Update online eval results (6 models)

app.py

@@ -12,7 +12,12 @@ from src.about import (
     TITLE,
 )
 from src.display.css_html_js import custom_css
-from src.utils import get_grouped_dfs, load_baseline_name, prepare_leaderboard_table
+from src.utils import (
+    get_grouped_dfs,
+    load_baseline_name,
+    prepare_ranks_table,
+    prepare_values_table,
+)
 
 RESULTS_PATH = os.getenv("TSFM_RESULTS_PATH", "results")
 REFRESH_SECONDS = int(os.getenv("TSFM_REFRESH_SECONDS", "300"))
@@ -25,22 +30,13 @@ def load_online_status(results_path: str = RESULTS_PATH) -> dict:
     return json.loads(status_path.read_text())
 
 
-def build_overall_table(result_df):
-    table = result_df.rename(columns={"model": "Model"})
-    if "MASE_Rank" in table.columns:
-        table = table.sort_values(by=["MASE_Rank"])
-
-    display_cols = ["Model", "MASE", "MASE_Rank", "CRPS", "CRPS_Rank"]
-    cols = [col for col in display_cols if col in table.columns]
-    return table[cols].reset_index(drop=True)
-
-
 def refresh_leaderboard():
     grouped = get_grouped_dfs(RESULTS_PATH)
     status = load_online_status(RESULTS_PATH)
     baseline = load_baseline_name(RESULTS_PATH)
 
-    overall = build_overall_table(prepare_leaderboard_table(grouped["overall"]))
+    values_table = prepare_values_table(grouped["overall_values"])
+    ranks_table = prepare_ranks_table(grouped["overall_ranks"])
 
     finished = status.get("finished_at", "n/a")
     data_source = status.get("data_source", "n/a")
@@ -49,10 +45,10 @@ def refresh_leaderboard():
         f"Baseline: `{baseline}` · Last eval: `{finished}` · "
         f"Auto-refresh every {REFRESH_SECONDS}s"
     )
-    return overall, status_line
+    return values_table, ranks_table, status_line
 
 
-overall_table, status_md = refresh_leaderboard()
+values_table, ranks_table, status_md = refresh_leaderboard()
 
 demo = gr.Blocks(css=custom_css)
 with demo:
@@ -63,11 +59,18 @@ with demo:
     with gr.Row():
         refresh_btn = gr.Button("Refresh now", variant="secondary")
 
-    overall_df = gr.Dataframe(
-        value=overall_table,
+    values_df = gr.Dataframe(
+        value=values_table,
         interactive=False,
         wrap=True,
-        label="Overall Leaderboard (normalized vs baseline, lower is better)",
+        label="Metric values (normalized vs baseline, lower is better)",
+    )
+
+    ranks_df = gr.Dataframe(
+        value=ranks_table,
+        interactive=False,
+        wrap=True,
+        label="Ranks (average per-dataset rank, lower is better)",
     )
 
     with gr.Accordion("About", open=False):
@@ -81,8 +84,10 @@ with demo:
             show_copy_button=True,
         )
 
-    refresh_btn.click(refresh_leaderboard, outputs=[overall_df, status])
-    demo.load(refresh_leaderboard, outputs=[overall_df, status])
-    gr.Timer(value=REFRESH_SECONDS).tick(refresh_leaderboard, outputs=[overall_df, status])
+    refresh_btn.click(refresh_leaderboard, outputs=[values_df, ranks_df, status])
+    demo.load(refresh_leaderboard, outputs=[values_df, ranks_df, status])
+    gr.Timer(value=REFRESH_SECONDS).tick(
+        refresh_leaderboard, outputs=[values_df, ranks_df, status]
+    )
 
 demo.queue(default_concurrency_limit=20).launch()

