diff --git a/requirements.txt b/requirements.txt
index 4cf59c3..81389a7 100644
--- a/requirements.txt
+++ b/requirements.txt
@@ -6,3 +6,5 @@ tqdm>=4.66.0
 piexif>=1.1.3
 Pillow>=10.0.0
 pytest>=8.0
+imageio>=2.34
+imageio-ffmpeg>=0.5
diff --git a/scripts/build_mosaic_movie.py b/scripts/build_mosaic_movie.py
new file mode 100644
index 0000000..8c420c7
--- /dev/null
+++ b/scripts/build_mosaic_movie.py
@@ -0,0 +1,710 @@
+#!/usr/bin/env python3
+"""
+Build a chronological MP4 from downloaded mosaic.jpg files for one machine ROI.
+
+Reads archives/scans.csv, filters by machine and mosaic_on_disk, optionally
+restricts to one (start_x, start_y, end_x, end_y) ROI, dedupes by scan_id
+(last row wins), sorts by scan_time, and encodes frames with imageio/ffmpeg.
+
+Usage:
+  .venv/bin/python scripts/build_mosaic_movie.py --machine "BW1-4 [AMR-15]"
+  .venv/bin/python scripts/build_mosaic_movie.py --machine "BW1-4 [AMR-15]" \\
+      --roi "195.65,219.22,219.73,235.04" --fps 8 --output /tmp/out.mp4
+  .venv/bin/python scripts/build_mosaic_movie.py --machine "BW1-4 [AMR-15]" --dry-run
+  # Lighter preview (caps tall full-tube mosaics by height — easier on players):
+  .venv/bin/python scripts/build_mosaic_movie.py --machine "BW1-4 [AMR-15]" \\
+      --roi "0.0,0.0,310.0,740.0" --preview
+  # Metadata is drawn on each frame by default (semi-transparent bar at the top);
+  # use --no-metadata-overlay to disable.
+"""
+
+from __future__ import annotations
+
+import argparse
+import csv
+import os
+import sys
+import time
+from collections import Counter, defaultdict
+from dataclasses import dataclass
+from pathlib import Path
+
+import imageio
+import numpy as np
+from PIL import Image, ImageDraw, ImageFont
+
+
+class MovieEncodeError(Exception):
+    """Raised from encoding helpers; caught by encode_movie for batch-safe handling."""
+
+
+@dataclass
+class EncodedMovieResult:
+    success: bool
+    machine: str
+    roi: tuple[float, float, float, float]
+    csv_frame_count: int
+    written: int
+    missing: int
+    dropped_read: int
+    output_path: Path | None
+    skipped_reason: str | None
+    size_mb: float | None
+    elapsed_s: float | None
+
+
+def sanitize_machine_label(label: str) -> str:
+    return label.replace("[", "").replace("]", "").replace(" ", "_").strip("_")
+
+
+def parse_roi(s: str) -> tuple[float, float, float, float]:
+    parts = [p.strip() for p in s.split(",")]
+    if len(parts) != 4:
+        sys.exit("--roi must be four comma-separated numbers: start_x,start_y,end_x,end_y")
+    try:
+        return tuple(float(p) for p in parts)  # type: ignore[return-value]
+    except ValueError as e:
+        sys.exit(f"Invalid --roi numbers: {e}")
+
+
+def extent_close(
+    row: dict,
+    roi: tuple[float, float, float, float],
+    *,
+    tol: float = 1e-4,
+) -> bool:
+    keys = ("start_x", "start_y", "end_x", "end_y")
+    try:
+        vals = tuple(float(row[k]) for k in keys)
+    except (KeyError, ValueError):
+        return False
+    return all(abs(a - b) < tol for a, b in zip(vals, roi))
+
+
+def extent_key(row: dict) -> tuple[str, str, str, str]:
+    """Stable grouping key from CSV string fields."""
