test(harness): keep the visual double's signature tied to production

Moves set_scrolling_state's frame_hold into the test double here, where
DisplayManager gains it, rather than in #534 where it arrived a PR early.
CodeRabbit flagged the #534 version correctly: a double that accepts an
argument production does not lets the call pass every harness run and
raise TypeError on the panel, which is the one failure a safety harness
exists to prevent.

The drift has now gone both ways across two branches -- double behind
production on this branch, double ahead of it on #534 -- so it is pinned
instead of remembered. test_display_double_parity.py compares the two
signatures and fails with the direction of the drift named. It reads the
files with ast rather than importing them, because display_manager
imports rgbmatrix at module scope and this check should hold on a laptop
and in CI as well as on a Pi.

Plugins begin passing frame_hold in ledmatrix-plugins#462, which is why
production and the double both need it before that lands.

Full suite: 3889 passed.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_014RRtqXDCnvnY6EQwhT5CV9

docs/SCROLL_PERFORMANCE.md

@@ -0,0 +1,266 @@
+# Scroll Performance
+
+How scrolling is paced on this hardware, what was wrong with it, and how to
+configure a plugin so its marquee is smooth.
+
+Measured on a Raspberry Pi 4 driving a 2×128×64 chain (256×64 logical) at
+`limit_refresh_rate_hz: 100`. Numbers below come from that panel.
+
+| | before | after |
+|---|---|---|
+| scroll frame rate | 44–46 fps | **100 fps, locked** |
+| frames ≥ 45 ms | 14–17% | none observed |
+| dominant frame time | 20 ms | **10 ms** |
+| disk cache write (~1 MB) | 14.8 ms | **5.4 ms** |
+
+---
+
+## The one rule that matters
+
+**Motion is smooth when the strip advances a whole number of pixels per panel
+refresh.**
+
+Advancing one pixel per refresh on a 100 Hz panel gives 100 px/s. Slower crisp
+speeds come from holding each frame for several refreshes -- 50 px/s is one
+pixel every second refresh -- which is covered under *Choosing a speed* below.
+A speed that lands on no such combination has to do one of two bad things:
+
+- **blend** two adjacent columns to render a half-step — on pixel-font text
+  this alternates crisp and smeared frames and reads as shimmer, or as the
+  text jumping a pixel ahead of itself;
+- **repeat** a frame — the strip stands still, then jumps, which reads as
+  judder.
+
+Neither is tunable away. Pick a speed that divides evenly.
+
+`src.common.scroll_config` solves this for you: `configure()` snaps a requested
+speed to the nearest one the panel can actually show in whole pixels, and
+`scripts/scroll_speeds.py` prints the full ladder for your hardware.
+
+## Choosing a speed
+
+The crisp speeds are not a fixed list -- they depend on how fast *your* panel
+refreshes, which depends on its size, `pwm_bits`, `gpio_slowdown` and the Pi
+model. A Pi Zero driving a long chain has a completely different set of good
+speeds from a Pi 4 driving a short one.
+
+```bash
+# what can this panel do? (reads your configured refresh rate)
+python3 scripts/scroll_speeds.py
+
+# what does it ACTUALLY manage, rather than what is configured?
+sudo systemctl stop ledmatrix
+sudo python3 scripts/scroll_speeds.py --measure
+sudo systemctl start ledmatrix
+
+# highlight the closest option to the speed you want
+python3 scripts/scroll_speeds.py --want 45
+
+# try one on the panel
+sudo systemctl stop ledmatrix
+sudo python3 scripts/scroll_speeds.py --demo 50
+sudo systemctl start ledmatrix
+```
+
+Sample ladder for a 100 Hz panel:
+
+```
+  20.0 px/s  (1px every 5 refreshes =  20.0 fps, slightly stepped)
+  25.0 px/s  (1px every 4 refreshes =  25.0 fps, slightly stepped)
+  33.3 px/s  (1px every 3 refreshes =  33.3 fps, smooth)
+  50.0 px/s  (1px every 2 refreshes =  50.0 fps, smooth)
+  66.7 px/s  (2px every 3 refreshes =  33.3 fps, smooth)
+ 100.0 px/s  (1px every 1 refresh  = 100.0 fps, smooth)
+```
+
+### How a slow speed stays crisp
+
+`SwapOnVSync(canvas, framerate_fraction)` holds each frame for N panel
+refreshes. **The panel keeps refreshing at its full rate either way**, so
+holding a frame costs nothing in flicker -- it only changes how often a *new*
+image is presented. That is what allows 50 px/s to be one whole pixel every
+second refresh, instead of half a pixel every refresh (which has no good
+rendering, only a choice between blur and judder).
+
+`scroll_config.configure()` snaps the requested speed to the nearest entry on
+the ladder and reports the hold that speed needs. It does **not** apply the
+hold: the hold belongs to a scroll, not to a plugin's lifetime, and plugins
+share one display manager -- one set at construction is reset the moment any
+other plugin finishes scrolling. Apply it yourself when the scroll starts:
+
+```python
+settings = scroll_config.configure(
+    self.scroll_helper,
+    plugin_config=self.config,
+    global_config=self.global_config,
+    display_manager=self.display_manager,   # supplies the panel refresh rate
+)
+
+# ...then, each time this plugin begins scrolling:
+self.display_manager.set_scrolling_state(True, frame_hold=settings.frame_hold)
+```
+
+Passing `display_manager` only lets `configure` read the true refresh rate from
+`display.hardware`, which a plugin config cannot see. Skipping the
+`set_scrolling_state` call is the mistake that matters: the speed still
+resolves, but the panel keeps presenting a new frame every refresh, so a slow
+snapped speed falls back to fractional pixels. Pass `snap_to_crisp=False` to
+keep an exact requested speed and accept the artefacts.
+
+Speeds slower than about 20 px/s are stepped no matter what, because a 1-pixel
+advance at 20 fps is simply a coarse increment. That is the pixel pitch, not a
+software limit; the only way to move in smaller increments is sub-pixel
+blending, which this display does not tolerate (see above).
+
+## Configuring a plugin
+
+Use the shared resolver rather than reading config keys yourself:
+
+```python
+from src.common import scroll_config
+
+settings = scroll_config.configure(
+    self.scroll_helper,
+    plugin_config=self.config,
+    global_config=self.global_config,
+    refresh_hz=scroll_config.refresh_hz_from_config(self.global_config),
+    plugin_logger=self.logger,
+)
+```
+
+It resolves every config shape in one place, applies the speed, and returns
+what it did. Precedence, highest first:
+
+1. `display_options.scroll_speed` + `scroll_delay` — **the recommended form**
+2. `display.scroll_speed` + `scroll_delay` — deprecated shape
+3. `scroll_speed` + `scroll_delay` at the root — legacy flat
+4. `scroll_pixels_per_second` — deprecated
+5. the global `display` block
+6. the built-in default (100 px/s)
+
+`scroll_speed` is pixels per frame and `scroll_delay` is the frame period in
+seconds, so the pair means `scroll_speed / scroll_delay` px/s. The recommended
+config for a 100 Hz panel:
+
+```json
+"display_options": { "scroll_speed": 1.0, "scroll_delay": 0.01 }
+```
+
+### Why the deprecated key ranks below the explicit pair
+
+Because some plugins give `scroll_pixels_per_second` a **schema default**, and
+schema defaults are merged into plugin config. Ranking it above the pair means
+it is always present and always wins, so the documented settings become
+unreachable. That is a real, shipped bug — see
+[ledmatrix-plugins#408](https://github.com/ChuckBuilds/ledmatrix-plugins/issues/408).
+
+If you are writing a plugin: do not give a deprecated key a schema default.
+
+## What was actually wrong
+
+Four independent faults, each found by measurement.
+
+### 1. The frame loop slept on top of a wait it had already done
+
+`display_controller.py` ran the high-FPS loop as `render → SwapOnVSync (blocks
+to the panel's refresh) → time.sleep(0.008) → plugin ticks`. The sleep was
+unconditional and added to a wait that had already happened. Render work
+measured ~4 ms, so each iteration cost ~12 ms against a 10 ms refresh grid —
+every swap missed a refresh and landed on the next one. The loop settled at
+exactly 50 fps while asking for 125, with no headroom, so ~14% of frames
+slipped a further refresh.
+
+Now the loop sleeps only the remainder of the frame budget, with a 1 ms floor
+so plugin threads still get the GIL.
+
+### 2. `SwapOnVSync` held the GIL while blocking
+
+The rgbmatrix binding declares it without `nogil` (unlike `SetPixel`, `Clear`
+and `Fill` immediately above it in `cppinc.pxd`), so the render thread held the
+GIL for the entire vsync wait — most of every frame. Background threads were
+starved into long uninterruptible bursts; a 1.5 MB API response costs ~17 ms to
+parse and ~18 ms to re-encode for the cache, and `json.raw_decode` cannot be
+preempted mid-document. Those bursts are what the render loop then waited on.
+
+Fixed by rebuilding the binding: `scripts/build_rgbmatrix_nogil.sh`.
+
+### 3. Sub-pixel blending was wrong for this display
+
+Enabling it made things worse, not better — see the rule at the top. It is off
+by default and only Vegas mode opts in via `set_sub_pixel_scrolling(True)`.
+
+### 4. Frame-based stepping raced the vsync clock
+
+Frame-based mode gated motion on a wall clock at `1/scroll_delay` steps per
+second. Plugins set `scroll_delay` to the frame period, which puts that
+comparison exactly on its own threshold: a frame arriving a hair early moved
+zero pixels and rendered an identical frame, which dirty-tracking skipped, so
+it returned in ~2 ms and the beat repeated. No `scroll_delay` value tunes this
+out — a shorter delay just trades stalled frames for periodic double-steps.
+
+`ScrollHelper` now accumulates elapsed time in both modes at the same
+configured speed, so position stays proportional to real time.
+
+## Diagnosing a juddery scroller
+
+**`Avg FPS` will lie to you.** It is a 100-frame moving average, and a 2 ms
+duplicate frame plus a 21 ms double-wait average to exactly 10 ms. A ticker
+that is stalling on half its frames still reports a healthy `100.0`.
+
+Look at the **distribution** instead:
+
+```bash
+journalctl -u ledmatrix --since "-10min" --no-pager \
+  | grep -oE "Frame time: [0-9.]+ms" | awk '{print $3}' | sed 's/ms//' \
+  | awk '{printf "%.0f\n", $1}' | sort -n | uniq -c
+```
+
+Reading it, on a 100 Hz panel:
+
+| you see | it means |
+|---|---|
+| everything at 10 ms | healthy |
+| a mode at ~2 ms | **duplicate frames** — the swap was skipped because the image did not change. The scroller is advancing less than one pixel per frame. |
+| a mode at 20/30/50 ms | frames missing refreshes — per-frame work is overrunning, or a background thread is holding the GIL |
+| `Avg FPS` above 100 | duplicates present, unless the scroll cycle has completed and is idling |
+
+Then confirm what the plugin actually loaded — config edits do not always reach
+the running code:
+
+```bash
+journalctl -u ledmatrix --since "-5min" --no-pager | grep -iE "px/s|px/frame"
+```
+
+If a plugin logs its scroll config **twice** with different modes, the second
+line is what is running.
+
+## Rebuilding the binding
+
+```bash
+bash scripts/build_rgbmatrix_nogil.sh              # build into a scratch dir
+sudo bash scripts/build_rgbmatrix_nogil.sh --install
+sudo bash scripts/build_rgbmatrix_nogil.sh --rollback
+```
+
+The build never touches the installed module. `--install` backs up the original
+to `~/rgbmatrix-core.so.ORIGINAL` first, and rolls back automatically if the
+service does not come back healthy. Requires `build-essential`; Cython is
+installed into a cached venv under `~/.cache/ledmatrix-cython`.
+
+Re-run it after upgrading `rpi-rgb-led-matrix`, since a library upgrade
+replaces the patched binding.
+
+## Faster JSON
+
+`src/cache/disk_cache.py` uses `orjson` when it is importable and falls back to
+the stdlib otherwise, so it is optional:
+
+```bash
+sudo pip3 install --break-system-packages orjson
+```
+
+Encoding is where it pays — about 7× on this hardware. Decoding gains far less
+(~1.3× on large payloads) because the cost there is building Python objects,
+not scanning text. That is also why moving parsing to a subprocess does not
+help: `pickle.loads` of the same payload costs 8.1 ms against `json.loads` at
+10.9 ms, so the work just moves rather than disappearing.

requirements.txt

@@ -55,6 +55,18 @@ packaging>=23.0,<27.0
 #                     range as a hard dependency — keep the two in sync.
 #   pip install 'psutil>=6.0.0,<7.0.0'
 #
+# orjson            — faster JSON for the disk cache
+#                     (src/cache/disk_cache.py). Encoding a ~1MB cache
+#                     record drops from ~12ms to ~1.6ms on a Pi 4, which
+#                     matters because that work holds the GIL and stalls
+#                     the render thread mid-scroll. Falls back to the
+#                     stdlib json when missing — see docs/SCROLL_PERFORMANCE.md.
+#                     The 3.11.6 floor is CVE-2025-67221: orjson.dumps did not
+#                     limit recursion on deeply nested documents, and the disk
+#                     cache encodes payloads parsed straight from third-party
+#                     APIs. 3.11.6 covers the Python range above.
+#   pip install 'orjson>=3.11.6,<4.0'
+#
 # Flask-Limiter     — request rate limiting in web_interface/app.py
 #                     (accidental-abuse protection, not security). The
 #                     web interface starts without rate limiting when