results/chronos_2/online_meta.json

@@ -1,23 +1,23 @@
 {
     "model": "Chronos-2",
     "api_model_id": "amazon/chronos-2",
-    "evaluated_at": "2026-05-28T09:58:41.598566+00:00",
+    "evaluated_at": "2026-05-28T10:12:27.369342+00:00",
     "datasets": [
         {
             "dataset": "berlin/H",
-            "data_fetched_at": "2026-05-28T09:58:16.749981+00:00",
+            "data_fetched_at": "2026-05-28T10:12:07.138621+00:00",
             "context_length": 504,
             "prediction_length": 24
         },
         {
             "dataset": "nyc/H",
-            "data_fetched_at": "2026-05-28T09:58:28.553123+00:00",
+            "data_fetched_at": "2026-05-28T10:12:16.626231+00:00",
             "context_length": 504,
             "prediction_length": 24
         },
         {
             "dataset": "tokyo/H",
-            "data_fetched_at": "2026-05-28T09:58:37.331184+00:00",
+            "data_fetched_at": "2026-05-28T10:12:22.950374+00:00",
             "context_length": 504,
             "prediction_length": 24
         }

results/chronos_bolt_base/online_meta.json

@@ -1,23 +1,23 @@
 {
     "model": "Chronos-Bolt-Base",
     "api_model_id": "amazon/chronos-bolt-base",
-    "evaluated_at": "2026-05-28T09:58:12.886980+00:00",
+    "evaluated_at": "2026-05-28T10:12:05.182036+00:00",
     "datasets": [
         {
             "dataset": "berlin/H",
-            "data_fetched_at": "2026-05-28T09:57:42.662907+00:00",
+            "data_fetched_at": "2026-05-28T10:10:35.736508+00:00",
             "context_length": 504,
             "prediction_length": 24
         },
         {
             "dataset": "nyc/H",
-            "data_fetched_at": "2026-05-28T09:57:56.819035+00:00",
+            "data_fetched_at": "2026-05-28T10:11:53.331496+00:00",
             "context_length": 504,
             "prediction_length": 24
         },
         {
             "dataset": "tokyo/H",
-            "data_fetched_at": "2026-05-28T09:58:06.485785+00:00",
+            "data_fetched_at": "2026-05-28T10:12:00.889157+00:00",
             "context_length": 504,
             "prediction_length": 24
         }

results/chronos_bolt_mini/all_results.csv

@@ -0,0 +1,4 @@
+dataset,model,eval_metrics/MSE[mean],eval_metrics/MSE[0.5],eval_metrics/MAE[0.5],eval_metrics/MASE[0.5],eval_metrics/MAPE[0.5],eval_metrics/sMAPE[0.5],eval_metrics/MSIS,eval_metrics/RMSE[mean],eval_metrics/NRMSE[mean],eval_metrics/ND[0.5],eval_metrics/mean_weighted_sum_quantile_loss,domain,num_variates
+berlin/H/short,Chronos-Bolt-Mini,23.02803795462124,23.02803795462124,3.1608052717314825,0.6035438633560637,0.15491954678594283,0.1448325903393575,3.115631327236734,4.798753791831921,0.19212093626207102,0.1265447019142281,0.1081969714859019,Energy,1
+nyc/H/short,Chronos-Bolt-Mini,67.67541915984299,67.67541915984299,5.3830602169036865,0.6020463455851162,0.13388074766003255,0.11707076108417061,4.7290140050230995,8.226507105682398,0.022892831572236993,0.014980050379299257,0.01192178705488713,Econ/Fin,1
+tokyo/H/short,Chronos-Bolt-Mini,5.129979933399418,5.129979933399418,1.3517586653534737,0.5165812565468629,0.37651829322351843,0.8238071342871242,3.8948858261463672,2.264945900766598,0.250040044563296,0.14922819869975965,0.106032322595721,Transport,1

results/chronos_bolt_mini/config.json

@@ -0,0 +1,11 @@
+{
+    "model": "Chronos-Bolt-Mini",
+    "model_type": "zero-shot",
+    "model_dtype": "float32",
+    "model_link": "https://tsfm.ai/models/amazon/chronos-bolt-mini",
+    "code_link": "https://github.com/zhouziyu02/TSFM_Bench/blob/main/scripts/run_online_eval.py",
+    "org": "Amazon / TSFM.ai",
+    "testdata_leakage": "No",
+    "replication_code_available": "Yes",
+    "api_model_id": "amazon/chronos-bolt-mini"
+}