+    return (
+        row.get("start_x", "").strip(),
+        row.get("start_y", "").strip(),
+        row.get("end_x", "").strip(),
+        row.get("end_y", "").strip(),
+    )
+
+
+def key_to_roi_floats(key: tuple[str, str, str, str]) -> tuple[float, float, float, float]:
+    return tuple(float(x) for x in key)  # type: ignore[return-value]
+
+
+def parse_args() -> argparse.Namespace:
+    p = argparse.ArgumentParser(description=__doc__)
+    p.add_argument("--machine", required=True, help='RootView machine label, e.g. "BW1-4 [AMR-15]"')
+    p.add_argument(
+        "--roi",
+        metavar="SX,SY,EX,EY",
+        help="Restrict to this extent (mm). If omitted, pick the ROI with the most on-disk mosaics.",
+    )
+    p.add_argument("--archive", default="archives", type=Path, help="Archive root (default: archives)")
+    p.add_argument(
+        "--scans-csv",
+        default=None,
+        type=Path,
+        help="Path to scans.csv (default: <archive>/scans.csv)",
+    )
+    p.add_argument(
+        "--output",
+        "-o",
+        type=Path,
+        default=None,
+        help="Output .mp4 path (default: <archive>/movies/<machine>/roi_<...>.mp4)",
+    )
+    p.add_argument("--fps", type=float, default=10.0, help="Frames per second (default: 10)")
+    p.add_argument(
+        "--max-height",
+        type=int,
+        default=None,
+        metavar="PX",
+        help="Scale each frame so height is at most PX pixels (width keeps aspect); "
+        "suited to tall full-tube mosaics. Both dimensions are rounded to even pixels for H.264.",
+    )
+    p.add_argument(
+        "--preview",
+        action="store_true",
+        help="Shorthand for --max-height 1080 (overridden if --max-height is also set).",
+    )
+    p.add_argument(
+        "--dry-run",
+        action="store_true",
+        help="List frames that would be written (no MP4)",
+    )
+    p.add_argument(
+        "--no-metadata-overlay",
+        action="store_true",
+        help="Do not draw scan metadata on each frame (default: overlay on).",
+    )
+    args = p.parse_args()
+    if args.preview and args.max_height is None:
+        args.max_height = 1080
+    return args
+
+
+def _csv_required_fieldnames() -> tuple[str, ...]:
+    return (
+        "machine",
+        "scan_id",
+        "scan_time",
+        "mosaic_on_disk",
+        "mosaic_local_path",
+        "start_x",
+        "start_y",
+        "end_x",
+        "end_y",
+    )
+
+
+def validate_scans_csv_header(reader: csv.DictReader, scans_csv: Path) -> None:
+    if reader.fieldnames is None:
+        sys.exit(f"Empty CSV: {scans_csv}")
+    required = _csv_required_fieldnames()
+    missing = [c for c in required if c not in reader.fieldnames]
+    if missing:
+        sys.exit(f"{scans_csv} missing columns: {missing}")
+
+
+def load_latest_rows(
+    scans_csv: Path,
+    machine: str,
+    roi: tuple[float, float, float, float] | None,
+) -> list[dict]:
+    """Last row per scan_id for matching machine; mosaic_on_disk True; optional ROI."""
+    latest: dict[str, dict] = {}
+    with scans_csv.open(newline="", encoding="utf-8") as fh:
+        reader = csv.DictReader(fh)
+        validate_scans_csv_header(reader, scans_csv)
+
+        for row in reader:
+            if row.get("machine", "") != machine:
+                continue
+            if row.get("mosaic_on_disk", "").strip() != "True":
+                continue
+            if roi is not None and not extent_close(row, roi):
+                continue
+            sid = row.get("scan_id", "").strip()
+            if not sid:
+                continue
+            latest[sid] = row
+
+    return list(latest.values())
+
+
+def load_on_disk_rows_by_machine(scans_csv: Path) -> dict[str, list[dict]]:
+    """One pass: last row per (machine, scan_id) where mosaic_on_disk True; group by machine."""
+    latest: dict[tuple[str, str], dict] = {}
+    with scans_csv.open(newline="", encoding="utf-8") as fh:
+        reader = csv.DictReader(fh)
+        validate_scans_csv_header(reader, scans_csv)
+        for row in reader:
+            if row.get("mosaic_on_disk", "").strip() != "True":
+                continue
+            sid = row.get("scan_id", "").strip()
+            m = row.get("machine", "").strip()
+            if not sid or not m:
+                continue
+            latest[(m, sid)] = row
+
+    by_machine: dict[str, list[dict]] = defaultdict(list)
+    for (m, _sid), r in latest.items():
+        by_machine[m].append(r)
+    return {k: v for k, v in by_machine.items()}
+
+
+def pick_top_rois(rows: list[dict], n: int) -> list[tuple[tuple[float, float, float, float], int]]:
+    """Top n distinct extents by count of deduped rows. Empty if rows empty or n < 1."""
