small fixes based on review

app.py

@@ -581,14 +581,19 @@ button.theme-toggle[data-mode="light"] .theme-icon-moon { display: block !import
     .prose .ranking-table td {
         padding: 8px 10px !important;
     }
+    .ranking-table,
+    .prose .ranking-table {
+        --rank-col-width: 3.25rem;
+    }
     .ranking-table .rank,
     .prose .ranking-table .rank,
     .ranking-table th.rank {
         position: sticky !important;
         left: 0 !important;
-        width: 2.4rem;
-        min-width: 2.4rem;
-        box-shadow: 6px 0 8px -6px rgba(0, 0, 0, 0.45);
+        width: var(--rank-col-width);
+        min-width: var(--rank-col-width);
+        max-width: var(--rank-col-width);
+        box-shadow: none;
     }
     .ranking-table .model-cell,
     .prose .ranking-table .model-cell {
@@ -598,6 +603,7 @@ button.theme-toggle[data-mode="light"] .theme-icon-moon { display: block !import
         min-width: 140px;
         max-width: none;
         background: transparent !important;
+        box-shadow: none !important;
     }
     .ranking-table th.model-cell,
     .prose .ranking-table th.model-cell {
@@ -608,6 +614,7 @@ button.theme-toggle[data-mode="light"] .theme-icon-moon { display: block !import
         min-width: 140px;
         max-width: none;
         background: var(--pruna-bg-header) !important;
+        box-shadow: 0 1px 0 var(--pruna-hairline) !important;
     }
     .ranking-table tbody tr:hover .model-cell {
         background: var(--pruna-table-hover) !important;
@@ -1107,11 +1114,11 @@ button.theme-toggle[data-mode="light"] .theme-icon-moon { display: block !import
     max-height: min(70vh, 720px);
     overflow-x: auto;
     overflow-y: auto;
-    -webkit-overflow-scrolling: touch;
-    overscroll-behavior-x: contain;
+    overscroll-behavior: none;
 }
 .ranking-table,
 .prose .ranking-table {
+    --rank-col-width: 4.25rem;
     width: 100%;
     margin: 0 !important;
     overflow: visible;
@@ -1187,10 +1194,14 @@ button.theme-toggle[data-mode="light"] .theme-icon-moon { display: block !import
 .prose .ranking-table .rank {
     position: sticky;
     left: 0;
-    z-index: 1;
+    z-index: 2;
     box-sizing: border-box;
-    width: 3.25rem;
-    min-width: 3.25rem;
+    width: var(--rank-col-width);
+    min-width: var(--rank-col-width);
+    max-width: var(--rank-col-width);
+    padding-left: 0.5rem !important;
+    padding-right: 0.5rem !important;
+    overflow: hidden;
     color: var(--pruna-lavender) !important;
     font-weight: 700 !important;
     text-align: center;
@@ -1207,18 +1218,21 @@ button.theme-toggle[data-mode="light"] .theme-icon-moon { display: block !import
 .ranking-table .model-cell,
 .prose .ranking-table .model-cell {
     position: sticky;
-    left: 3.25rem;
-    z-index: 1;
+    left: var(--rank-col-width);
+    z-index: 2;
+    box-sizing: border-box;
     min-width: 180px;
     max-width: 260px;
     background: var(--pruna-table-sticky) !important;
+    box-shadow: 8px 0 10px -8px rgba(0, 0, 0, 0.35) !important;
 }
 .ranking-table th.model-cell,
 .prose .ranking-table th.model-cell {
     top: 0;
-    left: 3.25rem;
+    left: var(--rank-col-width);
     z-index: 5;
     background: var(--pruna-bg-header) !important;
+    box-shadow: 0 1px 0 var(--pruna-hairline), 8px 0 10px -8px rgba(0, 0, 0, 0.35) !important;
 }
 .ranking-table tbody tr:hover .rank,
 .ranking-table tbody tr:hover .model-cell {
@@ -2307,8 +2321,7 @@ def load_video_editing_dataframe(path):
     df = df.rename(
         columns={
             "display_name": "Model",
-            "elo": "Video Edit Elo",
-            "n_generations": "Generations",
+            "elo": "Datapoint Elo",
             "min_generation_s": "Min Generation Time (s)",
             "median_generation_s": "Median Generation Time (s)",
             "p20_generation_s": "P20 Generation Time (s)",
@@ -2317,23 +2330,22 @@ def load_video_editing_dataframe(path):
             "model_execution_time_s_per_output_video_s": (
                 "Execution Time / Output Video Second (s)"
             ),
-            "price": "Price / Video (USD)",
+            "price": "Price / Second of Video (USD)",
         }
     )
-    df = df.drop(columns=["wandb_run_ids"], errors="ignore")
+    df = df.drop(columns=["wandb_run_ids", "n_generations"], errors="ignore")
     df["Model"] = df["Model"].astype(str).str.strip()
     df = _as_numeric(
         df,
         [
-            "Video Edit Elo",
-            "Generations",
+            "Datapoint Elo",
             "Min Generation Time (s)",
             "Median Generation Time (s)",
             "P20 Generation Time (s)",
             "Time / Output Video Second (s)",
             "Predict Time / Output Video Second (s)",
             "Execution Time / Output Video Second (s)",
-            "Price / Video (USD)",
+            "Price / Second of Video (USD)",
         ],
     )
     end_to_end = df.get("Time / Output Video Second (s)")
@@ -2460,12 +2472,11 @@ video_display_columns = [
     col
     for col in [
         "Model",
-        "Video Edit Elo",
-        "Generations",
+        "Datapoint Elo",
+        "Time / Output Video Second (s)",
         "Median Generation Time (s)",
         "Min Generation Time (s)",
-        "Time / Output Video Second (s)",
-        "Price / Video (USD)",
+        "Price / Second of Video (USD)",
     ]
     if col in video_df.columns
 ]
@@ -2490,7 +2501,6 @@ metrics = [
     {"id": "arena_art", "column": "Arena Art Elo"},
     {"id": "arena_portraits", "column": "Arena Portraits Elo"},
     {"id": "arena_text", "column": "Arena Text Rendering Elo"},
-    {"id": "video_edit_elo", "column": "Video Edit Elo"},
 ]
 
