diff --git a/app.py b/app.py
deleted file mode 100644
index 515d2e3a910569210ab8dd5c18a4918160bd6b37..0000000000000000000000000000000000000000
--- a/app.py
+++ /dev/null
@@ -1,548 +0,0 @@
-import json
-import os
-import sys
-from html import escape
-from pathlib import Path
-
-import gradio as gr
-
-# Gradio checks its own HTTP endpoint at startup; keep loopback off outbound proxies.
-os.environ["NO_PROXY"] = ",".join(filter(None, [os.getenv("NO_PROXY", os.getenv("no_proxy", "")),
- "localhost,127.0.0.1,0.0.0.0"]))
-
-SPACE_DIR = Path(__file__).resolve().parent
-if str(SPACE_DIR) not in sys.path:
- sys.path.insert(0, str(SPACE_DIR))
-
-from src.about import (
- CITATION_BUTTON_LABEL,
- CITATION_BUTTON_TEXT,
- INTRODUCTION_TEXT,
- LLM_BENCHMARKS_TEXT,
- TITLE,
-)
-from src.display.css_html_js import custom_css
-from src.utils import (
- build_leaderboard_summary_html,
- build_domain_status_md,
- dataset_section_title,
- get_grouped_dfs,
- group_datasets_by_domain,
- load_fixed_window_table,
- load_gift_aggregate_metadata_md,
- load_gift_aggregate_table,
- load_forecast_snapshots,
- forecast_snapshot_status,
- make_domain_pie_chart,
- make_forecast_plot,
- prepare_ranks_table,
- prepare_values_table,
-)
-
-RESULTS_PATH = os.getenv("TSFM_RESULTS_PATH", str(SPACE_DIR / "results"))
-from src.result_feed import ResultFeed
-RESULT_FEED = ResultFeed(RESULTS_PATH)
-RESULTS_PATH = RESULT_FEED.refresh()
-REFRESH_SECONDS = int(os.getenv("TSFM_REFRESH_SECONDS", "300"))
-DOMAIN_OUTPUT_OFFSET = 14
-MODEL_COLUMN_WIDTH = "180px"
-RANK_MODEL_COLUMN_WIDTH = MODEL_COLUMN_WIDTH
-RANK_INDEX_COLUMN_WIDTH = MODEL_COLUMN_WIDTH
-METRIC_COLUMN_WIDTH = "118px"
-MODEL_COLUMN_NAMES = {"model"}
-
-def _fixed_window_outputs(window: str) -> tuple[str, str]:
- return (
- _rank_table_html(load_fixed_window_table(RESULTS_PATH, window, "rank")),
- _rank_table_html(load_fixed_window_table(RESULTS_PATH, window, "overall")),
- )
-
-
-def _model_column_index(value) -> int | None:
- columns = getattr(value, "columns", None)
- if columns is None:
- return None
- for idx, column in enumerate(columns):
- if str(column).strip().lower() in MODEL_COLUMN_NAMES:
- return idx
- return None
-
-
-def _model_cell_html(value) -> str:
- text = "" if value is None else str(value)
- return (
- ''
- + escape(text)
- + "
"
- )
-
-
-def _rank_display_table(value):
- model_col_idx = _model_column_index(value)
- if model_col_idx is None:
- return value
- columns = getattr(value, "columns", None)
- if columns is None:
- return value
- table = value.copy()
- model_col = columns[model_col_idx]
- table[model_col] = table[model_col].map(_model_cell_html)
- return table
-
-
-def _dataframe_datatypes(value, *, rank_table: bool = False) -> list[str] | None:
- columns = getattr(value, "columns", None)
- if columns is None:
- return None
- try:
- column_count = len(columns)
- except TypeError:
- return None
- if column_count <= 0:
- return None
- datatypes = ["str"] * column_count
- model_col_idx = _model_column_index(value)
- if rank_table and model_col_idx is not None:
- datatypes[model_col_idx] = "html"
- return datatypes
-
-
-def _dataframe_column_widths(value, *, rank_table: bool = False) -> list[str] | None:
- columns = getattr(value, "columns", None)
- if columns is None:
- return None
- try:
- column_count = len(columns)
- except TypeError:
- return None
- if column_count <= 0:
- return None
- model_width = RANK_MODEL_COLUMN_WIDTH if rank_table else MODEL_COLUMN_WIDTH
- widths = [METRIC_COLUMN_WIDTH] * column_count
- model_col_idx = _model_column_index(value)
- if model_col_idx is None:
- model_col_idx = 0
- widths[model_col_idx] = model_width
- return widths
-
-
-def _leaderboard_dataframe(value, *, rank_table: bool = False, **kwargs):
- classes = kwargs.pop("elem_classes", []) or []
- if isinstance(classes, str):
- classes = [classes]
- if "leaderboard-dataframe" not in classes:
- classes.append("leaderboard-dataframe")
- if rank_table and "rank-leaderboard-dataframe" not in classes:
- classes.append("rank-leaderboard-dataframe")
- model_col_idx = _model_column_index(value)
- if model_col_idx is not None:
- model_col_class = f"leaderboard-model-col-{model_col_idx}"
- if model_col_class not in classes:
- classes.append(model_col_class)
- kwargs["elem_classes"] = classes
- kwargs["interactive"] = False
- kwargs["wrap"] = False
- kwargs["line_breaks"] = False