+    if not rows or n < 1:
+        return []
+    counts = Counter(extent_key(r) for r in rows)
+    return [(key_to_roi_floats(key), cnt) for key, cnt in counts.most_common(n)]
+
+
+def pick_top_roi(rows: list[dict]) -> tuple[float, float, float, float]:
+    if not rows:
+        sys.exit("No rows with mosaic_on_disk=True for this machine (and ROI filter, if any).")
+    return pick_top_rois(rows, 1)[0][0]
+
+
+def default_output_path(
+    archive: Path,
+    machine: str,
+    roi: tuple[float, float, float, float],
+    *,
+    max_height: int | None,
+    metadata_overlay: bool,
+    rank: int | None = None,
+) -> Path:
+    safe = sanitize_machine_label(machine)
+    sx, sy, ex, ey = roi
+    base = f"roi_{sx}_{sy}_{ex}_{ey}".replace(" ", "")
+    if rank is not None:
+        base = f"{base}_r{rank}"
+    if max_height is not None:
+        base = f"{base}_h{max_height}"
+    if metadata_overlay:
+        base = f"{base}_meta"
+    name = f"{base}.mp4"
+    return archive / "movies" / safe / name
+
+
+def resolve_mosaic_path(rel: str, archive: Path) -> Path:
+    """CSV paths are usually repo-relative, e.g. archives/BW1-4__AMR-15/.../mosaic.jpg."""
+    p = Path(rel)
+    if p.is_absolute():
+        return p.resolve()
+    ar = archive.resolve()
+    norm = rel.replace("\\", "/")
+    if norm.startswith("archives/") or norm.startswith("./archives/"):
+        return (ar.parent / rel).resolve()
+    return (ar / rel).resolve()
+
+
+def even_dimensions(w: int, h: int) -> tuple[int, int]:
+    """libx264 requires even width and height."""
+    w2 = w - (w % 2)
+    h2 = h - (h % 2)
+    if w2 < 2 or h2 < 2:
+        raise MovieEncodeError(f"Frame dimensions too small after evenizing: {w}x{h}")
+    return w2, h2
+
+
+def frame_size_mode(paths: list[Path]) -> tuple[int, int]:
+    sizes: list[tuple[int, int]] = []
+    for p in paths:
+        try:
+            with Image.open(p) as im:
+                sizes.append(im.size)
+        except OSError:
+            continue
+    if not sizes:
+        raise MovieEncodeError("No readable mosaic images to determine frame size.")
+    w, h = Counter(sizes).most_common(1)[0][0]
+    return even_dimensions(w, h)
+
+
+def encode_size(native_w: int, native_h: int, max_height: int | None) -> tuple[int, int]:
+    """Native size is already even; optional downscale for preview encodes (cap height)."""
+    if max_height is None:
+        return native_w, native_h
+    if max_height < 32:
+        raise MovieEncodeError("--max-height must be at least 32")
+    cap = max_height - (max_height % 2)
+    if cap < 2:
+        raise MovieEncodeError("--max-height must allow an even height of at least 2")
+    if native_h <= cap:
+        return native_w, native_h
+    new_h = cap
+    new_w = int(round(native_w * (new_h / native_h)))
+    new_w -= new_w % 2
+    if new_w < 2:
+        raise MovieEncodeError("Computed preview width too small; try a larger --max-height")
+    return new_w, new_h
+
+
+def _truetype_font_candidates() -> list[Path]:
+    windir = os.environ.get("WINDIR", r"C:\Windows")
+    return [
+        Path(windir) / "Fonts" / "arial.ttf",
+        Path(windir) / "Fonts" / "consola.ttf",
+        Path("/usr/share/fonts/truetype/dejavu/DejaVuSans.ttf"),
+        Path("/usr/share/fonts/truetype/liberation/LiberationSans-Regular.ttf"),
+    ]
+
+
+def get_overlay_font(size: int) -> ImageFont.FreeTypeFont | ImageFont.ImageFont:
+    for path in _truetype_font_candidates():
+        if path.is_file():
+            try:
+                return ImageFont.truetype(str(path), size=size)
+            except OSError:
+                continue
+    return ImageFont.load_default()
+
+
+def _truncate(s: str, max_len: int) -> str:
+    s = s.strip()
+    if len(s) <= max_len:
+        return s
+    if max_len <= 3:
+        return s[:max_len]
+    return s[: max_len - 3] + "..."
