Vegas mode: reclaim dead space and pace the rotation

On a wide panel Vegas mode spent much of its time showing black. At 50px/s
on a 512px display, one display width of blank is 10.2 seconds, which makes
several long-standing behaviours expensive:

- ScrollHelper prepended a full display width of black as an "initial gap",
  charged once per cycle — 10.2s of black at the start of every rotation.
- Plugins without get_vegas_content() are captured off a full-display canvas,
  so their blank margins entered the ticker too. Measured: of-the-day drew
  35px of "No Data" on a 512px canvas (92% blank), youtube-stats 142px of
  content with 185px of black either side. Only the scroll_helper path had
  any trimming.
- Cycle transitions deliberately pushed a blank frame and then recomposed
  synchronously: 84ms at best, 4.8s at worst, every millisecond of it black.
- buffer_ahead doubled as the cycle size, so a 21-plugin install showed 3
  plugins per cycle and took ~7 cycles to come around.
- separator_width was applied between every image rather than at plugin
  boundaries, so a per-row ticker like the F1 scoreboard (116 images, which
  it renders 4px apart internally) got a 32px chasm between each row — and
  the width budget didn't count those gaps, so the plugin quietly occupied
  far more of the panel than intended.

Changes:

- src/vegas_mode/geometry.py: numpy column-ink primitives shared by the
  trimmer and the audit tool, so the number reported is the number acted on.
  A Python per-column loop over a 17,000px strip is far too slow for the
  render path.
- PluginAdapter trims every content path, not just scroll_helper. Only outer
  edges are cropped: interior blank columns are the plugin's own layout
  (logo left, score right) and closing them would corrupt the design. A
  plugin on a non-black background is inherently unaffected.
- ScrollHelper.create_scrolling_image takes an explicit lead_gap, still
  defaulting to display_width so the many standalone-ticker callers are
  unchanged. Vegas passes lead_in_width (default 0).
- Cycle end holds the last rendered frame instead of blanking, turning the
  recompose into a brief freeze rather than the panel switching off.
- plugins_per_cycle (default 6) is split from buffer_ahead, which goes back
  to being only a prefetch low-water mark.
- max_plugin_width_ratio (default 3x display width) caps one plugin's share
  of a cycle. Overflow is deferred, not discarded: a rotation offset advances
  each fetch so later rows appear on subsequent cycles. Single oversized
  images are cropped at a blank column so the cut misses glyphs.
- Composition groups images by plugin: rows are joined by intra_plugin_gap
  (default 8) and separator_width applies only between plugins. The width
  budget now counts those gaps.
- Plugin data updates no longer run on the Vegas render path.

All new settings are user-configurable in Display -> Vegas Scroll, including
min/max cycle duration and dynamic duration, which previously existed in code
but were reachable only by hand-editing config.json.

Measured with scripts/dev/vegas_audit.py on a 512x64 panel:

  mean ink coverage    42.7% -> 69.4%
  fully blank           5.9% -> 0%
  reads as empty        13.6% -> 0%
  worst blank stretch    4.8s -> 0s
  full rotation          414s -> 123s
  plugins per cycle         3 -> 6

Note the metric choice: a "fully blank" scan (>=95% black viewport) reported
only 0.4% and badly understated the problem, because two full-width segments
with mid-canvas content never fully blank the viewport — they hold it at ~28%.
window_coverage_stats grades every viewport position by how much ink it
carries, which is what tracks perceived dead time.

Known remaining: cycle transitions still freeze ~3.5s while the next cycle is
fetched. Fixing that needs background prefetch, which is deferred because the
fallback-capture path mutates the shared display_manager.image and racing it
against the render loop risks torn frames.

Co-Authored-By: Claude <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01KEZK1P1Q1fu5pcuVrkrCFZ

config/config.template.json

@@ -130,7 +130,18 @@
             "plugin_order": [],
             "excluded_plugins": [],
             "target_fps": 125,
-            "buffer_ahead": 2
+            "buffer_ahead": 2,
+            "intra_plugin_gap": 8,
+            "auto_trim": true,
+            "trim_threshold": 10,
+            "content_padding": 8,
+            "min_plugin_width": 8,
+            "lead_in_width": 0,
+            "plugins_per_cycle": 6,
+            "max_plugin_width_ratio": 3.0,
+            "dynamic_duration_enabled": true,
+            "min_cycle_duration": 60,
+            "max_cycle_duration": 240
         }
     },
     "sync": {

scripts/dev/vegas_audit.py

@@ -0,0 +1,379 @@
+#!/usr/bin/env python3
+"""
+Vegas Mode Density Audit
+
+Reports how much of the Vegas ticker is actually showing something. Loads the
+real enabled plugins, pulls each one's content through the real
+``PluginAdapter``, composes the strip through the real ``ScrollHelper``, then
+measures the result.
+
+The headline number is the **dead-frame ratio**: the fraction of viewport
+positions across a full cycle that are effectively blank. Because the panel
+only ever shows ``display_width`` columns at a time, a blank stretch wider than
+the viewport is a stretch where the display looks switched off — so this ratio
+tracks perceived dead time rather than just counting unlit pixels.
+
+Runs entirely off-hardware, so it is safe to run alongside a live display.
+
+Usage:
+    # Audit every enabled plugin at the display size from config.json
+    python scripts/dev/vegas_audit.py
+
+    # Specific plugins, dump each segment as a PNG for eyeballing
+    python scripts/dev/vegas_audit.py -p of-the-day,youtube-stats --dump-dir /tmp/vg
+
+    # Machine-readable, for before/after comparison
+    python scripts/dev/vegas_audit.py --json > after.json
+"""
+
+import argparse
+import json
+import logging
+import os
+import sys
+import time
+from pathlib import Path
+from typing import Any, Dict, List, Optional
+
+PROJECT_ROOT = Path(__file__).resolve().parent.parent.parent
+sys.path.insert(0, str(PROJECT_ROOT))
+
+# Must precede any src import that may reach for hardware.
+os.environ.setdefault('EMULATOR', 'true')
+
+from PIL import Image  # noqa: E402
+
+from src.common.scroll_helper import ScrollHelper  # noqa: E402
+from src.plugin_system.testing.loading import (  # noqa: E402
+    build_full_config,
+    find_plugin_dir,
+    load_manifest,
+)
+from src.vegas_mode.config import VegasModeConfig  # noqa: E402
+from src.vegas_mode.geometry import (  # noqa: E402
+    DEFAULT_INK_THRESHOLD,
+    column_has_ink,
+    content_bounds,
+    dead_window_stats,
+    window_coverage_stats,
+)
+from src.vegas_mode.plugin_adapter import PluginAdapter  # noqa: E402
+
+# Sampling stride for the dead-window scan. A full cycle can be 30,000px wide;
+# 4px granularity keeps the scan instant while staying well under the ~10px a
+# single scroll step ever covers, so no dead stretch is missed.
+DEAD_SCAN_STEP = 4
+
+
+def load_main_config(path: Path) -> Dict[str, Any]:
+    with open(path, 'r') as fh:
+        return json.load(fh)
+
+
+def display_size_from_config(config: Dict[str, Any]) -> tuple:
+    """Derive the logical ticker size the way DisplayManager does."""
+    hw = config.get('display', {}).get('hardware', {})
+    cols = int(hw.get('cols', 64))
+    chain = int(hw.get('chain_length', 1))
+    rows = int(hw.get('rows', 32))
+    parallel = int(hw.get('parallel', 1))
+    return cols * chain, rows * parallel
+
+
+def enabled_plugin_ids(config: Dict[str, Any]) -> List[str]:
+    """Plugin IDs that are enabled in config, excluding non-plugin sections."""
+    ids = []
+    for key, value in config.items():
+        if isinstance(value, dict) and value.get('enabled') is True:
+            ids.append(key)
+    return ids
+
+
+def instantiate(plugin_id: str, display_manager, cache_manager, plugin_manager):
+    """Load one plugin offline. Returns the instance or None."""
+    from src.plugin_system.plugin_loader import PluginLoader
+
+    search_dirs = [
+        str(PROJECT_ROOT / 'plugin-repos'),
+        str(PROJECT_ROOT / 'plugins'),
+    ]
+    plugin_dir = find_plugin_dir(plugin_id, search_dirs)
+    if not plugin_dir:
+        return None
+
+    try:
+        manifest = load_manifest(Path(plugin_dir))
+        cfg = build_full_config(Path(plugin_dir))
+        instance, _ = PluginLoader().load_plugin(
+            plugin_id=plugin_id,
+            manifest=manifest,
+            plugin_dir=Path(plugin_dir),
+            config=cfg,
+            display_manager=display_manager,
+            cache_manager=cache_manager,
+            plugin_manager=plugin_manager,
+            install_deps=False,
+        )
+        return instance
+    except Exception as exc:  # noqa: BLE001 - audit tool must survive any plugin
+        print(f"  ! {plugin_id}: load failed ({type(exc).__name__}: {exc})",
+              file=sys.stderr)
+        return None
+
+
+def join_rows(images: List[Image.Image], gap: int) -> Image.Image:
+    """Concatenate one plugin's rows, matching RenderPipeline._join_plugin_rows."""
+    if len(images) == 1:
+        return images[0]
+    gap = max(0, gap)
+    width = sum(img.width for img in images) + gap * (len(images) - 1)
+    height = max(img.height for img in images)
+    block = Image.new('RGB', (width, height), (0, 0, 0))
+    x = 0
+    for img in images:
+        block.paste(img, (x, 0))
+        x += img.width + gap
+    return block
+
+
+def measure_segment(images: List[Image.Image], display_width: int,
+                    scroll_speed: float, threshold: int) -> Dict[str, Any]:
+    """Geometry of one plugin's contribution to the ticker."""
+    total_width = sum(img.width for img in images)
+    combined = Image.new('RGB', (max(1, total_width), images[0].height))
+    x = 0
+    for img in images:
+        combined.paste(img, (x, 0))
+        x += img.width
+
+    ink = column_has_ink(combined, threshold)
+    bounds = content_bounds(combined, threshold)
+    ink_cols = int(ink.sum())
+
+    return {
+        'images': len(images),
+        'width_px': total_width,
+        'ink_cols': ink_cols,
+        'ink_pct': round(100.0 * ink_cols / total_width, 1) if total_width else 0.0,
+        'lead_black_px': bounds[0] if bounds else total_width,
+        'trail_black_px': (total_width - 1 - bounds[1]) if bounds else 0,
+        'seconds_on_screen': round(total_width / scroll_speed, 1) if scroll_speed else 0.0,
+        'widths': [img.width for img in images],
+    }
+
+
+def main() -> int:
+    parser = argparse.ArgumentParser(
+        description='Audit Vegas mode content density')
+    parser.add_argument('--config', default=str(PROJECT_ROOT / 'config' / 'config.json'),
+                        help='Path to main config.json')
+    parser.add_argument('-p', '--plugins', default=None,
+                        help='Comma-separated plugin IDs (default: all enabled)')
+    parser.add_argument('--width', type=int, default=None,
+                        help='Override display width (default: from config hardware)')
+    parser.add_argument('--height', type=int, default=None,
+                        help='Override display height (default: from config hardware)')
+    parser.add_argument('--dump-dir', default=None,
+                        help='Write each segment and the composed strip as PNGs here')
+    parser.add_argument('--threshold', type=int, default=DEFAULT_INK_THRESHOLD,
+                        help=f'Ink threshold (default: {DEFAULT_INK_THRESHOLD})')
+    parser.add_argument('--per-cycle', type=int, default=None,
+                        help='Plugins composed per cycle '
+                             '(default: buffer_ahead + 1, matching production)')
+    parser.add_argument('--json', action='store_true',
+                        help='Emit JSON instead of a text report')
+    args = parser.parse_args()
+
+    config = load_main_config(Path(args.config))
+    vegas = VegasModeConfig.from_config(config)
+
+    cfg_w, cfg_h = display_size_from_config(config)
+    width = args.width or cfg_w
+    height = args.height or cfg_h
+    speed = vegas.scroll_speed
+
+    if args.plugins:
+        plugin_ids = [p.strip() for p in args.plugins.split(',') if p.strip()]
+    else:
+        plugin_ids = vegas.get_ordered_plugins(enabled_plugin_ids(config))
+
+    dump_dir = Path(args.dump_dir) if args.dump_dir else None
+    if dump_dir:
+        dump_dir.mkdir(parents=True, exist_ok=True)
+
+    from src.plugin_system.testing import (
+        MockCacheManager, MockPluginManager, VisualTestDisplayManager,
+    )
+
+    display_manager = VisualTestDisplayManager(width=width, height=height)
+    cache_manager = MockCacheManager()
+    plugin_manager = MockPluginManager()
+    adapter = PluginAdapter(display_manager)
+
+    if not args.json:
+        print(f"Vegas audit — display {width}x{height}, scroll {speed:g}px/s, "
+              f"separator {vegas.separator_width}px")
+        print(f"One display width = {width / speed:.1f}s of screen time\n")
+
+    results: List[Dict[str, Any]] = []
+    segments: List[Image.Image] = []
+
+    for plugin_id in plugin_ids:
+        started = time.time()
+        instance = instantiate(plugin_id, display_manager, cache_manager, plugin_manager)
+        if instance is None:
+            results.append({'plugin': plugin_id, 'status': 'load_failed'})
+            continue
+
+        plugin_manager.plugins[plugin_id] = instance
+        adapter.invalidate_cache(plugin_id)
+
+        try:
+            images = adapter.get_content(instance, plugin_id)
+        except Exception as exc:  # noqa: BLE001
+            results.append({'plugin': plugin_id, 'status': 'fetch_error',
+                            'error': f'{type(exc).__name__}: {exc}'})
+            continue
+
+        fetch_ms = round((time.time() - started) * 1000)
+
+        if not images:
+            results.append({'plugin': plugin_id, 'status': 'no_content',
+                            'fetch_ms': fetch_ms})
+            if not args.json:
+                print(f"  {plugin_id:28s} NO CONTENT            ({fetch_ms}ms)")
+            continue
+
+        entry = {'plugin': plugin_id, 'status': 'ok', 'fetch_ms': fetch_ms}
+        entry.update(measure_segment(images, width, speed, args.threshold))
+        results.append(entry)
+        segments.extend(images)
+
+        if dump_dir:
+            for idx, img in enumerate(images):
+                img.save(dump_dir / f"{plugin_id}__{idx:02d}.png")
+
+        if not args.json:
+            print(f"  {plugin_id:28s} {entry['width_px']:>6d}px  "
+                  f"{entry['images']:>2d} img  ink {entry['ink_pct']:>5.1f}%  "
+                  f"lead {entry['lead_black_px']:>4d}  tail {entry['trail_black_px']:>4d}  "
+                  f"{entry['seconds_on_screen']:>6.1f}s  ({fetch_ms}ms)")
+
+    summary: Dict[str, Any] = {
+        'display_width': width,
+        'display_height': height,
+        'scroll_speed': speed,
+        'separator_width': vegas.separator_width,
+        'plugins_audited': len(plugin_ids),
+        'plugins_with_content': sum(1 for r in results if r.get('status') == 'ok'),
+    }
+
+    # Production composes only the plugins sitting in the active buffer, so
+    # measuring one giant strip of every plugin would hide the per-cycle costs
+    # (most importantly the leading gap, which is charged once per cycle).
+    # Group the segments the way the running service does.
+    per_cycle = max(1, args.per_cycle or vegas.plugins_per_cycle)
+
+    cycles: List[Dict[str, Any]] = []
+    with_content = [r for r in results if r.get('status') == 'ok']
+
+    if segments:
+        logger = logging.getLogger('vegas_audit')
+        seg_index = 0
+        for start in range(0, len(with_content), per_cycle):
+            group = with_content[start:start + per_cycle]
+
+            # Mirror RenderPipeline: each plugin's rows are joined by
+            # intra_plugin_gap into one block, and separator_width is applied
+            # only between blocks. Measuring a flat list here would report gaps
+            # the service does not emit.
+            blocks: List[Image.Image] = []
+            for entry in group:
+                count = entry['images']
+                rows = segments[seg_index:seg_index + count]
+                seg_index += count
+                if rows:
+                    blocks.append(join_rows(rows, vegas.intra_plugin_gap))
+            if not blocks:
+                continue
+
+            # ScrollHelper logs unconditionally, so it needs a real logger.
+            helper = ScrollHelper(width, height, logger)
+            helper.create_scrolling_image(
+                content_items=blocks,
+                item_gap=vegas.separator_width,
+                element_gap=0,
+                # Must match RenderPipeline. Omitting this made the audit
+                # measure a full-display-width leading gap the service no
+                # longer emits, overstating dead space by 512px per cycle.
+                lead_gap=vegas.lead_in_width,
+            )
+            composed = helper.cached_image
+            if composed is None:
+                continue
+
+            dead = dead_window_stats(composed, width, args.threshold, step=DEAD_SCAN_STEP)
+            cover = window_coverage_stats(
+                composed, width, args.threshold, step=DEAD_SCAN_STEP)
+
+            if dump_dir:
+                composed.save(dump_dir / f"_cycle{len(cycles):02d}.png")
+
+            cycles.append({
+                'plugins': [e['plugin'] for e in group],
+                'width_px': composed.width,
+                'seconds': round(composed.width / speed, 1) if speed else 0.0,
+                'dead_pct': round(100 * dead.dead_ratio, 1),
+                'longest_dead_seconds': round(
+                    dead.longest_dead_run * DEAD_SCAN_STEP / speed, 1) if speed else 0.0,
+                'mean_ink_pct': round(100 * cover.mean_ink_ratio, 1),
+                'sparse_pct': round(100 * cover.sparse_ratio, 1),
+                'longest_sparse_seconds': round(
+                    cover.longest_sparse_run * DEAD_SCAN_STEP / speed, 1) if speed else 0.0,
+            })
+
+    if cycles:
+        total_px = sum(c['width_px'] for c in cycles)
+        # Weight each cycle by its width so a long cycle counts proportionally.
+        summary.update({
+            'cycles': len(cycles),
+            'total_px': total_px,
+            'full_rotation_seconds': round(total_px / speed, 1) if speed else 0.0,
+            'dead_pct': round(
+                sum(c['dead_pct'] * c['width_px'] for c in cycles) / total_px, 1),
+            'mean_ink_pct': round(
+                sum(c['mean_ink_pct'] * c['width_px'] for c in cycles) / total_px, 1),
+            'sparse_pct': round(
+                sum(c['sparse_pct'] * c['width_px'] for c in cycles) / total_px, 1),
+            'worst_dead_seconds': max(c['longest_dead_seconds'] for c in cycles),
+            'worst_sparse_seconds': max(c['longest_sparse_seconds'] for c in cycles),
+        })
+
+    if args.json:
+        print(json.dumps({'summary': summary, 'cycles': cycles, 'plugins': results},
+                         indent=2))
+    else:
+        print(f"\n  Cycles ({per_cycle} plugins each, as production composes them):")
+        for idx, cyc in enumerate(cycles):
+            print(f"    [{idx}] {cyc['width_px']:>6d}px {cyc['seconds']:>6.1f}s  "
+                  f"ink {cyc['mean_ink_pct']:>5.1f}%  blank {cyc['dead_pct']:>5.1f}%  "
+                  f"worst blank {cyc['longest_dead_seconds']:>5.1f}s  "
+                  f"| {', '.join(cyc['plugins'])}")
+
+        print(f"\n  {'-' * 66}")
+        print(f"  full rotation        {summary.get('full_rotation_seconds', 0):>7.1f}s "
+              f"over {summary.get('cycles', 0)} cycles")
+        print(f"  mean ink coverage    {summary.get('mean_ink_pct', 0):>7.1f}%  "
+              f"(higher is better; target >25%)")
+        print(f"  fully blank          {summary.get('dead_pct', 0):>7.1f}%  (target <2%)")
+        print(f"  reads as empty       {summary.get('sparse_pct', 0):>7.1f}%  (target <15%)")
+        print(f"  worst blank stretch  {summary.get('worst_dead_seconds', 0):>7.1f}s  "
+              f"(target <1.5s)")
+        print(f"  plugins w/ content   {summary.get('plugins_with_content', 0):>7d}"
+              f" of {summary['plugins_audited']}")
+
+    return 0
+
+
+if __name__ == '__main__':
+    raise SystemExit(main())

