+
+def task_count() -> int:
+    config = load_bench_config()
+    return int(config["benchmark"].get("task_count") or sum(
+        len(meta["tasks"]) for meta in config["sections"].values()
+    ))

src/display.py

@@ -0,0 +1,47 @@
+"""Display helpers for Gradio tables."""
+
+from __future__ import annotations
+
+import html
+
+import pandas as pd
+
+
+def display_dataframe(df: pd.DataFrame | None) -> pd.DataFrame:
+    if df is None or df.empty:
+        return pd.DataFrame([{"—": "No data"}])
+    out = df.copy().astype(object)
+    return out.where(pd.notnull(out), "—")
+
+
+def _cell_text(value: object) -> str:
+    if value is None or (isinstance(value, float) and pd.isna(value)):
+        return "—"
+    if pd.isna(value):
+        return "—"
+    return str(value)
+
+
+def dataframe_to_html(df: pd.DataFrame | None, *, empty_note: str = "No data") -> str:
+    if df is None or df.empty:
+        return f'<p class="wab-detail-note">{html.escape(empty_note)}</p>'
+
+    headers = "".join(f"<th>{html.escape(str(col))}</th>" for col in df.columns)
+    body_rows = []
+    for row in df.itertuples(index=False, name=None):
+        cells = []
+        for col, value in zip(df.columns, row, strict=True):
+            text = _cell_text(value)
+            css_class = ""
+            if col in {"Success %", "Success"} and text not in {"—", "✓", "✗"}:
+                css_class = ' class="wab-metric-value"'
+            elif col in {"Tasks", "Metric", "Harness", "Model", "Class", "Task"}:
+                css_class = ' class="wab-mono-cell"'
+            cells.append(f"<td{css_class}>{html.escape(text)}</td>")
+        body_rows.append(f"<tr>{''.join(cells)}</tr>")
+    rows = "".join(body_rows)
+    return (
+        '<div class="wab-table-wrap wab-detail-table-wrap">'
+        f'<table class="wab-table wab-detail-table"><thead><tr>{headers}</tr></thead>'
+        f"<tbody>{rows}</tbody></table></div>"
+    )

src/export_results.py

@@ -0,0 +1,366 @@
+"""Export latest eval results.json into liderboard/results/ (leaderboard feed)."""
+
+from __future__ import annotations
+
+import argparse
+import json
+import os
+import re
+import shutil
+import sys
+from collections import defaultdict
+from pathlib import Path
+from typing import Any
+
+from src.bench_config import load_bench_config
+from src.models import LeaderboardEntry, SectionStats, reconcile_success_metrics
+
+
+def _sanitize_name(value: str) -> str:
+    safe = re.sub(r"[^\w.\-]+", "_", value.strip())
+    return safe or "unknown"
+
+
+def _space_root() -> Path:
+    return Path(__file__).resolve().parent.parent
+
+
+def _repo_root() -> Path:
+    return _space_root().parent
+
+
+def _ensure_repo_on_path() -> Path:
+    root = _repo_root()
+    root_str = str(root)
+    if root_str not in sys.path:
+        sys.path.insert(0, root_str)
+    return root
+
+
+def _domain_for_test(test: dict[str, Any], config: dict[str, Any]) -> str | None:
+    ui = test.get("ui_taxonomy") or {}
+    domain = ui.get("domain")
+    if domain:
+        return str(domain)
+    patterns = ui.get("ui_patterns") or {}
+    if patterns.get("domain"):
+        return str(patterns["domain"])
+    name = test.get("test_name") or ""
+    for section_id, meta in config["sections"].items():
+        for task in meta["tasks"]:
+            if task == name:
+                return section_id
+    if name.startswith("bench_hub"):
+        return "hub"
+    if name.startswith("digital_books"):
+        return "books"
+    if name.startswith("grocery") or name.startswith("bench_grocery"):
+        return "grocery"
+    if name.startswith("hotel"):
+        return "hotels"
+    if name.startswith("government"):
+        return "gov"
+    if name.startswith("rail"):
+        return "rail"
+    if name.startswith("ecommerce"):
+        return "shop"
+    return None
+
+
+def _section_avg_rate(sections: dict[str, SectionStats]) -> float | None:
+    rates = [stats.success_rate for stats in sections.values() if stats.success_rate is not None]
+    return (sum(rates) / len(rates)) if rates else None
+
+
+def raw_results_path_for(entry: LeaderboardEntry, root: Path | None = None) -> Path:
+    base = (root or _space_root()) / "results" / "raw"
+    return base / entry.entry_id / "results.json"
+
+
+def _ui_classes(payload: dict[str, Any]) -> dict[str, dict[str, Any]]:
+    stats = payload.get("ui_taxonomy_stats") or {}
+    by_primary = stats.get("by_primary") or {}
+    classes: dict[str, dict[str, Any]] = {}
+    for class_id, row in by_primary.items():
+        if not isinstance(row, dict):
+            continue
+        classes[str(class_id)] = {
+            "success_rate": row.get("success_rate"),
+            "passed": row.get("passed"),
+            "total": row.get("total"),
+            "failed": row.get("failed"),
+        }
+    return classes
+
+
+ui_classes_from_payload = _ui_classes
+
+
+def _ui_badges(payload: dict[str, Any], limit: int = 4) -> list[str]:
+    stats = payload.get("ui_taxonomy_stats") or {}
+    by_primary = stats.get("by_primary") or {}
+    ranked: list[tuple[float, str]] = []
+    for class_id, row in by_primary.items():
+        if not isinstance(row, dict):
+            continue
+        rate = row.get("success_rate")
+        if isinstance(rate, (int, float)):
+            ranked.append((float(rate), str(class_id)))
+    ranked.sort(reverse=True)
+    return [class_id for _, class_id in ranked[:limit]]
+
+