- kwargs.setdefault("datatype", _dataframe_datatypes(value, rank_table=rank_table))
- kwargs.setdefault("column_widths", _dataframe_column_widths(value, rank_table=rank_table))
- return gr.Dataframe(value=value, **kwargs)
-
-
-def _rank_table_html(table) -> str:
- columns = list(getattr(table, "columns", []))
- if not columns:
- return ''
-
- model_col_idx = _model_column_index(table)
- if model_col_idx is None:
- fixed_col_idx = 0
- fixed_col_width = RANK_INDEX_COLUMN_WIDTH
- else:
- fixed_col_idx = model_col_idx
- fixed_col_width = RANK_MODEL_COLUMN_WIDTH
-
- colgroup = []
- for idx in range(len(columns)):
- width = fixed_col_width if idx == fixed_col_idx else METRIC_COLUMN_WIDTH
- colgroup.append(f'')
-
- header_cells = []
- for idx, column in enumerate(columns):
- class_name = ' class="rank-fixed-column"' if idx == fixed_col_idx else ""
- header_cells.append(f"{escape(str(column))}")
-
- body_rows = []
- for _, row in table.iterrows():
- cells = []
- for idx, column in enumerate(columns):
- text = "" if row[column] is None else str(row[column])
- if idx == fixed_col_idx:
- cells.append(
- ''
- + escape(text)
- + " | "
- )
- else:
- cells.append(f"{escape(text)} | ")
- body_rows.append("" + "".join(cells) + "
")
-
- return (
- '"
- )
-
-
-def _dataset_tables(grouped: dict, datasets: list[str]) -> tuple[list, list]:
- values_tables = [
- prepare_values_table(grouped["dataset_values"].get(dataset, None))
- for dataset in datasets
- ]
- rank_tables = [
- _rank_table_html(prepare_ranks_table(grouped["dataset_ranks"].get(dataset, None)))
- for dataset in datasets
- ]
- return values_tables, rank_tables
-
-
-def _model_label(root: Path, model_slug: str) -> str:
- config_path = root / model_slug / "config.json"
- if not config_path.exists():
- return model_slug
- try:
- config = json.loads(config_path.read_text())
- return str(config.get("model") or model_slug)
- except Exception:
- return model_slug
-
-
-def _forecast_choices(root_dir: str, datasets: list[str]) -> dict[str, dict[str, dict]]:
- dataset_keys = {dataset.split("/")[0] for dataset in datasets}
- choices: dict[str, dict[str, dict]] = {dataset_key: {} for dataset_key in dataset_keys}
- root = Path(root_dir)
- if not root.exists():
- return choices
-
- model_slugs = {p.name for p in root.iterdir() if p.is_dir()}
- model_slugs.update(p.name for p in (root / "public_snapshots").glob("*") if p.is_dir())
- for slug in sorted(model_slugs):
- for snapshot_dataset, snapshot in load_forecast_snapshots(root_dir, slug).items():
- dataset_key = str(snapshot_dataset).split("/")[0]
- if dataset_key not in choices:
- continue
- model_label = str(snapshot.get("model") or _model_label(root, slug))
- choices[dataset_key][model_label] = snapshot
-
- return choices
-
-
-def _available_forecast_models(dataset_key: str) -> list[str]:
- return sorted(FORECAST_CHOICES.get(dataset_key, {}).keys())
-
-
-def _forecast_plot_for(dataset_key: str, model_label: str):
- if not model_label:
- return None
- return make_forecast_plot(FORECAST_CHOICES.get(dataset_key, {}).get(model_label))
-
-
-def _make_forecast_plotter(dataset_key: str):
- def update_forecast_plot(model_label: str):
- choices = _forecast_choices(RESULT_FEED.refresh(), DATASETS)
- snapshot = choices.get(dataset_key, {}).get(model_label)
- return make_forecast_plot(snapshot), forecast_snapshot_status(snapshot)
-
- return update_forecast_plot
-
-
-def refresh_forecasts(*selected_models) -> tuple:
- """Reload the published catalog, preserving each visitor's model selections."""
- catalog = _forecast_choices(RESULT_FEED.refresh(), DATASETS)
- outputs = []
- for index, key in enumerate(FORECAST_DATASET_KEYS):
- choices = sorted(catalog.get(key, {}))
- selected = selected_models[index] if index < len(selected_models) else None
- selected = selected if selected in choices else (choices[0] if choices else None)
- snapshot = catalog.get(key, {}).get(selected)
- outputs.extend([gr.update(choices=choices, value=selected, interactive=bool(choices)),
- make_forecast_plot(snapshot), forecast_snapshot_status(snapshot)])
- return tuple(outputs)
-
-
-def refresh_leaderboard() -> tuple:
- global RESULTS_PATH
- RESULTS_PATH = RESULT_FEED.refresh()
- grouped = get_grouped_dfs(RESULTS_PATH)
- # Gradio outputs are fixed at startup; missing/new datasets must not change
- # callback arity or shift another dataset's values into an existing table.