src/common/scroll_helper.py

@@ -112,7 +112,8 @@ class ScrollHelper:
         
     def create_scrolling_image(self, content_items: list,
                              item_gap: int = 32,
-                             element_gap: int = 16) -> Image.Image:
+                             element_gap: int = 16,
+                             lead_gap: Optional[int] = None) -> Image.Image:
         """
         Create a wide image containing all content items for scrolling.
 
@@ -120,10 +121,19 @@ class ScrollHelper:
             content_items: List of PIL Images to include in scroll
             item_gap: Gap between different items
             element_gap: Gap between elements within an item
+            lead_gap: Blank columns before the first item. Defaults to a full
+                display width, which makes a standalone ticker scroll in from
+                off-screen. Callers that loop many plugins back-to-back (Vegas
+                mode) pass a smaller value, since a full display width of black
+                reads as the panel being switched off at the start of every
+                cycle.
 
         Returns:
             PIL Image containing all content arranged horizontally
         """
+        if lead_gap is None:
+            lead_gap = self.display_width
+        lead_gap = max(0, int(lead_gap))
         if not content_items:
             # Create empty image if no content
             # Still set total_scroll_width to 0 to indicate no scrollable content
@@ -144,13 +154,13 @@ class ScrollHelper:
         total_width += element_gap * len(content_items)
         
         # Add initial gap before first item
-        total_width += self.display_width
+        total_width += lead_gap
 
         # Create the full scrolling image
         full_image = Image.new('RGB', (total_width, self.display_height), (0, 0, 0))
 
         # Position items
-        current_x = self.display_width  # Start with initial gap
+        current_x = lead_gap  # Start with initial gap
         
         for i, img in enumerate(content_items):
             # Paste the item image

src/plugin_system/testing/visual_display_manager.py

@@ -15,6 +15,7 @@ PIL Image canvas and draws text using the actual project fonts.
 import math
 import os
 import time
+from contextlib import contextmanager
 from pathlib import Path
 from typing import Any, List, Optional, Tuple
 
@@ -62,6 +63,9 @@ class VisualTestDisplayManager:
         # Matrix proxy (plugins access display_manager.matrix.width/height)
         self.matrix = _MatrixProxy(width, height)
 
+        # Set while inside capture_mode(); mirrors DisplayManager's flag.
+        self._capture_mode_active = False
+
         # Scrolling state (interface compat, no-op)
         self._scrolling_state = {
             'is_scrolling': False,
@@ -174,6 +178,21 @@ class VisualTestDisplayManager:
         """No-op for hardware; marks that display was updated."""
         self.update_called = True
 
+    @contextmanager
+    def capture_mode(self):
+        """
+        Interface parity with DisplayManager.capture_mode().
+
+        There is no hardware to suppress here, but Vegas mode's PluginAdapter
+        wraps every off-screen content fetch in this context, so the harness
+        must provide it for that code path to be exercisable in tests.
+        """
+        self._capture_mode_active = True
+        try:
+            yield
+        finally:
+            self._capture_mode_active = False
+
     def draw_text(self, text: str, x: Optional[int] = None, y: Optional[int] = None,
                   color: Tuple[int, int, int] = (255, 255, 255), small_font: bool = False,
                   font: Optional[Any] = None, centered: bool = False) -> None:

src/vegas_mode/config.py

@@ -21,6 +21,41 @@ class VegasModeConfig:
     scroll_speed: float = 50.0  # Pixels per second
     separator_width: int = 32  # Gap between plugins (pixels)
 
+    # Gap between rows contributed by the *same* plugin. separator_width marks
+    # the handoff from one plugin to the next; applying it between every image
+    # forced a 32px chasm between each row of a per-row ticker (the F1
+    # scoreboard renders its own rows 4px apart), which both looked wrong and
+    # silently inflated the width that plugin occupied.
+    intra_plugin_gap: int = 8
+
+    # Content density
+    #
+    # Plugins that render onto a full-display canvas contribute that whole
+    # canvas to the ticker, blank margins included. On a wide panel that is the
+    # dominant source of dead air: a plugin drawing 35px of text on a 512px
+    # canvas otherwise buys 9.5s of black at 50px/s. Trimming reclaims it.
+    auto_trim: bool = True
+    trim_threshold: int = 10  # Per-channel value a pixel must exceed to be "ink"
+    content_padding: int = 8  # Blank columns kept either side of trimmed content
+    min_plugin_width: int = 8  # Segments narrower than this after trim are dropped
+
+    # Columns of blank lead-in before the first item of a cycle. ScrollHelper
+    # defaults this to a full display width, which reads as the display being
+    # switched off at the start of every cycle.
+    lead_in_width: int = 0
+
+    # How many plugins are composed into one scroll cycle. Kept separate from
+    # buffer_ahead (which is only a prefetch low-water mark) because the two
+    # were previously the same number: a buffer_ahead of 2 meant just 3 plugins
+    # per cycle, so a 20-plugin install took seven cycles to come around.
+    plugins_per_cycle: int = 6
+
+    # Cap on one plugin's share of a cycle, as a multiple of display width.
+    # A single ticker returning 7,000px would otherwise hold the panel for over
+    # two minutes. Overflow is deferred to later cycles rather than discarded.
+    # 0 disables the cap.
+    max_plugin_width_ratio: float = 3.0
+
     # Plugin management
     plugin_order: List[str] = field(default_factory=list)
     excluded_plugins: Set[str] = field(default_factory=set)
@@ -55,6 +90,15 @@ class VegasModeConfig:
             enabled=vegas_config.get('enabled', False),
             scroll_speed=float(vegas_config.get('scroll_speed', 50.0)),
             separator_width=int(vegas_config.get('separator_width', 32)),
+            intra_plugin_gap=int(vegas_config.get('intra_plugin_gap', 8)),
+            auto_trim=vegas_config.get('auto_trim', True),
+            trim_threshold=int(vegas_config.get('trim_threshold', 10)),
+            content_padding=int(vegas_config.get('content_padding', 8)),
+            min_plugin_width=int(vegas_config.get('min_plugin_width', 8)),
+            lead_in_width=int(vegas_config.get('lead_in_width', 0)),
+            plugins_per_cycle=int(vegas_config.get('plugins_per_cycle', 6)),
+            max_plugin_width_ratio=float(
+                vegas_config.get('max_plugin_width_ratio', 3.0)),
             plugin_order=list(vegas_config.get('plugin_order', [])),
             excluded_plugins=set(vegas_config.get('excluded_plugins', [])),
             target_fps=int(vegas_config.get('target_fps', 125)),
@@ -72,6 +116,14 @@ class VegasModeConfig:
             'enabled': self.enabled,
             'scroll_speed': self.scroll_speed,
             'separator_width': self.separator_width,
+            'intra_plugin_gap': self.intra_plugin_gap,
+            'auto_trim': self.auto_trim,
+            'trim_threshold': self.trim_threshold,
+            'content_padding': self.content_padding,
+            'min_plugin_width': self.min_plugin_width,
+            'lead_in_width': self.lead_in_width,
+            'plugins_per_cycle': self.plugins_per_cycle,
+            'max_plugin_width_ratio': self.max_plugin_width_ratio,
             'plugin_order': self.plugin_order,
             'excluded_plugins': list(self.excluded_plugins),
             'target_fps': self.target_fps,
@@ -157,6 +209,44 @@ class VegasModeConfig:
         if self.buffer_ahead > 5:
             errors.append(f"buffer_ahead must be <= 5, got {self.buffer_ahead}")
 