 
@@ -2531,20 +2541,20 @@ arena_metric_ids = _metric_ids_for(
         "arena_text",
     ],
 )
-video_metric_ids = _metric_ids_for(video_df, ["video_edit_elo"])
+video_metric_ids = _metric_ids_for(video_df, ["datapoint_elo"])
 
 datasets = [
     {
         "id": "video_editing",
-        "name": "Video Editing Dataset",
+        "name": "Pruna Internal Video-Edit Benchmark",
         "modality": "video_to_video",
         "data": video_df,
         "columns": video_display_columns,
         "metric_ids": video_metric_ids,
         "note": (
-            "Elo from pairwise video-edit preference. Price is USD per output "
-            "video. Seconds for 1s of video is how long it takes to produce "
-            "one second of output video (end-to-end wall time)."
+            "Datapoint Elo from pairwise video-edit preference. Price is USD "
+            "per second of output video. Generation time per second of video "
+            "is end-to-end wall time to produce one second of output."
         ),
         "samples": None,
     },

ui.py

@@ -25,18 +25,18 @@ MAX_COMPARE_PROMPTS = 8
 MAX_PARETO_METRICS = 8
 _PARETO_SLOT_COUNT = 1 + MAX_PARETO_METRICS * 8
 _PARETO_PRICE_COLUMN = "Price / Image (USD)"
-_PARETO_VIDEO_PRICE_COLUMN = "Price / Video (USD)"
+_PARETO_VIDEO_PRICE_COLUMN = "Price / Second of Video (USD)"
 _PARETO_PRICE_COLUMNS = (_PARETO_PRICE_COLUMN, _PARETO_VIDEO_PRICE_COLUMN)
 _PARETO_TIME_COLUMN = "Min Generation Time (s)"
 _PARETO_VIDEO_TIME_COLUMN = "Pareto Time / Output Video Second (s)"
 _PARETO_TIME_COLUMNS = (_PARETO_VIDEO_TIME_COLUMN, _PARETO_TIME_COLUMN)
 _PARETO_PRICE_TITLES = {
     _PARETO_PRICE_COLUMN: "Price per image (USD)",
-    _PARETO_VIDEO_PRICE_COLUMN: "Price per video (USD)",
+    _PARETO_VIDEO_PRICE_COLUMN: "Price per second of video (USD)",
 }
 _PARETO_TIME_TITLES = {
     _PARETO_TIME_COLUMN: "Min generation time (s)",
-    _PARETO_VIDEO_TIME_COLUMN: "Seconds needed for 1s of output video",
+    _PARETO_VIDEO_TIME_COLUMN: "Generation time per second of video",
 }
 _PARETO_SCALE_CHOICES = [
     ("Log", "Logarithmic"),
@@ -105,9 +105,16 @@ prompt suites, so samples are not shown.
 