scripts/build_rgbmatrix_nogil.sh

@@ -0,0 +1,330 @@
+#!/usr/bin/env bash
+#
+# Rebuild the rgbmatrix Python binding so it releases the GIL.
+#
+# WHY THIS EXISTS
+# ---------------
+# The upstream binding declares FrameCanvas::SwapOnVSync WITHOUT `nogil`
+# (cppinc.pxd), unlike SetPixel/Clear/Fill on the lines just above it.
+# SwapOnVSync blocks until the panel's next vertical sync -- up to a full
+# refresh period on every frame -- so the render thread was holding the GIL
+# for most of every frame. Background threads (API fetches, JSON parsing,
+# image decode) were starved into long uninterruptible bursts, which in turn
+# made the render loop miss refreshes.
+#
+# Measured on a Pi 4 driving a 2x128x64 chain at limit_refresh_rate_hz=100:
+#
+#     before   ~44 fps average, 14-17% of frames 41-53ms
+#     after    100 fps, median 10.00ms, p95 10.05ms, 0% stalls
+#
+# The per-pixel blit (SetPixelsPillow) can also release the GIL and walk the
+# Pillow buffer row-major, but that is OFF by default and you almost certainly
+# want to leave it that way. Row-major changes what a partially-written frame
+# looks like: column-major tearing shows as a vertical seam, row-major tearing
+# shows as a horizontal split between the panel's upper and lower halves. On a
+# 1/32 scan panel that reads as a one-pixel "fold" across the middle of every
+# panel -- reported on hardware, and it went away when the blit was reverted.
+# Enable with RGB_PATCH_BLIT=1 only if you have measured that you need it;
+# essentially all of the gain above comes from the SwapOnVSync change alone.
+#
+# SAFETY
+# ------
+# Builds into a scratch directory; touches the installed module only in the
+# --install step, and backs up the original first. Roll back at any time with:
+#
+#     sudo bash scripts/build_rgbmatrix_nogil.sh --rollback
+#
+# USAGE
+#     bash scripts/build_rgbmatrix_nogil.sh              # build only
+#     sudo bash scripts/build_rgbmatrix_nogil.sh --install
+#     sudo bash scripts/build_rgbmatrix_nogil.sh --rollback
+#
+set -uo pipefail
+
+# Resolve the invoking user's home, not root's. --install runs under sudo,
+# where $HOME is /root, so every default path below pointed somewhere the
+# build had never written and the install died with "no built module found".
+if [ -n "${SUDO_USER:-}" ]; then
+    OWNER_HOME="$(getent passwd "$SUDO_USER" | cut -d: -f6)"
+fi
+OWNER_HOME="${OWNER_HOME:-$HOME}"
+
+SRC_TREE="${RGB_SRC_TREE:-$OWNER_HOME/LEDMatrix/rpi-rgb-led-matrix-master}"
+BUILD_DIR="${RGB_BUILD_DIR:-$OWNER_HOME/rgbmatrix-nogil-build}"
+VENV="${RGB_CYTHON_VENV:-$OWNER_HOME/.cache/ledmatrix-cython}"
+BACKUP="${RGB_BACKUP:-$OWNER_HOME/rgbmatrix-core.so.ORIGINAL}"
+PATCH_BLIT="${RGB_PATCH_BLIT:-0}"
+
+die() { echo "FATAL: $*" >&2; exit 1; }
+
+# This script runs under `set -uo pipefail` -- no -e -- so an unchecked
+# systemctl failure is silently ignored. That matters most for `stop`: leaving
+# the old service running means cp overwrites a module the running process has
+# mapped, the following `start` succeeds as a no-op, and the health check sees
+# an active unit and reports SUCCESS for a binding that was never loaded.
+# A machine with no ledmatrix.service at all is a normal build host, so that
+# case is skipped rather than treated as a failure.
+service_present() { systemctl cat ledmatrix.service >/dev/null 2>&1; }
+
+service_do() {
+    local verb="$1"
+    if ! service_present; then
+        echo "   (no ledmatrix.service installed - skipping $verb)"
+        return 0
+    fi
+    systemctl "$verb" ledmatrix || die "systemctl $verb ledmatrix failed"
+}
+
+py_site() {
+    python3 -c 'import rgbmatrix, os; print(os.path.dirname(rgbmatrix.__file__))' 2>/dev/null
+}
+
+# The extension filename the interpreter that builds -- and then loads -- this
+# module actually uses, e.g. core.cpython-313-aarch64-linux-gnu.so. The build
+# venv is made with --system-site-packages from python3, so the two agree;
+# falling back keeps --install working when the venv has been cleaned up.
+abi_name() {
+    local py="$VENV/bin/python"
+    [ -x "$py" ] || py=python3
+    "$py" -c \
+        'import sysconfig; print("core" + sysconfig.get_config_var("EXT_SUFFIX"))' \
+        2>/dev/null
+}
+
+# Exactly the current interpreter's artifact, never merely the first one that
+# sorts. Staging copies $SRC_TREE wholesale, so a core.cpython-*.so left in the
+# source tree by an earlier build comes along for the ride; build_ext --inplace
+# only ever overwrites the current ABI's name, and a glob piped to `head -1`
+# sorts cpython-311 ahead of cpython-313. That installed a stale, unpatched
+# module as core.so while the GIL check below -- which reads the freshly
+# generated core.cpp, not the .so -- still reported success.
+abi_so() {
+    local name path
+    name="$(abi_name)" || return 1
+    [ -n "$name" ] || return 1
+    path="$BUILD_DIR/bindings/python/rgbmatrix/$name"
+    [ -f "$path" ] || return 1
+    printf '%s\n' "$path"
+}
+
+do_rollback() {
+    local dst; dst="$(py_site)"
+    [ -n "$dst" ] || die "could not locate the installed rgbmatrix package"
+    [ -f "$BACKUP" ] || die "no backup at $BACKUP"
+    service_do stop
+    cp -a "$BACKUP" "$dst/core.so" || die "restore failed"
+    find "$dst" -name __pycache__ -type d -exec rm -rf {} + 2>/dev/null
+    service_do start
+    echo "rolled back to the original core.so"
+    exit 0
+}
+
+do_install() {
+    local so dst
+    so="$(abi_so)"; [ -n "$so" ] || die "no built module found - run the build first"
+    dst="$(py_site)"; [ -n "$dst" ] || die "could not locate the installed rgbmatrix package"
+
+    if [ ! -f "$BACKUP" ]; then
+        cp -a "$dst/core.so" "$BACKUP" || die "could not back up the original"
+        echo "backed up original core.so -> $BACKUP"
+    else
+        echo "backup already present at $BACKUP (keeping the true original)"
+    fi
+
+    service_do stop
+    cp "$so" "$dst/core.so" || die "install failed"
+    find "$dst" -name __pycache__ -type d -exec rm -rf {} + 2>/dev/null
+    service_do start
+
+    echo "waiting 25s for the display to come back..."
+    sleep 25
+    local healthy=1
+    systemctl is-active --quiet ledmatrix || healthy=0
+    if journalctl -u ledmatrix --since "40 sec ago" --no-pager \
+         | grep -qiE "Traceback|ImportError|Segmentation fault|undefined symbol"; then
+        healthy=0
+    fi
+    if [ "$healthy" = "1" ]; then
+        echo "SUCCESS - running on the rebuilt binding"
+    else
+        echo "UNHEALTHY - rolling back"
+        cp -a "$BACKUP" "$dst/core.so" \
+            || echo "ROLLBACK FAILED: could not restore $BACKUP -> $dst/core.so" >&2
+        if service_present && ! systemctl restart ledmatrix; then
+            echo "ROLLBACK FAILED: ledmatrix did not restart - the display is" \
+                 "down; restore manually with 'sudo bash $0 --rollback'" >&2
+        fi
+        journalctl -u ledmatrix --since "90 sec ago" --no-pager | tail -25
+        exit 1
+    fi
+    exit 0
+}
+
+case "${1:-}" in
+    --rollback) do_rollback ;;
+    --install)  do_install ;;
+    "" )        ;;
+    *)          die "unknown option: $1" ;;
+esac
+
+# ---------------------------------------------------------------- build ----
+[ -d "$SRC_TREE" ] || die "matrix source tree not found at $SRC_TREE (set RGB_SRC_TREE)"
+command -v g++ >/dev/null || die "g++ not installed (apt install build-essential)"
+
+echo "==> staging a scratch copy at $BUILD_DIR"
+rm -rf "$BUILD_DIR"
+cp -r "$SRC_TREE" "$BUILD_DIR" || die "copy failed"
+
+# Drop any extension artifacts that came across from the source tree. Nothing
+# downstream should be able to pick one up, and build_ext --inplace can decide
+# a copied .so is already up to date and skip the compile entirely.
+find "$BUILD_DIR/bindings/python/rgbmatrix" -maxdepth 1 \
+     -name 'core*.so' -delete 2>/dev/null
+
+echo "==> patching the bindings to release the GIL"
+python3 - "$BUILD_DIR" "$PATCH_BLIT" <<'PYEOF' || die "patch failed"
+import io
+import sys
+
+base = sys.argv[1] + "/bindings/python/rgbmatrix/"
+patch_blit = len(sys.argv) > 2 and sys.argv[2] == "1"
+
+# --- declare SwapOnVSync as nogil ---------------------------------------
+p = base + "cppinc.pxd"
+s = io.open(p, encoding="utf-8").read()
+OLD_DECL = "        FrameCanvas *SwapOnVSync(FrameCanvas*, uint8_t)\n"
+NEW_DECL = "        FrameCanvas *SwapOnVSync(FrameCanvas*, uint8_t) nogil\n"
+if OLD_DECL in s:
+    io.open(p, "w", encoding="utf-8", newline="\n").write(s.replace(OLD_DECL, NEW_DECL, 1))
+    print("   cppinc.pxd: SwapOnVSync declared nogil")
+elif NEW_DECL in s:
+    print("   cppinc.pxd: already nogil")
+else:
+    sys.exit("could not find the SwapOnVSync declaration")
+
+# --- release the GIL across the vsync wait ------------------------------
+p = base + "core.pyx"
+s = io.open(p, encoding="utf-8").read()
+
+OLD_SWAP = (
+    "    def SwapOnVSync(self, FrameCanvas newFrame, uint8_t framerate_fraction = 1):\n"
+    "        return __createFrameCanvas("
+    "self.__matrix.SwapOnVSync(newFrame.__canvas, framerate_fraction))\n"
+)
+NEW_SWAP = (
+    "    def SwapOnVSync(self, FrameCanvas newFrame, uint8_t framerate_fraction = 1):\n"
+    "        # Blocks until the panel's next vertical sync. Holding the GIL\n"
+    "        # across that wait starves every other Python thread for most of\n"
+    "        # each frame. Pointers are hoisted into C locals so the blocking\n"
+    "        # call itself needs no Python state.\n"
+    "        cdef cppinc.RGBMatrix* matrix = self.__matrix\n"
+    "        cdef cppinc.FrameCanvas* frame = newFrame.__canvas\n"
+    "        cdef uint8_t fraction = framerate_fraction\n"
+    "        cdef cppinc.FrameCanvas* swapped\n"
+    "        with nogil:\n"
+    "            swapped = matrix.SwapOnVSync(frame, fraction)\n"
+    "        return __createFrameCanvas(swapped)\n"
+)
+if OLD_SWAP in s:
+    s = s.replace(OLD_SWAP, NEW_SWAP, 1)
+    print("   core.pyx: SwapOnVSync releases the GIL")
+elif "swapped = matrix.SwapOnVSync(frame, fraction)" in s:
+    print("   core.pyx: SwapOnVSync already patched")
+else:
+    sys.exit("could not find the SwapOnVSync body")
+
+# --- optional: release the GIL across the blit --------------------------
+OLD_BLIT = (
+    "        buffer = get_pillow_buffer(image_capsule)\n"
+    "\n"
+    "        for col in range(max(0, -xstart), min(width, frame_width - xstart)):\n"
+    "            for row in range(max(0, -ystart), min(height, frame_height - ystart)):\n"
+    "                pixel = buffer[row][col]\n"
+    "                r = (pixel ) & 0xFF\n"
+    "                g = (pixel >> 8) & 0xFF\n"
+    "                b = (pixel >> 16) & 0xFF\n"
+    "                my_canvas.SetPixel(xstart+col, ystart+row, r, g, b)\n"
+)
+NEW_BLIT = (
+    "        buffer = get_pillow_buffer(image_capsule)\n"
+    "\n"
+    "        # Bounds hoisted so the blit needs no Python state and can run\n"
+    "        # without the GIL: it touches only a C buffer and a C++ canvas.\n"
+    "        # NOTE: row-major order makes a torn frame show as a horizontal\n"
+    "        # split across the panel's halves. See the header before enabling.\n"
+    "        cdef int col_start = max(0, -xstart)\n"
+    "        cdef int col_end = min(width, frame_width - xstart)\n"
+    "        cdef int row_start = max(0, -ystart)\n"
+    "        cdef int row_end = min(height, frame_height - ystart)\n"
+    "\n"
+    "        with nogil:\n"
+    "            for row in range(row_start, row_end):\n"
+    "                for col in range(col_start, col_end):\n"
+    "                    pixel = buffer[row][col]\n"
+    "                    r = (pixel ) & 0xFF\n"
+    "                    g = (pixel >> 8) & 0xFF\n"
+    "                    b = (pixel >> 16) & 0xFF\n"
+    "                    my_canvas.SetPixel(xstart+col, ystart+row, r, g, b)\n"
+)
+if patch_blit:
+    if OLD_BLIT in s:
+        s = s.replace(OLD_BLIT, NEW_BLIT, 1)
+        print("   core.pyx: pixel blit releases the GIL, row-major")
+    elif "for row in range(row_start, row_end):" in s:
+        print("   core.pyx: blit already patched")
+    else:
+        sys.exit("could not find the SetPixelsPillow loop")
+else:
+    print("   core.pyx: blit left unpatched (RGB_PATCH_BLIT=1 to enable)")
+
+io.open(p, "w", encoding="utf-8", newline="\n").write(s)
+PYEOF
+
+echo "==> building librgbmatrix.a (this takes a few minutes)"
+nice -n 10 make -C "$BUILD_DIR/lib" -j2 >/dev/null 2>&1 \
+    || die "library build failed - rerun 'make -C $BUILD_DIR/lib' to see why"
+[ -f "$BUILD_DIR/lib/librgbmatrix.a" ] || die "librgbmatrix.a was not produced"
+
+echo "==> preparing Cython"
+[ -d "$VENV" ] || python3 -m venv --system-site-packages "$VENV" || die "venv failed"
+"$VENV/bin/pip" install --quiet cython || die "cython install failed"
+
+cat > "$BUILD_DIR/bindings/python/setup.py" <<'EOF'
+from setuptools import setup, Extension
+from Cython.Build import cythonize
+
+core = Extension(
+    "rgbmatrix.core",
+    sources=["rgbmatrix/core.pyx", "rgbmatrix/shims/pillow.c"],
+    include_dirs=["../../include", "rgbmatrix/shims"],
+    extra_objects=["../../lib/librgbmatrix.a"],
+    language="c++",
+    extra_compile_args=["-O3", "-Wall", "-fno-exceptions", "-std=c++11"],
+    extra_link_args=["-lrt", "-lm", "-lpthread"],
+)
+
+setup(name="rgbmatrix",
+      ext_modules=cythonize([core], language_level="3str",
+                            compiler_directives={"binding": False}))
+EOF
+
+echo "==> compiling the extension"
+( cd "$BUILD_DIR/bindings/python" && "$VENV/bin/python" setup.py build_ext --inplace ) \
+    >/dev/null 2>&1 || die "extension build failed"
+
+SO="$(abi_so)" || true
+[ -n "$SO" ] || die "no .so produced - expected $(abi_name) in $BUILD_DIR/bindings/python/rgbmatrix"
+
+# Verify the GIL really is released before anyone installs this.
+EXPECTED=1; [ "$PATCH_BLIT" = "1" ] && EXPECTED=2
+PAIRS=$(grep -c "PyEval_SaveThread\|Py_UNBLOCK_THREADS" \
+        "$BUILD_DIR/bindings/python/rgbmatrix/core.cpp")
+[ "$PAIRS" -ge "$EXPECTED" ] \
+    || die "generated C++ has $PAIRS GIL-release sites, expected >= $EXPECTED"
+
+echo
+echo "BUILT: $SO"
+echo "       ($PAIRS GIL-release site(s) in the generated C++)"
+echo
+echo "Install with:   sudo bash $0 --install"
+echo "Roll back with: sudo bash $0 --rollback"