+        if self.intra_plugin_gap < 0:
+            errors.append(
+                f"intra_plugin_gap must be >= 0, got {self.intra_plugin_gap}")
+        if self.intra_plugin_gap > 128:
+            errors.append(
+                f"intra_plugin_gap must be <= 128, got {self.intra_plugin_gap}")
+
+        if not 0 <= self.trim_threshold <= 254:
+            errors.append(
+                f"trim_threshold must be between 0 and 254, got {self.trim_threshold}")
+
+        if self.content_padding < 0:
+            errors.append(
+                f"content_padding must be >= 0, got {self.content_padding}")
+        if self.content_padding > 128:
+            errors.append(
+                f"content_padding must be <= 128, got {self.content_padding}")
+
+        if self.min_plugin_width < 0:
+            errors.append(
+                f"min_plugin_width must be >= 0, got {self.min_plugin_width}")
+
+        if self.lead_in_width < 0:
+            errors.append(
+                f"lead_in_width must be >= 0, got {self.lead_in_width}")
+
+        if self.plugins_per_cycle < 1:
+            errors.append(
+                f"plugins_per_cycle must be >= 1, got {self.plugins_per_cycle}")
+        if self.plugins_per_cycle > 50:
+            errors.append(
+                f"plugins_per_cycle must be <= 50, got {self.plugins_per_cycle}")
+
+        if self.max_plugin_width_ratio < 0:
+            errors.append(
+                "max_plugin_width_ratio must be >= 0 "
+                f"(0 disables the cap), got {self.max_plugin_width_ratio}")
+
         return errors
 
     def update(self, new_config: Dict[str, Any]) -> None:
@@ -174,6 +264,23 @@ class VegasModeConfig:
             self.scroll_speed = float(vegas_config['scroll_speed'])
         if 'separator_width' in vegas_config:
             self.separator_width = int(vegas_config['separator_width'])
+        if 'intra_plugin_gap' in vegas_config:
+            self.intra_plugin_gap = int(vegas_config['intra_plugin_gap'])
+        if 'auto_trim' in vegas_config:
+            self.auto_trim = vegas_config['auto_trim']
+        if 'trim_threshold' in vegas_config:
+            self.trim_threshold = int(vegas_config['trim_threshold'])
+        if 'content_padding' in vegas_config:
+            self.content_padding = int(vegas_config['content_padding'])
+        if 'min_plugin_width' in vegas_config:
+            self.min_plugin_width = int(vegas_config['min_plugin_width'])
+        if 'lead_in_width' in vegas_config:
+            self.lead_in_width = int(vegas_config['lead_in_width'])
+        if 'plugins_per_cycle' in vegas_config:
+            self.plugins_per_cycle = int(vegas_config['plugins_per_cycle'])
+        if 'max_plugin_width_ratio' in vegas_config:
+            self.max_plugin_width_ratio = float(
+                vegas_config['max_plugin_width_ratio'])
         if 'plugin_order' in vegas_config:
             self.plugin_order = list(vegas_config['plugin_order'])
         if 'excluded_plugins' in vegas_config:

src/vegas_mode/coordinator.py

@@ -64,7 +64,7 @@ class VegasModeCoordinator:
         self.plugin_manager = plugin_manager
 
         # Initialize components
-        self.plugin_adapter = PluginAdapter(display_manager)
+        self.plugin_adapter = PluginAdapter(display_manager, self.vegas_config)
         self.stream_manager = StreamManager(
             self.vegas_config,
             plugin_manager,
@@ -505,6 +505,10 @@ class VegasModeCoordinator:
             # Update components
             self.render_pipeline.update_config(new_vegas_config)
             self.stream_manager.config = new_vegas_config
+            self.plugin_adapter.config = new_vegas_config
+            # Cached segments were trimmed under the old settings, so drop them
+            # or a changed trim/padding value would not visibly take effect.
+            self.plugin_adapter.invalidate_cache()
 
             # Force refresh of stream manager to pick up plugin_order/buffer changes
             self.stream_manager._last_refresh = 0

src/vegas_mode/geometry.py

@@ -0,0 +1,341 @@
+"""
+Geometry primitives for Vegas Mode.
+
+Pure, side-effect-free measurements over PIL images. Two consumers:
+
+- ``PluginAdapter`` trims the blank margins plugins bake into their content
+  before it enters the ticker (see ``trim_to_content``).
+- ``scripts/dev/vegas_audit.py`` reports how much of the composed ticker is
+  dead space (see ``dead_window_stats``).
+
+Keeping both on the same primitives means the number the audit reports is the
+number the trimmer acted on.
+
+All column scans go through numpy: a Python-level per-column loop over a
+17,000px-wide ticker image takes seconds, which is far too slow for the render
+path.
+"""
+
+from typing import NamedTuple, Optional, Tuple
+
+import numpy as np
+from PIL import Image
+
+# A pixel counts as "ink" when any channel exceeds this. Chosen to ignore the
+# 1-2/255 noise that JPEG-sourced logos and alpha compositing leave behind in
+# nominally black areas, while still treating any deliberately drawn dark grey
+# as real content.
+DEFAULT_INK_THRESHOLD = 10
+
+# A window counts as "dead" when this fraction of its columns carry no ink.
+DEFAULT_DEAD_WINDOW_RATIO = 0.95
+
+
+def column_has_ink(img: Image.Image, threshold: int = DEFAULT_INK_THRESHOLD) -> np.ndarray:
+    """
+    Return a boolean array, one entry per image column, True where the column
+    contains at least one pixel brighter than ``threshold`` in any channel.
+
+    Args:
+        img: Image to scan (converted to RGB internally)
+        threshold: Per-channel value a pixel must exceed to count as ink
+
+    Returns:
+        Bool array of shape (width,)
+    """
+    arr = np.asarray(img if img.mode == 'RGB' else img.convert('RGB'))
+    if arr.ndim != 3:
+        # Degenerate/empty image — treat every column as blank.
+        return np.zeros(img.width, dtype=bool)
+    # Collapse rows and channels: a column is ink if any pixel in it is bright.
+    return arr.max(axis=(0, 2)) > threshold
+
+
+def content_bounds(
+    img: Image.Image, threshold: int = DEFAULT_INK_THRESHOLD
+) -> Optional[Tuple[int, int]]:
+    """
+    Find the first and last columns containing ink.
+
+    Args:
+        img: Image to measure
+        threshold: Ink threshold
+
+    Returns:
+        (first_col, last_col) inclusive, or None if the image is entirely blank
+    """
+    ink = column_has_ink(img, threshold)
+    if not ink.any():
+        return None
+    first = int(ink.argmax())
+    last = len(ink) - 1 - int(ink[::-1].argmax())
+    return first, last
+
+
+class TrimResult(NamedTuple):
+    """Outcome of a ``trim_to_content`` call."""
+
+    image: Optional[Image.Image]  # None when the source was entirely blank
+    original_width: int
+    trimmed_left: int
+    trimmed_right: int
+
+    @property
+    def is_blank(self) -> bool:
+        """True when the source image carried no ink at all."""
+        return self.image is None
+
+    @property
+    def width(self) -> int:
+        """Width after trimming (0 for a blank source)."""
+        return 0 if self.image is None else self.image.width
+
+    @property
+    def removed(self) -> int:
+        """Total columns removed."""
+        return self.trimmed_left + self.trimmed_right
+
+
+def trim_to_content(
+    img: Image.Image,
+    threshold: int = DEFAULT_INK_THRESHOLD,
+    padding: int = 0,
+) -> TrimResult:
+    """
+    Crop blank columns off the left and right edges of an image.
+
+    Only the outer edges are considered. Blank columns *between* two pieces of
+    content are deliberately preserved — those are the plugin's own layout
+    (e.g. a logo on the left and a score on the right), and closing them up
+    would corrupt the design rather than reclaim dead space.
+
+    A plugin drawing on a non-black background is unaffected: every column of a
+    filled background carries ink, so there is nothing to trim.
+
+    Args:
+        img: Image to trim
+        threshold: Ink threshold
+        padding: Columns of the original blank margin to keep on each side, as
+            breathing room. Capped at what the margin actually contains, so
+            this never widens the image beyond its original bounds.
+
+    Returns:
+        TrimResult. When the image is entirely blank, ``image`` is None and the
+        caller decides whether to skip the plugin.
+    """
+    bounds = content_bounds(img, threshold)
+    if bounds is None:
+        return TrimResult(None, img.width, 0, 0)
+
+    first, last = bounds
+    pad = max(0, padding)
+    left = max(0, first - pad)
+    right = min(img.width, last + 1 + pad)
+
+    if left == 0 and right == img.width:
+        return TrimResult(img, img.width, 0, 0)
+
+    cropped = img.crop((left, 0, right, img.height))
+    return TrimResult(cropped, img.width, left, img.width - right)
+
+
+def find_blank_cut(
+    img: Image.Image,
+    target: int,
+    search_radius: int,
+    threshold: int = DEFAULT_INK_THRESHOLD,
+) -> int:
+    """
+    Find a column near ``target`` that carries no ink, so an image can be cut
+    there without slicing through a glyph or logo.
+
+    Used when a single oversized segment has to be narrowed to fit a width
+    budget. Cutting at an arbitrary column would leave half a character
+    hanging at the panel edge; snapping to the nearest gap hides the cut.
+
+    Args:
+        img: Image to cut
+        target: Preferred cut column
+        search_radius: How far either side of ``target`` to look
+        threshold: Ink threshold
+
+    Returns:
+        A blank column within the search window, or ``target`` clamped to the
+        image bounds when the window contains no blank column at all.
+    """
+    width = img.width
+    target = max(0, min(target, width))
+    if search_radius <= 0 or width == 0:
+        return target
+
+    ink = column_has_ink(img, threshold)
+    lo = max(0, target - search_radius)
+    hi = min(width - 1, target + search_radius)
+
+    # Walk outwards from target so the nearest gap wins.
+    for offset in range(0, search_radius + 1):
+        right = target + offset
+        if right <= hi and not ink[right]:
+            return right
+        left = target - offset
+        if left >= lo and not ink[left]:
+            return left
+
+    return target
+
+
+class DeadWindowStats(NamedTuple):
+    """How much of a composed ticker reads as blank to a viewer."""
+
+    total_windows: int
+    dead_windows: int
+    longest_dead_run: int  # consecutive dead windows (i.e. scroll steps)
+
+    @property
+    def dead_ratio(self) -> float:
+        """Fraction of viewport positions that are effectively blank."""
+        if self.total_windows <= 0:
+            return 0.0
+        return self.dead_windows / self.total_windows
+
+
+def dead_window_stats(
+    img: Image.Image,
+    viewport_width: int,
+    threshold: int = DEFAULT_INK_THRESHOLD,
+    dead_ratio: float = DEFAULT_DEAD_WINDOW_RATIO,
+    step: int = 1,
+) -> DeadWindowStats:
+    """
+    Slide a viewport across a composed ticker image and count how many
+    positions are effectively blank.
+
+    This models what the viewer actually experiences: the ticker is only ever
+    seen ``viewport_width`` columns at a time, so a stretch of blank wider than
+    the viewport becomes a period where the panel looks switched off. Measuring
+    per-window rather than per-column is what makes the result correspond to
+    perceived dead time.
+
+    Args:
+        img: Composed ticker image
+        viewport_width: Display width in pixels
+        threshold: Ink threshold
+        dead_ratio: Fraction of blank columns for a window to count as dead
+        step: Column stride between sampled windows. 1 is exact; larger values
+            trade precision for speed on very wide images.
+
+    Returns:
+        DeadWindowStats. ``longest_dead_run`` is in units of ``step`` columns,
+        so multiply by ``step`` for pixels.
+    """
+    if viewport_width <= 0 or img.width <= 0:
+        return DeadWindowStats(0, 0, 0)
+
+    ink = column_has_ink(img, threshold)
+    step = max(1, step)
+
+    # Prefix sum of ink counts lets each window be evaluated in constant time,
+    # instead of re-summing viewport_width columns per position.
+    prefix = np.concatenate(([0], np.cumsum(ink)))
+
+    # Only whole windows are sampled; a partial tail window would report
+    # artificially dead because it has fewer columns to draw ink from.
+    last_start = img.width - viewport_width
+    if last_start < 0:
+        # Image narrower than the viewport — evaluate it as a single window.
+        blank_cols = len(ink) - int(prefix[-1])
+        is_dead = blank_cols >= dead_ratio * len(ink)
+        return DeadWindowStats(1, 1 if is_dead else 0, 1 if is_dead else 0)
+
+    starts = np.arange(0, last_start + 1, step)
+    ink_counts = prefix[starts + viewport_width] - prefix[starts]
+    blank_counts = viewport_width - ink_counts
+    dead = blank_counts >= dead_ratio * viewport_width
+
+    longest = _longest_true_run(dead)
+    return DeadWindowStats(len(starts), int(dead.sum()), longest)
+
+
+class CoverageStats(NamedTuple):
+    """How well-filled the viewport stays as the ticker scrolls past."""
+
+    total_windows: int
+    mean_ink_ratio: float       # average fraction of the viewport carrying ink
+    min_ink_ratio: float        # worst viewport position in the cycle
+    sparse_windows: int         # positions below the "looks empty" threshold
+    longest_sparse_run: int     # consecutive sparse positions, in steps
+
+    @property
+    def sparse_ratio(self) -> float:
+        """Fraction of viewport positions that read as near-empty."""
+        if self.total_windows <= 0:
+            return 0.0
+        return self.sparse_windows / self.total_windows
+
+
+def window_coverage_stats(
+    img: Image.Image,
+    viewport_width: int,
+    threshold: int = DEFAULT_INK_THRESHOLD,
+    sparse_ink_ratio: float = 0.10,
+    step: int = 1,
+) -> CoverageStats:
+    """
+    Measure how full the viewport stays across a whole scroll cycle.
+
+    ``dead_window_stats`` only catches viewport positions that are *entirely*
+    blank. That misses the more common complaint: a position holding one narrow
+    sliver of content at the very edge, with the other 90% black. Such a
+    position is not "dead" by that definition but still looks switched off.
+    This function grades every position by how much ink it carries, so
+    "there is always something to see" becomes measurable.
+
+    Args:
+        img: Composed ticker image
+        viewport_width: Display width in pixels
+        threshold: Ink threshold
+        sparse_ink_ratio: A position with less than this fraction of inked
+            columns counts as reading near-empty
+        step: Column stride between sampled positions
+
+    Returns:
+        CoverageStats
+    """
+    if viewport_width <= 0 or img.width <= 0:
+        return CoverageStats(0, 0.0, 0.0, 0, 0)
+
+    ink = column_has_ink(img, threshold)
+    step = max(1, step)
+    prefix = np.concatenate(([0], np.cumsum(ink)))
+
+    last_start = img.width - viewport_width
+    if last_start < 0:
+        ratio = float(prefix[-1]) / viewport_width
+        sparse = ratio < sparse_ink_ratio
+        return CoverageStats(1, ratio, ratio, 1 if sparse else 0, 1 if sparse else 0)
+
+    starts = np.arange(0, last_start + 1, step)
+    ratios = (prefix[starts + viewport_width] - prefix[starts]) / viewport_width
+    sparse_flags = ratios < sparse_ink_ratio
+
+    return CoverageStats(
+        total_windows=len(starts),
+        mean_ink_ratio=float(ratios.mean()),
+        min_ink_ratio=float(ratios.min()),
+        sparse_windows=int(sparse_flags.sum()),
+        longest_sparse_run=_longest_true_run(sparse_flags),
+    )
+
+
+def _longest_true_run(flags: np.ndarray) -> int:
+    """Length of the longest consecutive run of True in a boolean array."""
+    if flags.size == 0 or not flags.any():
+        return 0
+    # Reset a running counter at every False by subtracting the cumulative max
+    # of the counter's value at the preceding False positions.
+    idx = np.arange(len(flags))
+    not_flag = ~flags
+    # For each position, the index of the most recent False at or before it.
+    last_false = np.maximum.accumulate(np.where(not_flag, idx, -1))
+    run_lengths = idx - last_false
+    return int(run_lengths[flags].max())

src/vegas_mode/plugin_adapter.py

@@ -11,6 +11,8 @@ import time
 from typing import Optional, List, Any, Tuple, Union, TYPE_CHECKING
 from PIL import Image
 
+from src.vegas_mode.geometry import find_blank_cut, trim_to_content
+
 if TYPE_CHECKING:
     from src.plugin_system.base_plugin import BasePlugin
 
@@ -26,14 +28,21 @@ class PluginAdapter:
     2. Fallback: Capture display_manager.image after calling plugin.display()
     """
 
-    def __init__(self, display_manager: Any):
+    def __init__(self, display_manager: Any, config: Optional[Any] = None):
         """
         Initialize the plugin adapter.
 