- datasets = DATASETS
- one_day_rank, one_day_metrics = _fixed_window_outputs("1d")
- seven_day_rank, seven_day_metrics = _fixed_window_outputs("7d")
- thirty_day_rank, thirty_day_metrics = _fixed_window_outputs("30d")
-
- outputs: list = [
- build_leaderboard_summary_html(RESULTS_PATH), # [0]
- prepare_values_table(grouped["overall_values"]), # [1]
- _rank_table_html(prepare_ranks_table(grouped["overall_ranks"])), # [2]
- make_domain_pie_chart(RESULTS_PATH), # [3]
- one_day_rank, # [4]
- one_day_metrics, # [5]
- seven_day_rank, # [6]
- seven_day_metrics, # [7]
- thirty_day_rank, # [8]
- thirty_day_metrics, # [9]
- load_gift_aggregate_metadata_md(RESULTS_PATH), # [10]
- load_gift_aggregate_table(RESULTS_PATH, "prediction_length"), # [11]
- load_gift_aggregate_table(RESULTS_PATH, "domain"), # [12]
- load_gift_aggregate_table(RESULTS_PATH, "frequency"), # [13]
- ]
-
- # [14 .. 14+D-1] per-domain status lines
- for domain, domain_datasets in DOMAIN_GROUPS.items():
- outputs.append(build_domain_status_md(domain, domain_datasets, RESULTS_PATH))
-
- value_tables, rank_tables = _dataset_tables(grouped, datasets)
- # [14+D .. 14+D+N-1] value tables
- outputs.extend(value_tables)
- # [14+D+N .. 14+D+2N-1] rank tables
- outputs.extend(rank_tables)
- return tuple(outputs)
-
-
-grouped_initial = get_grouped_dfs(RESULTS_PATH)
-DATASETS: list[str] = grouped_initial["datasets"]
-FORECAST_CHOICES = _forecast_choices(RESULTS_PATH, DATASETS)
-DOMAIN_GROUPS = group_datasets_by_domain(DATASETS, RESULTS_PATH)
-N_DOMAINS = len(DOMAIN_GROUPS)
-FORECAST_DATASET_KEYS = [ds.split("/")[0] for datasets in DOMAIN_GROUPS.values() for ds in datasets]
-INITIAL_OUTPUTS = refresh_leaderboard()
-
-demo = gr.Blocks(css=custom_css)
-with demo:
- gr.HTML(TITLE)
- gr.Markdown(INTRODUCTION_TEXT, elem_classes="markdown-text")
- summary_html = gr.HTML(INITIAL_OUTPUTS[0])
-
- dataset_value_dfs: list[gr.Dataframe] = []
- dataset_rank_dfs: list[gr.HTML] = []
- domain_status_mds: list[gr.Markdown] = []
- forecast_dropdowns = []
- forecast_outputs = []
- dataset_idx = 0
-
- with gr.Tabs(elem_classes="tab-buttons"):
- # ── Overall Tab ──────────────────────────────────────────────────────
- with gr.TabItem("Overall"):
- gr.Markdown(
- "**Latest evaluation snapshot**",
- elem_classes="markdown-text",
- )
- gr.Markdown(
- "**Overall summary.** Geometric mean of absolute metric values across the latest dataset snapshot. MSE/CRPS are lower-is-better; RankScore is higher-is-better.",
- elem_classes="markdown-text",
- )
- overall_values_df = _leaderboard_dataframe(
- value=INITIAL_OUTPUTS[1],
- label="Metric values and RankScore",
- )
- gr.Markdown(
- "**Latest snapshot ranks.** Per-metric ranks derived from the snapshot values above; lower rank is better.",
- elem_classes="markdown-text",
- )
- overall_ranks_df = gr.HTML(
- value=INITIAL_OUTPUTS[2],
- elem_classes="rank-html-output",
- )
- gr.Markdown(
- "**Evaluated timestamps by domain** · hover for dataset names",
- elem_classes="markdown-text",
- )
- domain_pie_plot = gr.Plot(
- value=INITIAL_OUTPUTS[3],
- label="Domain distribution",
- )
-
- gr.Markdown(
- "**Pairwise historical ranking.** Every Ranking table uses all shared releases completed through the current cutoff, balanced equally across datasets. The 24-hour, 7-day, and 30-day windows apply only to their Metrics tables. Official ranks require at least 30 shared releases, 5 shared datasets, 7 days of shared coverage, 3 eligible opponents, and membership in the main comparison component. Absolute metrics remain descriptive and do not determine rank.",
- elem_classes="markdown-text",
- )
- gr.Markdown(
- "**Last 24 hours** · cumulative historical ranking; metrics use releases whose target window ended in the trailing 24 hours.",
- elem_classes="markdown-text",
- )
- gr.Markdown("**Ranking**", elem_classes="markdown-text")
- last_24h_rank_df = gr.HTML(
- value=INITIAL_OUTPUTS[4],
- elem_classes="rank-html-output",
- )
- gr.Markdown("**Metrics**", elem_classes="markdown-text")
- last_24h_metrics_df = gr.HTML(
- value=INITIAL_OUTPUTS[5],
- elem_classes="rank-html-output",
- )
- gr.Markdown(
- "**Last 7 days** · cumulative historical ranking; metrics use a fixed trailing seven-day window.",
- elem_classes="markdown-text",
- )
- gr.Markdown("**Ranking**", elem_classes="markdown-text")
- last_7d_rank_df = gr.HTML(
- value=INITIAL_OUTPUTS[6],
- elem_classes="rank-html-output",
- )
- gr.Markdown("**Metrics**", elem_classes="markdown-text")
- last_7d_metrics_df = gr.HTML(
- value=INITIAL_OUTPUTS[7],
- elem_classes="rank-html-output",
- )
- gr.Markdown(
- "**Last 30 days** · cumulative historical ranking; metrics use a fixed trailing thirty-day window.",
- elem_classes="markdown-text",
- )
- gr.Markdown("**Ranking**", elem_classes="markdown-text")
- last_30d_rank_df = gr.HTML(
- value=INITIAL_OUTPUTS[8],
- elem_classes="rank-html-output",
- )
- gr.Markdown("**Metrics**", elem_classes="markdown-text")
- last_30d_metrics_df = gr.HTML(
- value=INITIAL_OUTPUTS[9],
- elem_classes="rank-html-output",
- )
-
- # ── GIFT-style grouped result tables ────────────────────────────────
- with gr.TabItem("GIFT-style Aggregates"):