scripts/scroll_speeds.py

@@ -0,0 +1,238 @@
+#!/usr/bin/env python3
+"""Show and try the scroll speeds your panel can display cleanly.
+
+Motion looks smooth when the strip advances a WHOLE number of pixels per panel
+refresh. Anything else has to blend two columns (which on pixel-font text reads
+as shimmer) or repeat frames unevenly (which reads as judder). So the speeds
+worth using are not arbitrary -- they are
+
+    refresh_hz / frame_hold * pixels_per_frame
+
+for whole numbers of frame_hold and pixels_per_frame, and that ladder depends
+on how fast YOUR panel actually refreshes. A Pi Zero driving a big chain will
+have a completely different set of good speeds from a Pi 4 driving a small one.
+
+    # what can this panel do? (no hardware needed, uses your configured rate)
+    python3 scripts/scroll_speeds.py
+
+    # measure what the panel ACTUALLY manages, rather than what is configured
+    sudo systemctl stop ledmatrix
+    sudo python3 scripts/scroll_speeds.py --measure
+    sudo systemctl start ledmatrix
+
+    # what would a 60Hz panel offer?
+    python3 scripts/scroll_speeds.py --hz 60
+
+    # try one on the panel
+    sudo systemctl stop ledmatrix
+    sudo python3 scripts/scroll_speeds.py --demo 50
+    sudo systemctl start ledmatrix
+
+This script never starts or stops the display service itself -- that is left to
+you, so a crash here can never leave the panel dark.
+"""
+from __future__ import annotations
+
+import argparse
+import json
+import os
+import sys
+import time
+from pathlib import Path
+
+sys.path.insert(0, str(Path(__file__).resolve().parent.parent))
+
+from src.common import scroll_config  # noqa: E402
+
+CONFIG = Path(__file__).resolve().parent.parent / "config" / "config.json"
+
+
+def load_hardware():
+    try:
+        with open(CONFIG, encoding="utf-8") as handle:
+            return (json.load(handle).get("display") or {}).get("hardware") or {}
+    except (OSError, ValueError):
+        return {}
+
+
+def build_options(hardware, refresh_override=None):
+    from rgbmatrix import RGBMatrixOptions
+
+    o = RGBMatrixOptions()
+    o.rows = int(hardware.get("rows", 32))
+    o.cols = int(hardware.get("cols", 64))
+    o.chain_length = int(hardware.get("chain_length", 1))
+    o.parallel = int(hardware.get("parallel", 1))
+    o.brightness = int(hardware.get("brightness", 80))
+    o.hardware_mapping = hardware.get("hardware_mapping", "regular")
+    o.pwm_bits = int(hardware.get("pwm_bits", 11))
+    o.pwm_dither_bits = int(hardware.get("pwm_dither_bits", 0))
+    o.pwm_lsb_nanoseconds = int(hardware.get("pwm_lsb_nanoseconds", 130))
+    o.led_rgb_sequence = hardware.get("led_rgb_sequence", "RGB")
+    o.scan_mode = int(hardware.get("scan_mode", 0))
+    o.row_address_type = int(hardware.get("row_address_type", 0))
+    o.multiplexing = int(hardware.get("multiplexing", 0))
+    o.gpio_slowdown = int(hardware.get("gpio_slowdown", 2))
+    o.limit_refresh_rate_hz = (
+        int(refresh_override) if refresh_override is not None
+        else int(hardware.get("limit_refresh_rate_hz", 0))
+    )
+    return o
+
+
+def open_matrix(hardware, refresh_override=None):
+    """Construct the matrix, or explain why it will not open."""
+    if os.geteuid() != 0:
+        sys.exit("this needs root for GPIO access - rerun with sudo")
+    try:
+        from rgbmatrix import RGBMatrix
+    except ImportError:
+        sys.exit("rgbmatrix is not installed on this machine")
+    try:
+        return RGBMatrix(options=build_options(hardware, refresh_override))
+    except Exception as exc:  # pragma: no cover - hardware dependent
+        sys.exit(
+            "could not open the panel ({}).\n"
+            "If the display service is running it owns the GPIO - stop it first:\n"
+            "    sudo systemctl stop ledmatrix".format(exc)
+        )
+
+
+def measure_refresh(hardware, seconds=6.0):
+    """Actual refresh rate, by running uncapped and timing the swaps.
+
+    SwapOnVSync blocks until the panel's next refresh, so an unthrottled loop
+    runs at exactly the panel's rate. This is what an older Pi or a longer
+    chain will really give you, as opposed to whatever limit_refresh_rate_hz
+    optimistically asks for.
+    """
+    matrix = open_matrix(hardware, refresh_override=0)
+    canvas = matrix.CreateFrameCanvas()
+    canvas = matrix.SwapOnVSync(canvas)  # discard the first, it includes setup
+    frames = 0
+    started = time.perf_counter()
+    while time.perf_counter() - started < seconds:
+        canvas = matrix.SwapOnVSync(canvas)
+        frames += 1
+    measured = frames / (time.perf_counter() - started)
+    matrix.Clear()
+    return measured
+
+
+def demo(hardware, target, seconds):
+    """Scroll text at the crisp speed nearest `target`."""
+    from PIL import Image, ImageDraw, ImageFont
+
+    hz = float(hardware.get("limit_refresh_rate_hz") or scroll_config.DEFAULT_REFRESH_HZ)
+    choice = scroll_config.solve_crisp(target, hz)
+    print("asked for {:.0f} px/s -> {}".format(target, choice.describe()))
+
+    matrix = open_matrix(hardware)
+    canvas = matrix.CreateFrameCanvas()
+    W, H = canvas.width, canvas.height
+
+    font = None
+    for path, size in (
+        (str(Path(__file__).resolve().parent.parent / "assets/fonts/PressStart2P-Regular.ttf"), 16),
+        ("/usr/share/fonts/truetype/dejavu/DejaVuSansMono-Bold.ttf", 26),
+    ):
+        try:
+            font = ImageFont.truetype(path, size)
+            break
+        except OSError:
+            continue
+    if font is None:
+        font = ImageFont.load_default()
+
+    text = "  {:.0f} px/s  ***  THE QUICK BROWN FOX JUMPS OVER THE LAZY DOG  ***".format(
+        choice.pixels_per_second)
+    box = ImageDraw.Draw(Image.new("RGB", (8, 8))).textbbox((0, 0), text, font=font)
+    tw, th = box[2] - box[0], box[3] - box[1]
+    reps = max(2, (W * 3) // max(tw, 1) + 1)
+    strip = Image.new("RGB", (tw * reps, H), (0, 0, 0))
+    draw = ImageDraw.Draw(strip)
+    for i in range(reps):
+        draw.text((i * tw, (H - th) // 2 - box[1]), text, font=font, fill=(255, 210, 60))
+
+    offset = 0
+    frames = 0
+    started = time.time()
+    while time.time() - started < seconds:
+        window = strip.crop((offset, 0, offset + W, H))
+        if window.width < W:
+            whole = Image.new("RGB", (W, H), (0, 0, 0))
+            head = strip.crop((offset, 0, strip.width, H))
+            whole.paste(head, (0, 0))
+            whole.paste(strip.crop((0, 0, W - head.width, H)), (head.width, 0))
+            window = whole
+        canvas.SetImage(window)
+        canvas = matrix.SwapOnVSync(canvas, choice.frame_hold)
+        offset = (offset + choice.pixels_per_frame) % strip.width
+        frames += 1
+    elapsed = time.time() - started
+    print("  {} frames in {:.1f}s = {:.1f} fps = {:.1f} px/s actual".format(
+        frames, elapsed, frames / elapsed, frames * choice.pixels_per_frame / elapsed))
+    matrix.Clear()
+
+
+def print_ladder(hz, highlight=None):
+    print("")
+    print("Whole-pixel scroll speeds at {:.1f}Hz refresh".format(hz))
+    print("(the panel refreshes at {:.0f}Hz for every one of these - holding a "
+          "frame costs no flicker)".format(hz))
+    print("")
+    for entry in scroll_config.crisp_ladder(hz):
+        if entry.pixels_per_second > hz * 3:
+            break
+        mark = " <-- nearest to {:.0f}".format(highlight) if (
+            highlight is not None
+            and entry.pixels_per_second == scroll_config.solve_crisp(highlight, hz).pixels_per_second
+        ) else ""
+        print("  " + entry.describe() + mark)
+    print("")
+    print("Set one in config.json as pixels per second, e.g.")
+    print('  "display_options": {{"scroll_pixels_per_second": {:.0f}}}'.format(
+        scroll_config.solve_crisp(highlight if highlight else hz / 2, hz).pixels_per_second))
+
+
+def main():
+    ap = argparse.ArgumentParser(description=__doc__,
+                                 formatter_class=argparse.RawDescriptionHelpFormatter)
+    ap.add_argument("--hz", type=float,
+                    help="refresh rate to compute the ladder for (default: your config)")
+    ap.add_argument("--measure", action="store_true",
+                    help="measure the panel's real refresh rate (needs root, service stopped)")
+    ap.add_argument("--demo", type=float, metavar="PXPS",
+                    help="scroll text at the crisp speed nearest this (needs root)")
+    ap.add_argument("--seconds", type=float, default=15.0, help="demo duration")
+    ap.add_argument("--want", type=float, metavar="PXPS",
+                    help="highlight the entry nearest this speed")
+    args = ap.parse_args()
+
+    hardware = load_hardware()
+    configured = float(hardware.get("limit_refresh_rate_hz") or 0)
+
+    if args.demo is not None:
+        demo(hardware, args.demo, args.seconds)
+        return
+
+    if args.measure:
+        measured = measure_refresh(hardware)
+        print("measured panel refresh: {:.1f}Hz".format(measured))
+        if configured:
+            print("configured limit_refresh_rate_hz: {:.0f}".format(configured))
+            if measured < configured * 0.95:
+                print("  -> the panel cannot reach the configured rate; the ladder")
+                print("     below uses what it actually manages")
+        print_ladder(measured, args.want)
+        return
+
+    hz = args.hz or configured or scroll_config.DEFAULT_REFRESH_HZ
+    if not args.hz and not configured:
+        print("no limit_refresh_rate_hz in config; assuming {:.0f}Hz".format(hz))
+        print("run with --measure to find your panel's real rate")
+    print_ladder(hz, args.want)
+
+
+if __name__ == "__main__":
+    main()

src/cache/disk_cache.py

@@ -5,6 +5,7 @@ Handles persistent disk-based caching with atomic writes and error recovery.
 """
 
 import json
+import math
 import os
 import time
 import tempfile
@@ -14,6 +15,11 @@ import zlib
 from typing import Dict, Any, Optional, Protocol
 from datetime import datetime
 
+try:  # optional: large speedup on the cache write path, see _dumps below
+    import orjson
+except ImportError:  # pragma: no cover - exercised on hosts without the wheel
+    orjson = None
+
 # How old an abandoned write's temp file must be before the sweep removes it.
 # A real write holds its temp file for milliseconds, so an hour is far beyond
 # any in-flight write while still clearing the same day's debris. Deliberately
@@ -40,13 +46,97 @@ class CacheStrategyProtocol(Protocol):
 
 
 class DateTimeEncoder(json.JSONEncoder):
-    """JSON encoder that handles datetime objects."""
+    """JSON encoder that handles datetime objects.
+
+    Retained for the stdlib fallback path and for any caller importing it.
+    """
     def default(self, obj: Any) -> Any:
         if isinstance(obj, datetime):
             return obj.isoformat()
         return super().default(obj)
 
 
+def _datetime_default(obj: Any) -> Any:
+    """Serialise datetimes exactly as DateTimeEncoder did."""
+    if isinstance(obj, datetime):
+        return obj.isoformat()
+    raise TypeError(f"Object of type {type(obj).__name__} is not JSON serializable")
+
+
+def _replace_nonfinite(obj: Any) -> Any:
+    """Non-finite floats -> None, matching what ``orjson.dumps`` writes.
+
+    Only reached once a strict pass has proved there is something to replace,
+    so the ordinary write path never pays for this walk.
+    """
+    if isinstance(obj, float):
+        return obj if math.isfinite(obj) else None
+    if isinstance(obj, dict):
+        return {k: _replace_nonfinite(v) for k, v in obj.items()}
+    if isinstance(obj, (list, tuple)):
+        return [_replace_nonfinite(v) for v in obj]
+    return obj
+
+
+# NON-FINITE FLOATS
+# -----------------
+# JSON has no NaN or Infinity. The stdlib emits them anyway as an extension;
+# orjson refuses to and writes null. That divergence is not acceptable in a
+# cache whose files outlive the decision of which encoder is installed, so the
+# policy here is one behaviour on both paths:
+#
+#   writing  non-finite floats become null, whichever encoder is in use
+#   reading  files already on disk that carry the stdlib's NaN/Infinity
+#            tokens stay readable, whichever encoder is in use
+#
+# Without the write half, installing orjson silently changed cached values.
+# Without the read half, installing orjson turned every legacy record holding a
+# NaN into a "corrupted cache file" that DiskCache.get logged as an error and
+# deleted. Both halves are covered by test/test_cache_nonfinite_floats.py.
+
+
+if orjson is not None:
+    # Encoding the cache record dominated the background fetch worker: on a
+    # Pi 4, stdlib json.dumps runs ~12ms per MB and holds the GIL for all of
+    # it, which stalls the render thread mid-scroll. orjson measures ~7x
+    # faster on the same payloads (11.9ms -> 1.6ms for 985KB). Decoding gains
+    # far less (~1.3x on large payloads) because the cost there is building
+    # the Python objects, not scanning the text, but it is still free to take.
+    #
+    # OPT_NON_STR_KEYS: stdlib json coerces int/float dict keys to strings;
+    # orjson raises without this, and cache records do carry numeric keys.
+    # OPT_PASSTHROUGH_DATETIME: orjson would otherwise emit its own RFC 3339
+    # form for datetimes instead of calling default(). Routing them through
+    # _datetime_default keeps byte-for-byte parity with the records already
+    # on disk.
+    _DUMPS_OPTS = orjson.OPT_NON_STR_KEYS | orjson.OPT_PASSTHROUGH_DATETIME
+
+    def _dumps(data: Any) -> bytes:
+        return orjson.dumps(data, default=_datetime_default, option=_DUMPS_OPTS)
+
+    def _loads(raw: bytes) -> Any:
+        try:
+            return orjson.loads(raw)
+        except orjson.JSONDecodeError:
+            # Legacy record written by the stdlib path, carrying NaN or
+            # Infinity. Genuinely malformed files raise again from here, as
+            # json.JSONDecodeError, which is what DiskCache.get expects.
+            return json.loads(raw)
+else:
+    def _dumps(data: Any) -> bytes:
+        try:
+            return json.dumps(data, cls=DateTimeEncoder,
+                              allow_nan=False).encode("utf-8")
+        except ValueError:
+            # allow_nan=False is what detects the non-finite values; the walk
+            # runs only now that we know there is one to replace.
+            return json.dumps(_replace_nonfinite(data), cls=DateTimeEncoder,
+                              allow_nan=False).encode("utf-8")
+
+    def _loads(raw: bytes) -> Any:
+        return json.loads(raw)
+
+
 class DiskCache:
     """Manages persistent disk-based cache."""
     
@@ -99,8 +189,8 @@ class DiskCache:
         
         try:
             with self._lock:
-                with open(cache_path, 'r', encoding='utf-8') as f:
-                    record = json.load(f)
+                with open(cache_path, 'rb') as f:
+                    record = _loads(f.read())
             
             # Determine record timestamp (prefer embedded, else file mtime)
             record_ts = None
@@ -189,12 +279,12 @@ class DiskCache:
         # write path below, and cache files are machine-read only — indenting
         # them just multiplied the bytes written to the SD card.
         try:
-            payload = json.dumps(data, cls=DateTimeEncoder)
+            payload = _dumps(data)
         except (TypeError, ValueError) as e:
             self.logger.warning("Cache data for key '%s' not serializable: %s", key, e)
             return
 
-        digest = zlib.adler32(payload.encode('utf-8'))
+        digest = zlib.adler32(payload)
 
         try:
             # Atomic write to avoid partial/corrupt files
@@ -242,7 +332,7 @@ class DiskCache:
                         # wear source (dozens of fsyncs/min on API-heavy
                         # installs) for data that can be re-downloaded.
                         try:
-                            with os.fdopen(fd, 'w', encoding='utf-8') as tmp_file:
+                            with os.fdopen(fd, 'wb') as tmp_file:
                                 tmp_file.write(payload)
                             os.replace(tmp_path, cache_path)
                             self._write_digests[key] = digest
@@ -260,7 +350,7 @@ class DiskCache:
                     else:
                         # Fallback: direct write (not atomic, but better than failing)
                         try:
-                            with open(cache_path, 'w', encoding='utf-8') as cache_file:
+                            with open(cache_path, 'wb') as cache_file:
                                 cache_file.write(payload)
                             self._write_digests[key] = digest
                             # Set proper permissions: 660 (rw-rw----) for group-readable cache files
@@ -290,7 +380,7 @@ class DiskCache:
                             # is a different path, so future sets must keep
                             # retrying the primary location.
                             fallback_path = os.path.join(fallback_dir, os.path.basename(cache_path))
-                            with open(fallback_path, 'w', encoding='utf-8') as tmp_file:
+                            with open(fallback_path, 'wb') as tmp_file:
                                 tmp_file.write(payload)
                             # Set proper permissions: 660 (rw-rw----) for group-readable cache files
                             try:

src/common/__init__.py

@@ -23,6 +23,13 @@ from src.common.error_handler import (
 )
 from src.common.api_helper import APIHelper
 from src.common.scroll_helper import ScrollHelper
+from src.common import scroll_config
+from src.common.scroll_config import (
+    ScrollSettings,
+    configure as configure_scroll,
+    resolve as resolve_scroll_settings,
+    refresh_hz_from_config,
+)
 from src.common.logo_helper import LogoHelper
 from src.common.text_helper import TextHelper
 