         Args:
             display_manager: DisplayManager instance for fallback capture
+            config: VegasModeConfig controlling trim behaviour. When omitted,
+                trimming runs with the dataclass defaults, so existing callers
+                and tests keep working unchanged.
         """
         self.display_manager = display_manager
+        if config is None:
+            from src.vegas_mode.config import VegasModeConfig
+            config = VegasModeConfig()
+        self.config = config
         # Handle both property and method access patterns
         self.display_width = (
             display_manager.width() if callable(display_manager.width)
@@ -49,6 +58,11 @@ class PluginAdapter:
         self._cache_lock = threading.Lock()
         self._cache_ttl = 5.0  # Cache for 5 seconds
 
+        # Per-plugin rotation offset, so a plugin whose content exceeds its
+        # width budget shows a different slice on each cycle rather than
+        # always the same opening items.
+        self._item_offsets: dict = {}
+
         logger.info(
             "PluginAdapter initialized: display=%dx%d",
             self.display_width, self.display_height
@@ -93,8 +107,7 @@ class PluginAdapter:
                     "[%s] Native content SUCCESS: %d images, %dpx total",
                     plugin_id, len(content), total_width
                 )
-                self._cache_content(plugin_id, content)
+                return self._finalize(content, plugin_id, 'native')
-                return content
             logger.info("[%s] Native content returned None", plugin_id)
 
         # Try to get scroll_helper's cached image (for scrolling plugins like stocks/odds)
@@ -107,8 +120,7 @@ class PluginAdapter:
                 "[%s] ScrollHelper content SUCCESS: %d images, %dpx total",
                 plugin_id, len(content), total_width
             )
-            self._cache_content(plugin_id, content)
+            return self._finalize(content, plugin_id, 'scroll_helper')
-            return content
         if has_scroll_helper:
             logger.info("[%s] ScrollHelper content returned None", plugin_id)
 
@@ -121,8 +133,7 @@ class PluginAdapter:
                 "[%s] Fallback capture SUCCESS: %d images, %dpx total",
                 plugin_id, len(content), total_width
             )
-            self._cache_content(plugin_id, content)
+            return self._finalize(content, plugin_id, 'fallback')
-            return content
 
         logger.warning(
             "[%s] NO CONTENT from any method (native=%s, scroll_helper=%s, fallback=tried)",
@@ -130,6 +141,181 @@ class PluginAdapter:
         )
         return None
 
+    def _finalize(
+        self, images: List[Image.Image], plugin_id: str, source: str
+    ) -> Optional[List[Image.Image]]:
+        """
+        Trim dead space off a segment, then cache it.
+
+        Every content path funnels through here so trimming is applied
+        uniformly. Previously only the scroll_helper path had its margins
+        stripped, which left plugins that render onto a full-display canvas
+        contributing their entire blank canvas to the ticker.
+
+        Each image is trimmed independently because compose_scroll_content()
+        treats every image as its own item and inserts separator_width between
+        them — so a per-image trim is what makes that separator the real gap.
+
+        Args:
+            images: Raw content from one of the fetch paths
+            plugin_id: Plugin identifier for logging
+            source: Which path produced the content, for logging
+
+        Returns:
+            Trimmed image list, or None if nothing worth showing remains
+        """
+        if not self.config.auto_trim:
+            self._cache_content(plugin_id, images)
+            return images
+
+        original_width = sum(img.width for img in images)
+        kept: List[Image.Image] = []
+        dropped_blank = 0
+
+        for img in images:
+            result = trim_to_content(
+                img,
+                threshold=self.config.trim_threshold,
+                padding=self.config.content_padding,
+            )
+            if result.is_blank:
+                dropped_blank += 1
+                continue
+            kept.append(result.image)
+
+        if not kept:
+            logger.info(
+                "[%s] All %d image(s) from %s were blank — contributing nothing",
+                plugin_id, len(images), source
+            )
+            return None
+
+        trimmed_width = sum(img.width for img in kept)
+
+        if trimmed_width < self.config.min_plugin_width:
+            logger.info(
+                "[%s] Trimmed content %dpx is below min_plugin_width %dpx — skipping",
+                plugin_id, trimmed_width, self.config.min_plugin_width
+            )
+            return None
+
+        if trimmed_width != original_width or dropped_blank:
+            logger.info(
+                "[%s] Trimmed %s content: %dpx -> %dpx (%.0f%% reclaimed), "
+                "%d image(s) kept, %d blank dropped",
+                plugin_id, source, original_width, trimmed_width,
+                100.0 * (original_width - trimmed_width) / original_width
+                if original_width else 0.0,
+                len(kept), dropped_blank
+            )
+
+        kept = self._apply_width_budget(kept, plugin_id)
+
+        self._cache_content(plugin_id, kept)
+        return kept
+
+    def _width_budget(self) -> int:
+        """Maximum columns one plugin may occupy in a cycle. 0 means unlimited."""
+        ratio = self.config.max_plugin_width_ratio
+        if ratio <= 0:
+            return 0
+        return int(self.display_width * ratio)
+
+    def _apply_width_budget(
+        self, images: List[Image.Image], plugin_id: str
+    ) -> List[Image.Image]:
+        """
+        Hold one plugin to its share of a cycle.
+
+        A ticker returning 7,000px would otherwise own the panel for over two
+        minutes, which defeats the point of a rotation. Overflow is deferred
+        rather than discarded: the starting offset advances each time this
+        plugin is fetched, so later items appear on subsequent cycles instead
+        of never being seen.
+
+        Args:
+            images: Trimmed images for this plugin
+            plugin_id: Plugin identifier, used to track its rotation offset
+
+        Returns:
+            Images that fit the budget, starting from the plugin's current
+            rotation offset.
+        """
+        budget = self._width_budget()
+
+        # Count the gaps the compositor will insert between these rows, not
+        # just the pixels of the rows themselves — otherwise a plugin with many
+        # rows quietly occupies far more of the panel than its budget allows.
+        gap = max(0, self.config.intra_plugin_gap)
+        total = sum(img.width for img in images) + gap * (len(images) - 1)
+
+        if not budget or total <= budget:
+            # Fits, so reset rotation — the whole segment is being shown.
+            self._item_offsets.pop(plugin_id, None)
+            return images
+
+        if len(images) == 1:
+            return [self._crop_to_budget(images[0], budget, plugin_id)]
+
+        start = self._item_offsets.get(plugin_id, 0) % len(images)
+        selected: List[Image.Image] = []
+        used = 0
+        consumed = 0
+
+        # Walk forward from the rotation offset, taking whole items only, so a
+        # cut never lands in the middle of one.
+        for step in range(len(images)):
+            img = images[(start + step) % len(images)]
+            cost = img.width + (gap if selected else 0)
+            if selected and used + cost > budget:
+                break
+            selected.append(img)
+            used += cost
+            consumed += 1
+
+        self._item_offsets[plugin_id] = (start + consumed) % len(images)
+
+        logger.info(
+            "[%s] Width budget %dpx: showing %d of %d row(s) (%dpx incl. gaps) "
+            "from offset %d; remainder deferred to a later cycle",
+            plugin_id, budget, len(selected), len(images), used, start
+        )
+        return selected
+
+    def _crop_to_budget(
+        self, img: Image.Image, budget: int, plugin_id: str
+    ) -> Image.Image:
+        """
+        Narrow a single oversized image to the budget, advancing a window
+        through it across cycles.
+
+        The cut is snapped to the nearest blank column so it does not slice
+        through a glyph or logo and leave half a character at the panel edge.
+        """
+        offset = self._item_offsets.get(plugin_id, 0)
+        if offset >= img.width:
+            offset = 0
+
+        # Snap both edges to blank columns. The search radius is generous
+        # enough to clear a wide glyph but small enough not to distort the
+        # requested budget much.
+        snap = max(8, self.display_width // 16)
+        start = find_blank_cut(img, offset, snap, self.config.trim_threshold)
+        end = find_blank_cut(
+            img, min(start + budget, img.width), snap, self.config.trim_threshold)
+        if end <= start:
+            end = min(start + budget, img.width)
+
+        # Next cycle resumes where this one stopped; wrap when the strip ends.
+        self._item_offsets[plugin_id] = 0 if end >= img.width else end
+
+        logger.info(
+            "[%s] Width budget %dpx: cropped single %dpx image to [%d:%d] "
+            "(%dpx), window advances next cycle",
+            plugin_id, budget, img.width, start, end, end - start
+        )
+        return img.crop((start, 0, end, img.height))
+
     def _get_native_content(
         self, plugin: 'BasePlugin', plugin_id: str
     ) -> Optional[List[Image.Image]]:

src/vegas_mode/render_pipeline.py

@@ -66,10 +66,6 @@ class RenderPipeline:
             else display_manager.height
         )
 
-        # Reusable blank frame for cycle-end pushes (allocated lazily,
-        # re-blacked before each reuse)
-        self._blank_frame = None
-
         # ScrollHelper for optimized scrolling
         self.scroll_helper = ScrollHelper(
             self.display_width,
@@ -141,23 +137,37 @@ class RenderPipeline:
             True if composition successful
         """
         try:
-            # Get all buffered content
+            # Content grouped by plugin, so a separator can be placed at the
-            images = self.stream_manager.get_all_content_for_composition()
+            # plugin boundaries only.
+            grouped = self.stream_manager.get_grouped_content_for_composition()
 
-            if not images:
+            if not grouped:
                 logger.warning("No content available for composition")
                 return False
 
-            # Add separator gaps between images
+            # Collapse each plugin's rows into a single block, joined by
-            content_with_gaps = []
+            # intra_plugin_gap. ScrollHelper applies one uniform gap between the
-            for i, img in enumerate(images):
+            # items it is given, so handing it one item per plugin is what makes
-                content_with_gaps.append(img)
+            # separator_width mean "between plugins" instead of "between every
+            # row". Without this, a per-row ticker such as the F1 scoreboard got
+            # the full separator between each of its ~116 rows.
+            blocks = []
+            total_rows = 0
+            for plugin_id, images in grouped:
+                total_rows += len(images)
+                blocks.append(self._join_plugin_rows(images))
 
-            # Create scrolling image via ScrollHelper
+            # Create scrolling image via ScrollHelper.
+            #
+            # lead_gap is explicit because ScrollHelper otherwise prepends a
+            # full display width of black — appropriate for a standalone ticker
+            # scrolling in from off-screen, but in Vegas mode it is charged
+            # once per cycle and reads as the panel switching off.
             self.scroll_helper.create_scrolling_image(
-                content_items=content_with_gaps,
+                content_items=blocks,
                 item_gap=self.config.separator_width,
-                element_gap=0
+                element_gap=0,
+                lead_gap=self.config.lead_in_width
             )
 
             # Verify scroll image was created successfully
@@ -177,11 +187,14 @@ class RenderPipeline:
             self._cycle_complete = False
 
             logger.info(
-                "Composed scroll image: %dx%d, %d plugins, %d items",
+                "Composed scroll image: %dx%d, %d plugin block(s), %d rows, "
+                "separator=%dpx between plugins / %dpx within",
                 self.scroll_helper.cached_image.width if self.scroll_helper.cached_image else 0,
                 self.display_height,
-                len(self._segments_in_scroll),
+                len(blocks),
-                len(images)
+                total_rows,
+                self.config.separator_width,
+                self.config.intra_plugin_gap,
             )
 
             return True
@@ -191,6 +204,32 @@ class RenderPipeline:
             logger.exception("Error composing scroll content")
             return False
 
+    def _join_plugin_rows(self, images: List[Image.Image]) -> Image.Image:
+        """
+        Concatenate one plugin's images into a single block.
+
+        Args:
+            images: That plugin's content, in order
+
+        Returns:
+            A single image with the rows laid out left to right, separated by
+            ``intra_plugin_gap``. Returned unchanged when there is only one row,
+            which is the common case and avoids a pointless copy.
+        """
+        if len(images) == 1:
+            return images[0]
+
+        gap = max(0, self.config.intra_plugin_gap)
+        width = sum(img.width for img in images) + gap * (len(images) - 1)
+        height = max(img.height for img in images)
+
+        block = Image.new('RGB', (width, height), (0, 0, 0))
+        x = 0
+        for img in images:
+            block.paste(img, (x, 0))
+            x += img.width + gap
+        return block
+
     def render_frame(self) -> bool:
         """
         Render a single frame to the display.
@@ -236,24 +275,17 @@ class RenderPipeline:
                         "Scroll cycle complete after %.1fs",
                         time.time() - self._cycle_start_time
                     )
-                    # Push blank immediately so the hardware never shows any
+                    # Deliberately leave the last rendered frame on the panel.
-                    # post-wrap content while the coordinator recomposes the
+                    #
-                    # next cycle (~100 ms). The blank is allocated once and
+                    # This used to push a blank frame so no post-wrap content
-                    # reused across cycle wraps (fresh paste each time in case
+                    # could be seen while the next cycle was composed. But
-                    # a consumer drew on the previous one).
+                    # recomposing is synchronous and fetches plugin content:
-                    try:
+                    # measured 84ms at best and 4.8s at worst on a 512px panel,
-                        if self._blank_frame is None or self._blank_frame.size != (
+                    # and every millisecond of it was black. Holding the last
-                                self.display_width, self.display_height):
+                    # frame instead turns that into a brief freeze, which reads
-                            self._blank_frame = Image.new(
+                    # as far less broken than the display switching off. The
-                                'RGB', (self.display_width, self.display_height))
+                    # frame is already past the end of the content, so there is
-                        else:
+                    # no second-pass content to leak.
-                            self._blank_frame.paste(
-                                (0, 0, 0),
-                                (0, 0, self.display_width, self.display_height))
-                        self.display_manager.image = self._blank_frame
-                        self.display_manager.update_display()
-                    except Exception:
-                        logger.exception("Failed to write blank frame to display at cycle end")
                 return True  # Cycle done; coordinator starts new cycle next frame
 