- aggregate_metadata_md = gr.Markdown(
- INITIAL_OUTPUTS[10],
- elem_classes="markdown-text",
- )
- gr.Markdown(
- "MSE and CRPS are normalized per dataset configuration against Seasonal-Naive (1.0 = baseline). Rank is the mean per-configuration CRPS rank. Lower is better for all three metrics.",
- elem_classes="markdown-text",
- )
- with gr.Tabs():
- with gr.TabItem("Prediction Length"):
- prediction_length_aggregate_df = _leaderboard_dataframe(
- value=INITIAL_OUTPUTS[11],
- label="Results on TSFM_Bench aggregated by Prediction Length",
- )
- with gr.TabItem("Domain"):
- domain_aggregate_df = _leaderboard_dataframe(
- value=INITIAL_OUTPUTS[12],
- label="Results on TSFM_Bench aggregated by Domain",
- )
- with gr.TabItem("Frequency"):
- frequency_aggregate_df = _leaderboard_dataframe(
- value=INITIAL_OUTPUTS[13],
- label="Results on TSFM_Bench aggregated by Frequency",
- )
-
- # ── Domain Tabs ───────────────────────────────────────────────────────
- domain_idx = 0
- for domain, domain_datasets in DOMAIN_GROUPS.items():
- domain_label = f"{domain} ({len(domain_datasets)})"
- with gr.TabItem(domain_label):
- # Per-domain status line (data source, last eval, refresh intervals)
- domain_status_mds.append(
- gr.Markdown(
- INITIAL_OUTPUTS[DOMAIN_OUTPUT_OFFSET + domain_idx],
- elem_classes="markdown-text",
- )
- )
- domain_idx += 1
-
- for dataset in domain_datasets:
- gr.Markdown(
- dataset_section_title(dataset, RESULTS_PATH),
- elem_classes="dataset-heading",
- )
- dataset_value_dfs.append(
- _leaderboard_dataframe(
- value=INITIAL_OUTPUTS[DOMAIN_OUTPUT_OFFSET + N_DOMAINS + dataset_idx],
- show_label=False,
- )
- )
- gr.Markdown(
- dataset_section_title(dataset, RESULTS_PATH, ranks=True),
- elem_classes="dataset-heading",
- )
- dataset_rank_dfs.append(
- gr.HTML(
- value=INITIAL_OUTPUTS[DOMAIN_OUTPUT_OFFSET + N_DOMAINS + len(DATASETS) + dataset_idx],
- elem_classes="rank-html-output",
- )
- )
- dataset_key = dataset.split("/")[0]
- forecast_model_choices = _available_forecast_models(dataset_key)
- default_forecast_model = forecast_model_choices[0] if forecast_model_choices else None
- forecast_model_dropdown = gr.Dropdown(
- choices=forecast_model_choices,
- value=default_forecast_model,
- label="Model",
- interactive=bool(forecast_model_choices),
- )
- forecast_plot = gr.Plot(
- value=_forecast_plot_for(dataset_key, default_forecast_model),
- label="Forecast snapshot",
- )
- forecast_status = gr.Markdown(forecast_snapshot_status(
- FORECAST_CHOICES.get(dataset_key, {}).get(default_forecast_model)))
- forecast_dropdowns.append(forecast_model_dropdown)
- forecast_outputs.extend([forecast_model_dropdown, forecast_plot, forecast_status])
- forecast_model_dropdown.input(
- fn=_make_forecast_plotter(dataset_key),
- inputs=[forecast_model_dropdown],
- outputs=[forecast_plot, forecast_status],
- api_name=f"forecast_{dataset_key}",
- )
-
- dataset_idx += 1
-
- with gr.Accordion("About", open=False):
- gr.Markdown(LLM_BENCHMARKS_TEXT, elem_classes="markdown-text")
-
- with gr.Accordion("Citation", open=False):
- gr.Textbox(
- value=CITATION_BUTTON_TEXT,
- label=CITATION_BUTTON_LABEL,
- lines=12,
- show_copy_button=True,
- )
-
- refresh_outputs = [
- summary_html,
- overall_values_df,
- overall_ranks_df,
- domain_pie_plot,
- last_24h_rank_df,
- last_24h_metrics_df,
- last_7d_rank_df,
- last_7d_metrics_df,
- last_30d_rank_df,
- last_30d_metrics_df,
- aggregate_metadata_md,
- prediction_length_aggregate_df,
- domain_aggregate_df,
- frequency_aggregate_df,
- *domain_status_mds,
- *dataset_value_dfs,
- *dataset_rank_dfs,
- ]
-
- demo.load(refresh_leaderboard, inputs=[], outputs=refresh_outputs)
- demo.load(refresh_forecasts, inputs=forecast_dropdowns, outputs=forecast_outputs)
- refresh_timer = gr.Timer(value=REFRESH_SECONDS)
- refresh_timer.tick(refresh_leaderboard, inputs=[], outputs=refresh_outputs)
- refresh_timer.tick(refresh_forecasts, inputs=forecast_dropdowns, outputs=forecast_outputs)
-
-if __name__ == "__main__":
- demo.queue(default_concurrency_limit=20).launch()
diff --git a/cloud/acceptance.py b/cloud/acceptance.py
index 75c5697fe4493efe2ca85b8cc8813684f196ca44..a9b0d37ac4f1439d6a682e2f74987dbd4dcb5721 100644
--- a/cloud/acceptance.py
+++ b/cloud/acceptance.py
@@ -12,7 +12,7 @@ import tempfile
import time
ROOT = Path(__file__).resolve().parents[1]
-sys.path[:0] = [str(ROOT), str(ROOT / "src")]
+sys.path[:0] = [str(ROOT / "src"), str(ROOT)]
BRANCH = "acceptance"
diff --git a/cloud/worker.py b/cloud/worker.py
index 1eb1b2640ab24d43bd0ec7a4f7bbe3c456cbe5d4..de3afd2fa1e1caf7331a44e2758b9d2766b52ce3 100644
--- a/cloud/worker.py
+++ b/cloud/worker.py
@@ -4,7 +4,7 @@ from pathlib import Path
import sys
ROOT = Path(__file__).resolve().parents[1]
-sys.path[:0] = [str(ROOT), str(ROOT / "src")]
+sys.path[:0] = [str(ROOT / "src"), str(ROOT)]
os.chdir(ROOT)
from tsfm_bench.remote_state import RemoteState
diff --git a/constraints.txt b/constraints.txt
deleted file mode 100644
index 90c8500d61fa9a86c3643e8adf4eb279b3a021e4..0000000000000000000000000000000000000000
--- a/constraints.txt
+++ /dev/null
@@ -1,72 +0,0 @@
-# Tested deployment dependency constraints, generated 2026-09-05 on Python 3.11.