+
+
+def metadata_overlay_lines(row: dict, *, max_name_chars: int) -> list[str]:
+    sid = row.get("scan_id", "").strip()
+    st = row.get("scan_time", "").strip()
+    name = _truncate(row.get("name", "").strip(), max_name_chars)
+    nx = row.get("nx", "").strip()
+    ny = row.get("ny", "").strip()
+    dx = row.get("dx", "").strip()
+    dy = row.get("dy", "").strip()
+    lines = row.get("scan_lines", "").strip()
+    mode = row.get("scan_mode", "").strip()
+    user = row.get("user", "").strip()
+    status = row.get("status", "").strip()
+    sx = row.get("start_x", "").strip()
+    sy = row.get("start_y", "").strip()
+    ex = row.get("end_x", "").strip()
+    ey = row.get("end_y", "").strip()
+    machine = row.get("machine", "").strip()
+
+    grid = f"{nx}x{ny}" if nx and ny else ""
+    step = f"{dx}x{dy} mm" if dx and dy else ""
+    geom = "  ".join(p for p in (grid, step) if p)
+    orient = f"{lines} / {mode}" if lines or mode else ""
+
+    out: list[str] = []
+    if machine:
+        out.append(machine)
+    if sid or st:
+        parts = []
+        if sid:
+            parts.append(f"id {sid}")
+        if st:
+            parts.append(st)
+        out.append("  ".join(parts))
+    if name:
+        out.append(name)
+    if geom or orient:
+        out.append("  ".join(p for p in (geom, orient) if p))
+    if sx and sy and ex and ey:
+        out.append(f"ROI mm {sx},{sy} .. {ex},{ey}")
+    if user or status:
+        tail: list[str] = []
+        if user:
+            tail.append(f"user {user}")
+        if status:
+            tail.append(status)
+        out.append("  ".join(tail))
+    return out if out else ["(no metadata)"]
+
+
+def draw_metadata_overlay(
+    rgb: Image.Image,
+    row: dict,
+    *,
+    margin: int,
+) -> None:
+    """Draw a semi-transparent label block along the top; mutates rgb in place."""
+    # Panel alpha: ~50% so roots stay visible through the bar.
+    panel_fill = (0, 0, 0, 128)
+    panel_outline = (220, 220, 230, 100)
+
+    w, h = rgb.size
+    margin = max(4, min(margin, w // 8))
+    font_size = max(10, min(22, h // 50))
+    pad = max(4, font_size // 3)
+    line_gap = max(2, font_size // 6)
+
+    def measure_block(fs: int, name_max: int) -> tuple[ImageFont.FreeTypeFont | ImageFont.ImageFont, list[str], int, int]:
+        font = get_overlay_font(fs)
+        lines = metadata_overlay_lines(row, max_name_chars=name_max)
+        tmp = Image.new("RGB", (1, 1))
+        draw = ImageDraw.Draw(tmp)
+        max_tw = 0
+        total_h = 0
+        for line in lines:
+            bbox = draw.textbbox((0, 0), line, font=font)
+            tw = bbox[2] - bbox[0]
+            th = bbox[3] - bbox[1]
+            max_tw = max(max_tw, tw)
+            total_h += th
+        if len(lines) > 1:
+            total_h += line_gap * (len(lines) - 1)
+        block_w = max_tw + 2 * pad
+        block_h = total_h + 2 * pad
+        return font, lines, block_w, block_h
+
+    name_max = max(24, w // max(6, font_size // 2))
+    font, lines, block_w, block_h = measure_block(font_size, name_max)
+    max_block_w = w - 2 * margin
+    max_block_h = min(h // 2, h - 2 * margin)
+    while (block_w > max_block_w or block_h > max_block_h) and font_size > 9:
+        font_size -= 1
+        name_max = max(16, w // max(7, font_size // 2))
+        font, lines, block_w, block_h = measure_block(font_size, name_max)
+    while block_w > max_block_w and name_max > 12:
+        name_max -= 4
+        font, lines, block_w, block_h = measure_block(font_size, name_max)
+
+    bw = min(block_w, max_block_w)
+    bh = min(block_h, max_block_h)
+    x0 = margin
+    x1 = x0 + bw
+    y0 = margin
+    y1 = y0 + bh
+
+    overlay = Image.new("RGBA", (w, h), (0, 0, 0, 0))
+    draw_ov = ImageDraw.Draw(overlay)
+    draw_ov.rounded_rectangle(
+        [x0, y0, x1, y1],
+        radius=max(4, pad // 2),
+        fill=panel_fill,
+        outline=panel_outline,
+        width=1,
+    )
+
+    base = rgb.convert("RGBA")
+    composited = Image.alpha_composite(base, overlay)
+    draw = ImageDraw.Draw(composited)
+
+    cx, cy = x0 + pad, y0 + pad
+    text_bottom_limit = y1 - pad
+    for line in lines:
+        bbox = draw.textbbox((0, 0), line, font=font)
+        th = bbox[3] - bbox[1]
+        if cy + th > text_bottom_limit:
+            break
+        draw.text(
+            (cx, cy),
+            line,
+            fill=(245, 245, 245, 255),
+            font=font,
+            stroke_width=1,
+            stroke_fill=(0, 0, 0, 255),
+        )
+        cy += th + line_gap
+
+    rgb.paste(composited.convert("RGB"))
+
+
+def encode_movie(
+    *,
+    machine: str,
+    roi: tuple[float, float, float, float],
+    rows: list[dict],
+    archive: Path,
+    max_height: int | None,
+    metadata_overlay: bool,
+    fps: float,
+    output: Path | None = None,
+    dry_run: bool = False,
+    rank: int | None = None,
+    quiet: bool = False,
+) -> EncodedMovieResult:
+    """Build MP4 from pre-filtered rows for one ROI. Does not sys.exit on encode failures."""