             # Get visible portion
@@ -415,11 +447,12 @@ class RenderPipeline:
         result = self.compose_scroll_content()
 
         if result and self.sync_manager:
-            # When sync is active, start the leader at display_width instead of 0.
+            # When sync is active, start the leader past the lead-in gap so it
-            # This skips the initial black gap so the leader immediately shows content.
+            # immediately shows content, leaving the follower on the blank gap
-            # The follower starts at position 0 (the gap) which looks like a clean
+            # for a clean transition rather than near-end content wrapping
-            # blank transition rather than near-end content wrapping around.
+            # around. This tracks lead_in_width rather than assuming a full
-            self.scroll_helper.scroll_position = float(self.display_width)
+            # display width of gap, which is no longer the default.
+            self.scroll_helper.scroll_position = float(self.config.lead_in_width)
 
         if result and self.sync_manager:
             # Signal follower that a new cycle started (triggers its own rebuild)

src/vegas_mode/stream_manager.py

@@ -14,7 +14,7 @@ Supports three display modes:
 import logging
 import threading
 import time
-from typing import Optional, List, Dict, Any, Deque, TYPE_CHECKING
+from typing import Optional, List, Dict, Any, Deque, Tuple, TYPE_CHECKING
 from collections import deque
 from dataclasses import dataclass, field
 from PIL import Image
@@ -116,8 +116,11 @@ class StreamManager:
             logger.warning("No plugins available for Vegas scroll")
             return False
 
-        # Prefetch initial content
+        # Fill the buffer to a whole cycle's worth of plugins. This used to be
-        self._prefetch_content(count=min(self.config.buffer_ahead + 1, len(self._ordered_plugins)))
+        # buffer_ahead + 1, which conflated prefetch depth with cycle size and
+        # meant a 20-plugin install only showed 3 plugins before recomposing.
+        self._prefetch_content(
+            count=min(self.config.plugins_per_cycle, len(self._ordered_plugins)))
 
         logger.info(
             "StreamManager initialized with %d plugins, %d segments buffered",
@@ -385,7 +388,7 @@ class StreamManager:
                 return
 
             for _ in range(count):
-                if len(self._active_buffer) >= self.config.buffer_ahead + 1:
+                if len(self._active_buffer) >= self.config.plugins_per_cycle:
                     break
 
                 # Ensure index is valid (guard against empty list)
@@ -521,28 +524,61 @@ class StreamManager:
             logger.debug("Refreshed content for %s in staging buffer", plugin_id)
 
     def _ensure_buffer_filled(self) -> None:
-        """Ensure buffer has enough content prefetched."""
+        """
-        if len(self._active_buffer) < self.config.buffer_ahead:
+        Top the buffer back up after segments have been served.
-            needed = self.config.buffer_ahead - len(self._active_buffer)
+
-            self._prefetch_content(count=needed)
+        buffer_ahead is the low-water mark only; plugins_per_cycle is the
+        ceiling and is enforced inside _prefetch_content.
+        """
+        low_water = min(self.config.buffer_ahead, self.config.plugins_per_cycle)
+        if len(self._active_buffer) < low_water:
+            self._prefetch_content(count=low_water - len(self._active_buffer))
 
     def get_all_content_for_composition(self) -> List[Image.Image]:
         """
         Get all buffered content as a flat list of images.
 
-        Used when composing the full scroll image.
         Skips STATIC segments as they don't have images to compose.
 
+        Prefer get_grouped_content_for_composition(): flattening loses the
+        plugin boundaries, which is what tells the compositor where a
+        separator belongs and where it does not.
+
         Returns:
             List of all images in buffer order
         """
         all_images = []
+        for _plugin_id, images in self.get_grouped_content_for_composition():
+            all_images.extend(images)
+        return all_images
+
+    def get_grouped_content_for_composition(self) -> List[Tuple[str, List[Image.Image]]]:
+        """
+        Get buffered content grouped by the plugin that produced it.
+
+        The grouping matters: separator_width is meant to mark the handoff from
+        one plugin to the next, not to sit between every row a single plugin
+        contributes. A per-row ticker like the F1 scoreboard returns over a
+        hundred images that it renders 4px apart internally, so flattening them
+        into one list and applying a uniform gap forced 32px between each of
+        its rows — both inconsistent with how the plugin looks standalone, and
+        a large hidden addition to the width it occupies.
+
+        Skips STATIC segments, which trigger a pause rather than contributing
+        scroll content, and segments left with no images.
+
+        Returns:
+            List of (plugin_id, images) in buffer order
+        """
+        grouped: List[Tuple[str, List[Image.Image]]] = []
         with self._buffer_lock:
             for segment in self._active_buffer:
-                # Skip STATIC segments - they trigger pauses, not scroll content
+                if segment.display_mode == VegasDisplayMode.STATIC:
-                if segment.display_mode != VegasDisplayMode.STATIC:
+                    continue
-                    all_images.extend(segment.images)
+                if not segment.images:
-        return all_images
+                    continue
+                grouped.append((segment.plugin_id, list(segment.images)))
+        return grouped
 
     def advance_cycle(self) -> None:
         """