+def _cost_from_payload(payload: dict[str, Any]) -> tuple[float | None, float | None]:
+    _ensure_repo_on_path()
+    try:
+        from bench_eval.llm_cost import compute_results_cost
+
+        cost = compute_results_cost(payload)
+        return cost.total_cost_usd, cost.avg_cost_per_task_usd
+    except Exception:
+        return None, None
+
+
+def export_entry(payload: dict[str, Any], *, source_path: str | None = None) -> LeaderboardEntry:
+    config = load_bench_config()
+    run = payload.get("run") or {}
+    tests = payload.get("tests") or []
+
+    by_section: dict[str, dict[str, list[bool]]] = defaultdict(lambda: defaultdict(list))
+    for test in tests:
+        section_id = _domain_for_test(test, config)
+        if not section_id:
+            continue
+        task = test.get("test_name") or "unknown"
+        by_section[section_id][task].append(bool(test.get("success")))
+
+    sections: dict[str, SectionStats] = {}
+    for section_id, meta in config["sections"].items():
+        task_map = by_section.get(section_id, {})
+        tasks: dict[str, bool] = {}
+        for task in meta["tasks"]:
+            values = task_map.get(task)
+            if values:
+                tasks[task] = sum(values) / len(values) >= 0.5
+        passed = sum(1 for ok in tasks.values() if ok)
+        total = len(tasks)
+        success_rate = (passed / total) if total else None
+        sections[section_id] = SectionStats(
+            section_id=section_id,
+            label=meta["label"],
+            success_rate=success_rate,
+            passed=passed,
+            total=total,
+            tasks=tasks,
+        )
+
+    extended = run.get("extended_metrics") if isinstance(run.get("extended_metrics"), dict) else {}
+    pass_at_k = extended.get("pass_at_k")
+    pass_overall = pass_at_k.get("overall") if isinstance(pass_at_k, dict) else None
+    total_cost, avg_cost = _cost_from_payload(payload)
+
+    entry = LeaderboardEntry(
+        model=str(run.get("model") or "unknown"),
+        harness=str(run.get("agent_harness") or run.get("harness") or "unknown"),
+        provider=run.get("provider"),
+        mock=run.get("mock"),
+        finished=bool(run.get("finished", True)),
+        source_path=source_path,
+        submitted_at=run.get("finished_at") or run.get("started_at"),
+        success_rate=run.get("success_rate"),
+        passed_tasks=int(run.get("passed_tasks") or 0),
+        failed_tasks=int(run.get("failed_tasks") or 0),
+        total_tasks=int(run.get("total_tasks") or 0),
+        avg_duration_seconds=run.get("avg_duration_seconds"),
+        avg_agent_steps=run.get("avg_agent_steps"),
+        avg_tokens_per_task=run.get("avg_tokens_per_task"),
+        total_tokens=run.get("total_tokens"),
+        total_cost_usd=total_cost,
+        avg_cost_per_task_usd=avg_cost,
+        pass_at_k=float(pass_overall) if isinstance(pass_overall, (int, float)) else None,
+        agent_completion_rate=extended.get("agent_completion_rate"),
+        agent_dab_agreement_rate=extended.get("agent_dab_agreement_rate"),
+        sections=sections,
+        ui_badges=_ui_badges(payload),
+        ui_classes=_ui_classes(payload),
+        section_avg_rate=_section_avg_rate(sections),
+    )
+    reconcile_success_metrics(entry)
+    return entry
+
+
+def default_output_path(entry: LeaderboardEntry, output_dir: Path) -> Path:
+    model = _sanitize_name(entry.model)
+    harness = _sanitize_name(entry.harness)
+    return output_dir / f"{model}__{harness}.json"
+
+
+def clean_results_dir(output_dir: Path) -> None:
+    output_dir.mkdir(parents=True, exist_ok=True)
+    for child in output_dir.iterdir():
+        if child.name == ".gitkeep":
+            continue
+        if child.is_dir():
+            shutil.rmtree(child)
+        else:
+            child.unlink()
+
+
+def export_results_file(
+    results_json: Path,
+    *,
+    output_dir: Path,
+) -> Path:
+    payload = json.loads(results_json.read_text(encoding="utf-8"))
+    if not payload.get("run"):
+        raise ValueError(f"Not an eval results.json (missing run block): {results_json}")
+    entry = export_entry(payload, source_path=str(results_json.resolve()))
+    raw_path = raw_results_path_for(entry, output_dir.parent)
+    raw_path.parent.mkdir(parents=True, exist_ok=True)
+    shutil.copy2(results_json, raw_path)
+    entry.raw_results_path = str(raw_path.relative_to(output_dir.parent))
+    target = default_output_path(entry, output_dir)
+    target.parent.mkdir(parents=True, exist_ok=True)
+    target.write_text(json.dumps(entry.to_json(), ensure_ascii=False, indent=2) + "\n", encoding="utf-8")
+    print(f"Wrote {target}  ({results_json})")
+    return target
+
+
+def export_latest_runs(
+    base_dir: Path,
+    *,
+    output_dir: Path,
+    require_finished: bool = False,
+    min_total_tasks: int | None = None,
+) -> list[Path]:
+    _ensure_repo_on_path()
+    from bench_eval.aggregate_report import discover_latest_runs
+
+    discovered = discover_latest_runs(
+        base_dir,
+        require_finished=require_finished,
+        min_total_tasks=min_total_tasks,
+    )
+    clean_results_dir(output_dir)
+    written: list[Path] = []
+    for item in discovered:
+        written.append(export_results_file(item.results_path, output_dir=output_dir))
+    return written
+
+
+def backfill_summary_ui_classes(results_dir: Path | None = None) -> int:
+    """Write ui_classes into summary JSON from raw results when missing."""