+    rows_sorted = sorted(
+        rows,
+        key=lambda r: (r.get("scan_time") or "", r.get("scan_id") or ""),
+    )
+    csv_frame_count = len(rows_sorted)
+
+    row_path_candidates: list[tuple[dict, Path]] = []
+    for r in rows_sorted:
+        rel = (r.get("mosaic_local_path") or "").strip()
+        if not rel:
+            continue
+        row_path_candidates.append((r, resolve_mosaic_path(rel, archive)))
+
+    out: Path = output or default_output_path(
+        archive,
+        machine,
+        roi,
+        max_height=max_height,
+        metadata_overlay=metadata_overlay,
+        rank=rank,
+    )
+
+    if not quiet:
+        print(f"Machine: {machine}")
+        print(f"ROI (mm): {roi[0]}, {roi[1]}, {roi[2]}, {roi[3]}")
+        print(f"Frames (from CSV, deduped): {csv_frame_count}")
+        if max_height is not None:
+            print(f"Preview max height: {max_height}px")
+        print(f"Metadata overlay: {'on' if metadata_overlay else 'off'}")
+
+    on_disk = sum(1 for _r, p in row_path_candidates if p.is_file())
+    missing_paths = len(row_path_candidates) - on_disk
+
+    if dry_run:
+        if not quiet:
+            print(f"On-disk files among ordered list: {on_disk} / {len(row_path_candidates)}")
+            for i, (_r, p) in enumerate(row_path_candidates[:5]):
+                print(f"  [{i}] {p} exists={p.is_file()}")
+            if len(row_path_candidates) > 10:
+                print("  ...")
+            start = max(0, len(row_path_candidates) - 3)
+            for i, (_r, p) in enumerate(row_path_candidates[start:], start=start):
+                print(f"  [{i}] {p} exists={p.is_file()}")
+        return EncodedMovieResult(
+            success=True,
+            machine=machine,
+            roi=roi,
+            csv_frame_count=csv_frame_count,
+            written=0,
+            missing=missing_paths,
+            dropped_read=0,
+            output_path=out,
+            skipped_reason=None,
+            size_mb=None,
+            elapsed_s=None,
+        )
+
+    row_path_pairs = [(r, p) for r, p in row_path_candidates if p.is_file()]
+    if not row_path_pairs:
+        if not quiet:
+            print("No mosaic files on disk for the selected rows.")