-# Regenerate in a clean environment after running CI and smoke tests.
-aiofiles==23.2.1
-annotated-doc==0.0.5
-annotated-types==0.8.0
-anyio==4.15.0
-certifi==2026.7.22
-charset-normalizer==3.5.1
-click==8.5.0
-contourpy==1.3.3; python_version >= "3.11"
-contourpy==1.3.2; python_version < "3.11"
-cycler==0.12.1
-fastapi==0.112.4
-ffmpy==1.0.0
-filelock==3.32.5
-fonttools==4.64.0
-fsspec==2026.7.0
-gluonts==0.16.3
-gradio==4.44.0
-gradio_client==1.3.0
-h11==0.16.0
-hf-xet==1.6.0
-httpcore==1.0.9
-httpx==0.28.1
-huggingface-hub==0.35.2
-idna==3.19
-importlib_resources==6.5.2
-iniconfig==2.3.0
-Jinja2==3.1.6
-kiwisolver==1.5.1
-markdown-it-py==4.2.0
-MarkupSafe==2.1.5
-matplotlib==3.9.4
-mdurl==0.1.2
-numpy==1.26.4
-orjson==3.12.0
-packaging==26.3
-pandas==2.2.3
-patsy==1.0.3
-pillow==10.4.0
-plotly==5.24.1
-pluggy==1.6.0
-pydantic==2.10.6
-pydantic_core==2.27.2
-pydub==0.25.1
-Pygments==2.21.0
-pyparsing==3.3.2
-pytest==8.4.2
-python-dateutil==2.9.0.post0
-python-dotenv==1.2.3
-python-multipart==0.0.32
-pytz==2026.3.post1
-PyYAML==6.0.3
-requests==2.34.2
-rich==15.0.0
-ruff==0.16.6
-scipy==1.11.4
-semantic-version==2.10.0
-shellingham==1.5.4
-six==1.17.0
-starlette==0.37.2
-statsmodels==0.14.6
-tenacity==9.1.4
-tomlkit==0.12.0
-toolz==0.12.1
-tqdm==4.70.0
-typer==0.27.2
-typing_extensions==4.16.0
-tzdata==2026.3
-urllib3==2.7.0
-uvicorn==0.30.6
-websockets==12.0
diff --git a/deployment.json b/deployment.json
index 8f934ff257401dffab1beafc226a3a70da9bf1b1..128a4607515dfd2e89635bc336bdc559fb90581b 100644
--- a/deployment.json
+++ b/deployment.json
@@ -1,5 +1,5 @@
{
"source_repository": "https://github.com/Thinkcat-Lab/LiveHouse-TS",
"source_branch": "main",
- "source_revision": "0d2efcf4a46bffe86cd0bf965a279c46cb2ecc27"
+ "source_revision": "8c9fc27f33a5bfd1b3e4c4d1ff5e7d8c2aa05361"
}
diff --git a/paper/livehouse_up_to_now_tables.tex b/paper/livehouse_up_to_now_tables.tex
deleted file mode 100644
index 3bfb9a1d914e0cd457a709c0855a79104f6ac7bf..0000000000000000000000000000000000000000
--- a/paper/livehouse_up_to_now_tables.tex
+++ /dev/null
@@ -1,129 +0,0 @@
-\begin{table*}[!t]
-\centering
-\caption{Per-dataset RMSE ($\downarrow$) on the online benchmark. In each column the best available model is in \textcolor{red}{\textbf{red bold}} and the second best is in \textcolor{blue}{\underline{blue underline}}.}
-\label{tab_perdataset_rmse}
-\resizebox{0.8 \linewidth}{!}{%
-\begin{tabular}{l rrrrrrrrrr}
-\toprule
-Model & BTC & PM2.5 & Quake & Potomac & Water & Wave & T2M & KSFO & Temp2m & Wiki \\
-\midrule
-Chronos-2 & \textcolor{red}{\textbf{483.614}} & 11.420 & \textcolor{blue}{\underline{2.830}} & 19.675 & 0.318 & 0.892 & 1.293 & 1.078 & 1.074 & \textcolor{red}{\textbf{8.420}} \\
-TiRex & 497.749 & \textcolor{red}{\textbf{10.843}} & 2.882 & \textcolor{red}{\textbf{16.958}} & 0.339 & 0.899 & 1.329 & 1.003 & \textcolor{blue}{\underline{1.025}} & 9.889 \\
-TimesFM-2.5 & \textcolor{blue}{\underline{485.407}} & \textcolor{blue}{\underline{11.156}} & \textcolor{red}{\textbf{2.812}} & \textcolor{blue}{\underline{17.502}} & \textcolor{red}{\textbf{0.234}} & 0.844 & 1.332 & 1.060 & \textcolor{red}{\textbf{0.988}} & \textcolor{blue}{\underline{8.711}} \\
-Toto-1.0 & 1.162e+03 & 40.693 & 5.314 & 58.600 & 0.539 & 1.093 & 1.578 & 1.208 & 1.835 & 96.250 \\
-Moirai-2.0 & 569.016 & 19.226 & 3.619 & 24.663 & 0.361 & 0.308 & \textcolor{red}{\textbf{0.909}} & \textcolor{blue}{\underline{0.568}} & 1.511 & 13.782 \\
-Chronos-Bolt & 802.832 & 20.359 & 3.917 & 28.495 & 0.634 & \textcolor{red}{\textbf{0.299}} & \textcolor{blue}{\underline{0.943}} & 0.578 & 1.630 & 33.431 \\