@@ -60,6 +67,11 @@ __all__ = [
     'log_and_raise',
     'APIHelper',
     'ScrollHelper',
+    'scroll_config',
+    'ScrollSettings',
+    'configure_scroll',
+    'resolve_scroll_settings',
+    'refresh_hz_from_config',
     'LogoHelper',
     'TextHelper',
     # adaptive layout & images

src/common/scroll_config.py

@@ -0,0 +1,440 @@
+"""One place that turns plugin config into a configured ScrollHelper.
+
+Five ticker plugins each hand-rolled this resolution (odds-ticker, news and
+ledmatrix-leaderboard reference the deprecated ``scroll_pixels_per_second``
+key 16-18 times apiece), and they disagreed in ways that were invisible until
+someone watched the panel:
+
+* odds-ticker read ``scroll_pixels_per_second`` on the *recommended* config
+  path and let it override ``scroll_speed``/``scroll_delay``. Because that key
+  carries a schema default, the documented settings were dead for every user
+  -- see ChuckBuilds/ledmatrix-plugins#408.
+* ledmatrix-leaderboard read the same key only as a fallback, so identical
+  config produced different speeds in the two plugins.
+* stock-news derived px/frame from it via its own arithmetic.
+
+What matters on the hardware
+----------------------------
+Motion is smooth when the strip advances a **whole number of pixels per panel
+refresh**. On a 100Hz panel that means 100 px/s, 200 px/s, and so on. Anything
+else has to either blend adjacent columns (which on pixel-font text reads as
+shimmer) or repeat frames (which reads as judder). :func:`resolve` warns when
+the requested speed will not divide evenly, because that is a real display
+artefact and not a rounding detail.
+
+Speed is always expressed to the helper as pixels per second and applied in
+time-based mode. Frame-based stepping gates motion on a wall clock at
+``1/scroll_delay`` steps per second; plugins set ``scroll_delay`` to the frame
+period, which puts that comparison exactly on its own threshold and makes the
+step count flip on sub-millisecond jitter. Accumulating elapsed time keeps
+position proportional to real time instead.
+"""
+
+from __future__ import annotations
+
+import logging
+from dataclasses import dataclass, replace
+from typing import Any, Dict, Optional
+
+logger = logging.getLogger(__name__)
+
+#: Speed used when a plugin supplies nothing usable. One pixel per refresh on a
+#: 100Hz panel, which is the slowest crisp scroll that hardware can show.
+DEFAULT_PIXELS_PER_SECOND = 100.0
+
+#: Bounds accepted from config. Below the floor a marquee appears frozen;
+#: above the ceiling it outruns any panel's refresh and tears.
+MIN_PIXELS_PER_SECOND = 1.0
+MAX_PIXELS_PER_SECOND = 500.0
+
+#: Assumed refresh when the caller does not say. Matches the usual
+#: ``display.hardware.limit_refresh_rate_hz``.
+DEFAULT_REFRESH_HZ = 100.0
+
+#: How far px/s may sit from a whole number of pixels per refresh before it is
+#: worth warning about. 0.05px per frame is invisible; a third of a pixel is not.
+_WHOLE_PIXEL_TOLERANCE = 0.05
+
+
+#: Longest a frame may be held before motion reads as a slideshow rather than
+#: a scroll. 6 refreshes at 100Hz is ~17px/s, already visibly stepped.
+MAX_FRAME_HOLD = 8
+
+#: Largest whole-pixel jump per presented frame before motion looks like it is
+#: teleporting rather than sliding.
+MAX_PIXELS_PER_FRAME = 6
+
+
+@dataclass(frozen=True)
+class CrispSpeed:
+    """A speed the panel can show with whole-pixel motion.
+
+    ``pixels_per_second`` is always ``refresh_hz / frame_hold * pixels_per_frame``
+    exactly -- no rounding, no fractional pixel positions, so nothing has to be
+    blended or repeated unevenly.
+
+    :param frame_hold: refreshes each frame is held for. This is rgbmatrix's
+        ``SwapOnVSync(canvas, framerate_fraction)``. The panel keeps refreshing
+        at full rate either way, so holding a frame costs nothing in flicker.
+    :param pixels_per_frame: whole pixels advanced per presented frame.
+    """
+
+    pixels_per_second: float
+    frame_hold: int
+    pixels_per_frame: int
+    refresh_hz: float
+
+    @property
+    def frames_per_second(self) -> float:
+        return self.refresh_hz / self.frame_hold
+
+    @property
+    def steppiness(self) -> str:
+        """Rough readability hint for this combination."""
+        if self.pixels_per_frame > 2:
+            return "jumpy"
+        if self.frames_per_second < 20:
+            return "stepped"
+        if self.frames_per_second < 30:
+            return "slightly stepped"
+        return "smooth"
+
+    def describe(self) -> str:
+        return (
+            f"{self.pixels_per_second:6.1f} px/s  "
+            f"({self.pixels_per_frame}px every {self.frame_hold} refresh"
+            f"{'es' if self.frame_hold != 1 else ' '} = "
+            f"{self.frames_per_second:5.1f} fps, {self.steppiness})"
+        )
+
+
+def crisp_ladder(
+    refresh_hz: float = DEFAULT_REFRESH_HZ,
+    max_frame_hold: int = MAX_FRAME_HOLD,
+    max_pixels_per_frame: int = MAX_PIXELS_PER_FRAME,
+):
+    """Every whole-pixel speed this panel can show, slowest first.
+
+    Duplicates are collapsed keeping the gentlest option: 100 px/s is reachable
+    as 1px every refresh or 2px every 2nd refresh, and the former moves in
+    smaller increments, so that is the one worth offering.
+    """
+    best = {}
+    for hold in range(1, max_frame_hold + 1):
+        for ppf in range(1, max_pixels_per_frame + 1):
+            pps = refresh_hz / hold * ppf
+            key = round(pps, 3)
+            candidate = CrispSpeed(pps, hold, ppf, refresh_hz)
+            incumbent = best.get(key)
+            if incumbent is None or ppf < incumbent.pixels_per_frame:
+                best[key] = candidate
+    return [best[k] for k in sorted(best)]
+
+
+#: How much a bigger pixel step costs, as a fraction of the target speed.
+#: Tuned so 66.7px/s (2px at 33fps) beats 50px/s (1px at 50fps) when 60 was
+#: asked for, but 33.3px/s (1px, smooth) still beats 28.6px/s (2px at 14fps)
+#: when 30 was asked for -- being 11% slow is worth far less than looking bad.
+_STEP_PENALTY = 0.05
+_SLOW_FPS_PENALTY = 0.25   # below 20fps
+_LOWISH_FPS_PENALTY = 0.10  # below 25fps
+
+
+def _quality_cost(candidate: "CrispSpeed", target: float) -> float:
+    """Lower is better. Numeric closeness alone picks bad-looking speeds.
+
+    Nearest-by-value would answer "30 px/s" with 28.6 px/s -- which is 2px
+    jumps at 14fps -- over 33.3 px/s, which is single-pixel motion at 33fps and
+    obviously better on the panel. Proximity has to be traded against how the
+    motion actually reads.
+    """
+    error = abs(candidate.pixels_per_second - target) / max(target, 1e-6)
+    cost = error + _STEP_PENALTY * (candidate.pixels_per_frame - 1)
+    fps = candidate.frames_per_second
+    if fps < 20:
+        cost += _SLOW_FPS_PENALTY
+    elif fps < 25:
+        cost += _LOWISH_FPS_PENALTY
+    return cost
+
+
+def solve_crisp(
+    target_pixels_per_second: float,
+    refresh_hz: float = DEFAULT_REFRESH_HZ,
+    max_frame_hold: int = MAX_FRAME_HOLD,
+    max_pixels_per_frame: int = MAX_PIXELS_PER_FRAME,
+) -> CrispSpeed:
+    """The whole-pixel speed that will look best for what was asked for.
+
+    Not simply the nearest -- see :func:`_quality_cost`. Ties break toward the
+    smaller pixel step and the shorter hold.
+    """
+    ladder = crisp_ladder(refresh_hz, max_frame_hold, max_pixels_per_frame)
+    # Clamp into the ladder's range first. Relative error saturates near 1.0
+    # for a target far outside it, so the quality penalty would dominate and
+    # answer "10000 px/s" with the *slowest* entry -- smooth, and useless.
+    target = min(max(target_pixels_per_second, ladder[0].pixels_per_second),
+                 ladder[-1].pixels_per_second)
+    return min(
+        ladder,
+        key=lambda c: (round(_quality_cost(c, target), 6),
+                       c.pixels_per_frame, c.frame_hold),
+    )
+
+
+@dataclass(frozen=True)
+class ScrollSettings:
+    """The resolved outcome, and which config key produced it."""
+
+    pixels_per_second: float
+    source: str
+    target_fps: Optional[float] = None
+    pixels_per_frame: Optional[float] = None
+    warning: Optional[str] = None
+    #: The whole-pixel speed actually applied, when snapping was enabled.
+    crisp: Optional[CrispSpeed] = None
+    #: What the config asked for, before snapping.
+    requested_pixels_per_second: Optional[float] = None
+
+    @property
+    def frame_hold(self) -> int:
+        """Refreshes to hold each frame for; pass to set_scrolling_state()."""
+        return self.crisp.frame_hold if self.crisp else 1
+
+    def describe(self) -> str:
+        text = f"{self.pixels_per_second:.1f} px/s (from {self.source})"
+        if self.pixels_per_frame is not None:
+            text += f" = {self.pixels_per_frame:.2f} px/frame"
+        if self.target_fps:
+            text += f" at {self.target_fps:.0f} fps"
+        return text
+
+
+def _coerce(value: Any) -> Optional[float]:
+    """A positive float, or None. Config reaches us with nulls and strings."""
+    if value is None or isinstance(value, bool):
+        return None
+    try:
+        number = float(value)
+    except (TypeError, ValueError):
+        return None
+    return number if number > 0 else None
+
+
+def _from_speed_and_delay(block: Any) -> Optional[float]:
+    """px/s from a ``scroll_speed`` (px/frame) + ``scroll_delay`` (s) pair."""
+    if not isinstance(block, dict):
+        return None
+    speed = _coerce(block.get("scroll_speed"))
+    delay = _coerce(block.get("scroll_delay"))
+    if speed is None or delay is None:
+        return None
+    return speed / delay
+
+
+def resolve(
+    plugin_config: Optional[Dict[str, Any]] = None,
+    global_config: Optional[Dict[str, Any]] = None,
+    default_pixels_per_second: float = DEFAULT_PIXELS_PER_SECOND,
+    refresh_hz: Optional[float] = None,
+) -> ScrollSettings:
+    """Resolve one scroll speed from the several shapes plugins accept.
+
+    Precedence, highest first. The deprecated flat key sits *below* the
+    explicit pairs deliberately: it carries schema defaults in some plugins, so
+    ranking it above them silently disables the documented settings.
+
+    1. ``display_options.scroll_speed`` + ``scroll_delay``  (current)
+    2. ``display.scroll_speed`` + ``scroll_delay``          (deprecated shape)
+    3. ``scroll_speed`` + ``scroll_delay`` at the root      (legacy flat)
+    4. ``scroll_pixels_per_second``, nested or flat         (deprecated)
+    5. the global ``display`` block
+    6. ``default_pixels_per_second``
+
+    :param refresh_hz: panel refresh, used only to check whether the resolved
+        speed lands on whole pixels per frame and to fill in ``target_fps``.
+    """
+    plugin_config = plugin_config or {}
+    global_config = global_config or {}
+    refresh = _coerce(refresh_hz) or DEFAULT_REFRESH_HZ
+
+    display_options = plugin_config.get("display_options")
+    display_block = plugin_config.get("display")
+
+    candidates = [
+        (_from_speed_and_delay(display_options), "display_options.scroll_speed/delay"),
+        (_from_speed_and_delay(display_block), "display.scroll_speed/delay"),
+        (_from_speed_and_delay(plugin_config), "scroll_speed/delay (root)"),
+    ]
+    for block, label in (
+        (display_options, "display_options.scroll_pixels_per_second"),
+        (display_block, "display.scroll_pixels_per_second"),
+        (plugin_config, "scroll_pixels_per_second"),
+    ):
+        if isinstance(block, dict):
+            candidates.append((_coerce(block.get("scroll_pixels_per_second")), label))
+
+    global_display = global_config.get("display")
+    candidates.append((_from_speed_and_delay(global_display), "global display.scroll_speed/delay"))
+
+    pixels_per_second = None
+    source = "default"
+    for value, label in candidates:
+        if value is not None:
+            pixels_per_second, source = value, label
+            break
+    if pixels_per_second is None:
+        pixels_per_second = default_pixels_per_second
+
+    clamped = max(MIN_PIXELS_PER_SECOND, min(MAX_PIXELS_PER_SECOND, pixels_per_second))
+    warning = None
+    if clamped != pixels_per_second:
+        warning = (
+            f"scroll speed {pixels_per_second:.1f} px/s out of range, "
+            f"clamped to {clamped:.1f}"
+        )
+        pixels_per_second = clamped
+
+    pixels_per_frame = pixels_per_second / refresh if refresh > 0 else None
+    if warning is None and pixels_per_frame is not None:
+        offset = abs(pixels_per_frame - round(pixels_per_frame))
+        if pixels_per_frame < 1.0 - _WHOLE_PIXEL_TOLERANCE or offset > _WHOLE_PIXEL_TOLERANCE:
+            suggestion = max(1.0, round(pixels_per_frame)) * refresh
+            warning = (
+                f"{pixels_per_second:.1f} px/s is {pixels_per_frame:.2f} px per "
+                f"refresh at {refresh:.0f}Hz, so some frames repeat and the "
+                f"scroll will judder; {suggestion:.0f} px/s divides evenly"
+            )
+
+    return ScrollSettings(
+        pixels_per_second=pixels_per_second,
+        source=source,
+        target_fps=refresh,
+        pixels_per_frame=pixels_per_frame,
+        warning=warning,
+    )
+
+
+def configure(
+    scroll_helper: Any,
+    plugin_config: Optional[Dict[str, Any]] = None,
+    global_config: Optional[Dict[str, Any]] = None,
+    default_pixels_per_second: float = DEFAULT_PIXELS_PER_SECOND,
+    refresh_hz: Optional[float] = None,
+    plugin_logger: Optional[logging.Logger] = None,
+    display_manager: Any = None,
+    snap_to_crisp: bool = True,
+) -> ScrollSettings:
+    """Resolve the config and apply it to ``scroll_helper``.
+
+    Applied in time-based mode: see the module docstring for why frame-based
+    stepping is not used. ``hasattr`` guards keep this usable against older
+    ScrollHelper builds that a plugin may be running on.
+
+    :param display_manager: consulted for the panel's refresh rate only (it can
+        see display.hardware; a plugin cannot). The frame hold is NOT applied
+        here -- see the note in the body. The caller must pass
+        ``settings.frame_hold`` to ``display_manager.set_scrolling_state(True,
+        ...)`` when it starts scrolling, or a sub-refresh speed still presents
+        a new frame every refresh and the motion falls back to fractional
+        pixels.
+    :param snap_to_crisp: move the requested speed to the nearest speed the
+        panel can show in whole pixels. On by default because a speed that does
+        not divide evenly has no good rendering, only a choice of artefacts.
+
+    :returns: the settings applied, so the caller can log or assert on them.
+    """
+    log = plugin_logger or logger
+
+    # Refresh rate, most authoritative first: what the caller passed, then the
+    # display manager (which can see display.hardware; a plugin cannot), then
+    # the global config, then the default.
+    #
+    # This has to be settled BEFORE resolve(), not after. resolve() uses the
+    # refresh to fill in target_fps, pixels_per_frame and the judder warning,
+    # so deriving it afterwards described a 100Hz panel to everyone running at
+    # 60 -- and with snap_to_crisp=False nothing downstream corrected it, so
+    # set_target_fps() paced the helper to 100 FPS on a 60Hz panel.
+    hz = _coerce(refresh_hz)
+    if hz is None and display_manager is not None:
+        hz = _coerce(getattr(display_manager, "refresh_hz", None))
+    if hz is None:
+        hz = refresh_hz_from_config(global_config)
+
+    settings = resolve(
+        plugin_config,
+        global_config,
+        default_pixels_per_second=default_pixels_per_second,
+        refresh_hz=hz,
+    )
+    applied = settings.pixels_per_second
+    choice = None
+
+    if snap_to_crisp:
+        choice = solve_crisp(settings.pixels_per_second, hz)
+        applied = choice.pixels_per_second
+        settings = replace(
+            settings,
+            pixels_per_second=applied,
+            requested_pixels_per_second=settings.pixels_per_second,
+            crisp=choice,
+            pixels_per_frame=float(choice.pixels_per_frame),
+            # Snapping resolves the whole-pixel problem the warning describes.
+            warning=None if settings.warning and "judder" in settings.warning
+            else settings.warning,
+        )
+
+    if hasattr(scroll_helper, "set_frame_based_scrolling"):
+        scroll_helper.set_frame_based_scrolling(False)
+    scroll_helper.set_scroll_speed(applied)
+    if choice and hasattr(scroll_helper, "set_target_fps"):
+        scroll_helper.set_target_fps(choice.frames_per_second)
+    elif settings.target_fps and hasattr(scroll_helper, "set_target_fps"):
+        scroll_helper.set_target_fps(settings.target_fps)
+
+    # Deliberately NOT applied here. The hold belongs to a scroll, not to a
+    # plugin's lifetime: plugins share one display manager, and one left set at
+    # construction is reset the moment any other plugin finishes scrolling.
+    # Callers pass settings.frame_hold to set_scrolling_state(True, ...) when
+    # they start scrolling. configure() only reports what is needed.
+
+    if choice:
+        requested = settings.requested_pixels_per_second
+        if abs(requested - applied) > 0.05:
+            log.info(
+                "Scroll configured: %s (asked for %.1f px/s from %s; "
+                "nearest whole-pixel speed on a %.0fHz panel)",
+                choice.describe(), requested, settings.source, hz,
+            )
+        else:
+            log.info("Scroll configured: %s (from %s)",
+                     choice.describe(), settings.source)
+        if choice.frame_hold > 1:
+            log.debug(
+                "Scroll needs a frame hold of %d - pass settings.frame_hold to "
+                "display_manager.set_scrolling_state(True, ...) each scroll",
+                choice.frame_hold,
+            )
+    else:
+        log.info("Scroll configured: %s", settings.describe())
+
+    if settings.warning:
+        log.warning("Scroll speed: %s", settings.warning)
+    return settings
+
+
+def refresh_hz_from_config(global_config: Optional[Dict[str, Any]]) -> float:
+    """The panel's refresh cap from the global config, or the default."""
+    if not isinstance(global_config, dict):
+        return DEFAULT_REFRESH_HZ
+    # Each level is checked for being a mapping rather than merely truthy: a
+    # malformed config where display or display.hardware is a string or a list
+    # raised AttributeError out of what is meant to be a total function with a
+    # default, taking down every caller that asked for the refresh rate.
+    display = global_config.get("display")
+    if not isinstance(display, dict):
+        return DEFAULT_REFRESH_HZ
+    hardware = display.get("hardware")
+    if not isinstance(hardware, dict):
+        return DEFAULT_REFRESH_HZ
+    return _coerce(hardware.get("limit_refresh_rate_hz")) or DEFAULT_REFRESH_HZ