test/test_vegas_density.py

@@ -0,0 +1,527 @@
+"""
+Tests for the Vegas mode density work: dead-space trimming in PluginAdapter
+and the configurable lead-in gap in ScrollHelper.
+"""
+
+import pytest
+from PIL import Image
+
+from src.common.scroll_helper import ScrollHelper
+from src.vegas_mode.config import VegasModeConfig
+from src.vegas_mode.geometry import column_has_ink
+from src.vegas_mode.plugin_adapter import PluginAdapter
+
+DISPLAY_W = 512
+DISPLAY_H = 64
+
+
+class FakeDisplayManager:
+    """Minimal stand-in; the native content path never touches the canvas."""
+
+    width = DISPLAY_W
+    height = DISPLAY_H
+
+    def __init__(self):
+        self.image = Image.new('RGB', (DISPLAY_W, DISPLAY_H))
+
+
+class NativePlugin:
+    """Plugin that returns pre-rendered Vegas content."""
+
+    def __init__(self, images):
+        self._images = images
+
+    def get_vegas_content(self):
+        return self._images
+
+
+def canvas(content_spans, width=DISPLAY_W, height=DISPLAY_H):
+    """Full-display canvas with white content in the given [x0, x1) spans."""
+    img = Image.new('RGB', (width, height), (0, 0, 0))
+    for x0, x1 in content_spans:
+        img.paste(Image.new('RGB', (x1 - x0, height), (255, 255, 255)), (x0, 0))
+    return img
+
+
+def adapter_with(**overrides):
+    cfg = VegasModeConfig(**overrides)
+    return PluginAdapter(FakeDisplayManager(), cfg)
+
+
+class TestAdapterTrimming:
+    def test_of_the_day_case_is_reclaimed(self):
+        # Measured on devpi: "No Data" occupying 35px of a 512px canvas bought
+        # 9.5s of black at 50px/s.
+        adapter = adapter_with(content_padding=0)
+        plugin = NativePlugin([canvas([(4, 39)])])
+        images = adapter.get_content(plugin, 'of-the-day')
+        assert len(images) == 1
+        assert images[0].width == 35
+
+    def test_youtube_stats_case_is_reclaimed(self):
+        # Centred 142px of content on a 512px canvas: 185px black each side.
+        adapter = adapter_with(content_padding=0)
+        plugin = NativePlugin([canvas([(185, 327)])])
+        images = adapter.get_content(plugin, 'youtube-stats')
+        assert images[0].width == 142
+
+    def test_padding_is_applied_within_available_margin(self):
+        adapter = adapter_with(content_padding=8)
+        plugin = NativePlugin([canvas([(185, 327)])])
+        images = adapter.get_content(plugin, 'youtube-stats')
+        assert images[0].width == 142 + 16
+
+    def test_interior_layout_gap_survives(self):
+        # A logo far left and a score far right is deliberate layout; closing
+        # the gap would corrupt the design rather than reclaim dead space.
+        adapter = adapter_with(content_padding=0)
+        plugin = NativePlugin([canvas([(10, 40), (400, 460)])])
+        images = adapter.get_content(plugin, 'scoreboard')
+        assert images[0].width == 450  # 10..459
+        assert int(column_has_ink(images[0]).sum()) == 90
+
+    def test_wide_legitimate_content_is_left_alone(self):
+        # geochron uses 444 of 512 columns; only the real tail should go.
+        adapter = adapter_with(content_padding=0)
+        plugin = NativePlugin([canvas([(0, 444)])])
+        images = adapter.get_content(plugin, 'geochron')
+        assert images[0].width == 444
+
+    def test_non_black_background_is_untouched(self):
+        adapter = adapter_with(content_padding=0)
+        bg = Image.new('RGB', (DISPLAY_W, DISPLAY_H), (0, 0, 40))
+        plugin = NativePlugin([bg])
+        images = adapter.get_content(plugin, 'weather')
+        assert images[0].width == DISPLAY_W
+
+    def test_each_image_of_a_multi_item_segment_is_trimmed(self):
+        # compose_scroll_content treats every image as its own item, so a
+        # per-image trim is what makes separator_width the real gap.
+        adapter = adapter_with(content_padding=0)
+        plugin = NativePlugin([
+            canvas([(168, 344)]),
+            canvas([(200, 300)]),
+            canvas([(0, 512)]),
+        ])
+        images = adapter.get_content(plugin, 'ledmatrix-flights')
+        assert [img.width for img in images] == [176, 100, 512]
+
+    def test_fully_blank_segment_contributes_nothing(self):
+        adapter = adapter_with()
+        plugin = NativePlugin([Image.new('RGB', (DISPLAY_W, DISPLAY_H))])
+        assert adapter.get_content(plugin, 'empty') is None
+
+    def test_blank_images_are_dropped_but_others_kept(self):
+        adapter = adapter_with(content_padding=0)
+        plugin = NativePlugin([
+            canvas([(100, 150)]),
+            Image.new('RGB', (DISPLAY_W, DISPLAY_H)),
+            canvas([(200, 260)]),
+        ])
+        images = adapter.get_content(plugin, 'mixed')
+        assert [img.width for img in images] == [50, 60]
+
+    def test_min_plugin_width_rejects_noise(self):
+        adapter = adapter_with(content_padding=0, min_plugin_width=32)
+        plugin = NativePlugin([canvas([(10, 14)])])
+        assert adapter.get_content(plugin, 'sliver') is None
+
+    def test_min_plugin_width_of_zero_keeps_everything(self):
+        adapter = adapter_with(content_padding=0, min_plugin_width=0)
+        plugin = NativePlugin([canvas([(10, 14)])])
+        images = adapter.get_content(plugin, 'sliver')
+        assert images[0].width == 4
+
+    def test_auto_trim_off_preserves_original_behaviour(self):
+        adapter = adapter_with(auto_trim=False)
+        plugin = NativePlugin([canvas([(4, 39)])])
+        images = adapter.get_content(plugin, 'of-the-day')
+        assert images[0].width == DISPLAY_W
+
+    def test_trim_threshold_ignores_near_black_noise(self):
+        # A very dark band should not be mistaken for content.
+        img = Image.new('RGB', (DISPLAY_W, DISPLAY_H), (0, 0, 0))
+        img.paste(Image.new('RGB', (100, DISPLAY_H), (3, 3, 3)), (0, 0))
+        img.paste(Image.new('RGB', (50, DISPLAY_H), (255, 255, 255)), (200, 0))
+        adapter = adapter_with(content_padding=0, trim_threshold=10)
+        images = adapter.get_content(NativePlugin([img]), 'noisy')
+        assert images[0].width == 50
+
+    def test_height_is_preserved_through_trim(self):
+        adapter = adapter_with(content_padding=0)
+        plugin = NativePlugin([canvas([(100, 200)])])
+        images = adapter.get_content(plugin, 'x')
+        assert images[0].height == DISPLAY_H
+
+    def test_trimmed_result_is_cached(self):
+        adapter = adapter_with(content_padding=0)
+        plugin = NativePlugin([canvas([(100, 200)])])
+        first = adapter.get_content(plugin, 'cached')
+        # Swap the plugin's content; the cache should still serve the old size.
+        plugin._images = [canvas([(0, 512)])]
+        second = adapter.get_content(plugin, 'cached')
+        assert first[0].width == second[0].width == 100
+
+    def test_default_adapter_construction_still_works(self):
+        # Existing callers pass only the display manager.
+        adapter = PluginAdapter(FakeDisplayManager())
+        assert adapter.config.auto_trim is True
+
+
+class TestWidthBudget:
+    def test_segment_within_budget_is_untouched(self):
+        adapter = adapter_with(content_padding=0, max_plugin_width_ratio=3.0)
+        plugin = NativePlugin([canvas([(0, 400)])])
+        assert adapter.get_content(plugin, 'small')[0].width == 400
+
+    def test_oversized_multi_item_segment_is_capped(self):
+        # 10 items of 400px = 4000px against a 1536px budget (3 x 512).
+        adapter = adapter_with(content_padding=0, max_plugin_width_ratio=3.0)
+        items = [canvas([(0, 400)], width=400) for _ in range(10)]
+        images = adapter.get_content(NativePlugin(items), 'stock-news')
+        assert sum(i.width for i in images) <= 3 * DISPLAY_W
+        assert len(images) == 3  # 1200px; a 4th would exceed 1536
+
+    def test_deferred_items_appear_on_the_next_cycle(self):
+        adapter = adapter_with(content_padding=0, max_plugin_width_ratio=1.0)
+        # Distinguish items by width so the rotation is observable.
+        items = [canvas([(0, w)], width=w) for w in (200, 210, 220, 230, 240)]
+        plugin = NativePlugin(items)
+
+        first = adapter.get_content(plugin, 'ticker')
+        adapter.invalidate_cache('ticker')
+        second = adapter.get_content(plugin, 'ticker')
+        assert [i.width for i in first] != [i.width for i in second]
+
+    def test_rotation_eventually_covers_every_item(self):
+        adapter = adapter_with(content_padding=0, max_plugin_width_ratio=1.0)
+        widths = (200, 210, 220, 230, 240)
+        items = [canvas([(0, w)], width=w) for w in widths]
+        plugin = NativePlugin(items)
+
+        seen = set()
+        for _ in range(10):
+            adapter.invalidate_cache('ticker')
+            for img in adapter.get_content(plugin, 'ticker'):
+                seen.add(img.width)
+        assert seen == set(widths)
+
+    def test_single_oversized_image_is_cropped_to_budget(self):
+        adapter = adapter_with(content_padding=0, max_plugin_width_ratio=1.0)
+        # A 6898px leaderboard strip, as measured on devpi. Solid ink means
+        # there is no blank column to snap to, so the cut lands on the budget.
+        plugin = NativePlugin([canvas([(0, 6898)], width=6898)])
+        images = adapter.get_content(plugin, 'ledmatrix-leaderboard')
+        assert len(images) == 1
+        assert images[0].width == DISPLAY_W
+
+    def test_single_image_crop_snaps_to_a_blank_column(self):
+        adapter = adapter_with(content_padding=0, max_plugin_width_ratio=1.0)
+        # Content blocks with gaps; the cut should land in a gap, not mid-block.
+        spans = [(x, x + 90) for x in range(0, 2000, 100)]
+        plugin = NativePlugin([canvas(spans, width=2000)])
+        images = adapter.get_content(plugin, 'gapped')
+        assert images[0].width != DISPLAY_W
+        assert abs(images[0].width - DISPLAY_W) <= DISPLAY_W // 16 + 1
+
+    def test_single_image_window_advances_across_cycles(self):
+        adapter = adapter_with(content_padding=0, max_plugin_width_ratio=1.0)
+        plugin = NativePlugin([canvas([(0, 3000)], width=3000)])
+        adapter.get_content(plugin, 'strip')
+        assert adapter._item_offsets['strip'] == DISPLAY_W
+
+    def test_budget_of_zero_disables_the_cap(self):
+        adapter = adapter_with(content_padding=0, max_plugin_width_ratio=0)
+        plugin = NativePlugin([canvas([(0, 6898)], width=6898)])
+        assert adapter.get_content(plugin, 'huge')[0].width == 6898
+
+    def test_rotation_resets_when_content_shrinks_to_fit(self):
+        adapter = adapter_with(content_padding=0, max_plugin_width_ratio=1.0)
+        big = [canvas([(0, 300)], width=300) for _ in range(5)]
+        adapter.get_content(NativePlugin(big), 'shrink')
+        assert 'shrink' in adapter._item_offsets
+
+        adapter.invalidate_cache('shrink')
+        adapter.get_content(NativePlugin([canvas([(0, 100)], width=100)]), 'shrink')
+        assert 'shrink' not in adapter._item_offsets
+
+    def test_one_item_wider_than_budget_is_still_shown(self):
+        # Never return nothing just because the first whole item overflows.
+        adapter = adapter_with(content_padding=0, max_plugin_width_ratio=1.0)
+        items = [canvas([(0, 900)], width=900), canvas([(0, 100)], width=100)]
+        images = adapter.get_content(NativePlugin(items), 'wide-first')
+        assert len(images) >= 1
+        assert images[0].width == 900
+
+
+class TestPluginBoundaryGaps:
+    """separator_width belongs between plugins; intra_plugin_gap within one."""
+
+    def _pipeline(self, grouped, **cfg):
+        from src.vegas_mode.render_pipeline import RenderPipeline
+
+        class FakeStream:
+            def get_grouped_content_for_composition(self):
+                return grouped
+
+            def get_active_plugin_ids(self):
+                return [pid for pid, _ in grouped]
+
+        class DM:
+            width = DISPLAY_W
+            height = DISPLAY_H
+
+            def set_scrolling_state(self, *a):
+                pass
+
+        return RenderPipeline(VegasModeConfig(**cfg), DM(), FakeStream())
+
+    def test_separator_only_at_plugin_boundaries(self):
+        # Two plugins, two rows each. Expect: row row [sep] row row, with the
+        # small intra gap inside each pair.
+        rows = [Image.new('RGB', (100, DISPLAY_H), (255, 255, 255)) for _ in range(4)]
+        pipeline = self._pipeline(
+            [('a', rows[:2]), ('b', rows[2:])],
+            separator_width=32, intra_plugin_gap=8, lead_in_width=0,
+        )
+        assert pipeline.compose_scroll_content()
+
+        ink = column_has_ink(pipeline.scroll_helper.cached_image)
+        assert ink[:100].all()
+        assert not ink[100:108].any()      # intra gap inside plugin a
+        assert ink[108:208].all()
+        assert not ink[208:240].any()      # separator between a and b
+        assert ink[240:340].all()
+        assert not ink[340:348].any()      # intra gap inside plugin b
+        assert ink[348:448].all()
+
+    def test_total_width_uses_both_gap_sizes(self):
+        rows = [Image.new('RGB', (100, DISPLAY_H), (255, 255, 255)) for _ in range(4)]
+        pipeline = self._pipeline(
+            [('a', rows[:2]), ('b', rows[2:])],
+            separator_width=32, intra_plugin_gap=8, lead_in_width=0,
+        )
+        pipeline.compose_scroll_content()
+        # 4 rows + 2 intra gaps + 1 separator
+        assert pipeline.scroll_helper.cached_image.width == 400 + 16 + 32
+
+    def test_f1_shaped_case_reclaims_the_chasms(self):
+        # 12 rows from one plugin: previously 11 separators at 32px = 352px of
+        # gap; now 11 intra gaps at 8px = 88px.
+        rows = [Image.new('RGB', (128, DISPLAY_H), (255, 255, 255)) for _ in range(12)]
+        pipeline = self._pipeline(
+            [('f1-scoreboard', rows)],
+            separator_width=32, intra_plugin_gap=8, lead_in_width=0,
+        )
+        pipeline.compose_scroll_content()
+        assert pipeline.scroll_helper.cached_image.width == 12 * 128 + 11 * 8
+
+    def test_single_row_plugin_image_is_not_copied(self):
+        row = Image.new('RGB', (100, DISPLAY_H), (255, 255, 255))
+        pipeline = self._pipeline([('solo', [row])], lead_in_width=0)
+        assert pipeline._join_plugin_rows([row]) is row
+
+    def test_zero_intra_gap_butts_rows_together(self):
+        rows = [Image.new('RGB', (50, DISPLAY_H), (255, 255, 255)) for _ in range(3)]
+        pipeline = self._pipeline(
+            [('a', rows)], separator_width=32, intra_plugin_gap=0, lead_in_width=0)
+        pipeline.compose_scroll_content()
+        assert pipeline.scroll_helper.cached_image.width == 150
+        assert column_has_ink(pipeline.scroll_helper.cached_image).all()
+
+    def test_empty_grouping_fails_composition(self):
+        pipeline = self._pipeline([])
+        assert pipeline.compose_scroll_content() is False
+
+
+class TestWidthBudgetCountsGaps:
+    def test_budget_accounts_for_intra_plugin_gaps(self):
+        # 8 rows of 200px = 1600px of pixels, but with 8px gaps the real
+        # occupancy is 1600 + 56 = 1656px. Against a 512px budget the row count
+        # must be chosen using the gap-inclusive cost.
+        adapter = adapter_with(
+            content_padding=0, max_plugin_width_ratio=1.0, intra_plugin_gap=8)
+        items = [canvas([(0, 200)], width=200) for _ in range(8)]
+        images = adapter.get_content(NativePlugin(items), 'rows')
+        n = len(images)
+        assert 200 * n + 8 * (n - 1) <= DISPLAY_W
+
+    def test_gap_free_config_fits_more_rows(self):
+        items = [canvas([(0, 200)], width=200) for _ in range(8)]
+        with_gap = adapter_with(
+            content_padding=0, max_plugin_width_ratio=1.0, intra_plugin_gap=64)
+        without = adapter_with(
+            content_padding=0, max_plugin_width_ratio=1.0, intra_plugin_gap=0)
+        assert len(without.get_content(NativePlugin(items), 'r')) >= \
+            len(with_gap.get_content(NativePlugin(items), 'r'))
+
+
+class TestStreamGrouping:
+    def _stream(self, segments):
+        from src.vegas_mode.stream_manager import StreamManager
+        from collections import deque
+
+        sm = StreamManager.__new__(StreamManager)
+        import threading
+        sm._buffer_lock = threading.RLock()
+        sm._active_buffer = deque(segments)
+        return sm
+
+    def _seg(self, plugin_id, count, mode=None):
+        from src.vegas_mode.stream_manager import ContentSegment
+        from src.plugin_system.base_plugin import VegasDisplayMode
+        imgs = [Image.new('RGB', (10, 8)) for _ in range(count)]
+        return ContentSegment(
+            plugin_id=plugin_id, images=imgs, total_width=10 * count,
+            display_mode=mode or VegasDisplayMode.SCROLL)
+
+    def test_grouping_preserves_plugin_boundaries(self):
+        sm = self._stream([self._seg('a', 3), self._seg('b', 1)])
+        grouped = sm.get_grouped_content_for_composition()
+        assert [(pid, len(imgs)) for pid, imgs in grouped] == [('a', 3), ('b', 1)]
+
+    def test_static_segments_are_skipped(self):
+        from src.plugin_system.base_plugin import VegasDisplayMode
+        sm = self._stream([
+            self._seg('a', 2),
+            self._seg('paused', 1, VegasDisplayMode.STATIC),
+            self._seg('b', 1),
+        ])
+        assert [pid for pid, _ in sm.get_grouped_content_for_composition()] == ['a', 'b']
+
+    def test_imageless_segments_are_skipped(self):
+        sm = self._stream([self._seg('a', 0), self._seg('b', 2)])
+        assert [pid for pid, _ in sm.get_grouped_content_for_composition()] == ['b']
+
+    def test_flat_accessor_still_matches_grouped_total(self):
+        sm = self._stream([self._seg('a', 3), self._seg('b', 2)])
+        assert len(sm.get_all_content_for_composition()) == 5
+
+
+class TestCycleSizing:
+    def test_plugins_per_cycle_defaults_above_buffer_ahead(self):
+        cfg = VegasModeConfig()
+        assert cfg.plugins_per_cycle == 6
+        assert cfg.plugins_per_cycle > cfg.buffer_ahead + 1
+
+    def test_plugins_per_cycle_parses(self):
+        cfg = VegasModeConfig.from_config(
+            {'display': {'vegas_scroll': {'plugins_per_cycle': 10}}})
+        assert cfg.plugins_per_cycle == 10
+
+    def test_max_plugin_width_ratio_parses(self):
+        cfg = VegasModeConfig.from_config(
+            {'display': {'vegas_scroll': {'max_plugin_width_ratio': 1.5}}})
+        assert cfg.max_plugin_width_ratio == 1.5
+
+    @pytest.mark.parametrize('overrides,bad_key', [
+        ({'plugins_per_cycle': 0}, 'plugins_per_cycle'),
+        ({'plugins_per_cycle': 99}, 'plugins_per_cycle'),
+        ({'max_plugin_width_ratio': -1.0}, 'max_plugin_width_ratio'),
+    ])
+    def test_validate_rejects_out_of_range(self, overrides, bad_key):
+        errors = VegasModeConfig(**overrides).validate()
+        assert any(bad_key in e for e in errors), errors
+
+
+class TestScrollHelperLeadGap:
+    def test_default_lead_gap_is_display_width(self):
+        # Standalone tickers rely on scrolling in from off-screen; that
+        # behaviour must not change for the many non-Vegas callers.
+        sh = ScrollHelper(128, 32)
+        sh.create_scrolling_image([Image.new('RGB', (100, 32), (255, 0, 0))],
+                                  item_gap=0, element_gap=0)
+        assert sh.cached_image.width == 128 + 100
+        assert not column_has_ink(sh.cached_image)[:128].any()
+
+    def test_zero_lead_gap_starts_on_content(self):
+        sh = ScrollHelper(128, 32)
+        sh.create_scrolling_image([Image.new('RGB', (100, 32), (255, 0, 0))],
+                                  item_gap=0, element_gap=0, lead_gap=0)
+        assert sh.cached_image.width == 100
+        assert column_has_ink(sh.cached_image)[0]
+
+    def test_explicit_lead_gap_is_honoured(self):
+        sh = ScrollHelper(128, 32)
+        sh.create_scrolling_image([Image.new('RGB', (100, 32), (255, 0, 0))],
+                                  item_gap=0, element_gap=0, lead_gap=16)
+        assert sh.cached_image.width == 116
+        ink = column_has_ink(sh.cached_image)
+        assert not ink[:16].any()
+        assert ink[16:].all()
+
+    def test_negative_lead_gap_is_clamped(self):
+        sh = ScrollHelper(128, 32)
+        sh.create_scrolling_image([Image.new('RGB', (100, 32), (255, 0, 0))],
+                                  item_gap=0, element_gap=0, lead_gap=-50)
+        assert sh.cached_image.width == 100
+
+    def test_total_scroll_width_matches_image(self):
+        # The cycle-complete check compares against total_scroll_width, so a
+        # mismatch here would cut cycles short or overrun them.
+        sh = ScrollHelper(128, 32)
+        items = [Image.new('RGB', (60, 32), (255, 0, 0)) for _ in range(3)]
+        sh.create_scrolling_image(items, item_gap=32, element_gap=0, lead_gap=0)
+        assert sh.total_scroll_width == sh.cached_image.width
+        assert sh.cached_image.width == 60 * 3 + 32 * 2
+
+    def test_item_gaps_are_unaffected_by_lead_gap(self):
+        sh = ScrollHelper(128, 32)
+        items = [Image.new('RGB', (10, 32), (255, 0, 0)) for _ in range(2)]
+        sh.create_scrolling_image(items, item_gap=20, element_gap=0, lead_gap=0)
+        ink = column_has_ink(sh.cached_image)
+        assert ink[:10].all()
+        assert not ink[10:30].any()
+        assert ink[30:40].all()
+
+
+class TestConfigSurface:
+    def test_new_keys_parse_from_config(self):
+        cfg = VegasModeConfig.from_config({'display': {'vegas_scroll': {
+            'auto_trim': False,
+            'trim_threshold': 25,
+            'content_padding': 4,
+            'min_plugin_width': 64,
+            'lead_in_width': 32,
+        }}})
+        assert cfg.auto_trim is False
+        assert cfg.trim_threshold == 25
+        assert cfg.content_padding == 4
+        assert cfg.min_plugin_width == 64
+        assert cfg.lead_in_width == 32
+
+    def test_defaults_favour_trimming(self):
+        cfg = VegasModeConfig.from_config({})
+        assert cfg.auto_trim is True
+        assert cfg.lead_in_width == 0
+        assert cfg.content_padding == 8
+
+    def test_round_trips_through_to_dict(self):
+        cfg = VegasModeConfig(trim_threshold=20, lead_in_width=64)
+        restored = VegasModeConfig.from_config(
+            {'display': {'vegas_scroll': cfg.to_dict()}})
+        assert restored.trim_threshold == 20
+        assert restored.lead_in_width == 64
+
+    def test_update_applies_new_keys(self):
+        cfg = VegasModeConfig()
+        cfg.update({'display': {'vegas_scroll': {'content_padding': 16}}})
+        assert cfg.content_padding == 16
+
+    @pytest.mark.parametrize('overrides,bad_key', [
+        ({'trim_threshold': 300}, 'trim_threshold'),
+        ({'trim_threshold': -1}, 'trim_threshold'),
+        ({'content_padding': -5}, 'content_padding'),
+        ({'content_padding': 500}, 'content_padding'),
+        ({'min_plugin_width': -1}, 'min_plugin_width'),
+        ({'lead_in_width': -1}, 'lead_in_width'),
+    ])
+    def test_validate_rejects_out_of_range(self, overrides, bad_key):
+        errors = VegasModeConfig(**overrides).validate()
+        assert any(bad_key in e for e in errors), errors
+
+    def test_valid_config_has_no_errors(self):
+        assert VegasModeConfig(
+            trim_threshold=10, content_padding=8,
+            min_plugin_width=8, lead_in_width=0,
+        ).validate() == []