 ## Current datasets
 
-### Video Editing Dataset
-Video-to-video edit Elo from pairwise preference, with price per output
-video and generation time. Samples are not shown yet.
+### Pruna Internal Video-Edit Benchmark
+Pruna's internal video-to-video editing benchmark, collected by our
+research engineers. It combines prompts from public video-editing
+benchmarks with use-case examples we gathered for advertisement,
+e-commerce, real estate, concept art, and similar work. The suite also
+covers camera-angle and movement changes, lighting, and text in video
+(altering, adding, or removing it). Quality is Datapoint Elo from
+pairwise preference. Price is USD per second of output video;
+generation time is wall time per second of output video. Samples are
+not shown yet.
 
 ### Qwen Image Dataset
 100 prompts from the 1,000-prompt Qwen Image Bench set, sampled for coverage
@@ -151,12 +158,14 @@ ABOUT_DETAILS_CONTENT = """
 - **Arena Elo**: Elo published by Arena AI on their own dataset, plus
   category Elos (branding, 3D, cartoon/anime, photorealistic, art, portraits,
   text rendering).
-- **Generation time**: median and minimum generation time in seconds, as
-  reported in the evaluation table. This is not a p95, and we do not state
-  warm vs cold or concurrent load. Not available for Arena AI.
-- **Price**: USD per image in the evaluation table. We do not state list
-  price vs amount paid, or whether failed generations are included. Not
-  available for Arena AI.
+- **Generation time**: median and minimum generation time in seconds for
+  images, as reported in the evaluation table. For video, generation time
+  per second of output video is the more informative figure (end-to-end
+  wall time). This is not a p95, and we do not state warm vs cold or
+  concurrent load. Not available for Arena AI.
+- **Price**: USD per image for text-to-image, or USD per second of output
+  video for video-to-video. We do not state list price vs amount paid, or
+  whether failed generations are included. Not available for Arena AI.
 
 Scores from different datasets or metrics are **not interchangeable**. A high
 OneIG alignment score is not the same quantity as a Datapoint Elo. Compare
@@ -456,12 +465,11 @@ _LEADERBOARD_IDENTITY_COLUMNS = [
     "Optimized",
 ]
 _LEADERBOARD_META_COLUMNS = [
+    "Time / Output Video Second (s)",
     "Median Generation Time (s)",
     "Min Generation Time (s)",
-    "Time / Output Video Second (s)",
     "Price / Image (USD)",
-    "Price / Video (USD)",
-    "Generations",
+    "Price / Second of Video (USD)",
     "Evaluation Date (UTC)",
     "Date",
 ]
@@ -513,7 +521,7 @@ def _format_leaderboard_cell(column, value):
     if pd.isna(value) or value is None or value == "":
         return "-"
     label = str(column).lower()
-    if label in {"rank", "generations"}:
+    if label == "rank":
         return str(int(value))
     if "price" in label:
         return _format_price(value)
@@ -688,11 +696,9 @@ def _display_label(column):
         "Arena Text Rendering Elo": "Text Rendering",
         "Median Generation Time (s)": "Median generation time",
         "Min Generation Time (s)": "Min generation time",
-        "Time / Output Video Second (s)": "Seconds for 1s of video",
+        "Time / Output Video Second (s)": "Generation time per second of video",
         "Price / Image (USD)": "Price per image",
-        "Price / Video (USD)": "Price per video",
-        "Video Edit Elo": "Elo",
-        "Generations": "Generations",
+        "Price / Second of Video (USD)": "Price per second of video",
         "Evaluation Date (UTC)": "Date",
         "Date": "Date",
     }
@@ -1035,7 +1041,7 @@ def _pareto_pair(
     price_column = _pareto_price_column(data) or _PARETO_PRICE_COLUMN
     price_title = _PARETO_PRICE_TITLES.get(price_column, "Price (USD)")
     price_missing = (
-        "Price per video isn't available for this dataset."
+        "Price per second of video isn't available for this dataset."
         if price_column == _PARETO_VIDEO_PRICE_COLUMN
         else "Price per image isn't available for this dataset."
     )
@@ -1052,12 +1058,12 @@ def _pareto_pair(
     time_column = _pareto_time_column(data) or _PARETO_TIME_COLUMN
     time_title = _PARETO_TIME_TITLES.get(time_column, "Generation time (s)")
     time_missing = (
-        "Seconds needed for 1s of output video isn't available for this dataset."
+        "Generation time per second of video isn't available for this dataset."
         if time_column == _PARETO_VIDEO_TIME_COLUMN
         else "Min generation time isn't available for this dataset."
     )
     time_empty = (
-        "No models have both a score and seconds needed for 1s of output "
+        "No models have both a score and generation time per second of "
         "video for this metric."
         if time_column == _PARETO_VIDEO_TIME_COLUMN
         else "No models have both a score and a min generation time for this metric."