+        return EncodedMovieResult(
+            success=False,
+            machine=machine,
+            roi=roi,
+            csv_frame_count=csv_frame_count,
+            written=0,
+            missing=missing_paths,
+            dropped_read=0,
+            output_path=out,
+            skipped_reason="No mosaic files on disk for the selected rows.",
+            size_mb=None,
+            elapsed_s=None,
+        )
+
+    t0 = time.perf_counter()
+    try:
+        ordered_paths = [p for _r, p in row_path_pairs]
+        target_w, target_h = frame_size_mode(ordered_paths)
+        enc_w, enc_h = encode_size(target_w, target_h, max_height)
+    except MovieEncodeError as e:
+        if not quiet:
+            print(str(e))
+        return EncodedMovieResult(
+            success=False,
+            machine=machine,
+            roi=roi,
+            csv_frame_count=csv_frame_count,
+            written=0,
+            missing=missing_paths,
+            dropped_read=0,
+            output_path=out,
+            skipped_reason=str(e),
+            size_mb=None,
+            elapsed_s=None,
+        )
+
+    if not quiet:
+        print(f"Target frame size (mode): {target_w} x {target_h}")
+        if (enc_w, enc_h) != (target_w, target_h):
+            print(f"Encode size (after max-height): {enc_w} x {enc_h}")
+
+    out.parent.mkdir(parents=True, exist_ok=True)
+    written = 0
+    dropped = 0
+    resized = 0
+    scaled_preview = 0
+    try:
+        writer = imageio.get_writer(
+            str(out),
+            fps=float(fps),
+            codec="libx264",
+            quality=8,
+            macro_block_size=1,
+        )
+        try:
+            for row, p in row_path_pairs:
+                try:
+                    with Image.open(p) as im:
+                        rgb = im.convert("RGB")
+                        if rgb.size != (target_w, target_h):
+                            rgb = rgb.resize((target_w, target_h), Image.Resampling.LANCZOS)
+                            resized += 1
+                        if rgb.size != (enc_w, enc_h):
+                            rgb = rgb.resize((enc_w, enc_h), Image.Resampling.LANCZOS)
+                            scaled_preview += 1
+                        if metadata_overlay:
+                            draw_metadata_overlay(rgb, row, margin=max(6, enc_w // 80))
+                        frame = np.asarray(rgb)
+                    writer.append_data(frame)
+                    written += 1
+                except OSError:
+                    dropped += 1
+        finally:
+            writer.close()
+    except Exception as e:
+        if not quiet:
+            print(f"Encode error: {e}")
+        return EncodedMovieResult(
+            success=False,
+            machine=machine,
+            roi=roi,
+            csv_frame_count=csv_frame_count,
+            written=written,
+            missing=missing_paths,
+            dropped_read=dropped,
+            output_path=out,
+            skipped_reason=f"encode_error: {e}",
+            size_mb=None,
+            elapsed_s=None,
+        )
+
+    elapsed = time.perf_counter() - t0
+    size_mb = out.stat().st_size / (1024 * 1024) if out.is_file() else 0.0
+    if not quiet:
+        print(
+            f"Written: {written} frames  (normalized to mode: {resized}, "
+            f"preview scale: {scaled_preview})"
+        )
+        print(f"Dropped (read error): {dropped}")
+        print(f"Missing paths (not on disk): {missing_paths}")
+        print(f"Output: {out.resolve()}  ({size_mb:.2f} MB)")
+
+    return EncodedMovieResult(
+        success=True,
+        machine=machine,
+        roi=roi,
+        csv_frame_count=csv_frame_count,
+        written=written,
+        missing=missing_paths,
+        dropped_read=dropped,
+        output_path=out,
+        skipped_reason=None,
+        size_mb=size_mb,
+        elapsed_s=elapsed,
+    )
+
+
+def main() -> None:
+    args = parse_args()
+    archive: Path = args.archive
+    scans_csv: Path = args.scans_csv or (archive / "scans.csv")
+    if not scans_csv.is_file():
+        sys.exit(f"scans.csv not found: {scans_csv}")
+
+    roi_sel: tuple[float, float, float, float] | None
+    if args.roi:
+        roi_sel = parse_roi(args.roi)
+    else:
+        roi_sel = None
+
+    rows = load_latest_rows(scans_csv, args.machine, roi_sel)
+    if roi_sel is None:
+        roi_sel = pick_top_roi(rows)
+        rows = [r for r in rows if extent_close(r, roi_sel)]
+
+    assert roi_sel is not None
+    max_height: int | None = args.max_height
+    metadata_overlay = not args.no_metadata_overlay
+
+    res = encode_movie(
+        machine=args.machine,
+        roi=roi_sel,
+        rows=rows,
+        archive=archive,
+        max_height=max_height,
+        metadata_overlay=metadata_overlay,
+        fps=float(args.fps),
+        output=args.output,
+        dry_run=bool(args.dry_run),
+        rank=None,
+        quiet=False,
+    )
+    if not res.success and not args.dry_run:
+        sys.exit(1 if res.skipped_reason else 1)
+
+
+if __name__ == "__main__":
+    main()
diff --git a/scripts/build_mosaic_movies_batch.py b/scripts/build_mosaic_movies_batch.py
new file mode 100644
index 0000000..bf5569f
--- /dev/null
+++ b/scripts/build_mosaic_movies_batch.py
@@ -0,0 +1,269 @@
+#!/usr/bin/env python3
+"""
+Build preview MP4s for the top N ROIs per machine (default N=2, max-height 1080).