results/chronos_bolt_mini/online_meta.json

@@ -0,0 +1,25 @@
+{
+    "model": "Chronos-Bolt-Mini",
+    "api_model_id": "amazon/chronos-bolt-mini",
+    "evaluated_at": "2026-05-28T10:10:33.484071+00:00",
+    "datasets": [
+        {
+            "dataset": "berlin/H",
+            "data_fetched_at": "2026-05-28T10:10:08.921860+00:00",
+            "context_length": 504,
+            "prediction_length": 24
+        },
+        {
+            "dataset": "nyc/H",
+            "data_fetched_at": "2026-05-28T10:10:20.642074+00:00",
+            "context_length": 504,
+            "prediction_length": 24
+        },
+        {
+            "dataset": "tokyo/H",
+            "data_fetched_at": "2026-05-28T10:10:28.780465+00:00",
+            "context_length": 504,
+            "prediction_length": 24
+        }
+    ]
+}

results/chronos_bolt_tiny/online_meta.json

@@ -1,23 +1,23 @@
 {
     "model": "Chronos-Bolt-Tiny",
     "api_model_id": "amazon/chronos-bolt-tiny",
-    "evaluated_at": "2026-05-28T09:57:39.194711+00:00",
+    "evaluated_at": "2026-05-28T10:09:59.581604+00:00",
     "datasets": [
         {
             "dataset": "berlin/H",
-            "data_fetched_at": "2026-05-28T09:56:56.023164+00:00",
+            "data_fetched_at": "2026-05-28T10:09:00.615970+00:00",
             "context_length": 504,
             "prediction_length": 24
         },
         {
             "dataset": "nyc/H",
-            "data_fetched_at": "2026-05-28T09:57:26.649671+00:00",
+            "data_fetched_at": "2026-05-28T10:09:46.973725+00:00",
             "context_length": 504,
             "prediction_length": 24
         },
         {
             "dataset": "tokyo/H",
-            "data_fetched_at": "2026-05-28T09:57:33.990365+00:00",
+            "data_fetched_at": "2026-05-28T10:09:54.263751+00:00",
             "context_length": 504,
             "prediction_length": 24
         }

results/moirai_1_1_small/all_results.csv

@@ -0,0 +1,4 @@
+dataset,model,eval_metrics/MSE[mean],eval_metrics/MSE[0.5],eval_metrics/MAE[0.5],eval_metrics/MASE[0.5],eval_metrics/MAPE[0.5],eval_metrics/sMAPE[0.5],eval_metrics/MSIS,eval_metrics/RMSE[mean],eval_metrics/NRMSE[mean],eval_metrics/ND[0.5],eval_metrics/mean_weighted_sum_quantile_loss,domain,num_variates
+berlin/H/short,Moirai-1.1-Small,25.942115142478016,25.914808326787288,4.134999871253967,0.8981191360243419,0.2807477289705378,0.2263154350433037,4.74372356414242,5.0933402735805915,0.20391487967713806,0.16554715685998675,0.13114703596375427,Energy,1
+nyc/H/short,Moirai-1.1-Small,115.85427334839227,120.1817511764708,7.214917302131653,0.797558713399391,0.19310537226504765,0.15686588488083114,4.257123747918823,10.763562298253877,0.029952981945521127,0.02007776623583399,0.014418479379084874,Econ/Fin,1
+tokyo/H/short,Moirai-1.1-Small,5.042828195691895,4.650267012598303,1.3240045196531962,0.49590462426539694,0.38585838156843166,0.8294245926005037,4.263774493409405,2.2456242329677276,0.2479070175996799,0.1461642633424671,0.11333818582137908,Transport,1

results/moirai_1_1_small/config.json

@@ -0,0 +1,11 @@
+{
+    "model": "Moirai-1.1-Small",
+    "model_type": "zero-shot",
+    "model_dtype": "float32",
+    "model_link": "https://tsfm.ai/models/Salesforce/moirai-1.1-R-small",
+    "code_link": "https://github.com/zhouziyu02/TSFM_Bench/blob/main/scripts/run_online_eval.py",
+    "org": "Salesforce / TSFM.ai",
+    "testdata_leakage": "No",
+    "replication_code_available": "Yes",
+    "api_model_id": "Salesforce/moirai-1.1-R-small"
+}