+    root = results_dir or (_space_root() / "results")
+    if not root.is_dir():
+        return 0
+    updated = 0
+    for path in sorted(root.glob("*.json")):
+        payload = json.loads(path.read_text(encoding="utf-8"))
+        if payload.get("ui_classes"):
+            continue
+        if "metrics" not in payload:
+            continue
+        entry = LeaderboardEntry.from_json(payload)
+        raw_path = raw_results_path_for(entry, root.parent)
+        if not raw_path.is_file() and entry.raw_results_path:
+            raw_path = root.parent / entry.raw_results_path
+        if not raw_path.is_file():
+            continue
+        raw = json.loads(raw_path.read_text(encoding="utf-8"))
+        classes = _ui_classes(raw)
+        if not classes:
+            continue
+        entry.ui_classes = classes
+        path.write_text(json.dumps(entry.to_json(), ensure_ascii=False, indent=2) + "\n", encoding="utf-8")
+        updated += 1
+    return updated
+
+
+def main() -> None:
+    _ensure_repo_on_path()
+    from bench_eval.aggregate_report import FULL_BENCH_MIN_TASKS, discover_latest_runs
+
+    parser = argparse.ArgumentParser(
+        description=(
+            "Export latest eval results.json for every harness × model "
+            "into liderboard/results/ (summary + copied raw results.json)"
+        ),
+    )
+    parser.add_argument(
+        "--base-dir",
+        default=os.getenv("EVAL_RESULTS_BASE_DIR", "tests/eval"),
+        help="Base directory with eval runs (default: tests/eval)",
+    )
+    parser.add_argument(
+        "--output-dir",
+        default=None,
+        help="Leaderboard results directory (default: liderboard/results)",
+    )
+    parser.add_argument(
+        "--require-finished",
+        action="store_true",
+        help="Skip combinations whose latest run has finished=false",
+    )
+    parser.add_argument(
+        "--min-tasks-for-aggregate",
+        type=int,
+        default=FULL_BENCH_MIN_TASKS,
+        metavar="N",
+        help=(
+            "Include only runs with at least N tasks "
+            f"(default: {FULL_BENCH_MIN_TASKS}, i.e. full bench; smoke runs excluded). "
+            "Use 0 to include all runs."
+        ),
+    )
+    parser.add_argument(
+        "--backfill-ui-classes",
+        action="store_true",
+        help="Add ui_classes to existing summary JSON from copied raw results.json",
+    )
+    parser.add_argument(
+        "--list",
+        action="store_true",
+        help="Print discovered harness × model paths and exit without writing results",
+    )
+    args = parser.parse_args()
+
+    base_dir = Path(args.base_dir)
+    if not base_dir.is_absolute():
+        base_dir = _repo_root() / base_dir
+    min_total_tasks = args.min_tasks_for_aggregate or None
+    output_dir = Path(args.output_dir) if args.output_dir else _space_root() / "results"
+
+    if args.backfill_ui_classes:
+        count = backfill_summary_ui_classes(output_dir)
+        print(f"Backfilled ui_classes in {count} summary file(s) under {output_dir}")
+        return
+
+    if args.list:
+        discovered = discover_latest_runs(
+            base_dir,
+            require_finished=args.require_finished,
+            min_total_tasks=min_total_tasks,
+        )
+        if not discovered:
+            print(f"No results under {base_dir}")
+            raise SystemExit(1)
+        for item in sorted(discovered, key=lambda row: (row.harness, row.model_dir)):
+            status = "finished" if item.finished else "unfinished"
+            print(f"{item.harness}\t{item.model}\t{status}\t{item.results_path}")
+        return
+
+    written = export_latest_runs(
+        base_dir,
+        output_dir=output_dir,
+        require_finished=args.require_finished,
+        min_total_tasks=min_total_tasks,
+    )
+    if not written:
+        clean_results_dir(output_dir)
+        print(f"No runs exported from {base_dir} (min_tasks={min_total_tasks})")
+        raise SystemExit(1)
+    print(f"Exported {len(written)} harness × model combination(s) to {output_dir}")
+
+
+if __name__ == "__main__":
+    main()

src/load_entries.py

@@ -0,0 +1,93 @@
+"""Load leaderboard entries from liderboard/results."""
+
+from __future__ import annotations
+
+import json
+from pathlib import Path
+from typing import Any
+
+from src.bench_config import load_bench_config
+from src.export_results import raw_results_path_for, ui_classes_from_payload
+from src.models import LeaderboardEntry
+
+
+def space_root(root: Path | None = None) -> Path:
+    return root or Path(__file__).resolve().parent.parent
+
+
+def results_dir(root: Path | None = None) -> Path:
+    return space_root(root) / "results"
+
+
+def load_entry_file(path: Path) -> LeaderboardEntry | None:
+    payload = json.loads(path.read_text(encoding="utf-8"))
+    if "metrics" not in payload:
+        return None
+    entry = LeaderboardEntry.from_json(payload)
+    if not entry.raw_results_path:
+        raw_path = raw_results_path_for(entry, space_root(path.parent.parent))
+        if raw_path.is_file():
+            entry.raw_results_path = str(raw_path.relative_to(space_root(path.parent.parent)))
+    if not entry.ui_classes:
+        raw = load_raw_results(entry, space_root(path.parent.parent))
+        if raw:
+            entry.ui_classes = ui_classes_from_payload(raw)
+    return entry
+
+
+def load_entries(results_path: Path | None = None) -> list[LeaderboardEntry]:
+    root = results_path or results_dir()
+    if not root.is_dir():
+        return []
+    entries: list[LeaderboardEntry] = []
+    for path in sorted(root.glob("*.json")):
+        entry = load_entry_file(path)
+        if entry is not None:
+            entries.append(entry)
+    return sorted(
+        entries,
+        key=lambda entry: entry.success_rate if entry.success_rate is not None else -1,
+        reverse=True,
+    )
+
+
+def entry_label(entry: LeaderboardEntry) -> str:
+    return f"{entry.model} · {entry.harness}"
+
+
+def entry_choices(entries: list[LeaderboardEntry]) -> list[tuple[str, str]]:
+    return [(entry_label(entry), entry.entry_id) for entry in entries]
+
+
+def entry_by_id(entries: list[LeaderboardEntry], entry_id: str) -> LeaderboardEntry | None:
+    for entry in entries:
+        if entry.entry_id == entry_id:
+            return entry
+    return None
+
+
+def load_raw_results(entry: LeaderboardEntry, root: Path | None = None) -> dict[str, Any] | None:
+    base = space_root(root)
+    if entry.raw_results_path:
+        raw_path = base / entry.raw_results_path
+        if raw_path.is_file():
+            return json.loads(raw_path.read_text(encoding="utf-8"))
+    fallback = raw_results_path_for(entry, base)
+    if fallback.is_file():
+        return json.loads(fallback.read_text(encoding="utf-8"))
+    return None
+
+
+def sort_entries(entries: list[LeaderboardEntry], view_id: str) -> list[LeaderboardEntry]:
+    config = load_bench_config()
+    view = next((row for row in config["views"] if row["id"] == view_id), config["views"][0])
+    key = view["sort_key"]
+    reverse = not view.get("ascending", False)
+
+    def sort_value(entry: LeaderboardEntry) -> float:
+        value = entry.metric(key)
+        if value is None:
+            return float("inf") if view.get("ascending") else -1.0
+        return float(value)
+
+    return sorted(entries, key=sort_value, reverse=reverse)

src/models.py

@@ -0,0 +1,252 @@
+"""Leaderboard entry schema for Web Agent Bench."""