-TabPFN-TS & 716.183 & 19.586 & 3.934 & 27.656 & 0.518 & \textcolor{blue}{\underline{0.307}} & 1.141 & \textcolor{red}{\textbf{0.515}} & 1.538 & 20.472 \\
-Sundial & 1.344e+03 & 19.184 & 4.298 & 24.235 & \textcolor{blue}{\underline{0.294}} & 0.410 & 1.313 & 0.890 & 1.539 & 20.726 \\
-\midrule
-Moving-Average & 774.684 & 16.576 & 3.263 & 24.945 & 0.731 & 0.808 & 2.119 & 1.285 & 1.702 & 26.906 \\
-ETS & 543.172 & 18.520 & 3.176 & 23.910 & 0.865 & 0.979 & 5.053 & 0.964 & 7.677 & 28.427 \\
-ARIMA & 665.233 & 15.283 & 3.271 & 26.375 & 0.731 & 0.807 & 3.096 & 0.930 & 2.616 & 29.245 \\
-Seasonal-Naive & 847.250 & 18.334 & 3.983 & 28.209 & 0.755 & 0.873 & 1.500 & 1.329 & 1.373 & 53.966 \\
-\bottomrule
-\end{tabular}}
-\end{table*}
-
-\begin{table*}[!t]
-\centering
-\caption{Per-dataset MAPE ($\downarrow$) on the online benchmark. In each column the best available model is in \textcolor{red}{\textbf{red bold}} and the second best is in \textcolor{blue}{\underline{blue underline}}.}
-\label{tab_perdataset_mape}
-\resizebox{0.8 \linewidth}{!}{%
-\begin{tabular}{l rrrrrrrrrr}
-\toprule
-Model & BTC & PM2.5 & Quake & Potomac & Water & Wave & T2M & KSFO & Temp2m & Wiki \\
-\midrule
-Chronos-2 & 0.010 & 0.329 & 0.565 & 0.009 & 0.469 & 0.158 & 0.043 & 0.043 & 0.043 & 0.061 \\
-TiRex & 0.010 & 0.315 & 0.578 & \textcolor{red}{\textbf{0.007}} & 0.591 & 0.164 & 0.043 & 0.038 & \textcolor{blue}{\underline{0.042}} & \textcolor{blue}{\underline{0.057}} \\
-TimesFM-2.5 & 0.010 & \textcolor{blue}{\underline{0.314}} & \textcolor{blue}{\underline{0.493}} & \textcolor{blue}{\underline{0.008}} & \textcolor{red}{\textbf{0.283}} & \textcolor{red}{\textbf{0.145}} & 0.046 & 0.040 & 0.049 & 0.060 \\
-Toto-1.0 & 0.011 & 0.333 & 0.509 & 0.010 & \textcolor{blue}{\underline{0.426}} & 0.171 & 0.034 & 0.031 & \textcolor{red}{\textbf{0.040}} & 0.210 \\
-Moirai-2.0 & \textcolor{red}{\textbf{0.008}} & 0.341 & \textcolor{red}{\textbf{0.483}} & 0.009 & 0.437 & 0.178 & \textcolor{blue}{\underline{0.031}} & \textcolor{blue}{\underline{0.031}} & 0.045 & \textcolor{red}{\textbf{0.046}} \\
-Chronos-Bolt & 0.011 & 0.364 & 0.546 & 0.010 & 0.722 & 0.176 & \textcolor{red}{\textbf{0.030}} & 0.031 & 0.045 & 0.134 \\
-TabPFN-TS & 0.010 & 0.348 & 0.556 & 0.009 & 0.547 & 0.179 & 0.035 & \textcolor{red}{\textbf{0.026}} & 0.044 & 0.066 \\
-Sundial & 0.020 & \textcolor{red}{\textbf{0.282}} & 0.534 & 0.008 & 0.600 & 0.204 & 0.040 & 0.046 & 0.045 & 0.073 \\
-\midrule
-Moving-Average & 0.014 & 0.477 & 0.628 & 0.011 & 1.858 & 0.158 & 0.075 & 0.067 & 0.068 & 0.200 \\
-ETS & \textcolor{blue}{\underline{0.010}} & 0.713 & 0.585 & 0.010 & 1.340 & 0.213 & 0.176 & 0.044 & 0.321 & 0.203 \\
-ARIMA & 0.014 & 0.439 & 0.580 & 0.012 & 1.378 & \textcolor{blue}{\underline{0.155}} & 0.100 & 0.038 & 0.115 & 0.191 \\
-Seasonal-Naive & 0.015 & 0.518 & 0.641 & 0.011 & 1.863 & 0.166 & 0.052 & 0.065 & 0.052 & 0.361 \\
-\bottomrule
-\end{tabular}}
-\end{table*}
-
-\begin{table*}[!t]
-\centering
-\caption{Per-dataset CRPS ($\downarrow$) on the online benchmark. In each column the best available model is in \textcolor{red}{\textbf{red bold}} and the second best is in \textcolor{blue}{\underline{blue underline}}.}
-\label{tab_perdataset_crps}
-\resizebox{0.8 \linewidth}{!}{%
-\begin{tabular}{l rrrrrrrrrr}
-\toprule
-Model & BTC & PM2.5 & Quake & Potomac & Water & Wave & T2M & KSFO & Temp2m & Wiki \\
-\midrule
-Chronos-2 & 0.112 & 0.445 & 0.345 & 0.139 & 0.166 & 0.445 & 0.059 & 0.438 & 0.092 & 0.122 \\