results/moirai_1_1_small/online_meta.json

@@ -0,0 +1,25 @@
+{
+    "model": "Moirai-1.1-Small",
+    "api_model_id": "Salesforce/moirai-1.1-R-small",
+    "evaluated_at": "2026-05-28T10:18:27.617821+00:00",
+    "datasets": [
+        {
+            "dataset": "berlin/H",
+            "data_fetched_at": "2026-05-28T10:18:02.944773+00:00",
+            "context_length": 504,
+            "prediction_length": 24
+        },
+        {
+            "dataset": "nyc/H",
+            "data_fetched_at": "2026-05-28T10:18:14.313951+00:00",
+            "context_length": 504,
+            "prediction_length": 24
+        },
+        {
+            "dataset": "tokyo/H",
+            "data_fetched_at": "2026-05-28T10:18:22.281838+00:00",
+            "context_length": 504,
+            "prediction_length": 24
+        }
+    ]
+}

results/online_status.json

@@ -1,85 +1,61 @@
 {
     "status": "ok",
-    "started_at": "2026-05-28T09:56:53.592771+00:00",
-    "finished_at": "2026-05-28T09:58:41.599812+00:00",
+    "started_at": "2026-05-28T10:17:24.399004+00:00",
+    "finished_at": "2026-05-28T10:18:27.618686+00:00",
     "data_source": "open_meteo",
     "data_config": "configs/datasets/open_meteo.yaml",
     "baseline_model": "Chronos-Bolt-Base",
     "models": {
-        "Chronos-Bolt-Tiny": {
-            "model": "Chronos-Bolt-Tiny",
-            "api_model_id": "amazon/chronos-bolt-tiny",
-            "evaluated_at": "2026-05-28T09:57:39.194711+00:00",
+        "TimesFM-2.5": {
+            "model": "TimesFM-2.5",
+            "api_model_id": "google/timesfm-2.5-200m-pytorch",
+            "evaluated_at": "2026-05-28T10:18:01.120947+00:00",
             "datasets": [
                 {
                     "dataset": "berlin/H",
-                    "data_fetched_at": "2026-05-28T09:56:56.023164+00:00",
+                    "data_fetched_at": "2026-05-28T10:17:26.673802+00:00",
                     "context_length": 504,
                     "prediction_length": 24
                 },
                 {
                     "dataset": "nyc/H",
-                    "data_fetched_at": "2026-05-28T09:57:26.649671+00:00",
+                    "data_fetched_at": "2026-05-28T10:17:42.004619+00:00",
                     "context_length": 504,
                     "prediction_length": 24
                 },
                 {
                     "dataset": "tokyo/H",
-                    "data_fetched_at": "2026-05-28T09:57:33.990365+00:00",
+                    "data_fetched_at": "2026-05-28T10:17:52.661277+00:00",
                     "context_length": 504,
                     "prediction_length": 24
                 }
             ]
         },
-        "Chronos-Bolt-Base": {
-            "model": "Chronos-Bolt-Base",
-            "api_model_id": "amazon/chronos-bolt-base",
-            "evaluated_at": "2026-05-28T09:58:12.886980+00:00",
+        "Moirai-1.1-Small": {
+            "model": "Moirai-1.1-Small",
+            "api_model_id": "Salesforce/moirai-1.1-R-small",
+            "evaluated_at": "2026-05-28T10:18:27.617821+00:00",
             "datasets": [
                 {
                     "dataset": "berlin/H",
-                    "data_fetched_at": "2026-05-28T09:57:42.662907+00:00",
+                    "data_fetched_at": "2026-05-28T10:18:02.944773+00:00",
                     "context_length": 504,
                     "prediction_length": 24
                 },
                 {
                     "dataset": "nyc/H",
-                    "data_fetched_at": "2026-05-28T09:57:56.819035+00:00",
+                    "data_fetched_at": "2026-05-28T10:18:14.313951+00:00",
                     "context_length": 504,
                     "prediction_length": 24
                 },
                 {
                     "dataset": "tokyo/H",
-                    "data_fetched_at": "2026-05-28T09:58:06.485785+00:00",
-                    "context_length": 504,
-                    "prediction_length": 24
-                }
-            ]
-        },
-        "Chronos-2": {
-            "model": "Chronos-2",
-            "api_model_id": "amazon/chronos-2",
-            "evaluated_at": "2026-05-28T09:58:41.598566+00:00",
-            "datasets": [
-                {
-                    "dataset": "berlin/H",
-                    "data_fetched_at": "2026-05-28T09:58:16.749981+00:00",
-                    "context_length": 504,
-                    "prediction_length": 24
-                },
-                {
-                    "dataset": "nyc/H",
-                    "data_fetched_at": "2026-05-28T09:58:28.553123+00:00",
-                    "context_length": 504,
-                    "prediction_length": 24
-                },
-                {
-                    "dataset": "tokyo/H",
-                    "data_fetched_at": "2026-05-28T09:58:37.331184+00:00",
+                    "data_fetched_at": "2026-05-28T10:18:22.281838+00:00",
                     "context_length": 504,
                     "prediction_length": 24
                 }
             ]
         }
-    }
+    },
+    "failed_models": []
 }