+
+from __future__ import annotations
+
+import re
+from dataclasses import dataclass, field
+from typing import Any
+
+
+def entry_id_for(model: str, harness: str) -> str:
+    safe = re.sub(r"[^\w.\-]+", "_", f"{model}__{harness}".strip())
+    return safe or "unknown"
+
+
+def reconcile_success_metrics(entry: LeaderboardEntry) -> None:
+    """Primary success rate is always passed_tasks / total_tasks (DAB)."""
+    section_passed = sum(section.passed for section in entry.sections.values())
+    section_total = sum(section.total for section in entry.sections.values())
+    if section_total > 0 and entry.total_tasks <= 0:
+        entry.total_tasks = section_total
+        entry.passed_tasks = section_passed
+        entry.failed_tasks = section_total - section_passed
+    if entry.total_tasks > 0:
+        entry.success_rate = entry.passed_tasks / entry.total_tasks
+        entry.failed_tasks = entry.total_tasks - entry.passed_tasks
+
+
+@dataclass
+class SectionStats:
+    section_id: str
+    label: str
+    success_rate: float | None
+    passed: int
+    total: int
+    tasks: dict[str, bool] = field(default_factory=dict)
+
+
+@dataclass
+class LeaderboardEntry:
+    model: str
+    harness: str
+    provider: str | None = None
+    mock: str | None = None
+    finished: bool = True
+    source_path: str | None = None
+    submitted_at: str | None = None
+
+    success_rate: float | None = None
+    passed_tasks: int = 0
+    failed_tasks: int = 0
+    total_tasks: int = 0
+
+    avg_duration_seconds: float | None = None
+    avg_agent_steps: float | None = None
+    avg_tokens_per_task: float | None = None
+    total_tokens: int | None = None
+    total_cost_usd: float | None = None
+    avg_cost_per_task_usd: float | None = None
+
+    pass_at_k: float | None = None
+    agent_completion_rate: float | None = None
+    agent_dab_agreement_rate: float | None = None
+
+    sections: dict[str, SectionStats] = field(default_factory=dict)
+    ui_badges: list[str] = field(default_factory=list)
+    ui_classes: dict[str, dict[str, Any]] = field(default_factory=dict)
+    section_avg_rate: float | None = None
+    raw_results_path: str | None = None
+
+    @property
+    def entry_id(self) -> str:
+        return entry_id_for(self.model, self.harness)
+
+    @property
+    def success_pct(self) -> float | None:
+        if self.success_rate is None:
+            return None
+        return round(self.success_rate * 100, 1)
+
+    @property
+    def full_bench_success_rate(self) -> float | None:
+        if self.total_tasks >= 51:
+            return self.success_rate
+        return None
+
+    @property
+    def value_score(self) -> float | None:
+        if self.success_rate is None or not self.total_cost_usd or self.total_cost_usd <= 0:
+            return None
+        return self.success_rate / self.total_cost_usd
+
+    def metric(self, name: str) -> float | None:
+        if name.startswith("domain:"):
+            section_id = name.split(":", 1)[1]
+            section = self.sections.get(section_id)
+            return section.success_rate if section else None
+        if name.startswith("taxonomy:"):
+            class_id = name.split(":", 1)[1]
+            row = self.ui_classes.get(class_id)
+            if not row:
+                return None
+            rate = row.get("success_rate")
+            return float(rate) if isinstance(rate, (int, float)) else None
+        if name == "success_rate":
+            return self.success_rate
+        if name == "full_bench_success_rate":
+            return self.full_bench_success_rate
+        if name == "avg_duration_seconds":
+            return self.avg_duration_seconds
+        if name == "total_cost_usd":
+            return self.total_cost_usd
+        if name == "value_score":
+            return self.value_score
+        return getattr(self, name, None)
+
+    def to_json(self) -> dict[str, Any]:
+        return {
+            "model": self.model,
+            "harness": self.harness,
+            "provider": self.provider,
+            "mock": self.mock,
+            "finished": self.finished,
+            "source_path": self.source_path,
+            "submitted_at": self.submitted_at,
+            "metrics": {
+                "success_rate": self.success_rate,
+                "passed_tasks": self.passed_tasks,
+                "failed_tasks": self.failed_tasks,
+                "total_tasks": self.total_tasks,
+                "avg_duration_seconds": self.avg_duration_seconds,
+                "avg_agent_steps": self.avg_agent_steps,
+                "avg_tokens_per_task": self.avg_tokens_per_task,
+                "total_tokens": self.total_tokens,
+                "total_cost_usd": self.total_cost_usd,
+                "avg_cost_per_task_usd": self.avg_cost_per_task_usd,
+                "pass_at_k": self.pass_at_k,
+                "agent_completion_rate": self.agent_completion_rate,
+                "agent_dab_agreement_rate": self.agent_dab_agreement_rate,
+                "section_avg_rate": self.section_avg_rate,
+            },
+            "sections": {
+                section_id: {
+                    "label": stats.label,
+                    "success_rate": stats.success_rate,
+                    "passed": stats.passed,
+                    "total": stats.total,
+                    "tasks": stats.tasks,
+                }
+                for section_id, stats in self.sections.items()
+            },
+            "ui_badges": self.ui_badges,
+            "ui_classes": self.ui_classes,
+            "section_avg_rate": self.section_avg_rate,
+            "raw_results_path": self.raw_results_path,
+            "entry_id": self.entry_id,
+        }
+
+    @classmethod
+    def from_json(cls, payload: dict[str, Any]) -> LeaderboardEntry:
+        if "metrics" in payload:
+            metrics = payload.get("metrics") or {}
+            sections_raw = payload.get("sections") or {}
+            sections = {
+                section_id: SectionStats(
+                    section_id=section_id,
+                    label=(data.get("label") or section_id),
+                    success_rate=data.get("success_rate"),
+                    passed=int(data.get("passed") or 0),
+                    total=int(data.get("total") or 0),
+                    tasks={k: bool(v) for k, v in (data.get("tasks") or {}).items()},
+                )
+                for section_id, data in sections_raw.items()
+            }
+            entry = cls(
+                model=str(payload.get("model") or "unknown"),
+                harness=str(payload.get("harness") or "unknown"),
+                provider=payload.get("provider"),
+                mock=payload.get("mock"),
+                finished=bool(payload.get("finished", True)),
+                source_path=payload.get("source_path"),