test/test_vegas_geometry.py

@@ -0,0 +1,273 @@
+"""Tests for Vegas mode geometry primitives."""
+
+import numpy as np
+import pytest
+from PIL import Image
+
+from src.vegas_mode.geometry import (
+    DEFAULT_INK_THRESHOLD,
+    column_has_ink,
+    content_bounds,
+    dead_window_stats,
+    trim_to_content,
+    window_coverage_stats,
+)
+
+
+def make_img(width, height=8, fill=(0, 0, 0)):
+    return Image.new('RGB', (width, height), fill)
+
+
+def paint(img, x0, x1, color=(255, 255, 255)):
+    """Fill columns [x0, x1) with a colour."""
+    block = Image.new('RGB', (x1 - x0, img.height), color)
+    img.paste(block, (x0, 0))
+    return img
+
+
+class TestColumnHasInk:
+    def test_all_black_has_no_ink(self):
+        assert not column_has_ink(make_img(16)).any()
+
+    def test_marks_only_painted_columns(self):
+        img = paint(make_img(16), 4, 8)
+        ink = column_has_ink(img)
+        assert ink.tolist() == [False] * 4 + [True] * 4 + [False] * 8
+
+    def test_threshold_is_exclusive(self):
+        # A pixel exactly at the threshold is not ink; one above it is.
+        at = paint(make_img(4), 0, 4, (DEFAULT_INK_THRESHOLD,) * 3)
+        above = paint(make_img(4), 0, 4, (DEFAULT_INK_THRESHOLD + 1,) * 3)
+        assert not column_has_ink(at).any()
+        assert column_has_ink(above).all()
+
+    def test_single_bright_channel_counts(self):
+        img = paint(make_img(4), 1, 2, (0, 0, 200))
+        assert column_has_ink(img).tolist() == [False, True, False, False]
+
+    def test_one_lit_pixel_lights_the_column(self):
+        img = make_img(4, height=8)
+        img.putpixel((2, 5), (255, 255, 255))
+        assert column_has_ink(img).tolist() == [False, False, True, False]
+
+
+class TestContentBounds:
+    def test_blank_returns_none(self):
+        assert content_bounds(make_img(16)) is None
+
+    def test_finds_inclusive_bounds(self):
+        assert content_bounds(paint(make_img(20), 5, 12)) == (5, 11)
+
+    def test_full_width_content(self):
+        assert content_bounds(paint(make_img(10), 0, 10)) == (0, 9)
+
+    def test_spans_interior_gap(self):
+        img = paint(make_img(30), 2, 5)
+        paint(img, 20, 25)
+        assert content_bounds(img) == (2, 24)
+
+
+class TestTrimToContent:
+    def test_blank_image_reports_blank(self):
+        result = trim_to_content(make_img(512))
+        assert result.is_blank
+        assert result.image is None
+        assert result.width == 0
+        assert result.original_width == 512
+
+    def test_trims_both_edges(self):
+        result = trim_to_content(paint(make_img(512), 100, 150))
+        assert not result.is_blank
+        assert result.width == 50
+        assert result.trimmed_left == 100
+        assert result.trimmed_right == 362
+        assert result.removed == 462
+
+    def test_preserves_interior_gap(self):
+        # Two content blocks with a wide blank between them: the gap is the
+        # plugin's layout and must survive trimming.
+        img = paint(make_img(400), 50, 80)
+        paint(img, 300, 330)
+        result = trim_to_content(img)
+        assert result.width == 280  # 50..329 inclusive
+        assert column_has_ink(result.image).sum() == 60
+
+    def test_full_width_content_is_returned_unchanged(self):
+        img = paint(make_img(128), 0, 128)
+        result = trim_to_content(img)
+        assert result.image is img
+        assert result.removed == 0
+
+    def test_non_black_background_is_never_trimmed(self):
+        # A plugin drawing on a dark-but-not-black background fills every
+        # column with ink, so there is nothing to reclaim.
+        result = trim_to_content(make_img(256, fill=(0, 0, 40)))
+        assert result.removed == 0
+        assert result.width == 256
+
+    def test_padding_keeps_margin_up_to_what_exists(self):
+        result = trim_to_content(paint(make_img(512), 100, 150), padding=8)
+        assert result.trimmed_left == 92
+        assert result.width == 66  # 50 content + 8 each side
+
+    def test_padding_cannot_widen_beyond_original(self):
+        # Content starts 2px in; padding of 8 can only reclaim the 2 available.
+        result = trim_to_content(paint(make_img(64), 2, 60), padding=8)
+        assert result.trimmed_left == 0
+        assert result.trimmed_right == 0
+        assert result.width == 64
+
+    def test_height_is_preserved(self):
+        result = trim_to_content(paint(make_img(200, height=64), 10, 20))
+        assert result.image.height == 64
+
+    def test_real_world_of_the_day_case(self):
+        # Measured on devpi: "No Data" occupying 35px of a 512px canvas.
+        result = trim_to_content(paint(make_img(512, height=64), 4, 39))
+        assert result.width == 35
+        assert result.removed == 477
+
+
+class TestDeadWindowStats:
+    def test_fully_inked_ticker_has_no_dead_windows(self):
+        stats = dead_window_stats(paint(make_img(400), 0, 400), viewport_width=100)
+        assert stats.dead_windows == 0
+        assert stats.dead_ratio == 0.0
+        assert stats.longest_dead_run == 0
+
+    def test_fully_blank_ticker_is_all_dead(self):
+        stats = dead_window_stats(make_img(400), viewport_width=100)
+        assert stats.total_windows == 301
+        assert stats.dead_windows == 301
+        assert stats.dead_ratio == 1.0
+        assert stats.longest_dead_run == 301
+
+    def test_leading_blank_run_is_measured(self):
+        # 512px of black then solid content: windows fully inside the black
+        # stretch are dead. With a 100px viewport, starts 0..412 exist and a
+        # window is dead while it holds >=95 blank columns.
+        img = paint(make_img(1024), 512, 1024)
+        stats = dead_window_stats(img, viewport_width=100)
+        assert stats.dead_windows == 418  # starts 0..417 keep >=95 blank cols
+        assert stats.longest_dead_run == 418
+
+    def test_narrow_content_island_still_leaves_dead_windows(self):
+        # 35px of content in a 512px field, viewed 100px at a time: no window
+        # can be 95% blank once it overlaps 35 lit columns, but the windows
+        # clear of it are dead.
+        img = paint(make_img(512), 100, 135)
+        stats = dead_window_stats(img, viewport_width=100)
+        assert stats.dead_windows > 0
+        assert stats.dead_ratio == pytest.approx(
+            stats.dead_windows / stats.total_windows
+        )
+
+    def test_step_reduces_sampling(self):
+        img = paint(make_img(1000), 500, 1000)
+        exact = dead_window_stats(img, viewport_width=100, step=1)
+        strided = dead_window_stats(img, viewport_width=100, step=10)
+        assert strided.total_windows < exact.total_windows
+        # Same underlying shape, so the ratios should stay close.
+        assert strided.dead_ratio == pytest.approx(exact.dead_ratio, abs=0.02)
+
+    def test_image_narrower_than_viewport_is_one_window(self):
+        stats = dead_window_stats(make_img(50), viewport_width=100)
+        assert stats.total_windows == 1
+        assert stats.dead_windows == 1
+
+    def test_zero_viewport_is_handled(self):
+        stats = dead_window_stats(make_img(50), viewport_width=0)
+        assert stats.total_windows == 0
+        assert stats.dead_ratio == 0.0
+
+    def test_longest_run_picks_the_larger_of_two_gaps(self):
+        # Short blank gap, content, then a long blank gap.
+        img = make_img(1000)
+        paint(img, 150, 400)
+        paint(img, 500, 520)
+        stats = dead_window_stats(img, viewport_width=100)
+        # The 400..500 gap is only 100 wide; the tail from 520 is 480 wide.
+        assert stats.longest_dead_run >= 380
+
+
+class TestWindowCoverageStats:
+    def test_solid_content_is_fully_covered(self):
+        stats = window_coverage_stats(paint(make_img(600), 0, 600), viewport_width=100)
+        assert stats.mean_ink_ratio == 1.0
+        assert stats.min_ink_ratio == 1.0
+        assert stats.sparse_windows == 0
+
+    def test_blank_strip_is_entirely_sparse(self):
+        stats = window_coverage_stats(make_img(600), viewport_width=100)
+        assert stats.mean_ink_ratio == 0.0
+        assert stats.sparse_ratio == 1.0
+
+    def test_catches_sliver_windows_that_dead_ratio_misses(self):
+        # Narrow content islands separated by more than the viewport. A window
+        # holding one whole 40px island carries 472 blank columns — under the
+        # 486 needed to count as "dead" — yet only 7.8% ink, so it still reads
+        # as an empty panel. Coverage must flag strictly more positions than
+        # the dead-window scan does.
+        img = paint(make_img(2000), 0, 40)
+        paint(img, 1000, 1040)
+        dead = dead_window_stats(img, viewport_width=512)
+        cover = window_coverage_stats(img, viewport_width=512, sparse_ink_ratio=0.10)
+        assert cover.sparse_windows > dead.dead_windows
+        assert cover.min_ink_ratio == 0.0
+
+    def test_adjacent_full_width_segments_stay_partially_covered(self):
+        # Documents why the dead-window scan alone understated the problem:
+        # two 512px segments with mid-canvas content never fully blank the
+        # viewport, they just hold it at a thin ~28%.
+        img = paint(make_img(1024), 185, 330)
+        paint(img, 697, 842)
+        dead = dead_window_stats(img, viewport_width=512)
+        cover = window_coverage_stats(img, viewport_width=512)
+        assert dead.dead_windows == 0
+        assert cover.mean_ink_ratio == pytest.approx(0.283, abs=0.01)
+
+    def test_min_ink_ratio_finds_the_worst_position(self):
+        # A wide blank tail guarantees at least one totally empty viewport.
+        img = paint(make_img(1200), 0, 200)
+        stats = window_coverage_stats(img, viewport_width=200)
+        assert stats.min_ink_ratio == 0.0
+        assert stats.mean_ink_ratio > 0.0
+
+    def test_sparse_threshold_is_respected(self):
+        # 40 inked columns in a 200px viewport = 20% coverage everywhere the
+        # island is fully inside the window.
+        img = paint(make_img(400), 100, 140)
+        lenient = window_coverage_stats(img, viewport_width=200, sparse_ink_ratio=0.05)
+        strict = window_coverage_stats(img, viewport_width=200, sparse_ink_ratio=0.50)
+        assert strict.sparse_windows > lenient.sparse_windows
+
+    def test_step_approximates_exact_scan(self):
+        img = paint(make_img(2000), 300, 500)
+        paint(img, 1200, 1400)
+        exact = window_coverage_stats(img, viewport_width=512, step=1)
+        strided = window_coverage_stats(img, viewport_width=512, step=4)
+        assert strided.mean_ink_ratio == pytest.approx(exact.mean_ink_ratio, abs=0.01)
+
+    def test_zero_viewport_is_handled(self):
+        stats = window_coverage_stats(make_img(50), viewport_width=0)
+        assert stats.total_windows == 0
+        assert stats.sparse_ratio == 0.0
+
+    def test_image_narrower_than_viewport(self):
+        stats = window_coverage_stats(paint(make_img(50), 0, 50), viewport_width=100)
+        assert stats.total_windows == 1
+        assert stats.mean_ink_ratio == pytest.approx(0.5)
+
+
+class TestLongestRunHelper:
+    @pytest.mark.parametrize("flags,expected", [
+        ([], 0),
+        ([False, False], 0),
+        ([True], 1),
+        ([True, True, False, True], 2),
+        ([False, True, True, True, False, True], 3),
+        ([True, True, True], 3),
+    ])
+    def test_run_lengths(self, flags, expected):
+        from src.vegas_mode.geometry import _longest_true_run
+        assert _longest_true_run(np.array(flags, dtype=bool)) == expected

web_interface/blueprints/api_v3.py

@@ -918,7 +918,12 @@ def save_main_config():
 