+
+Reads archives/scans.csv once, groups on-disk mosaic rows by machine, then for
+each machine picks the most frequent ROI extents and calls encode_movie().
+
+Usage:
+  python scripts/build_mosaic_movies_batch.py
+  python scripts/build_mosaic_movies_batch.py --dry-run
+  python scripts/build_mosaic_movies_batch.py --skip-existing
+  python scripts/build_mosaic_movies_batch.py --machine "BW2-10 [AMR-22]"
+  python scripts/build_mosaic_movies_batch.py --full-res   # no max-height cap
+"""
+
+from __future__ import annotations
+
+import argparse
+import sys
+import time
+from dataclasses import dataclass
+from pathlib import Path
+
+_SCRIPTS_DIR = Path(__file__).resolve().parent
+
+# Import sibling module (run as python scripts/build_mosaic_movies_batch.py from repo root)
+sys.path.insert(0, str(_SCRIPTS_DIR))
+import build_mosaic_movie as bmm  # noqa: E402
+
+
+def read_machine_labels(path: Path) -> list[str]:
+    out: list[str] = []
+    with path.open(encoding="utf-8") as fh:
+        for line in fh:
+            s = line.strip()
+            if not s or s.startswith("#"):
+                continue
+            out.append(s)
+    return out
+
+
+@dataclass
+class Job:
+    machine: str
+    rank: int
+    roi: tuple[float, float, float, float]
+    extent_count: int
+    rows: list[dict]
+    output_path: Path
+
+
+def collect_jobs(
+    *,
+    machines: list[str],
+    by_machine: dict[str, list[dict]],
+    archive: Path,
+    top_rois: int,
+    max_height: int | None,
+    metadata_overlay: bool,
+) -> list[Job]:
+    jobs: list[Job] = []
+    for machine in machines:
+        rows = by_machine.get(machine, [])
+        if not rows:
+            continue
+        picks = bmm.pick_top_rois(rows, top_rois)
+        for rank, (roi, extent_count) in enumerate(picks, start=1):
+            rows_roi = [r for r in rows if bmm.extent_close(r, roi)]
+            out = bmm.default_output_path(
+                archive,
+                machine,
+                roi,
+                max_height=max_height,
+                metadata_overlay=metadata_overlay,
+                rank=rank,
+            )
+            jobs.append(
+                Job(
+                    machine=machine,
+                    rank=rank,
+                    roi=roi,
+                    extent_count=extent_count,
+                    rows=rows_roi,
+                    output_path=out,
+                )
+            )
+    return jobs
+
+
+def parse_args() -> argparse.Namespace:
+    p = argparse.ArgumentParser(description=__doc__)
+    p.add_argument(
+        "--machines-file",
+        type=Path,
+        default=_SCRIPTS_DIR / "machines.example.txt",
+        help="One machine label per line (default: scripts/machines.example.txt next to this script)",
+    )
+    p.add_argument(
+        "--machine",
+        metavar="LABEL",
+        help='If set, only this machine (overrides list to a single job set), e.g. "BW2-10 [AMR-22]"',
+    )
+    p.add_argument("--archive", type=Path, default=Path("archives"))
+    p.add_argument("--scans-csv", type=Path, default=None)
+    p.add_argument("--top-rois", type=int, default=2, help="How many top extents per machine (default: 2)")
+    p.add_argument("--max-height", type=int, default=1080, help="Preview cap in px (default: 1080)")
+    p.add_argument(
+        "--full-res",
+        action="store_true",
+        help="Disable max-height cap (full mosaic resolution; can be huge)",
+    )
+    p.add_argument("--fps", type=float, default=10.0)
+    p.add_argument("--dry-run", action="store_true")
+    p.add_argument(
+        "--skip-existing",
+        action="store_true",
+        help="Skip encode if output MP4 exists and is non-empty",
+    )
+    p.add_argument("--no-metadata-overlay", action="store_true")
+    args = p.parse_args()
+    if args.full_res:
+        args.max_height = None
+    return args
+
+
+def main() -> None:
+    args = parse_args()
+    archive: Path = args.archive
+    scans_csv: Path = args.scans_csv or (archive / "scans.csv")
+    if not scans_csv.is_file():
+        sys.exit(f"scans.csv not found: {scans_csv}")
+
+    if args.machine:
+        machines = [args.machine.strip()]
+    else:
+        if not args.machines_file.is_file():
+            sys.exit(f"Machines file not found: {args.machines_file}")
+        machines = read_machine_labels(args.machines_file)
+        if not machines:
+            sys.exit(f"No machine labels in {args.machines_file}")
+
+    t_load0 = time.perf_counter()
+    by_machine = bmm.load_on_disk_rows_by_machine(scans_csv)
+    load_s = time.perf_counter() - t_load0
+
+    max_height: int | None = args.max_height
+    metadata_overlay = not args.no_metadata_overlay
+
+    jobs = collect_jobs(
+        machines=machines,
+        by_machine=by_machine,
+        archive=archive,
+        top_rois=args.top_rois,
+        max_height=max_height,
+        metadata_overlay=metadata_overlay,
+    )
+    total = len(jobs)
+    if total == 0:
+        sys.exit("No jobs (no on-disk mosaics for selected machines).")