+                submitted_at=payload.get("submitted_at"),
+                success_rate=metrics.get("success_rate"),
+                passed_tasks=int(metrics.get("passed_tasks") or 0),
+                failed_tasks=int(metrics.get("failed_tasks") or 0),
+                total_tasks=int(metrics.get("total_tasks") or 0),
+                avg_duration_seconds=metrics.get("avg_duration_seconds"),
+                avg_agent_steps=metrics.get("avg_agent_steps"),
+                avg_tokens_per_task=metrics.get("avg_tokens_per_task"),
+                total_tokens=metrics.get("total_tokens"),
+                total_cost_usd=metrics.get("total_cost_usd"),
+                avg_cost_per_task_usd=metrics.get("avg_cost_per_task_usd"),
+                pass_at_k=metrics.get("pass_at_k"),
+                agent_completion_rate=metrics.get("agent_completion_rate"),
+                agent_dab_agreement_rate=metrics.get("agent_dab_agreement_rate"),
+                sections=sections,
+                ui_badges=list(payload.get("ui_badges") or []),
+                ui_classes=dict(payload.get("ui_classes") or {}),
+                section_avg_rate=metrics.get("section_avg_rate") or payload.get("section_avg_rate"),
+                raw_results_path=payload.get("raw_results_path"),
+            )
+            reconcile_success_metrics(entry)
+            return entry
+        entry = from_legacy_payload(payload)
+        reconcile_success_metrics(entry)
+        return entry
+
+
+def from_legacy_payload(payload: dict[str, Any]) -> LeaderboardEntry:
+    """Convert pre-refactor LIBRA-style JSON."""
+    from src.bench_config import load_bench_config
+
+    config = load_bench_config()
+    sections: dict[str, SectionStats] = {}
+    for section_id, meta in config["sections"].items():
+        block = payload.get(section_id) or payload.get(meta["domain"]) or {}
+        if not isinstance(block, dict):
+            continue
+        tasks = {
+            task: bool(block.get(task, 0) >= 0.5)
+            for task in meta["tasks"]
+            if isinstance(block.get(task), (int, float))
+        }
+        passed = sum(1 for ok in tasks.values() if ok)
+        total = len(tasks)
+        domain_total = block.get("domain_total_score")
+        success_rate = float(domain_total) if isinstance(domain_total, (int, float)) else None
+        if success_rate is None and total:
+            success_rate = passed / total
+        sections[section_id] = SectionStats(
+            section_id=section_id,
+            label=meta["label"],
+            success_rate=success_rate,
+            passed=passed,
+            total=total,
+            tasks=tasks,
+        )
+
+    section_avg_rate = payload.get("total_score")
+    if not isinstance(section_avg_rate, (int, float)) and sections:
+        rates = [s.success_rate for s in sections.values() if s.success_rate is not None]
+        section_avg_rate = sum(rates) / len(rates) if rates else None
+
+    entry = LeaderboardEntry(
+        model=str(payload.get("model") or "unknown"),
+        harness=str(payload.get("harness") or "unknown"),
+        total_tasks=sum(section.total for section in sections.values()),
+        passed_tasks=sum(section.passed for section in sections.values()),
+        sections=sections,
+        section_avg_rate=float(section_avg_rate) if isinstance(section_avg_rate, (int, float)) else None,
+    )
+    reconcile_success_metrics(entry)
+    return entry

src/tables.py

@@ -0,0 +1,100 @@
+"""Category leaderboard tables and helpers."""
+
+from __future__ import annotations
+
+import html
+
+import pandas as pd
+
+from src.bench_config import (
+    load_bench_config,
+    section_description,
+    section_label,
+    section_order,
+    taxonomy_description,
+    taxonomy_label,
+    taxonomy_order,
+)
+from src.models import LeaderboardEntry
+
+
+def section_leaderboard_dataframe(entries: list[LeaderboardEntry], section_id: str) -> pd.DataFrame:
+    rows = []
+    for entry in entries:
+        stats = entry.sections.get(section_id)
+        rate = stats.success_rate if stats else None
+        rows.append(
+            {
+                "Model": entry.model,
+                "Harness": entry.harness,
+                "Tasks": f"{stats.passed}/{stats.total}" if stats else None,
+                "Success %": round(rate * 100, 1) if rate is not None else None,
+            }
+        )
+    if not rows:
+        return pd.DataFrame()
+    return pd.DataFrame(rows).sort_values("Success %", ascending=False, na_position="last")
+
+
+def taxonomy_leaderboard_dataframe(entries: list[LeaderboardEntry], class_id: str) -> pd.DataFrame:
+    rows = []
+    for entry in entries:
+        row = entry.ui_classes.get(class_id) or {}
+        rate = row.get("success_rate")
+        passed = row.get("passed")
+        total = row.get("total")
+        rows.append(
+            {
+                "Model": entry.model,
+                "Harness": entry.harness,
+                "Tasks": f"{passed}/{total}" if passed is not None and total is not None else None,
+                "Success %": round(float(rate) * 100, 1) if isinstance(rate, (int, float)) else None,
+            }
+        )
+    if not rows:
+        return pd.DataFrame()
+    return pd.DataFrame(rows).sort_values("Success %", ascending=False, na_position="last")
+
+
+def category_legend_html(kind: str) -> str:
+    config = load_bench_config()
+    if kind == "section":
+        items = [
+            (section_label(section_id), section_description(section_id))
+            for section_id in section_order()
+        ]
+    else:
+        items = [
+            (taxonomy_label(class_id), taxonomy_description(class_id))
+            for class_id in taxonomy_order()
+        ]
+    spans = "".join(
+        f'<span class="wab-legend-item" title="{html.escape(desc)}">{html.escape(label)}</span>'
+        for label, desc in items
+    )
+    return f'<div class="wab-category-legend">{spans}</div>'
+
+
+def category_description_html(kind: str, item_id: str) -> str:
+    if kind == "section":
+        desc = section_description(item_id)
+        label = section_label(item_id)
+    else:
+        desc = taxonomy_description(item_id)
+        label = taxonomy_label(item_id)
+    body = f"<strong>{html.escape(label)}</strong>"
+    if desc:
+        body += f" — {html.escape(desc)}"
+    return f'<p class="wab-category-desc">{body}</p>'
+
+
+def category_description(kind: str, item_id: str) -> str:
+    if kind == "section":
+        desc = section_description(item_id)
+        label = section_label(item_id)
+    else:
+        desc = taxonomy_description(item_id)
+        label = taxonomy_label(item_id)
+    if not desc:
+        return f"**{label}**"
+    return f"**{label}** — {desc}"

src/ui.py

@@ -0,0 +1,455 @@
+"""PinchBench-inspired HTML UI for Web Agent Bench."""