src/common/scroll_helper.py

@@ -16,6 +16,7 @@ Features:
 """
 
 import logging
+import math
 import time
 from typing import Optional, Dict, Any
 from PIL import Image
@@ -29,6 +30,49 @@ except ImportError:
     HAS_SCIPY = False
 
 
+def frame_stats(frame_times: list) -> Dict[str, Any]:
+    """Summary statistics over one window of frame durations (seconds).
+
+    Split out of log_frame_rate() so the arithmetic can be tested without a
+    clock. Median and p95 are the real ones: the median takes both middle
+    samples on an even window, and p95 is nearest-rank, so a 100-frame window
+    reports the 95th sorted sample rather than the 96th. That matters twice
+    over, because the median is also the threshold the stall and skip counts
+    are measured against.
+    """
+    window = sorted(frame_times)
+    n = len(window)
+    median = (window[n // 2] if n % 2
+              else (window[n // 2 - 1] + window[n // 2]) / 2.0)
+    mean = sum(window) / n
+    # Anything past 1.5x the median missed a panel refresh; anything under
+    # half of it never reached the panel at all (dirty tracking skipped the
+    # swap, so the frame did not wait for vsync).
+    return {
+        "frames": n,
+        "fps": (1.0 / mean) if mean > 0 else 0.0,
+        "median": median,
+        "p95": window[max(0, math.ceil(0.95 * n) - 1)],
+        "max": window[-1],
+        "min": window[0],
+        "stalls": sum(1 for f in window if f > median * 1.5),
+        "skips": sum(1 for f in window if f < median * 0.5),
+    }
+
+
+def format_frame_stats(frame_times: list) -> str:
+    """The one-line rendering of frame_stats(), in milliseconds."""
+    s = frame_stats(frame_times)
+    n = s["frames"]
+    return (
+        f"{s['fps']:.1f} fps over {n} frames | "
+        f"median {s['median'] * 1000:.2f}ms p95 {s['p95'] * 1000:.2f}ms "
+        f"max {s['max'] * 1000:.2f}ms min {s['min'] * 1000:.2f}ms | "
+        f"stalls {s['stalls']} ({100.0 * s['stalls'] / n:.1f}%) "
+        f"skips {s['skips']} ({100.0 * s['skips'] / n:.1f}%)"
+    )
+
+
 class ScrollHelper:
     """
     Helper class for scrolling text and image content on LED displays.
@@ -75,8 +119,19 @@ class ScrollHelper:
         # Pre-allocated buffer for output frame (reused to avoid allocations)
         self._frame_buffer: Optional[np.ndarray] = None
         
-        # Sub-pixel scrolling settings (disabled - using high FPS integer scrolling instead)
-        self.sub_pixel_scrolling = False  # Disabled - use high frame rate for smoothness
+        # Sub-pixel scrolling: OFF by default, and that is deliberate.
+        # Blending renders a half-step by mixing two adjacent columns 50/50.
+        # On a high-resolution screen that reads as smooth motion; on a coarse
+        # LED matrix showing pixel-font text it does not. A one-pixel stroke
+        # becomes two half-brightness pixels, so frames alternate between crisp
+        # and smeared and the text appears to shimmer and jump a pixel ahead --
+        # tested on a 2x128x64 panel and clearly worse than integer stepping.
+        #
+        # The rule this display obeys: motion is smooth when it advances a
+        # whole number of pixels per refresh. Anything slower must either
+        # blend (blur) or repeat frames (judder); blending is the worse of the
+        # two here. Vegas mode still opts in via set_sub_pixel_scrolling().
+        self.sub_pixel_scrolling = False
         self._last_integer_position = 0  # Cache for integer position to avoid repeated calculations
         
         # Frame-based scrolling settings
@@ -105,6 +160,9 @@ class ScrollHelper:
         self.last_frame_time = time.time()
         self.last_fps_log_time = time.time()
         self.frame_times = []
+        # Every frame time since the last stats line, so the 5s summary can
+        # report the tail rather than one arbitrary sample. Cleared on log.
+        self._window: list = []
         
         # Scrolling state management
         self.is_scrolling = False
@@ -244,19 +302,31 @@ class ScrollHelper:
             if self.last_step_time == 0.0:
                 self.last_step_time = current_time
             
-            # Check if scroll_delay has passed
-            time_since_last_step = current_time - self.last_step_time
-            if time_since_last_step >= self.scroll_delay:
-                # Move pixels (can move multiple steps if lag occurred, but cap to prevent huge jumps)
-                steps = int(time_since_last_step / self.scroll_delay)
-                # Cap at reasonable number to prevent huge jumps from lag
-                max_steps = max(1, int(0.04 / self.scroll_delay))  # Limit to 0.04s (2 steps at 50 FPS) for smoother scrolling
-                steps = min(steps, max_steps)
-                pixels_to_move = self.scroll_speed * steps
-                # Update last_step_time, preserving fractional delay for smooth timing
-                self.last_step_time = current_time - (time_since_last_step % self.scroll_delay)
+            # Frame-based mode advances by elapsed time, exactly like the
+            # time-based branch below, at the same configured speed
+            # (scroll_speed px per scroll_delay seconds).
+            #
+            # It used to step discretely: 0, 1 or 2 whole pixels depending on
+            # whether a wall clock had passed scroll_delay. Plugins set
+            # scroll_delay to the target frame period, so that comparison sits
+            # exactly on its own threshold and the decision flips on sub-
+            # millisecond jitter -- a frame a hair early moved nothing and
+            # rendered an identical frame, a frame a hair late moved two
+            # pixels. Rounding the step count fixed the stalls but still
+            # discarded the remainder, so the error never corrected.
+            #
+            # Accumulating elapsed time keeps position exactly proportional to
+            # real time: jitter shifts a pixel boundary by a fraction of a
+            # frame instead of flipping a whole step, and nothing is lost or
+            # gained. This is what the one visibly smooth scroller on the
+            # hardware (the stock ticker) was already doing by virtue of never
+            # enabling frame-based mode.
+            if self.scroll_delay > 0:
+                pixels_per_second = self.scroll_speed / self.scroll_delay
             else:
-                pixels_to_move = 0.0
+                pixels_per_second = self.scroll_speed * 100.0
+            pixels_to_move = pixels_per_second * delta_time
+            self.last_step_time = current_time
         else:
             # Time-based: move based on time delta (correct speed over time)
             # scroll_speed is pixels per second
@@ -1017,18 +1087,25 @@ class ScrollHelper:
         # Keep only last 100 frames for average
         if len(self.frame_times) > 100:
             self.frame_times.pop(0)
+
+        # Every frame since the last log, not just the last 100 and not just
+        # the one that happens to land on the 5s boundary. The old line
+        # reported a single instantaneous sample -- roughly 1 frame in 500 --
+        # which cannot see a stall that hits 1% of frames, and reported it
+        # next to an average that hides the same stall by construction (a 2ms
+        # duplicate and a 21ms double-wait mean exactly 10ms). Chasing scroll
+        # judder needs the tail, so keep the window and report percentiles.
+        self._window.append(frame_time)
         
         # Log FPS every 5 seconds to avoid spam
         if current_time - self.last_fps_log_time >= 5.0:
-            avg_frame_time = sum(self.frame_times) / len(self.frame_times)
-            avg_fps = 1.0 / avg_frame_time if avg_frame_time > 0 else 0
-            instant_fps = 1.0 / frame_time if frame_time > 0 else 0
-            
-            self.logger.info(f"Scroll frame stats - Avg FPS: {avg_fps:.1f}, "
-                           f"Current FPS: {instant_fps:.1f}, "
-                           f"Frame time: {frame_time*1000:.2f}ms")
+            self.logger.info(
+                "Scroll frame stats - %s",
+                format_frame_stats(self._window or [frame_time]),
+            )
             self.last_fps_log_time = current_time
             self.frame_count = 0
+            self._window = []
         
         self.last_frame_time = current_time
         self.frame_count += 1

src/display_controller.py

@@ -2405,6 +2405,7 @@ class DisplayController:
                             )
 
                             while True:
+                                _frame_start = time.perf_counter()
                                 try:
                                     with self._display_lock_or_skip(plugin_id) as can_display:
                                         if can_display:
@@ -2425,11 +2426,26 @@ class DisplayController:
                                 # Multi-display sync: send follower frame after each render
                                 self._send_follower_frame(manager_to_display)
 
-                                time.sleep(display_interval)
                                 self._tick_plugin_updates()
                                 self._poll_on_demand_requests()
                                 self._check_on_demand_expiration()
 
+                                # Pace to the frame deadline rather than sleeping a flat
+                                # interval on top of the work. display() has already
+                                # blocked on the panel's vsync by this point, so an
+                                # unconditional sleep is added to a wait that already
+                                # happened. Measured on a 2x128x64 chain at
+                                # limit_refresh_rate_hz=100: ~4ms of render plus a flat
+                                # 8ms put each iteration at ~12ms against a 10ms refresh
+                                # grid, so every swap missed a refresh and the loop
+                                # settled at 50fps where display_interval asks for 125 --
+                                # and with zero headroom, ~14% of frames slipped a
+                                # further refresh, which is what reads as scroll stutter.
+                                _remaining = display_interval - (time.perf_counter() - _frame_start)
+                                # Yield even when the frame overran its budget, so plugin
+                                # update threads and the web UI are not starved of the GIL.
+                                time.sleep(_remaining if _remaining > 0 else 0.001)
+
                                 if self.current_display_mode != active_mode:
                                     logger.debug("Mode changed during high-FPS loop, breaking early")
                                     break

src/display_manager.py

@@ -240,6 +240,14 @@ class DisplayManager:
         self._update_lock = threading.RLock()
         
         # Scrolling state tracking for graceful updates
+        # How many panel refreshes each pushed frame is held for. 1 means a new
+        # frame every refresh. Higher values are how a scroll runs slower than
+        # one pixel per refresh WITHOUT fractional pixel positions: the panel
+        # keeps refreshing at full rate (so flicker is unchanged) but motion
+        # advances a whole pixel every Nth refresh instead of every one.
+        # See src/common/scroll_config.py and scripts/scroll_speeds.py.
+        self._frame_hold = 1
+
         self._scrolling_state = {
             'is_scrolling': False,
             'last_scroll_activity': 0,
@@ -768,16 +776,33 @@ class DisplayManager:
                     return  # Skip hardware write — content is being captured off-screen
 
                 digest = None
+                frame_checksum = None
                 if self._dirty_tracking_enabled:
                     try:
                         brightness = getattr(self.matrix, 'brightness', None)
                     except AttributeError:
                         brightness = None
-                    digest = (zlib.adler32(self.image.tobytes()), brightness)
-                    if digest == self._last_pushed_digest:
+                    frame_checksum = zlib.adler32(self.image.tobytes())
+                    digest = (frame_checksum, brightness)
+                    if digest == self._last_pushed_digest and not self.is_currently_scrolling():
                         # Nothing changed since the last push — the panel is
                         # already showing exactly this frame.
-                        self._write_snapshot_if_due()
+                        #
+                        # Never taken mid-scroll, and that exception is the
+                        # point. SwapOnVSync is what paces the render loop, so
+                        # skipping it also skips the wait: a duplicate frame
+                        # returns in ~8ms instead of ~10ms on a 100Hz panel,
+                        # advances only 0.8px instead of 1.0px, and so makes
+                        # the *next* frame more likely to repeat as well. That
+                        # is self-sustaining -- measured at ~20% duplicate
+                        # frames mid-scroll on the odds ticker, against
+                        # essentially zero on a lighter plugin with identical
+                        # scroll settings. Swapping an identical frame costs
+                        # one canvas copy and keeps the loop locked to the
+                        # panel; falling out of that lock costs smooth motion.
+                        # Static content is unaffected: is_currently_scrolling()
+                        # expires on its own inactivity threshold.
+                        self._write_snapshot_if_due(frame_checksum)
                         return
 
                 # Copy the current image to the offscreen canvas. In double-sided
@@ -787,8 +812,10 @@ class DisplayManager:
                 else:
                     self.offscreen_canvas.SetImage(self.image)
 
-                # Swap buffers immediately
-                self.matrix.SwapOnVSync(self.offscreen_canvas)
+                # Swap buffers immediately. framerate_fraction holds the frame
+                # for N refreshes; SwapOnVSync blocks for all of them, which is
+                # what paces the render loop to the chosen frame rate.
+                self.matrix.SwapOnVSync(self.offscreen_canvas, self._frame_hold)
 