results/timesfm_2_5/all_results.csv

@@ -0,0 +1,4 @@
+dataset,model,eval_metrics/MSE[mean],eval_metrics/MSE[0.5],eval_metrics/MAE[0.5],eval_metrics/MASE[0.5],eval_metrics/MAPE[0.5],eval_metrics/sMAPE[0.5],eval_metrics/MSIS,eval_metrics/RMSE[mean],eval_metrics/NRMSE[mean],eval_metrics/ND[0.5],eval_metrics/mean_weighted_sum_quantile_loss,domain,num_variates
+berlin/H/short,TimesFM-2.5,26.54830902205753,31.322212017679387,3.443761388460795,0.6303993503146956,0.1386495615012084,0.13621565370038174,18.01169611505954,5.152505121012257,0.20628357921361393,0.13787301681124278,0.1312424613907817,Energy,1
+nyc/H/short,TimesFM-2.5,66.51865488618301,44.39263497040171,4.487805022133721,0.4947887756747412,0.09551055813760116,0.08531003986906531,15.677110160428015,8.155896939404213,0.02269633667797287,0.012488722513809137,0.014150956333775139,Econ/Fin,1
+tokyo/H/short,TimesFM-2.5,4.1984207631854,2.872721300592553,1.081747318436909,0.411514142106336,0.31230229584942987,0.6226950610386188,18.191892843171864,2.0490048226359545,0.2262011013105863,0.11942013609095183,0.13494832121177208,Transport,1

results/timesfm_2_5/config.json

@@ -0,0 +1,11 @@
+{
+    "model": "TimesFM-2.5",
+    "model_type": "zero-shot",
+    "model_dtype": "float32",
+    "model_link": "https://tsfm.ai/models/google/timesfm-2.5-200m-pytorch",
+    "code_link": "https://github.com/zhouziyu02/TSFM_Bench/blob/main/scripts/run_online_eval.py",
+    "org": "Google / TSFM.ai",
+    "testdata_leakage": "No",
+    "replication_code_available": "Yes",
+    "api_model_id": "google/timesfm-2.5-200m-pytorch"
+}

results/timesfm_2_5/online_meta.json

@@ -0,0 +1,25 @@
+{
+    "model": "TimesFM-2.5",
+    "api_model_id": "google/timesfm-2.5-200m-pytorch",
+    "evaluated_at": "2026-05-28T10:18:01.120947+00:00",
+    "datasets": [
+        {
+            "dataset": "berlin/H",
+            "data_fetched_at": "2026-05-28T10:17:26.673802+00:00",
+            "context_length": 504,
+            "prediction_length": 24
+        },
+        {
+            "dataset": "nyc/H",
+            "data_fetched_at": "2026-05-28T10:17:42.004619+00:00",
+            "context_length": 504,
+            "prediction_length": 24
+        },
+        {
+            "dataset": "tokyo/H",
+            "data_fetched_at": "2026-05-28T10:17:52.661277+00:00",
+            "context_length": 504,
+            "prediction_length": 24
+        }
+    ]
+}