+
+from __future__ import annotations
+
+import html
+from pathlib import Path
+from typing import Any
+
+from src.bench_config import load_bench_config, task_count
+from src.load_entries import sort_entries
+from src.models import LeaderboardEntry
+
+RANK_ICONS = ("🦞", "🦐", "🦀")
+
+_DETAIL_TAB_LABELS = {
+    "submission": "Submission",
+    "sections": "Sections",
+    "taxonomy": "UI taxonomy",
+    "metrics": "Metrics",
+    "about": "About",
+}
+
+
+def load_css() -> str:
+    return (Path(__file__).resolve().parent.parent / "assets" / "pinchbench.css").read_text(encoding="utf-8")
+
+
+def score_tier(pct: float | None) -> str:
+    if pct is None:
+        return "danger"
+    if pct >= 70:
+        return "success"
+    if pct >= 40:
+        return "warning"
+    return "danger"
+
+
+def fmt_pct(value: float | None) -> str:
+    if value is None:
+        return "—"
+    return f"{value * 100:.1f}%" if value <= 1 else f"{value:.1f}%"
+
+
+def fmt_pct_display(entry: LeaderboardEntry) -> str:
+    if entry.success_rate is None:
+        return "—"
+    return f"{entry.success_rate * 100:.1f}%"
+
+
+def fmt_duration(seconds: float | None) -> str:
+    if seconds is None:
+        return "—"
+    if seconds < 60:
+        return f"{seconds:.1f}s"
+    return f"{seconds / 60:.1f}m"
+
+
+def fmt_cost(value: float | None) -> str:
+    if value is None:
+        return "—"
+    if value <= 0:
+        return "FREE"
+    return f"${value:.3f}"
+
+
+def provider_from_model(model: str, provider: str | None = None) -> str:
+    if provider:
+        return provider
+    if "/" in model:
+        return model.split("/", 1)[0]
+    return model.split("-", 1)[0]
+
+
+def _badge_link(label: str, passed: object, total: object, css_class: str, title: str, nav: str) -> str:
+    stat_html = ""
+    try:
+        passed_i = int(passed) if passed is not None else None
+        total_i = int(total) if total is not None else None
+    except (TypeError, ValueError):
+        passed_i = total_i = None
+    if passed_i is not None and total_i is not None:
+        stat_html = f'<span class="wab-badge-stat">{passed_i}/{total_i}</span>'
+
+    return (
+        f'<span role="button" tabindex="0" class="wab-badge wab-badge-link {css_class}" '
+        f'title="{html.escape(title)}" data-wab-nav="{html.escape(nav)}">'
+        f"{html.escape(label)}{stat_html}</span>"
+    )
+
+
+def _badge(label: str, css_class: str, title: str) -> str:
+    return (
+        f'<span class="wab-badge {css_class}" title="{html.escape(title)}">'
+        f"{html.escape(label)}</span>"
+    )
+
+
+def _row_badges(entry: LeaderboardEntry) -> str:
+    config = load_bench_config()
+    badges: list[str] = []
+
+    ranked_sections = sorted(
+        (
+            (stats.success_rate, stats.section_id, stats.label, stats.passed, stats.total)
+            for stats in entry.sections.values()
+            if stats.success_rate is not None
+        ),
+        reverse=True,
+    )[:2]
+    for _, section_id, label, passed, total in ranked_sections:
+        desc = config["sections"].get(section_id, {}).get("description", label)
+        nav = f"sections:{section_id}"
+        badges.append(_badge_link(label, passed, total, "domain", str(desc), nav))
+
+    ranked_taxonomy = sorted(
+        (
+            (float(row.get("success_rate")), class_id)
+            for class_id, row in entry.ui_classes.items()
+            if isinstance(row.get("success_rate"), (int, float))
+        ),
+        reverse=True,
+    )[:2]
+    if not ranked_taxonomy and entry.ui_badges:
+        for class_id in entry.ui_badges[:2]:
+            meta = config.get("ui_classes", {}).get(class_id, {})
+            label = str(meta.get("label") or class_id)
+            row = entry.ui_classes.get(class_id) or {}
+            nav = f"taxonomy:{class_id}"
+            badges.append(
+                _badge_link(
+                    label,
+                    row.get("passed"),
+                    row.get("total"),
+                    "taxonomy",
+                    str(meta.get("description") or class_id),
+                    nav,
+                )
+            )
+    else:
+        for _, class_id in ranked_taxonomy:
+            meta = config.get("ui_classes", {}).get(class_id, {})
+            row = entry.ui_classes.get(class_id) or {}
+            label = str(meta.get("label") or class_id)
+            nav = f"taxonomy:{class_id}"
+            badges.append(
+                _badge_link(
+                    label,
+                    row.get("passed"),
+                    row.get("total"),
+                    "taxonomy",
+                    str(meta.get("description") or class_id),
+                    nav,
+                )
+            )
+
+    return "".join(badges)
+
+
+def top_section_badges(entry: LeaderboardEntry, limit: int = 2) -> list[str]:
+    ranked = [
+        (stats.success_rate, stats.label)
+        for stats in entry.sections.values()
+        if stats.success_rate is not None
+    ]
+    ranked.sort(reverse=True)
+    return [label for _, label in ranked[:limit]]
+
+
+def pick_winner(entries: list[LeaderboardEntry], metric: str, *, ascending: bool = False) -> LeaderboardEntry | None:
+    if not entries:
+        return None
+    values = [(entry, entry.metric(metric)) for entry in entries]
+    values = [(entry, value) for entry, value in values if value is not None]