                 # Swap our canvas references
                 self.offscreen_canvas, self.current_canvas = self.current_canvas, self.offscreen_canvas
@@ -796,7 +823,7 @@ class DisplayManager:
                 self._last_pushed_digest = digest
 
                 # Write a snapshot for the web preview (throttled)
-                self._write_snapshot_if_due()
+                self._write_snapshot_if_due(frame_checksum)
         except Exception as e:
             logger.error(f"Error updating display: {e}")
 
@@ -1274,12 +1301,65 @@ class DisplayManager:
         
         return dt.strftime(f"%b %-d{suffix}") 
 
-    def set_scrolling_state(self, is_scrolling: bool):
-        """Set the current scrolling state. Call this when a display starts/stops scrolling."""
+    @property
+    def refresh_hz(self) -> float:
+        """The panel's refresh rate in Hz, from the hardware config.
+
+        The authoritative place to ask, because a plugin only receives its own
+        config section and cannot see display.hardware. Scroll pacing needs
+        this: the speeds a panel can show in whole pixels are refresh_hz
+        divided by the frame hold, so getting it wrong silently produces
+        fractional-pixel motion. See src/common/scroll_config.py.
+
+        Note this is the configured *cap*, not necessarily what the panel
+        achieves -- scripts/scroll_speeds.py --measure reports the real rate.
+        """
+        hardware = (self.config.get('display') or {}).get('hardware') or {}
+        try:
+            value = float(hardware.get('limit_refresh_rate_hz') or 0)
+        except (TypeError, ValueError):
+            value = 0.0
+        return value if value > 0 else 100.0
+
+    def set_frame_hold(self, refreshes: int) -> None:
+        """Hold each pushed frame for this many panel refreshes (>=1).
+
+        Set by the scroll configuration so a plugin can run at, say, 50px/s on
+        a 100Hz panel as one whole pixel every second refresh, rather than half
+        a pixel every refresh (which has to be blended or repeated unevenly).
+
+        Reset to 1 whenever scrolling stops, so one plugin's pacing cannot
+        leak into the next thing on screen.
+        """
+        try:
+            value = int(refreshes)
+        except (TypeError, ValueError):
+            logger.warning("Ignoring unusable frame hold: %r", refreshes)
+            return
+        self._frame_hold = max(1, min(255, value))
+
+    def set_scrolling_state(self, is_scrolling: bool, frame_hold: int = 1):
+        """Set the current scrolling state, and this scroll's frame pacing.
+
+        Call this when a display starts or stops scrolling. ``frame_hold`` is
+        how many panel refreshes each frame is held for -- 2 gives one whole
+        pixel every second refresh, which is how a scroll runs at half the
+        refresh rate without fractional pixel positions.
+
+        The hold is set here rather than once at plugin construction because
+        it must not outlive the scroll that asked for it: plugins share one
+        display manager, so a hold left set by whoever scrolled last would
+        silently re-pace the next plugin. Passing it alongside the state makes
+        the lifetime exactly the scroll, and the default of 1 means any caller
+        that does not care gets a new frame every refresh.
+        """
         current_time = time.time()
         self._scrolling_state['is_scrolling'] = is_scrolling
         if is_scrolling:
             self._scrolling_state['last_scroll_activity'] = current_time
+            self.set_frame_hold(frame_hold)
+        else:
+            self._frame_hold = 1
         logger.debug(f"Scrolling state set to: {is_scrolling}")
 
     def is_currently_scrolling(self) -> bool:
@@ -1410,11 +1490,20 @@ class DisplayManager:
                 self._viewer_fresh = False
         return self._viewer_fresh
 
-    def _write_snapshot_if_due(self) -> None:
+    def _write_snapshot_if_due(self, frame_checksum: Optional[int] = None) -> None:
         """Mirror the current frame to the preview snapshot when the policy
         says it's worth it — see src/common/snapshot_policy.py. Unchanged
         frames are never re-encoded; without viewers the cadence drops to
-        the idle keepalive."""
+        the idle keepalive.
+
+        Args:
+            frame_checksum: adler32 of the current frame, when the caller has
+                already computed one. Dirty tracking checksums every frame a
+                few lines above the call site, and re-deriving it here meant a
+                second tobytes() plus a second pass over the whole framebuffer
+                on every single frame — ~0.17ms per frame of the two combined
+                at 256x64, paid 100 times a second to reach the same number.
+        """
         try:
             now = time.time()
             viewer_fresh = self._viewer_is_fresh(now)
@@ -1424,7 +1513,8 @@ class DisplayManager:
                 self._last_snapshot_ts = 0.0
             self._viewer_was_fresh = viewer_fresh
 
-            digest = zlib.adler32(self.image.tobytes())
+            digest = (frame_checksum if frame_checksum is not None
+                      else zlib.adler32(self.image.tobytes()))
             action = snapshot_policy.decide(
                 now, self._last_snapshot_ts, self._last_snapshot_touch_ts,
                 viewer_fresh, digest != self._last_snapshot_digest)

src/plugin_system/testing/visual_display_manager.py

@@ -503,9 +503,18 @@ class VisualTestDisplayManager:
     # Scrolling state (no-op interface compat)
     # ------------------------------------------------------------------
 
-    def set_scrolling_state(self, is_scrolling: bool):
-        """Set the current scrolling state (no-op for testing)."""
+    def set_scrolling_state(self, is_scrolling: bool, frame_hold: int = 1):
+        """Set the current scrolling state (no-op for testing).
+
+        ``frame_hold`` mirrors the DisplayManager signature this change adds.
+        The two are kept in step deliberately: a double that accepts arguments
+        production does not lets a call pass every harness run and then raise
+        TypeError on the panel, and a double that lacks one production has
+        fails every render of a plugin that legitimately paces its scroll.
+        Plugins begin passing it in ledmatrix-plugins#462.
+        """
         self._scrolling_state['is_scrolling'] = is_scrolling
+        self._scrolling_state['frame_hold'] = frame_hold
         if is_scrolling:
             self._scrolling_state['last_scroll_activity'] = time.time()
 

test/test_cache_nonfinite_floats.py

@@ -0,0 +1,158 @@
+"""Non-finite floats round-trip identically with and without orjson.
+
+JSON has no NaN or Infinity. The stdlib emits them anyway as an extension;
+orjson refuses to and writes null. Cache files outlive the decision of which
+encoder is installed, so both halves of that gap are pinned here:
+
+  * writing -- installing orjson must not silently change what gets cached,
+    so the stdlib path writes null too;
+  * reading -- records already on disk carrying NaN or Infinity must stay
+    readable, or installing orjson turns each of them into a "corrupted cache
+    file" that DiskCache.get logs as an error and deletes.
+"""
+
+import importlib
+import json
+import math
+import sys
+from pathlib import Path
+from unittest import mock
+
+import pytest
+
+sys.path.insert(0, str(Path(__file__).resolve().parent.parent))
+
+from src.cache import disk_cache as disk_cache_module  # noqa: E402
+
+
+@pytest.fixture
+def stdlib_cache():
+    """The module as it loads on a host with no orjson wheel."""
+    # A None entry in sys.modules makes `import orjson` raise ImportError,
+    # which is the branch we want, whether or not orjson is really installed.
+    with mock.patch.dict(sys.modules, {"orjson": None}):
+        module = importlib.reload(disk_cache_module)
+        assert module.orjson is None
+        yield module
+    importlib.reload(disk_cache_module)
+
+
+@pytest.fixture
+def orjson_cache():
+    """The module as it loads with orjson available."""
+    module = importlib.reload(disk_cache_module)
+    if module.orjson is None:
+        pytest.skip("orjson is not installed on this host")
+    return module
+
+
+NON_FINITE = {"nan": float("nan"), "inf": float("inf"), "ninf": float("-inf")}
+
+
+def _reject_constant(name):
+    """Make json.loads as strict as orjson about NaN/Infinity tokens."""
+    raise AssertionError(f"non-spec JSON constant in output: {name}")
+
+
+class TestWritePolicy:
+    """Non-finite floats become null on whichever encoder is in use."""
+
+    def _assert_nulled(self, module):
+        record = module._loads(module._dumps(dict(NON_FINITE, ok=1.5)))
+        assert record["nan"] is None
+        assert record["inf"] is None
+        assert record["ninf"] is None
+        # Finite values are untouched.
+        assert record["ok"] == 1.5
+
+    def test_stdlib_writes_null(self, stdlib_cache):
+        self._assert_nulled(stdlib_cache)
+
+    def test_orjson_writes_null(self, orjson_cache):
+        self._assert_nulled(orjson_cache)
+
+    def test_stdlib_emits_spec_compliant_json(self, stdlib_cache):
+        # The point of the write half: bytes written without orjson must still
+        # parse once orjson is installed later. json.loads accepts the
+        # extension tokens, so it cannot show this -- assert on the bytes, and
+        # on a strict reader when there is one.
+        raw = stdlib_cache._dumps(dict(NON_FINITE))
+        assert b"NaN" not in raw
+        assert b"Infinity" not in raw
+        assert json.loads(raw, parse_constant=_reject_constant) == {
+            "nan": None, "inf": None, "ninf": None}
+
+    def test_nested_non_finite_are_replaced(self, stdlib_cache):
+        data = {"a": [1.0, float("nan"), {"b": float("inf")}], "c": (float("-inf"),)}
+        record = stdlib_cache._loads(stdlib_cache._dumps(data))
+        assert record["a"] == [1.0, None, {"b": None}]
+        assert record["c"] == [None]
+
+    def test_finite_payloads_are_byte_identical_to_the_old_encoder(self, stdlib_cache):
+        # allow_nan=False must not change ordinary output.
+        data = {"x": 1, "y": [1.5, "s", True, None], "z": {"k": 2.25}}
+        assert stdlib_cache._dumps(data) == json.dumps(
+            data, cls=stdlib_cache.DateTimeEncoder).encode("utf-8")
+
+
+class TestReplaceNonFinite:
+    def test_leaves_ordinary_values_alone(self):
+        for value in (1, 1.5, "s", True, None, [], {}):
+            assert disk_cache_module._replace_nonfinite(value) == value
+
+    def test_replaces_every_non_finite_float(self):
+        for value in NON_FINITE.values():
+            assert disk_cache_module._replace_nonfinite(value) is None
+
+    def test_walks_nested_containers(self):
+        assert disk_cache_module._replace_nonfinite(
+            {"a": [{"b": float("nan")}]}) == {"a": [{"b": None}]}
+
+
+class TestLegacyRecordsStayReadable:
+    """Files written before orjson arrived still load."""
+
+    LEGACY = b'{"timestamp": 1000.0, "value": NaN, "other": Infinity}'
+
+    def test_stdlib_reads_legacy_tokens(self, stdlib_cache):
+        record = stdlib_cache._loads(self.LEGACY)
+        assert math.isnan(record["value"])
+        assert math.isinf(record["other"])
+
+    def test_orjson_falls_back_for_legacy_tokens(self, orjson_cache):
+        record = orjson_cache._loads(self.LEGACY)
+        assert math.isnan(record["value"])
+        assert math.isinf(record["other"])
+
+    def test_genuinely_malformed_files_still_raise(self, stdlib_cache):
+        with pytest.raises(json.JSONDecodeError):
+            stdlib_cache._loads(b'{"a": ')
+
+    def test_orjson_still_raises_for_malformed_files(self, orjson_cache):
+        with pytest.raises(json.JSONDecodeError):
+            orjson_cache._loads(b'{"a": ')
+
+
+class TestDiskCacheEndToEnd:
+    def _cache(self, module, tmp_path):
+        return module.DiskCache(str(tmp_path))
+
+    def test_legacy_file_is_not_deleted_as_corrupt(self, orjson_cache, tmp_path):
+        cache = self._cache(orjson_cache, tmp_path)
+        path = cache.get_cache_path("legacy")
+        Path(path).write_bytes(
+            b'{"timestamp": %d, "value": NaN}' % int(__import__("time").time()))
+
+        record = cache.get("legacy", max_age=None)
+
+        assert record is not None, "legacy NaN record was treated as corrupt"
+        assert math.isnan(record["value"])
+        assert Path(path).exists(), "legacy NaN record was deleted"
+
+    def test_round_trip_through_set_and_get(self, stdlib_cache, tmp_path):
+        cache = self._cache(stdlib_cache, tmp_path)
+        cache.set("k", {"timestamp": __import__("time").time(),
+                        "value": float("nan")})
+        record = cache.get("k", max_age=None)
+        assert record is not None
+        assert record["value"] is None

test/test_display_dirty_tracking.py

@@ -47,12 +47,17 @@ class _SwapSpy:
     def __init__(self, matrix):
         self.matrix = matrix
         self.count = 0
+        self.last_frame_hold = None
         self._orig = matrix.SwapOnVSync
 
     def __enter__(self):
-        def counting(canvas):
+        def counting(canvas, *args):
+            # *args carries framerate_fraction, which display_manager passes so
+            # a frame can be held for several refreshes. Signature must match
+            # the real binding or the spy hides a TypeError as a failed push.
             self.count += 1
-            return self._orig(canvas)
+            self.last_frame_hold = args[0] if args else 1
+            return self._orig(canvas, *args)
         self.matrix.SwapOnVSync = counting
         return self
 
@@ -109,6 +114,11 @@ class TestDirtyTracking:
         dm.draw.rectangle([0, 0, 30, 8], fill=(255, 255, 0))
         dm.update_display()   # push + snapshot write (first frame)
         assert os.path.exists(dm._snapshot_path)
+        # Backdate the file so the "was it bumped?" check below cannot be
+        # defeated by filesystem mtime granularity -- on Windows two writes in
+        # the same tick get identical timestamps, which made this test fail
+        # roughly two runs in three regardless of the code under test.
+        os.utime(dm._snapshot_path, (time.time() - 60, time.time() - 60))
         first_mtime = os.path.getmtime(dm._snapshot_path)
 
         # Age the write/touch bookkeeping past TOUCH_INTERVAL so the next
@@ -125,6 +135,62 @@ class TestDirtyTracking:
         assert os.path.getmtime(dm._snapshot_path) > first_mtime
 