src/about.py

@@ -3,7 +3,9 @@ TITLE = """ TSFM Realworld Bench Leaderboard """
 INTRODUCTION_TEXT = """
 **TSFM Realworld Bench** evaluates time series foundation models on **live Open-Meteo
 weather data** with **zero-shot API inference** via [TSFM.ai](https://tsfm.ai/).
-Scores are normalized against **Chronos-Bolt-Base**; lower is better.
+Scores are normalized against **Chronos-Bolt-Base**. The leaderboard shows
+**metric values** and **ranks** in separate tables (MASE, CRPS, MAE, RMSE, sMAPE,
+MSIS, ND, NRMSE, MAPE).
 """
 
 LLM_BENCHMARKS_TEXT = """

src/utils.py

@@ -6,8 +6,6 @@ from pathlib import Path
 import pandas as pd
 from scipy import stats
 
-from src.leaderboard.read_evals import get_model_info
-
 METRIC_COLUMNS = [
     "eval_metrics/MSE[mean]",
     "eval_metrics/MSE[0.5]",
@@ -21,9 +19,33 @@ METRIC_COLUMNS = [
     "eval_metrics/ND[0.5]",
     "eval_metrics/mean_weighted_sum_quantile_loss",
 ]
-RANKING_METRIC = "eval_metrics/mean_weighted_sum_quantile_loss"
-MASE_RANKING_METRIC = "eval_metrics/MASE[0.5]"
-METRIC_CHOICES = [MASE_RANKING_METRIC, RANKING_METRIC]
+
+DISPLAY_METRICS = [
+    "eval_metrics/MASE[0.5]",
+    "eval_metrics/mean_weighted_sum_quantile_loss",
+    "eval_metrics/MAE[0.5]",
+    "eval_metrics/RMSE[mean]",
+    "eval_metrics/sMAPE[0.5]",
+    "eval_metrics/MSIS",
+    "eval_metrics/ND[0.5]",
+    "eval_metrics/NRMSE[mean]",
+    "eval_metrics/MAPE[0.5]",
+]
+
+METRIC_LABELS = {
+    "eval_metrics/MASE[0.5]": "MASE",
+    "eval_metrics/mean_weighted_sum_quantile_loss": "CRPS",
+    "eval_metrics/MAE[0.5]": "MAE",
+    "eval_metrics/RMSE[mean]": "RMSE",
+    "eval_metrics/sMAPE[0.5]": "sMAPE",
+    "eval_metrics/MSIS": "MSIS",
+    "eval_metrics/ND[0.5]": "ND",
+    "eval_metrics/NRMSE[mean]": "NRMSE",
+    "eval_metrics/MAPE[0.5]": "MAPE",
+}
+
+VALUE_COLUMNS = [METRIC_LABELS[m] for m in DISPLAY_METRICS]
+RANK_COLUMNS = [f"{METRIC_LABELS[m]}_Rank" for m in DISPLAY_METRICS]
 
 LEGACY_MODEL_NAMES = {
     "TSFM1": "Chronos-Bolt-Tiny",
@@ -49,17 +71,6 @@ def format_df(df: pd.DataFrame) -> pd.DataFrame:
     return formatted
 
 
-def rename_metrics(df: pd.DataFrame) -> pd.DataFrame:
-    return df.rename(
-        columns={
-            "eval_metrics/MASE[0.5]": "MASE",
-            "eval_metrics/mean_weighted_sum_quantile_loss": "CRPS",
-            "rank": "CRPS_Rank",
-            "Rank_MASE": "MASE_Rank",
-        }
-    )
-
-
 def unify_freq(df: pd.DataFrame) -> pd.DataFrame:
     df = df.copy()
     df["frequency"] = df["frequency"].str.replace(r"\d+", "", regex=True)
@@ -88,7 +99,8 @@ def standardize_df(df: pd.DataFrame, root_dir: str = "results") -> pd.DataFrame:
     if baseline_name not in set(df["model"]):
         return df
 