+    if not values:
+        return None
+    return min(values, key=lambda pair: pair[1])[0] if ascending else max(values, key=lambda pair: pair[1])[0]
+
+
+def _quick_card(label: str, headline: str, entry: LeaderboardEntry, subtitle: str) -> str:
+    return f"""
+    <article class="wab-quick-card">
+      <div class="wab-quick-kicker">{html.escape(label)}</div>
+      <div class="wab-quick-score">{html.escape(headline)}</div>
+      <div class="wab-quick-model">{html.escape(entry.model)}</div>
+      <div class="wab-quick-meta">{html.escape(subtitle)}<br>{html.escape(entry.harness)}</div>
+    </article>
+    """
+
+
+def build_quick_picks(entries: list[LeaderboardEntry], view_id: str) -> str:
+    config = load_bench_config()
+    picks = config["quick_picks"].get(view_id, [])
+    cards: list[str] = []
+    for pick in picks:
+        metric = pick["metric"]
+        ascending = metric in {"avg_duration_seconds", "total_cost_usd"}
+        winner = pick_winner(entries, metric, ascending=ascending)
+        if winner is None:
+            continue
+        if metric == "success_rate":
+            headline = fmt_pct_display(winner)
+        elif metric == "avg_duration_seconds":
+            headline = fmt_duration(winner.avg_duration_seconds)
+        elif metric in {"total_cost_usd", "avg_cost_per_task_usd"}:
+            headline = fmt_cost(winner.total_cost_usd)
+        elif metric == "value_score":
+            value = winner.value_score
+            headline = f"{value:.1f}" if value is not None else "—"
+        elif metric.startswith("domain:"):
+            rate = winner.metric(metric)
+            headline = fmt_pct(rate)
+        else:
+            headline = fmt_pct_display(winner)
+        cards.append(_quick_card(pick["label"], headline, winner, pick["subtitle"]))
+    if not cards:
+        return ""
+    return f'<div class="wab-quick-grid">{"".join(cards)}</div>'
+
+
+def _success_row(idx: int, entry: LeaderboardEntry) -> str:
+    pct = entry.success_pct or 0
+    tier = score_tier(pct)
+    badges = _row_badges(entry)
+    return f"""
+    <tr>
+      <td class="wab-rank">{RANK_ICONS[idx] if idx < len(RANK_ICONS) else idx + 1}</td>
+      <td class="wab-model-cell">
+        <div class="wab-model-name">{html.escape(entry.model)}</div>
+        <div class="wab-badges">{badges}</div>
+      </td>
+      <td><span class="wab-harness">{html.escape(entry.harness)}</span></td>
+      <td>{entry.passed_tasks}/{entry.total_tasks}</td>
+      <td class="wab-score-cell">
+        <div class="wab-score-value wab-score-{tier}">{html.escape(fmt_pct_display(entry))}</div>
+        <div class="wab-bar"><span class="wab-bar-{tier}" style="width:{max(pct, 0):.1f}%"></span></div>
+      </td>
+    </tr>
+    """
+
+
+def _speed_row(idx: int, entry: LeaderboardEntry) -> str:
+    pct = entry.success_pct or 0
+    tier = score_tier(pct)
+    return f"""
+    <tr>
+      <td class="wab-rank">{RANK_ICONS[idx] if idx < len(RANK_ICONS) else idx + 1}</td>
+      <td class="wab-model-cell"><div class="wab-model-name">{html.escape(entry.model)}</div>
+        <div class="wab-badges"><span class="wab-badge">{html.escape(entry.harness)}</span></div></td>
+      <td>{html.escape(fmt_duration(entry.avg_duration_seconds))}</td>
+      <td>{entry.avg_agent_steps if entry.avg_agent_steps is not None else "—"}</td>
+      <td class="wab-score-cell">
+        <div class="wab-score-value wab-score-{tier}">{html.escape(fmt_pct_display(entry))}</div>
+        <div class="wab-bar"><span class="wab-bar-{tier}" style="width:{max(pct, 0):.1f}%"></span></div>
+      </td>
+    </tr>
+    """
+
+
+def _cost_row(idx: int, entry: LeaderboardEntry) -> str:
+    pct = entry.success_pct or 0
+    tier = score_tier(pct)
+    value = entry.value_score
+    return f"""
+    <tr>
+      <td class="wab-rank">{RANK_ICONS[idx] if idx < len(RANK_ICONS) else idx + 1}</td>
+      <td class="wab-model-cell"><div class="wab-model-name">{html.escape(entry.model)}</div>
+        <div class="wab-badges"><span class="wab-badge">{html.escape(entry.harness)}</span></div></td>
+      <td>{html.escape(fmt_cost(entry.total_cost_usd))}</td>
+      <td>{html.escape(fmt_cost(entry.avg_cost_per_task_usd))}</td>
+      <td>{f"{value:.1f}" if value is not None else "—"}</td>
+      <td class="wab-score-cell">
+        <div class="wab-score-value wab-score-{tier}">{html.escape(fmt_pct_display(entry))}</div>
+        <div class="wab-bar"><span class="wab-bar-{tier}" style="width:{max(pct, 0):.1f}%"></span></div>
+      </td>
+    </tr>
+    """
+
+
+def build_table(entries: list[LeaderboardEntry], view_id: str) -> str:
+    config = load_bench_config()
+    view = next(row for row in config["views"] if row["id"] == view_id)
+    sorted_entries = sort_entries(entries, view_id)
+
+    if view_id == "speed":
+        head = "<tr><th>#</th><th>Model</th><th>Avg time</th><th>Steps</th><th>Success %</th></tr>"
+        rows = [_speed_row(i, entry) for i, entry in enumerate(sorted_entries)]
+        subtitle = "Average task duration from results.json run.avg_duration_seconds."
+    elif view_id == "cost":
+        head = "<tr><th>#</th><th>Model</th><th>Total $</th><th>$/task</th><th>Value</th><th>Success %</th></tr>"
+        rows = [_cost_row(i, entry) for i, entry in enumerate(sorted_entries)]
+        subtitle = "LLM cost estimated via bench_eval.llm_cost. Value = success_rate / total_cost_usd."