+
+    print(f"Loaded scans.csv grouped by machine in {load_s:.2f}s  ({total} job(s))")
+    if max_height is not None:
+        print(f"Max height: {max_height}px")
+    else:
+        print("Max height: (full resolution)")
+    print(f"Metadata overlay: {'on' if metadata_overlay else 'off'}")
+    print()
+
+    summary_rows: list[tuple[str, ...]] = []
+
+    for idx, job in enumerate(jobs, start=1):
+        sx, sy, ex, ey = job.roi
+        roi_s = f"{sx},{sy}..{ex},{ey}"
+        print(
+            f"[{idx}/{total}] {job.machine}  rank={job.rank}  ROI {roi_s}  "
+            f"({job.extent_count} CSV rows this extent, {len(job.rows)} deduped rows)"
+        )
+
+        if args.skip_existing and job.output_path.is_file() and job.output_path.stat().st_size > 0:
+            sz = job.output_path.stat().st_size / (1024 * 1024)
+            print(f"  SKIP (exists): {job.output_path}")
+            summary_rows.append(
+                (
+                    job.machine,
+                    str(job.rank),
+                    roi_s,
+                    str(len(job.rows)),
+                    "-",
+                    "-",
+                    str(job.output_path),
+                    "SKIP (exists)",
+                    f"{sz:.2f}",
+                    "-",
+                )
+            )
+            continue
+
+        enc_t0 = time.perf_counter()
+        res = bmm.encode_movie(
+            machine=job.machine,
+            roi=job.roi,
+            rows=job.rows,
+            archive=archive,
+            max_height=max_height,
+            metadata_overlay=metadata_overlay,
+            fps=float(args.fps),
+            output=None,
+            dry_run=args.dry_run,
+            rank=job.rank,
+            quiet=True,
+        )
+        enc_elapsed = time.perf_counter() - enc_t0
+
+        if res.success:
+            if args.dry_run:
+                print(f"  dry-run OK  -> {res.output_path}  (missing files: {res.missing})")
+            else:
+                print(
+                    f"  OK  {res.written} frames  "
+                    f"{(res.size_mb or 0):.2f} MB  {enc_elapsed:.1f}s"
+                )
+        else:
+            print(f"  FAIL  {res.skipped_reason or 'unknown'}")
+
+        status = "OK" if res.success else "FAIL"
+        if not res.success and res.skipped_reason:
+            status = f"FAIL: {res.skipped_reason[:40]}"
+        mb = f"{res.size_mb:.2f}" if res.size_mb is not None else "-"
+        es = f"{enc_elapsed:.1f}" if not args.dry_run else "-"
+        w = str(res.written) if not args.dry_run else "0"
+        if args.dry_run:
+            status = "dry-run"
+            mb = "-"
+
+        summary_rows.append(
+            (
+                job.machine,
+                str(job.rank),
+                roi_s,
+                str(len(job.rows)),
+                w,
+                str(res.missing),
+                str(res.output_path or job.output_path),
+                status,
+                mb,
+                es,
+            )
+        )
+        print()
+
+    print("=" * 120)
+    hdr = (
+        f"{'machine':<26} {'rk':>2} {'ROI (mm)':<36} {'csv':>4} {'out':>5} "
+        f"{'miss':>4} {'MB':>7} {'s':>6}  {'status':<22}"
+    )
+    print(hdr)
+    print("-" * 120)
+    for row in summary_rows:
+        m, rk, roi_s, csv_n, wrt, miss, path, status, mb, es = row
+        print(
+            f"{m:<26} {rk:>2} {roi_s:<36} {csv_n:>4} {wrt:>5} {miss:>4} "
+            f"{mb:>7} {es:>6}  {status:<22}"
+        )
+        print(f"  -> {path}")
+    print("=" * 120)
+
+
+if __name__ == "__main__":
+    main()