-    for _, group in df.groupby(["dataset", "term_length", "frequency"]):
+    group_keys = ["dataset", "term_length", "frequency"]
+    for _, group in df.groupby(group_keys):
         baseline = group[group["model"] == baseline_name]
         if baseline.empty:
             continue
@@ -131,46 +143,51 @@ def enrich_with_properties(df: pd.DataFrame, ds_properties_path: str) -> pd.Data
     return unify_freq(df)
 
 
+def _add_per_dataset_ranks(df: pd.DataFrame) -> pd.DataFrame:
+    df = df.copy()
+    group_keys = ["dataset", "term_length", "frequency"]
+    for metric in DISPLAY_METRICS:
+        rank_col = f"Rank_{metric}"
+        df[rank_col] = df.groupby(group_keys)[metric].rank(method="first", ascending=True)
+    return df
+
+
 def get_grouped_dfs(
     root_dir: str = "results",
     ds_properties: str | None = None,
 ) -> dict[str, pd.DataFrame]:
     if ds_properties is None:
         ds_properties = str(Path(root_dir) / "dataset_properties.csv")
+
     df = load_all_results(root_dir)
     df = enrich_with_properties(df, ds_properties)
     df = standardize_df(df, root_dir)
+    df = _add_per_dataset_ranks(df)
 
-    df["rank"] = df.groupby(["dataset", "term_length", "frequency"])[RANKING_METRIC].rank(
-        method="first", ascending=True
+    rank_cols = [f"Rank_{metric}" for metric in DISPLAY_METRICS]
+    overall_values = df.groupby(["model"])[DISPLAY_METRICS].agg(stats.gmean)
+    overall_ranks = df.groupby(["model"])[rank_cols].mean()
+
+    overall_values = overall_values.rename(columns=METRIC_LABELS)
+    overall_ranks = overall_ranks.rename(
+        columns={f"Rank_{metric}": f"{METRIC_LABELS[metric]}_Rank" for metric in DISPLAY_METRICS}
     )
-    df["Rank_MASE"] = df.groupby(["dataset", "term_length", "frequency"])[
-        MASE_RANKING_METRIC
-    ].rank(method="first", ascending=True)
-
-    overall = df.groupby(["model"])[METRIC_CHOICES].agg(stats.gmean)
-    overall_rank = df.groupby(["model"])[["rank", "Rank_MASE"]].mean()
-    overall = pd.concat([overall, overall_rank], axis=1)
-    overall = overall[
-        [
-            "eval_metrics/MASE[0.5]",
-            "Rank_MASE",
-            "eval_metrics/mean_weighted_sum_quantile_loss",
-            "rank",
-        ]
-    ]
-
-    return {"overall": overall}
-
-
-def get_merged_df(result_df: pd.DataFrame, model_info_df: pd.DataFrame) -> pd.DataFrame:
-    merged = pd.merge(model_info_df, result_df, left_on="model", right_on="model", how="inner")
-    merged = merged.drop(columns=["model"])
-    merged = merged.rename(columns={"model_w_link": "Model"})
-    return merged
-
-
-def prepare_leaderboard_table(grouped_df: pd.DataFrame) -> pd.DataFrame:
-    table = rename_metrics(grouped_df.reset_index())
+
+    return {
+        "overall_values": overall_values,
+        "overall_ranks": overall_ranks,
+    }
+
+
+def prepare_values_table(values_df: pd.DataFrame) -> pd.DataFrame:
+    table = values_df.reset_index().rename(columns={"model": "Model"})
+    table = format_df(table)
+    cols = ["Model"] + [col for col in VALUE_COLUMNS if col in table.columns]
+    return table[cols].sort_values(by=["MASE"])
+
+
+def prepare_ranks_table(ranks_df: pd.DataFrame) -> pd.DataFrame:
+    table = ranks_df.reset_index().rename(columns={"model": "Model"})
     table = format_df(table)
-    return table.sort_values(by=["MASE_Rank"])
+    cols = ["Model"] + [col for col in RANK_COLUMNS if col in table.columns]
+    return table[cols].sort_values(by=["MASE_Rank"])