+    else:
+        head = "<tr><th>#</th><th>Model</th><th>Harness</th><th>Tasks</th><th>Success %</th></tr>"
+        rows = [_success_row(i, entry) for i, entry in enumerate(sorted_entries)]
+        subtitle = (
+            "DAB success rate: passed_tasks / total_tasks where dab_check.all_passed is true. "
+            "See docs/EVAL_README.md."
+        )
+
+    body = "".join(rows) if rows else f'<tr><td colspan="6">No results yet. Run export_results.py.</td></tr>'
+    return f"""
+    <div class="wab-panel">
+      <div class="wab-panel-head">
+        <h3>{html.escape(view["emoji"])} {html.escape(view["label"])} by model</h3>
+        <p>{html.escape(subtitle)}</p>
+      </div>
+      <div class="wab-table-wrap">
+        <table class="wab-table"><thead>{head}</thead><tbody>{body}</tbody></table>
+      </div>
+      <div class="wab-footnote">
+        Tasks and grading criteria:
+        <span class="wab-links"><a href="https://github.com/ai-forever/agent-bench/tree/open_bench/tests/bench" target="_blank">tests/bench</a></span>
+        · Metrics:
+        <span class="wab-links"><a href="https://github.com/ai-forever/agent-bench/blob/open_bench/docs/EVAL_README.md" target="_blank">EVAL_README.md</a></span>
+      </div>
+    </div>
+    """
+
+
+def build_shell_top(entries: list[LeaderboardEntry]) -> str:
+    """Header + hero (view switcher is a separate Gradio control)."""
+    config = load_bench_config()
+    benchmark = config["benchmark"]
+    return f"""
+    <div class="wab-shell">
+      <div class="wab-topbar">
+        <div class="wab-brand">
+          <div class="wab-logo">🌐</div>
+          <div>
+            <h1>{html.escape(benchmark["name"])}</h1>
+            <p>Web agent benchmark leaderboard</p>
+          </div>
+        </div>
+        <div class="wab-stats">
+          <div class="wab-stat"><strong>{len(entries)}</strong> models</div>
+          <div class="wab-stat"><strong>{task_count()}</strong> tasks</div>
+          <div class="wab-stat"><strong>{len(config["sections"])}</strong> sections</div>
+        </div>
+      </div>
+
+      <section class="wab-hero">
+        <h2>The best models for your web agent.</h2>
+        <p>Compare harness × LLM combinations on anonymized mock websites with DAB condition checks.</p>
+        <p class="wab-hero-links">
+          <a href="#wab-detail-anchor" class="wab-jump-link">Sections &amp; UI taxonomy breakdown ↓</a>
+        </p>
+      </section>
+    </div>
+    """
+
+
+def build_board(entries: list[LeaderboardEntry], view_id: str = "success") -> str:
+    return f"""
+    <div class="wab-board">
+      <section>
+        <div class="wab-section-label">Quick Picks</div>
+        {build_quick_picks(entries, view_id)}
+      </section>
+      {build_table(entries, view_id)}
+    </div>
+    """
+
+
+def build_shell(entries: list[LeaderboardEntry], view_id: str = "success") -> str:
+    return build_shell_top(entries) + build_board(entries, view_id)
+
+
+def build_nav_js() -> str:
+    """Client-side navigation for badge deep links (passed to demo.load js=)."""
+    labels = ", ".join(f'"{key}": "{value}"' for key, value in _DETAIL_TAB_LABELS.items())
+    return f"""
+() => {{
+  if (window.__wabNavInit) return [];
+  window.__wabNavInit = true;
+
+  const TAB_LABELS = {{{labels}}};
+
+  function setPayload(value) {{
+    const root = document.getElementById("wab-nav-payload");
+    if (!root) return;
+    const input = root.querySelector("textarea, input");
+    if (!input) return;
+    const proto = input.tagName === "TEXTAREA" ? HTMLTextAreaElement.prototype : HTMLInputElement.prototype;
+    const setter = Object.getOwnPropertyDescriptor(proto, "value")?.set;
+    if (setter) setter.call(input, value);
+    else input.value = value;
+    input.dispatchEvent(new Event("input", {{ bubbles: true }}));
+  }}
+
+  window.wabOpenDetailTab = function (tabId) {{
+    const label = TAB_LABELS[tabId];
+    if (!label) return;
+    const root = document.getElementById("wab-detail-tabs") || document.querySelector(".wab-tabs");
+    if (!root) return;
+    const buttons = root.querySelectorAll("button");
+    for (const btn of buttons) {{
+      if ((btn.textContent || "").trim() === label) {{
+        btn.click();
+        return;
+      }}
+    }}
+    const order = ["submission", "sections", "taxonomy", "metrics", "about"];
+    const idx = order.indexOf(tabId);
+    if (idx >= 0 && buttons[idx]) buttons[idx].click();
+  }};
+
+  function clickNavButton() {{
+    const root = document.getElementById("wab-nav-go");
+    if (!root) return false;
+    const btn = root.querySelector("button");
+    if (!btn) return false;
+    btn.click();
+    return true;
+  }}
+
+  function scrollToDetail() {{
+    const anchor = document.getElementById("wab-detail-anchor");
+    if (anchor) anchor.scrollIntoView({{ behavior: "smooth", block: "start" }});
+  }}
+
+  window.wabNavigate = function (payload) {{
+    if (!payload) return;
+    window.__wabPendingNav = payload;
+    setPayload(payload);
+    const tab = payload.split(":", 1)[0] || "submission";
+    window.wabOpenDetailTab(tab);
+    scrollToDetail();
+    setTimeout(clickNavButton, 60);
+  }};
+
+  document.addEventListener("click", function (event) {{
+    const jump = event.target.closest(".wab-jump-link");
+    if (jump) {{
+      event.preventDefault();
+      scrollToDetail();
+      return;
+    }}
+    const badge = event.target.closest("[data-wab-nav]");
+    if (!badge) return;
+    event.preventDefault();
+    event.stopPropagation();
+    window.wabNavigate(badge.getAttribute("data-wab-nav"));
+  }});
+
+  document.addEventListener("keydown", function (event) {{
+    if (event.key !== "Enter" && event.key !== " ") return;
+    const badge = event.target.closest("[data-wab-nav]");
+    if (!badge) return;
+    event.preventDefault();
+    window.wabNavigate(badge.getAttribute("data-wab-nav"));
+  }});
+
+  return [];
+}}
+"""