 
+class TestScrollLock:
+    """Dirty tracking must not skip the panel push while a scroll is running.
+
+    SwapOnVSync is what paces the render loop, so skipping it also skips the
+    wait for the panel. A duplicate frame therefore returns early -- ~8ms
+    instead of ~10ms on a 100Hz panel -- which advances the strip only 0.8px
+    instead of 1.0px, which makes the NEXT frame more likely to be a duplicate
+    too. That is self-sustaining: measured at ~20% duplicate frames mid-scroll
+    on the odds ticker against essentially zero on a lighter plugin with
+    identical scroll settings. Pushing an identical frame costs one canvas
+    copy; falling out of vsync lock costs smooth motion.
+    """
+
+    def test_identical_frames_still_push_while_scrolling(self, dm):
+        dm.draw.rectangle([0, 0, 12, 12], fill=(0, 0, 255))
+        dm.update_display()
+        dm.set_scrolling_state(True)
+        try:
+            with _SwapSpy(dm.matrix) as spy:
+                dm.update_display()
+                dm.update_display()
+                dm.update_display()
+            assert spy.count == 3, "scrolling must stay locked to the panel"
+        finally:
+            dm.set_scrolling_state(False)
+
+    def test_identical_frames_are_skipped_when_not_scrolling(self, dm):
+        """The optimisation still applies to static content."""
+        dm.set_scrolling_state(False)
+        dm.draw.rectangle([0, 0, 14, 14], fill=(255, 0, 255))
+        dm.update_display()
+        with _SwapSpy(dm.matrix) as spy:
+            dm.update_display()
+            dm.update_display()
+        assert spy.count == 0
+
+    def test_stale_scrolling_state_stops_forcing_pushes(self, dm):
+        """A plugin that stops scrolling without saying so must not pin the
+        panel into always-push forever. is_currently_scrolling() expires on
+        its own inactivity threshold, and the skip has to come back with it."""
+        dm.draw.rectangle([0, 0, 16, 16], fill=(0, 255, 255))
+        dm.update_display()
+        dm.set_scrolling_state(True)
+        try:
+            # Backdate the activity marker past the inactivity threshold.
+            dm._scrolling_state['last_scroll_activity'] = (
+                time.time() - dm._scrolling_state['scroll_inactivity_threshold'] - 1.0)
+            assert dm.is_currently_scrolling() is False
+            with _SwapSpy(dm.matrix) as spy:
+                dm.update_display()
+                dm.update_display()
+            assert spy.count == 0
+        finally:
+            dm.set_scrolling_state(False)
+
+
 class TestKillSwitch:
     def test_dirty_tracking_can_be_disabled(self, dm):
         dm._dirty_tracking_enabled = False
@@ -160,3 +226,89 @@ class TestKillSwitch:
 
 if __name__ == "__main__":
     sys.exit(pytest.main([__file__, "-v"]))
+
+
+class TestFrameHold:
+    """Holding a frame for N refreshes is how a scroll runs slower than one
+    pixel per refresh without fractional pixel positions."""
+
+    def test_hold_reaches_swap_on_vsync(self, dm):
+        dm.set_scrolling_state(True)
+        dm.set_frame_hold(3)
+        try:
+            dm.draw.rectangle([0, 0, 9, 9], fill=(120, 0, 200))
+            with _SwapSpy(dm.matrix) as spy:
+                dm.update_display()
+            assert spy.count == 1
+            assert spy.last_frame_hold == 3
+        finally:
+            dm.set_scrolling_state(False)
+
+    def test_default_is_every_refresh(self, dm):
+        dm.set_scrolling_state(True)
+        try:
+            dm.draw.rectangle([0, 0, 11, 11], fill=(0, 120, 200))
+            with _SwapSpy(dm.matrix) as spy:
+                dm.update_display()
+            assert spy.last_frame_hold == 1
+        finally:
+            dm.set_scrolling_state(False)
+
+    def test_hold_resets_when_scrolling_stops(self, dm):
+        """One plugin's pacing must not leak into whatever is on screen next."""
+        dm.set_scrolling_state(True)
+        dm.set_frame_hold(5)
+        dm.set_scrolling_state(False)
+        dm.set_scrolling_state(True)
+        try:
+            dm.draw.rectangle([0, 0, 13, 13], fill=(200, 120, 0))
+            with _SwapSpy(dm.matrix) as spy:
+                dm.update_display()
+            assert spy.last_frame_hold == 1
+        finally:
+            dm.set_scrolling_state(False)
+
+    @pytest.mark.parametrize("bad,expected", [(0, 1), (-4, 1), (None, 1), ("x", 1)])
+    def test_unusable_holds_are_ignored_or_floored(self, dm, bad, expected):
+        dm.set_frame_hold(bad)
+        assert dm._frame_hold == expected
+
+
+class TestFrameHoldLifetime:
+    """The hold must last exactly as long as the scroll that asked for it.
+
+    Plugins share one display manager. A hold applied at plugin construction is
+    wiped the moment any *other* plugin finishes scrolling, so by the time the
+    first plugin renders it is back to one pixel per refresh -- the speed reads
+    correct in the log and is wrong on the panel.
+    """
+
+    def test_scrolling_state_carries_the_hold(self, dm):
+        dm.set_scrolling_state(True, frame_hold=4)
+        try:
+            dm.draw.rectangle([0, 0, 7, 7], fill=(10, 200, 10))
+            with _SwapSpy(dm.matrix) as spy:
+                dm.update_display()
+            assert spy.last_frame_hold == 4
+        finally:
+            dm.set_scrolling_state(False)
+
+    def test_another_plugin_stopping_does_not_strand_a_hold(self, dm):
+        dm.set_scrolling_state(True, frame_hold=3)
+        dm.set_scrolling_state(False)          # some other plugin finishes
+        dm.set_scrolling_state(True)           # a plugin that wants no hold
+        try:
+            dm.draw.rectangle([0, 0, 6, 6], fill=(200, 10, 10))
+            with _SwapSpy(dm.matrix) as spy:
+                dm.update_display()
+            assert spy.last_frame_hold == 1
+        finally:
+            dm.set_scrolling_state(False)
+
+    def test_default_keeps_previous_behaviour(self, dm):
+        """Callers that never heard of frame holds get one frame per refresh."""
+        dm.set_scrolling_state(True)
+        try:
+            assert dm._frame_hold == 1
+        finally:
+            dm.set_scrolling_state(False)

test/test_display_double_parity.py

@@ -0,0 +1,69 @@
+"""The visual-test double must not drift from the real DisplayManager.
+
+Both directions of drift are silent and both are damaging:
+
+  * the double accepts an argument production does not -- the call passes
+    every harness run and raises TypeError on the panel, which is precisely
+    the failure a safety harness exists to prevent;
+  * the double lacks an argument production has -- every plugin that
+    legitimately uses it fails every render, and the harness blames the
+    plugin.
+
+`set_scrolling_state` has been each of those in turn across two branches, so
+the parity is asserted rather than remembered.
+
+Read with ast rather than imported: src/display_manager.py imports rgbmatrix
+at module scope, which is absent anywhere without the panel library, and this
+check should hold on a laptop and in CI as well as on a Pi.
+"""
+
+import ast
+from pathlib import Path
+
+import pytest
+
+ROOT = Path(__file__).resolve().parent.parent
+REAL = ROOT / "src" / "display_manager.py"
+DOUBLE = ROOT / "src" / "plugin_system" / "testing" / "visual_display_manager.py"
+
+#: Methods a plugin calls on whichever manager it is handed.
+SHARED_METHODS = ["set_scrolling_state", "is_currently_scrolling"]
+
+
+def _signature(path: Path, class_hint: str, method: str):
+    """(name, default-repr) pairs for `method`, or None if it is absent."""
+    tree = ast.parse(path.read_text(encoding="utf-8"))
+    for node in ast.walk(tree):
+        if not isinstance(node, ast.ClassDef) or class_hint not in node.name:
+            continue
+        for item in node.body:
+            if isinstance(item, (ast.FunctionDef, ast.AsyncFunctionDef)) and item.name == method:
+                args = [a.arg for a in item.args.args if a.arg != "self"]
+                pad = [None] * (len(args) - len(item.args.defaults))
+                defaults = pad + [ast.unparse(d) for d in item.args.defaults]
+                return list(zip(args, defaults))
+    return None
+
+
+@pytest.mark.parametrize("method", SHARED_METHODS)
+def test_the_double_matches_production(method):
+    real = _signature(REAL, "DisplayManager", method)
+    double = _signature(DOUBLE, "DisplayManager", method)
+
+    assert real is not None, f"DisplayManager lost {method}"
+    assert double is not None, \
+        f"the test double is missing {method}, so every plugin using it fails to render"
+    assert double == real, (
+        f"{method} has drifted: production takes {real}, the double takes {double}. "
+        "A double that is more permissive hides a production TypeError; one that "
+        "is less permissive fails plugins that are actually correct."
+    )
+
+
+def test_frame_hold_is_accepted_by_both():
+    """The specific argument that has drifted twice."""
+    for path, label in ((REAL, "DisplayManager"), (DOUBLE, "VisualTestDisplayManager")):
+        params = dict(_signature(path, "DisplayManager", "set_scrolling_state") or [])
+        assert "frame_hold" in params, f"{label} does not accept frame_hold"
+        assert params["frame_hold"] == "1", \
+            f"{label} must default frame_hold to 1 so existing callers are unaffected"