         # Handle Vegas scroll mode settings
         vegas_fields = ['vegas_scroll_enabled', 'vegas_scroll_speed', 'vegas_separator_width',
-                       'vegas_target_fps', 'vegas_buffer_ahead', 'vegas_plugin_order', 'vegas_excluded_plugins']
+                       'vegas_target_fps', 'vegas_buffer_ahead', 'vegas_plugin_order', 'vegas_excluded_plugins',
+                       'vegas_auto_trim', 'vegas_trim_threshold', 'vegas_content_padding',
+                       'vegas_min_plugin_width', 'vegas_lead_in_width', 'vegas_plugins_per_cycle',
+                       'vegas_max_plugin_width_ratio', 'vegas_dynamic_duration_enabled',
+                       'vegas_min_cycle_duration', 'vegas_max_cycle_duration',
+                       'vegas_intra_plugin_gap']
 
         if any(k in data for k in vegas_fields):
             if 'display' not in current_config:
@@ -933,13 +938,43 @@ def save_main_config():
             # was submitted (any vegas field present) but enabled key is missing,
             # the checkbox was unchecked and we should set enabled=False
             vegas_config['enabled'] = _coerce_to_bool(data.get('vegas_scroll_enabled'))
+            vegas_config['auto_trim'] = _coerce_to_bool(data.get('vegas_auto_trim'))
+            vegas_config['dynamic_duration_enabled'] = _coerce_to_bool(
+                data.get('vegas_dynamic_duration_enabled'))
+
+            # max_plugin_width_ratio is the one fractional setting, so it is
+            # handled outside the integer loop below.
+            if data.get('vegas_max_plugin_width_ratio') not in ('', None):
+                try:
+                    ratio = float(data['vegas_max_plugin_width_ratio'])
+                except (ValueError, TypeError):
+                    return jsonify({
+                        'status': 'error',
+                        'message': "Invalid value for vegas_max_plugin_width_ratio: "
+                                   "must be a number"
+                    }), 400
+                if not (0 <= ratio <= 20):
+                    return jsonify({
+                        'status': 'error',
+                        'message': "Invalid value for vegas_max_plugin_width_ratio: "
+                                   "must be between 0 and 20 (0 disables the cap)"
+                    }), 400
+                vegas_config['max_plugin_width_ratio'] = ratio
 
             # Handle numeric settings with validation
             numeric_fields = {
                 'vegas_scroll_speed': ('scroll_speed', 1, 100),
                 'vegas_separator_width': ('separator_width', 0, 500),
+                'vegas_intra_plugin_gap': ('intra_plugin_gap', 0, 128),
                 'vegas_target_fps': ('target_fps', 1, 200),
                 'vegas_buffer_ahead': ('buffer_ahead', 1, 20),
+                'vegas_trim_threshold': ('trim_threshold', 0, 254),
+                'vegas_content_padding': ('content_padding', 0, 128),
+                'vegas_min_plugin_width': ('min_plugin_width', 0, 512),
+                'vegas_lead_in_width': ('lead_in_width', 0, 2048),
+                'vegas_plugins_per_cycle': ('plugins_per_cycle', 1, 50),
+                'vegas_min_cycle_duration': ('min_cycle_duration', 5, 3600),
+                'vegas_max_cycle_duration': ('max_cycle_duration', 10, 3600),
             }
             for field_name, (config_key, min_val, max_val) in numeric_fields.items():
                 if field_name in data:

web_interface/templates/v3/partials/display.html

@@ -425,7 +425,7 @@
                     </div>
 
                     <div class="form-group" id="setting-display-vegas_separator_width" data-setting-key="display.vegas_scroll.separator_width">
-                        <label for="vegas_separator_width" class="block text-sm font-medium text-gray-700">Separator Width (pixels){{ ui.help_tip('Blank gap inserted between each plugin block in the ticker (0–128 px).\nDefault: 32. Larger values make the boundary between plugins clearer.', 'Separator Width') }}</label>
+                        <label for="vegas_separator_width" class="block text-sm font-medium text-gray-700">Separator Width (pixels){{ ui.help_tip('Blank gap where one plugin hands off to the next (0–128 px).\nDefault: 32. Larger values make the boundary between plugins clearer. This does not apply between rows of the same plugin — see Row Gap for that.', 'Separator Width') }}</label>
                         <input type="number"
                                id="vegas_separator_width"
                                name="vegas_separator_width"
@@ -436,6 +436,19 @@
                     </div>
                 </div>
 
+                <div class="grid grid-cols-1 md:grid-cols-2 gap-4">
+                    <div class="form-group" id="setting-display-vegas_intra_plugin_gap" data-setting-key="display.vegas_scroll.intra_plugin_gap">
+                        <label for="vegas_intra_plugin_gap" class="block text-sm font-medium text-gray-700">Row Gap (pixels){{ ui.help_tip('Gap between rows contributed by the same plugin (0–128 px).\nDefault: 8. Multi-row plugins such as sports scoreboards, news feeds and the F1 standings return one image per row; this keeps those rows close together while Separator Width still marks the jump to the next plugin. Set 0 to butt rows directly together.', 'Row Gap') }}</label>
+                        <input type="number"
+                               id="vegas_intra_plugin_gap"
+                               name="vegas_intra_plugin_gap"
+                               value="{{ main_config.display.get('vegas_scroll', {}).get('intra_plugin_gap', 8) }}"
+                               min="0"
+                               max="128"
+                               class="form-control">
+                    </div>
+                </div>
+
                 <div class="grid grid-cols-1 md:grid-cols-2 gap-4">
                     <div class="form-group" id="setting-display-vegas_target_fps" data-setting-key="display.vegas_scroll.target_fps">
                         <label for="vegas_target_fps" class="block text-sm font-medium text-gray-700">Target FPS{{ ui.help_tip('Frames per second the Vegas ticker aims to render.\nHigher = smoother scrolling but more CPU. Default: 125 (smoothest). Drop to 60/90 if the Pi runs hot.', 'Target FPS') }}</label>
@@ -456,6 +469,137 @@
                     </div>
                 </div>
 
+                <!-- Cycle Pacing -->
+                <div class="mt-4 pt-4 border-t border-gray-200">
+                    <h4 class="text-sm font-medium text-gray-900 mb-3">Cycle Pacing</h4>
+                    <p class="text-sm text-gray-600 mb-3">How long one pass through the ticker lasts, and how many plugins it covers.</p>
+
+                    <div class="grid grid-cols-1 md:grid-cols-2 gap-4">
+                        <div class="form-group" id="setting-display-vegas_plugins_per_cycle" data-setting-key="display.vegas_scroll.plugins_per_cycle">
+                            <label for="vegas_plugins_per_cycle" class="block text-sm font-medium text-gray-700">Plugins Per Cycle{{ ui.help_tip('How many plugins are composed into one pass of the ticker (1–50).\nDefault: 6. Higher means more variety before the ticker restarts, and fewer recompose pauses. Lower means each plugin comes around sooner.', 'Plugins Per Cycle') }}</label>
+                            <input type="number"
+                                   id="vegas_plugins_per_cycle"
+                                   name="vegas_plugins_per_cycle"
+                                   value="{{ main_config.display.get('vegas_scroll', {}).get('plugins_per_cycle', 6) }}"
+                                   min="1"
+                                   max="50"
+                                   class="form-control">
+                        </div>
+
+                        <div class="form-group" id="setting-display-vegas_max_plugin_width_ratio" data-setting-key="display.vegas_scroll.max_plugin_width_ratio">
+                            <label for="vegas_max_plugin_width_ratio" class="block text-sm font-medium text-gray-700">Max Plugin Width (screens){{ ui.help_tip('Caps how much of one cycle a single plugin may occupy, measured in screen widths (0–20).\nDefault: 3. A long ticker such as a news feed or leaderboard is trimmed to this and the remainder shown on later cycles, so one plugin cannot hold the display for minutes. Set 0 for no limit.', 'Max Plugin Width') }}</label>
+                            <input type="number"
+                                   id="vegas_max_plugin_width_ratio"
+                                   name="vegas_max_plugin_width_ratio"
+                                   value="{{ main_config.display.get('vegas_scroll', {}).get('max_plugin_width_ratio', 3.0) }}"
+                                   min="0"
+                                   max="20"
+                                   step="0.5"
+                                   class="form-control">
+                        </div>
+                    </div>
+
+                    <div class="form-group mt-4" id="setting-display-vegas_dynamic_duration_enabled" data-setting-key="display.vegas_scroll.dynamic_duration_enabled">
+                        <label class="flex items-center">
+                            <input type="checkbox"
+                                   id="vegas_dynamic_duration_enabled"
+                                   name="vegas_dynamic_duration_enabled"
+                                   {% if main_config.display.get('vegas_scroll', {}).get('dynamic_duration_enabled', True) %}checked{% endif %}
+                                   class="form-checkbox">
+                            <span class="ml-2 text-sm text-gray-700">Size cycle time to the content{{ ui.help_tip('When on, each cycle runs just long enough to scroll all its content past, clamped to the min and max below.\nWhen off, the max is always used. Default: on.', 'Dynamic Cycle Duration') }}</span>
+                        </label>
+                    </div>
+
+                    <div class="grid grid-cols-1 md:grid-cols-2 gap-4 mt-2">
+                        <div class="form-group" id="setting-display-vegas_min_cycle_duration" data-setting-key="display.vegas_scroll.min_cycle_duration">
+                            <label for="vegas_min_cycle_duration" class="block text-sm font-medium text-gray-700">Min Cycle Time (seconds){{ ui.help_tip('Shortest a single ticker pass may last (5–3600 s).\nDefault: 60.', 'Min Cycle Time') }}</label>
+                            <input type="number"
+                                   id="vegas_min_cycle_duration"
+                                   name="vegas_min_cycle_duration"
+                                   value="{{ main_config.display.get('vegas_scroll', {}).get('min_cycle_duration', 60) }}"
+                                   min="5"
+                                   max="3600"
+                                   class="form-control">
+                        </div>
+
+                        <div class="form-group" id="setting-display-vegas_max_cycle_duration" data-setting-key="display.vegas_scroll.max_cycle_duration">
+                            <label for="vegas_max_cycle_duration" class="block text-sm font-medium text-gray-700">Max Cycle Time (seconds){{ ui.help_tip('Longest a single ticker pass may last before it restarts with fresh content (10–3600 s).\nThis is the setting that caps total Vegas scroll time. Default: 240. Lower it if the ticker feels like it takes too long to come back around.', 'Max Cycle Time') }}</label>
+                            <input type="number"
+                                   id="vegas_max_cycle_duration"
+                                   name="vegas_max_cycle_duration"
+                                   value="{{ main_config.display.get('vegas_scroll', {}).get('max_cycle_duration', 240) }}"
+                                   min="10"
+                                   max="3600"
+                                   class="form-control">
+                        </div>
+                    </div>
+                </div>
+
+                <!-- Dead Space -->
+                <div class="mt-4 pt-4 border-t border-gray-200">
+                    <h4 class="text-sm font-medium text-gray-900 mb-3">Dead Space</h4>
+                    <p class="text-sm text-gray-600 mb-3">Plugins that draw onto a full-screen canvas contribute all the empty space around their content. Trimming reclaims it so the ticker stays full.</p>
+
+                    <div class="form-group mb-4" id="setting-display-vegas_auto_trim" data-setting-key="display.vegas_scroll.auto_trim">
+                        <label class="flex items-center">
+                            <input type="checkbox"
+                                   id="vegas_auto_trim"
+                                   name="vegas_auto_trim"
+                                   {% if main_config.display.get('vegas_scroll', {}).get('auto_trim', True) %}checked{% endif %}
+                                   class="form-checkbox">
+                            <span class="ml-2 text-sm text-gray-700">Trim empty edges from plugin content{{ ui.help_tip('Crops blank columns from the left and right of each plugin block before it enters the ticker. Space between two pieces of content inside a block is left alone, so layouts are not altered. Default: on.', 'Auto Trim') }}</span>
+                        </label>
+                    </div>
+
+                    <div class="grid grid-cols-1 md:grid-cols-2 gap-4">
+                        <div class="form-group" id="setting-display-vegas_content_padding" data-setting-key="display.vegas_scroll.content_padding">
+                            <label for="vegas_content_padding" class="block text-sm font-medium text-gray-700">Content Padding (pixels){{ ui.help_tip('Blank columns kept either side of trimmed content, so it does not butt against the separator (0–128 px).\nDefault: 8.', 'Content Padding') }}</label>
+                            <input type="number"
+                                   id="vegas_content_padding"
+                                   name="vegas_content_padding"
+                                   value="{{ main_config.display.get('vegas_scroll', {}).get('content_padding', 8) }}"
+                                   min="0"
+                                   max="128"
+                                   class="form-control">
+                        </div>
+
+                        <div class="form-group" id="setting-display-vegas_lead_in_width" data-setting-key="display.vegas_scroll.lead_in_width">
+                            <label for="vegas_lead_in_width" class="block text-sm font-medium text-gray-700">Lead-In Gap (pixels){{ ui.help_tip('Blank space before the first plugin of each cycle (0–2048 px).\nDefault: 0. Anything approaching your screen width reads as the display switching off at the start of every cycle.', 'Lead-In Gap') }}</label>
+                            <input type="number"
+                                   id="vegas_lead_in_width"
+                                   name="vegas_lead_in_width"
+                                   value="{{ main_config.display.get('vegas_scroll', {}).get('lead_in_width', 0) }}"
+                                   min="0"
+                                   max="2048"
+                                   class="form-control">
+                        </div>
+                    </div>
+
+                    <div class="grid grid-cols-1 md:grid-cols-2 gap-4 mt-4">
+                        <div class="form-group" id="setting-display-vegas_trim_threshold" data-setting-key="display.vegas_scroll.trim_threshold">
+                            <label for="vegas_trim_threshold" class="block text-sm font-medium text-gray-700">Trim Threshold{{ ui.help_tip('How bright a pixel must be to count as content rather than empty space (0–254).\nDefault: 10, which ignores the near-black noise left by image compression. Raise it if very dark artwork is being kept; lower it if dark detail is being cropped.', 'Trim Threshold') }}</label>
+                            <input type="number"
+                                   id="vegas_trim_threshold"
+                                   name="vegas_trim_threshold"
+                                   value="{{ main_config.display.get('vegas_scroll', {}).get('trim_threshold', 10) }}"
+                                   min="0"
+                                   max="254"
+                                   class="form-control">
+                        </div>
+
+                        <div class="form-group" id="setting-display-vegas_min_plugin_width" data-setting-key="display.vegas_scroll.min_plugin_width">
+                            <label for="vegas_min_plugin_width" class="block text-sm font-medium text-gray-700">Min Plugin Width (pixels){{ ui.help_tip('Plugin blocks narrower than this after trimming are skipped for that cycle (0–512 px).\nDefault: 8. Raise it to hide plugins showing only a tiny placeholder such as "No Data" until they have real content.', 'Min Plugin Width') }}</label>
+                            <input type="number"
+                                   id="vegas_min_plugin_width"
+                                   name="vegas_min_plugin_width"
+                                   value="{{ main_config.display.get('vegas_scroll', {}).get('min_plugin_width', 8) }}"
+                                   min="0"
+                                   max="512"
+                                   class="form-control">
+                        </div>
+                    </div>
+                </div>
+
                 <!-- Plugin Order Section -->
                 <div class="mt-4 pt-4 border-t border-gray-200">
                     <h4 class="text-sm font-medium text-gray-900 mb-3">Plugin Order</h4>