-TiRex & 0.118 & 0.436 & 0.347 & 0.139 & 0.174 & 0.428 & 0.056 & 0.408 & 0.100 & 0.118 \\
-TimesFM-2.5 & 0.136 & 0.518 & 0.481 & 0.172 & \textcolor{red}{\textbf{0.147}} & 0.505 & 0.071 & 0.499 & 0.111 & 0.161 \\
-Toto-1.0 & \textcolor{blue}{\underline{0.010}} & \textcolor{red}{\textbf{0.245}} & \textcolor{blue}{\underline{0.273}} & 0.010 & 0.186 & \textcolor{blue}{\underline{0.128}} & 0.027 & \textcolor{blue}{\underline{0.027}} & \textcolor{red}{\textbf{0.036}} & 0.132 \\
-Moirai-2.0 & \textcolor{red}{\textbf{0.008}} & 0.260 & \textcolor{red}{\textbf{0.262}} & \textcolor{blue}{\underline{0.008}} & 0.207 & \textcolor{red}{\textbf{0.124}} & \textcolor{red}{\textbf{0.027}} & 0.028 & 0.040 & \textcolor{red}{\textbf{0.050}} \\
-Chronos-Bolt & 0.010 & 0.272 & 0.278 & 0.010 & 0.365 & 0.129 & \textcolor{blue}{\underline{0.027}} & 0.028 & 0.041 & 0.111 \\
-TabPFN-TS & 0.010 & 0.334 & 0.371 & 0.009 & 0.339 & 0.171 & 0.036 & \textcolor{red}{\textbf{0.026}} & 0.045 & 0.063 \\
-Sundial & 0.018 & \textcolor{blue}{\underline{0.257}} & 0.296 & \textcolor{red}{\textbf{0.007}} & \textcolor{blue}{\underline{0.151}} & 0.164 & 0.033 & 0.039 & \textcolor{blue}{\underline{0.040}} & \textcolor{blue}{\underline{0.054}} \\
-\midrule
-Moving-Average & 0.107 & 0.490 & 0.348 & 0.132 & 0.597 & 0.451 & 0.106 & 0.435 & 0.178 & 0.321 \\
-ETS & 0.099 & 0.829 & 0.348 & 0.111 & 0.575 & 0.516 & 0.279 & 0.301 & 0.903 & 0.323 \\
-ARIMA & 0.117 & 0.457 & 0.338 & 0.117 & 0.514 & 0.458 & 0.174 & 0.345 & 0.356 & 0.284 \\
-Seasonal-Naive & 0.100 & 0.583 & 0.556 & 0.114 & 0.621 & 0.569 & 0.073 & 0.416 & 0.136 & 0.731 \\
-\bottomrule
-\end{tabular}}
-\end{table*}
-
-\begin{table*}[!t]
-\centering
-\caption{Per-dataset Temporal Stability ($\downarrow$) on the online benchmark, computed from release-level MSE histories. Cells with fewer than two releases are shown as --. In each column the best available model is in \textcolor{red}{\textbf{red bold}} and the second best is in \textcolor{blue}{\underline{blue underline}}.}
-\label{tab_perdataset_stability}
-\resizebox{0.8 \linewidth}{!}{%
-\begin{tabular}{l rrrrrrrrrr}
-\toprule
-Model & BTC & PM2.5 & Quake & Potomac & Water & Wave & T2M & KSFO & Temp2m & Wiki \\
-\midrule
-Chronos-2 & 7.329e+05 & 494.984 & 8.908 & 2.860e+03 & 0.271 & 3.756 & 1.756 & 27.049 & 1.953 & 129.300 \\
-TiRex & 6.805e+05 & 463.452 & 8.575 & \textcolor{blue}{\underline{2.414e+03}} & 0.310 & 3.479 & 1.653 & 27.449 & 1.788 & 167.781 \\
-TimesFM-2.5 & 6.407e+05 & 499.814 & 8.574 & 2.685e+03 & \textcolor{blue}{\underline{0.137}} & 3.345 & 1.964 & 27.191 & 1.673 & 133.699 \\
-Toto-1.0 & 3.767e+06 & 8.411e+03 & 26.727 & 9.212e+03 & 0.653 & 7.889 & 7.376 & 3.933 & 6.619 & 8.228e+03 \\
-Moirai-2.0 & \textcolor{blue}{\underline{4.540e+05}} & \textcolor{blue}{\underline{185.713}} & \textcolor{blue}{\underline{5.250}} & 2.734e+03 & 0.192 & 0.088 & \textcolor{red}{\textbf{0.375}} & \textcolor{blue}{\underline{0.737}} & 1.538 & \textcolor{blue}{\underline{64.521}} \\
-Chronos-Bolt & 6.818e+05 & \textcolor{red}{\textbf{169.701}} & 5.282 & 3.085e+03 & 0.358 & \textcolor{red}{\textbf{0.052}} & 1.235 & 0.981 & 2.238 & 238.712 \\
-TabPFN-TS & \textcolor{red}{\textbf{4.069e+05}} & 307.897 & 5.972 & 2.749e+03 & 0.288 & \textcolor{blue}{\underline{0.081}} & 0.928 & \textcolor{red}{\textbf{0.701}} & \textcolor{red}{\textbf{1.273}} & 68.714 \\
-Sundial & 1.307e+06 & 645.449 & \textcolor{red}{\textbf{5.049}} & \textcolor{red}{\textbf{2.020e+03}} & \textcolor{red}{\textbf{0.117}} & 0.102 & \textcolor{blue}{\underline{0.666}} & 2.253 & \textcolor{blue}{\underline{1.509}} & \textcolor{red}{\textbf{26.124}} \\