test/test_scroll_config.py

@@ -0,0 +1,467 @@
+"""Tests for the shared scroll configuration resolver."""
+
+import logging
+import sys
+from pathlib import Path
+
+import pytest
+
+sys.path.insert(0, str(Path(__file__).resolve().parent.parent))
+
+from src.common.scroll_config import (  # noqa: E402
+    DEFAULT_PIXELS_PER_SECOND,
+    MAX_PIXELS_PER_FRAME,
+    crisp_ladder,
+    solve_crisp,
+    MAX_PIXELS_PER_SECOND,
+    MIN_PIXELS_PER_SECOND,
+    ScrollSettings,
+    configure,
+    refresh_hz_from_config,
+    resolve,
+)
+
+
+class FakeHelper:
+    """Records what configure() applied."""
+
+    def __init__(self, with_optional=True):
+        self.speed = None
+        self.frame_based = None
+        self.target_fps = None
+        if not with_optional:
+            del FakeHelper.set_frame_based_scrolling
+            del FakeHelper.set_target_fps
+
+    def set_scroll_speed(self, speed):
+        self.speed = speed
+
+    def set_frame_based_scrolling(self, enabled):
+        self.frame_based = enabled
+
+    def set_target_fps(self, fps):
+        self.target_fps = fps
+
+
+class FakeDisplayManager:
+    """Records the frame hold configure() applies."""
+
+    def __init__(self):
+        self.hold = None
+
+    def set_frame_hold(self, refreshes):
+        self.hold = refreshes
+
+
+class MinimalHelper:
+    """An older helper exposing only set_scroll_speed."""
+
+    def __init__(self):
+        self.speed = None
+
+    def set_scroll_speed(self, speed):
+        self.speed = speed
+
+
+class TestPrecedence:
+    def test_display_options_pair_wins(self):
+        s = resolve({"display_options": {"scroll_speed": 1.0, "scroll_delay": 0.01}})
+        assert s.pixels_per_second == 100.0
+        assert s.source == "display_options.scroll_speed/delay"
+
+    def test_display_block_used_when_options_absent(self):
+        s = resolve({"display": {"scroll_speed": 2.0, "scroll_delay": 0.01}})
+        assert s.pixels_per_second == 200.0
+        assert s.source == "display.scroll_speed/delay"
+
+    def test_root_pair_used_when_both_blocks_absent(self):
+        s = resolve({"scroll_speed": 1.0, "scroll_delay": 0.02})
+        assert s.pixels_per_second == 50.0
+
+    def test_pixels_per_second_used_when_no_pair_given(self):
+        s = resolve({"display_options": {"scroll_pixels_per_second": 120.0}})
+        assert s.pixels_per_second == 120.0
+        assert s.source == "display_options.scroll_pixels_per_second"
+
+    def test_global_display_is_the_last_resort_before_default(self):
+        s = resolve({}, {"display": {"scroll_speed": 1.0, "scroll_delay": 0.005}})
+        assert s.pixels_per_second == 200.0
+
+    def test_default_when_nothing_configured(self):
+        s = resolve({}, {})
+        assert s.pixels_per_second == DEFAULT_PIXELS_PER_SECOND
+        assert s.source == "default"
+
+
+class TestDeprecatedKeyCannotOverrideExplicitPair:
+    """Regression for ChuckBuilds/ledmatrix-plugins#408.
+
+    odds-ticker ranked scroll_pixels_per_second above the documented
+    scroll_speed/scroll_delay pair. Because that key carries a schema default,
+    the documented settings became unreachable for every user and the ticker
+    silently ran at the default speed. The pair must win.
+    """
+
+    def test_pair_beats_pixels_per_second_in_the_same_block(self):
+        s = resolve(
+            {
+                "display_options": {
+                    "scroll_speed": 1.0,
+                    "scroll_delay": 0.01,
+                    "scroll_pixels_per_second": 50.0,  # schema default
+                }
+            }
+        )
+        assert s.pixels_per_second == 100.0
+        assert "scroll_speed/delay" in s.source
+
+    def test_pair_beats_pixels_per_second_in_an_outer_block(self):
+        s = resolve(
+            {
+                "display_options": {"scroll_speed": 1.0, "scroll_delay": 0.01},
+                "scroll_pixels_per_second": 50.0,
+            }
+        )
+        assert s.pixels_per_second == 100.0
+
+
+class TestMalformedValues:
+    @pytest.mark.parametrize("bad", [None, "fast", "", {}, [], float("nan")])
+    def test_unusable_speed_falls_through(self, bad):
+        s = resolve({"display_options": {"scroll_speed": bad, "scroll_delay": 0.01}})
+        assert s.pixels_per_second == DEFAULT_PIXELS_PER_SECOND
+
+    @pytest.mark.parametrize("bad", [0, -5, 0.0])
+    def test_non_positive_values_fall_through(self, bad):
+        s = resolve({"display_options": {"scroll_pixels_per_second": bad}})
+        assert s.pixels_per_second == DEFAULT_PIXELS_PER_SECOND
+
+    def test_booleans_are_not_treated_as_numbers(self):
+        s = resolve({"display_options": {"scroll_pixels_per_second": True}})
+        assert s.pixels_per_second == DEFAULT_PIXELS_PER_SECOND
+
+    def test_zero_delay_does_not_divide_by_zero(self):
+        s = resolve({"display_options": {"scroll_speed": 1.0, "scroll_delay": 0}})
+        assert s.pixels_per_second == DEFAULT_PIXELS_PER_SECOND
+
+    def test_non_dict_blocks_are_ignored(self):
+        s = resolve({"display_options": "nonsense", "display": 5})
+        assert s.pixels_per_second == DEFAULT_PIXELS_PER_SECOND
+
+
+class TestClamping:
+    def test_absurdly_fast_is_clamped(self):
+        s = resolve({"display_options": {"scroll_pixels_per_second": 100000.0}})
+        assert s.pixels_per_second == MAX_PIXELS_PER_SECOND
+        assert s.warning and "clamped" in s.warning
+
+    def test_absurdly_slow_is_clamped(self):
+        s = resolve({"display_options": {"scroll_pixels_per_second": 0.01}})
+        assert s.pixels_per_second == MIN_PIXELS_PER_SECOND
+
+
+class TestWholePixelWarning:
+    """The display rule: whole pixels per refresh, or it judders."""
+
+    def test_no_warning_when_speed_divides_evenly(self):
+        s = resolve({"display_options": {"scroll_pixels_per_second": 100.0}}, refresh_hz=100)
+        assert s.warning is None
+        assert s.pixels_per_frame == pytest.approx(1.0)
+
+    def test_no_warning_at_an_integer_multiple(self):
+        s = resolve({"display_options": {"scroll_pixels_per_second": 200.0}}, refresh_hz=100)
+        assert s.warning is None
+
+    def test_warns_on_a_half_pixel_per_frame(self):
+        s = resolve({"display_options": {"scroll_pixels_per_second": 50.0}}, refresh_hz=100)
+        assert s.warning is not None
+        assert "judder" in s.warning
+        assert "100 px/s" in s.warning
+
+    def test_warns_on_a_fractional_multiple(self):
+        s = resolve({"display_options": {"scroll_pixels_per_second": 130.0}}, refresh_hz=100)
+        assert s.warning is not None
+
+    def test_respects_a_non_default_refresh_rate(self):
+        s = resolve({"display_options": {"scroll_pixels_per_second": 150.0}}, refresh_hz=150)
+        assert s.warning is None
+        assert s.pixels_per_frame == pytest.approx(1.0)
+
+
+class TestConfigure:
+    def test_applies_time_based_mode_and_speed(self):
+        helper = FakeHelper()
+        s = configure(helper, {"display_options": {"scroll_speed": 1.0, "scroll_delay": 0.01}})
+        assert helper.speed == 100.0
+        assert helper.frame_based is False, "must not use the wall-clock step gate"
+        assert helper.target_fps == 100.0
+        assert s.pixels_per_second == 100.0
+
+    def test_works_against_a_helper_without_optional_methods(self):
+        helper = MinimalHelper()
+        configure(helper, {"display_options": {"scroll_pixels_per_second": 100.0}})
+        assert helper.speed == 100.0
+
+    def test_snapping_removes_the_judder_warning(self, caplog):
+        """50px/s cannot be shown in whole pixels at 100Hz *per refresh*, but it
+        can as 1px every 2nd refresh -- so once snapped there is nothing to warn
+        about. resolve() alone still warns; configure() resolves it."""
+        assert "judder" in resolve(
+            {"display_options": {"scroll_pixels_per_second": 50.0}},
+            refresh_hz=100).warning
+        helper, dm = FakeHelper(), FakeDisplayManager()
+        with caplog.at_level(logging.WARNING):
+            settings = configure(
+                helper, {"display_options": {"scroll_pixels_per_second": 50.0}},
+                display_manager=dm)
+        assert settings.warning is None
+        assert not [r for r in caplog.records if "judder" in r.getMessage()]
+
+    def test_reports_the_hold_without_applying_it(self):
+        """The hold belongs to a scroll, not a plugin's lifetime -- one left set
+        at construction is wiped as soon as any other plugin stops scrolling.
+        configure() reports it; the caller passes it to set_scrolling_state."""
+        dm = FakeDisplayManager()
+        s = configure(FakeHelper(), {"display_options": {"scroll_pixels_per_second": 25.0}},
+                      display_manager=dm)
+        assert s.frame_hold == 4, "25px/s at 100Hz is 1px every 4th refresh"
+        assert dm.hold is None, "must not apply the hold behind the caller's back"
+
+    def test_full_speed_needs_no_hold(self):
+        s = configure(FakeHelper(), {"display_options": {"scroll_pixels_per_second": 100.0}},
+                      display_manager=FakeDisplayManager())
+        assert s.frame_hold == 1
+
+    def test_frame_hold_is_one_when_snapping_is_off(self):
+        s = configure(FakeHelper(), {"display_options": {"scroll_pixels_per_second": 50.0}},
+                      snap_to_crisp=False)
+        assert s.frame_hold == 1
+
+    def test_snaps_to_the_nearest_crisp_speed_and_reports_both(self):
+        s = configure(FakeHelper(), {"display_options": {"scroll_pixels_per_second": 45.0}},
+                      display_manager=FakeDisplayManager())
+        assert s.requested_pixels_per_second == 45.0
+        assert s.pixels_per_second == 50.0
+        assert s.crisp is not None and s.crisp.pixels_per_frame == 1
+
+    def test_snapping_can_be_turned_off(self):
+        s = configure(FakeHelper(), {"display_options": {"scroll_pixels_per_second": 45.0}},
+                      snap_to_crisp=False)
+        assert s.pixels_per_second == 45.0
+        assert s.crisp is None
+
+    def test_helper_target_fps_matches_the_presentation_rate(self):
+        """At a hold of 4 the panel still refreshes at 100Hz, but frames are
+        presented at 25/s -- that is the rate the helper should pace to."""
+        helper = FakeHelper()
+        configure(helper, {"display_options": {"scroll_pixels_per_second": 25.0}},
+                  display_manager=FakeDisplayManager())
+        assert helper.target_fps == pytest.approx(25.0)
+
+    def test_returns_settings_describing_the_source(self):
+        s = configure(FakeHelper(), {"display_options": {"scroll_speed": 2.0, "scroll_delay": 0.01}})
+        assert isinstance(s, ScrollSettings)
+        assert "200.0 px/s" in s.describe()
+
+
+class TestRefreshFromConfig:
+    def test_reads_the_hardware_limit(self):
+        assert refresh_hz_from_config(
+            {"display": {"hardware": {"limit_refresh_rate_hz": 150}}}
+        ) == 150.0
+
+    @pytest.mark.parametrize("cfg", [None, {}, {"display": {}}, {"display": {"hardware": {}}},
+                                     "nonsense", {"display": {"hardware": {"limit_refresh_rate_hz": None}}}])
+    def test_falls_back_to_the_default(self, cfg):
+        assert refresh_hz_from_config(cfg) == 100.0
+
+    @pytest.mark.parametrize("cfg", [
+        {"display": "nonsense"},
+        {"display": ["nonsense"]},
+        {"display": 60},
+        {"display": {"hardware": "nonsense"}},
+        {"display": {"hardware": ["nonsense"]}},
+        {"display": {"hardware": 60}},
+    ])
+    def test_malformed_nesting_falls_back_rather_than_raising(self, cfg):
+        # Truthy-but-not-a-mapping at either level used to reach .get() on a
+        # str/list and raise AttributeError out of a function whose whole
+        # contract is "a refresh rate, or the default".
+        assert refresh_hz_from_config(cfg) == 100.0
+
+
+class TestCrispLadder:
+    """Whole-pixel speeds available on a given panel."""
+
+    def test_every_entry_is_exactly_reachable(self):
+        for c in crisp_ladder(100):
+            assert c.pixels_per_second == pytest.approx(
+                c.refresh_hz / c.frame_hold * c.pixels_per_frame)
+
+    def test_sorted_slowest_first(self):
+        speeds = [c.pixels_per_second for c in crisp_ladder(100)]
+        assert speeds == sorted(speeds)
+
+    def test_no_duplicate_speeds(self):
+        speeds = [round(c.pixels_per_second, 3) for c in crisp_ladder(100)]
+        assert len(speeds) == len(set(speeds))
+
+    def test_duplicates_resolve_to_the_smaller_step(self):
+        """100px/s is 1px every refresh or 2px every 2nd; prefer the former."""
+        entry = next(c for c in crisp_ladder(100)
+                     if c.pixels_per_second == pytest.approx(100.0))
+        assert entry.pixels_per_frame == 1
+        assert entry.frame_hold == 1
+
+    def test_ladder_scales_with_the_panel(self):
+        assert any(c.pixels_per_second == pytest.approx(60.0)
+                   for c in crisp_ladder(60))
+        assert any(c.pixels_per_second == pytest.approx(30.0)
+                   for c in crisp_ladder(60))
+
+    def test_full_refresh_speed_is_present(self):
+        for hz in (60, 75, 100, 120):
+            assert any(c.pixels_per_second == pytest.approx(float(hz))
+                       for c in crisp_ladder(hz))
+
+
+class TestSolveCrisp:
+    def test_exact_targets_are_matched_exactly(self):
+        for target in (100, 50, 25, 20):
+            assert solve_crisp(target, 100).pixels_per_second == pytest.approx(target)
+
+    def test_prefers_smooth_motion_over_raw_proximity(self):
+        """30 -> 33.3 (1px, 33fps), not 28.6 (2px at 14fps) which is nearer."""
+        got = solve_crisp(30, 100)
+        assert got.pixels_per_second == pytest.approx(33.333, abs=0.01)
+        assert got.pixels_per_frame == 1
+
+    def test_does_not_chase_a_jumpy_exact_match(self):
+        """45 -> 50 (1px, smooth) beats 42.9 (3px at 14fps)."""
+        assert solve_crisp(45, 100).pixels_per_frame == 1
+
+    def test_allows_a_two_pixel_step_when_it_is_clearly_closer(self):
+        """60 -> 66.7 (2px at 33fps) rather than 50 (1px) which is 17% slow."""
+        got = solve_crisp(60, 100)
+        assert got.pixels_per_second == pytest.approx(66.667, abs=0.01)
+        assert got.pixels_per_frame == 2
+
+    def test_result_is_always_on_the_ladder(self):
+        ladder = {round(c.pixels_per_second, 3) for c in crisp_ladder(100)}
+        for target in range(5, 205, 5):
+            assert round(solve_crisp(target, 100).pixels_per_second, 3) in ladder
+
+    def test_adapts_to_a_slower_panel(self):
+        got = solve_crisp(30, 60)
+        assert got.pixels_per_second == pytest.approx(30.0)
+        assert got.pixels_per_frame == 1
+        assert got.frame_hold == 2
+
+    def test_speeds_beyond_the_panel_clamp_to_the_fastest_available(self):
+        got = solve_crisp(10_000, 100)
+        assert got.frame_hold == 1
+        assert got.pixels_per_frame == MAX_PIXELS_PER_FRAME
+
+    def test_steppiness_labels_are_sane(self):
+        assert solve_crisp(100, 100).steppiness == "smooth"
+        assert crisp_ladder(100)[0].steppiness == "stepped"
+
+
+class TestRefreshFromDisplayManager:
+    """A plugin sees only its own config section, so the display manager is the
+    authoritative source for the panel's refresh rate."""
+
+    class DM:
+        def __init__(self, hz):
+            self.refresh_hz = hz
+            self.hold = None
+
+        def set_frame_hold(self, refreshes):
+            self.hold = refreshes
+
+    def test_uses_the_display_manager_rate(self):
+        dm = self.DM(60.0)
+        s = configure(FakeHelper(), {"display_options": {"scroll_pixels_per_second": 30.0}},
+                      display_manager=dm)
+        # 30px/s on a 60Hz panel is 1px every 2nd refresh, exactly.
+        assert s.pixels_per_second == pytest.approx(30.0)
+        assert s.frame_hold == 2
+
+    def test_explicit_refresh_hz_wins_over_the_display_manager(self):
+        dm = self.DM(60.0)
+        s = configure(FakeHelper(), {"display_options": {"scroll_pixels_per_second": 25.0}},
+                      display_manager=dm, refresh_hz=100)
+        assert s.frame_hold == 4, "25px/s at 100Hz is 1px every 4th refresh"
+
+    def test_missing_attribute_falls_back_to_the_default(self):
+        class Bare:
+            def set_frame_hold(self, refreshes):
+                self.hold = refreshes
+
+        bare = Bare()
+        s = configure(FakeHelper(), {"display_options": {"scroll_pixels_per_second": 50.0}},
+                      display_manager=bare)
+        assert s.frame_hold == 2, "assumed 100Hz"
+
+    def test_reported_settings_describe_the_display_manager_rate(self):
+        """Without snapping, nothing downstream corrects a wrong refresh rate.
+
+        The rate used to be read *after* resolve() had already filled in
+        target_fps, pixels_per_frame and the judder warning from the 100Hz
+        default -- so on a 60Hz panel every one of those described 100Hz, and
+        set_target_fps() paced the helper to 100 FPS.
+        """
+        dm = self.DM(60.0)
+        helper = FakeHelper()
+        s = configure(helper, {"display_options": {"scroll_pixels_per_second": 30.0}},
+                      display_manager=dm, snap_to_crisp=False)
+
+        assert s.target_fps == pytest.approx(60.0)
+        assert s.pixels_per_frame == pytest.approx(0.5), "30px/s over 60 frames"
+        assert helper.target_fps == pytest.approx(60.0)
+
+    def test_judder_warning_is_computed_at_the_real_refresh_rate(self):
+        dm = self.DM(60.0)
+        # 60px/s is exactly 1px per refresh at 60Hz -- crisp, no warning. At
+        # the 100Hz default it is 0.6px per refresh and would be flagged.
+        s = configure(FakeHelper(), {"display_options": {"scroll_pixels_per_second": 60.0}},
+                      display_manager=dm, snap_to_crisp=False)
+        assert s.warning is None, s.warning
+
+    def test_global_config_supplies_the_rate_without_a_display_manager(self):
+        s = configure(
+            FakeHelper(),
+            {"display_options": {"scroll_pixels_per_second": 30.0}},
+            {"display": {"hardware": {"limit_refresh_rate_hz": 60}}},
+            snap_to_crisp=False,
+        )
+        assert s.target_fps == pytest.approx(60.0)
+
+
+class TestFrameHoldIsReportedNotApplied:
+    """configure() reports the hold; the caller applies it when it scrolls.
+
+    The hold belongs to a scroll, not to a plugin's lifetime -- plugins share
+    one display manager, so one set at construction is reset the moment any
+    other plugin stops scrolling.
+    """
+
+    class RecordingDM:
+        refresh_hz = 100.0
+
+        def __init__(self):
+            self.calls = []
+
+        def set_scrolling_state(self, is_scrolling, frame_hold=1):
+            self.calls.append((is_scrolling, frame_hold))
+
+        def set_frame_hold(self, refreshes):
+            self.calls.append(("set_frame_hold", refreshes))
+
+    def test_configure_does_not_touch_the_display_manager(self):
+        dm = self.RecordingDM()
+        settings = configure(
+            FakeHelper(), {"display_options": {"scroll_pixels_per_second": 25.0}},
+            display_manager=dm)
+        assert dm.calls == [], "configure() must not apply the hold itself"
+        assert settings.frame_hold == 4, "but it must report what to apply"

test/test_scroll_helper.py

@@ -11,7 +11,11 @@ import time
 from unittest.mock import patch
 from PIL import Image
 
-from src.common.scroll_helper import ScrollHelper
+from src.common.scroll_helper import (
+    ScrollHelper,
+    format_frame_stats,
+    frame_stats,
+)
 
 
 DISPLAY_W = 64
@@ -315,3 +319,77 @@ class TestGetScrollInfo:
         helper.scroll_position = 42.0
         info = helper.get_scroll_info()
         assert info["scroll_position"] == 42.0
+
+
+class TestFrameStatsPercentiles:
+    """The stats line is the instrument this whole scroll change is measured
+    with, so its median and p95 have to be the real ones.
+
+    Both are also thresholds: stalls are counted at 1.5x the median and skips
+    at 0.5x, so an off-by-one in the median biases the counts as well as the
+    printed numbers.
+    """
+
+    # 100 samples of 1..100ms. True median 50.5ms (the mean of the two middle
+    # samples, not the upper one at 51ms); nearest-rank p95 is the 95th
+    # sample at 95ms, not the 96th at 96ms.
+    HUNDRED = [i / 1000.0 for i in range(1, 101)]
+
+    def test_even_window_median_averages_both_middle_samples(self):
+        assert frame_stats(self.HUNDRED)["median"] == pytest.approx(0.0505)
+
+    def test_p95_uses_nearest_rank(self):
+        assert frame_stats(self.HUNDRED)["p95"] == pytest.approx(0.095)
+
+    def test_odd_window_median_is_the_middle_sample(self):
+        times = [i / 1000.0 for i in range(1, 102)]  # 101 samples
+        assert frame_stats(times)["median"] == pytest.approx(0.051)
+
+    def test_single_sample_window_does_not_index_out_of_range(self):
+        stats = frame_stats([0.010])
+        assert stats["median"] == pytest.approx(0.010)
+        assert stats["p95"] == pytest.approx(0.010)
+        assert stats["min"] == stats["max"] == pytest.approx(0.010)
+
+    def test_two_sample_window(self):
+        stats = frame_stats([0.010, 0.020])
+        assert stats["median"] == pytest.approx(0.015)
+        assert stats["p95"] == pytest.approx(0.020)
+
+    def test_input_order_does_not_matter(self):
+        assert frame_stats(list(reversed(self.HUNDRED))) == frame_stats(self.HUNDRED)
+
+    def test_caller_window_is_not_mutated(self):
+        times = [0.030, 0.010, 0.020]
+        frame_stats(times)
+        assert times == [0.030, 0.010, 0.020]
+
+    def test_stall_threshold_follows_the_median(self):
+        # Ten 10ms frames and two 30ms stalls: median 10ms, so >15ms is a
+        # stall -- exactly the two.
+        stats = frame_stats([0.010] * 10 + [0.030] * 2)
+        assert stats["stalls"] == 2
+        assert stats["skips"] == 0
+
+    def test_skips_are_frames_that_never_reached_the_panel(self):
+        stats = frame_stats([0.010] * 10 + [0.002] * 3)
+        assert stats["skips"] == 3
+        assert stats["stalls"] == 0
+
+    def test_fps_is_the_reciprocal_of_the_mean(self):
+        assert frame_stats([0.010] * 50)["fps"] == pytest.approx(100.0)
+
+    def test_formatted_line_reports_the_corrected_values(self):
+        line = format_frame_stats(self.HUNDRED)
+        assert "median 50.50ms" in line, line
+        assert "p95 95.00ms" in line, line
+        assert "over 100 frames" in line, line
+
+    def test_log_frame_rate_emits_the_line_and_clears_the_window(self, helper):
+        helper._window = [0.010] * 20
+        helper.last_fps_log_time = 0.0  # force the 5s boundary
+        with patch.object(helper.logger, "info") as info:
+            helper.log_frame_rate()
+        assert info.called
+        assert "Scroll frame stats" in info.call_args[0][0]
+        assert helper._window == []