-\midrule
-Moving-Average & 9.664e+05 & 571.085 & 9.040 & 3.870e+03 & 1.352 & 3.555 & 5.080 & 26.020 & 2.855 & 1.196e+03 \\
-ETS & 9.012e+05 & 7.217e+03 & 8.041 & 2.763e+03 & 1.431 & 4.353 & 29.082 & 26.072 & 209.632 & 1.770e+03 \\
-ARIMA & 7.312e+05 & 943.393 & 8.244 & 3.046e+03 & 1.027 & 3.510 & 11.488 & 25.728 & 10.261 & 1.427e+03 \\
-Seasonal-Naive & 1.002e+06 & 644.216 & 12.657 & 4.223e+03 & 1.389 & 3.750 & 2.021 & 25.672 & 1.829 & 4.084e+03 \\
-\bottomrule
-\end{tabular}}
-\end{table*}
-
-\begin{table*}[!t]
-\centering
-\caption{Per-dataset Improvement ($\downarrow$; more negative is better) on the online benchmark, computed as Kendall $\tau$ on release-level MSE histories. Cells with fewer than two releases are shown as --. In each column the best available model is in \textcolor{red}{\textbf{red bold}} and the second best is in \textcolor{blue}{\underline{blue underline}}.}
-\label{tab_perdataset_improvement}
-\resizebox{0.8 \linewidth}{!}{%
-\begin{tabular}{l rrrrrrrrrr}
-\toprule
-Model & BTC & PM2.5 & Quake & Potomac & Water & Wave & T2M & KSFO & Temp2m & Wiki \\
-\midrule
-Chronos-2 & 0.560 & 0.588 & 0.398 & 0.224 & -0.033 & -0.500 & -0.138 & -0.025 & 0.527 & 0.704 \\
-TiRex & 0.565 & 0.545 & 0.429 & 0.242 & -0.010 & \textcolor{blue}{\underline{-0.504}} & 0.144 & -0.016 & 0.547 & 0.873 \\
-TimesFM-2.5 & 0.531 & 0.548 & 0.431 & 0.247 & -0.051 & -0.428 & -0.002 & -0.034 & 0.585 & 0.479 \\
-Toto-1.0 & -0.309 & \textcolor{blue}{\underline{0.059}} & -0.273 & \textcolor{red}{\textbf{0.072}} & 0.022 & -0.085 & 0.344 & 0.123 & 0.022 & 1.000 \\
-Moirai-2.0 & \textcolor{blue}{\underline{-0.355}} & \textcolor{red}{\textbf{-0.013}} & -0.351 & 0.132 & -0.028 & \textcolor{red}{\textbf{-0.639}} & -0.048 & \textcolor{red}{\textbf{-0.195}} & \textcolor{blue}{\underline{-0.385}} & \textcolor{red}{\textbf{-1.000}} \\
-Chronos-Bolt & 0.090 & 0.307 & \textcolor{blue}{\underline{-0.371}} & 0.151 & 0.014 & -0.098 & -0.048 & \textcolor{blue}{\underline{-0.143}} & \textcolor{red}{\textbf{-0.429}} & 1.000 \\
-TabPFN-TS & \textcolor{red}{\textbf{-0.505}} & 0.200 & \textcolor{red}{\textbf{-0.415}} & 0.121 & 0.000 & -0.237 & 0.023 & -0.086 & -0.114 & 0.333 \\
-Sundial & 0.503 & 0.066 & -0.271 & \textcolor{blue}{\underline{0.084}} & -0.080 & -0.221 & \textcolor{red}{\textbf{-0.505}} & -0.134 & -0.096 & \textcolor{blue}{\underline{0.000}} \\
-\midrule
-Moving-Average & 0.458 & 0.520 & 0.351 & 0.176 & \textcolor{blue}{\underline{-0.321}} & -0.395 & -0.391 & -0.081 & 0.525 & 0.423 \\
-ETS & 0.345 & 0.533 & 0.374 & 0.221 & -0.267 & -0.476 & -0.379 & 0.054 & 0.296 & 0.366 \\
-ARIMA & 0.393 & 0.555 & 0.414 & 0.191 & -0.313 & -0.464 & \textcolor{blue}{\underline{-0.456}} & -0.079 & 0.364 & 0.592 \\
-Seasonal-Naive & 0.484 & 0.459 & 0.381 & 0.134 & \textcolor{red}{\textbf{-0.329}} & -0.419 & -0.036 & -0.074 & 0.491 & 0.477 \\
-\bottomrule
-\end{tabular}}
-\end{table*}
diff --git a/requirements.txt b/requirements.txt
deleted file mode 100644
index 79066974868f0eb2a13f8708bfb47f71613386bd..0000000000000000000000000000000000000000
--- a/requirements.txt
+++ /dev/null
@@ -1,64 +0,0 @@
-# Self-contained: HF mounts only this file during dependency installation.
-# Validated Python 3.11 runtime; refresh these pins after deployment tests.
-aiofiles==23.2.1
-annotated-doc==0.0.5
-annotated-types==0.8.0
-anyio==4.15.0
-certifi==2026.7.22
-charset-normalizer==3.5.1
-click==8.5.0
-contourpy==1.3.3; python_version >= "3.11"
-contourpy==1.3.2; python_version <