Merge branch 'main' of https://github.com/ChuckBuilds/LEDMatrix into fix/plugin-update-keeps-local-files

# Conflicts:
#	test/test_on_demand_live_and_restore.py

CHANGELOG.md

@@ -19,6 +19,68 @@ accepts both, but the store flags the old spelling as deprecated
 
 ## Unreleased
 
+### A scrolling screen held by its plugin's update() is reported
+
+- While a plugin's `update()` runs it holds the plugin's lock, and that
+  plugin's frames are skipped: on a scroller, a frozen strip, with nothing
+  logged (and a freeze of 5 s or more is a gap, not a freeze, to the frame
+  stats). The high-FPS loop now times each run of skipped frames; one of
+  250 ms or more logs `Display of <plugin> held N ms by its update()`
+  (rate-limited per plugin) when it ends, and is recorded on the plugin's
+  health as a `display hold` busy skip, which never counts toward the
+  circuit breaker. The 1 Hz loop is left out: its frames are a second apart,
+  so one skipped frame there measures nothing and freezes nothing visible.
+
+### Fixed
+
+- The web preview and `/api/v3/display/current` no longer stay black for a
+  whole screen that draws its card once and then holds it. The snapshot is
+  written from `update_display()` at most once per write interval, so a frame
+  pushed inside that interval was skipped and left for the next
+  `update_display()` -- which such a screen never makes. Soccer's
+  recent/upcoming cards skip redundant redraws, and the first one after an
+  on-demand start lands a few milliseconds after the start's clear wrote a
+  black frame: on ledpi the preview showed 0 lit pixels for the whole 15 s
+  while the panel showed the card. `DisplayManager` now remembers a skipped
+  changed frame, and the render loop writes it (`write_owed_snapshot()`)
+  once the interval has passed. The cadence is unchanged, and nothing extra
+  runs when no frame is owed.
+
+### ESPN date-range fetches: fewer requests, fewer at once
+
+A soccer board (8 leagues, ESPN rejecting `dates=` ranges) logged ~90
+`NameResolutionError` lines and an `update() timed out` at every start on a
+Pi: each league's fortnight-either-side window was 29 day requests, fetched
+by several managers at once, ~40 in flight. Measured against live ESPN with
+soccer-scoreboard 2.39.2, alternating runs: **~450 requests per start, peak
+~45 in flight, ~75 DNS lookups -> 46 requests, peak 13, ~30 lookups**.
+
+- `fetch_espn_date_chunks()` asks for a window's partial edge month whole
+  when the window covers `ESPN_MONTH_COVER_MIN_DAYS` (7) or more of its days,
+  and trims the answer to the window's days by each event's US Eastern start
+  date -- the day ESPN's `dates=YYYYMMDD` means (417 of 417 live soccer
+  events matched). A 29-day window spanning two months is 2 requests instead
+  of 29. Short windows (a live poll's 1-2 days) stay day by day. A trimmed
+  month that comes back at the 500-event cap re-asks only the window's days.
+  An event with no readable date is kept. New: `espn_request_chunks()`.
+- Chunk requests share one process-wide cap of `ESPN_CHUNK_WORKERS` (6) in
+  flight, across every window being fetched, instead of six per window.
+- A new process starts as if a range had just been rejected, so it no longer
+  spends one doomed 400 per window at every start (eleven at once from a
+  soccer board); the range is still retried `RANGE_RETRY_SECONDS` in.
+
+### Fetch stats: bytes on the wire, not just decoded
+
+`GET /api/v3/plugins/fetch-stats` reported only `bytes`, the decoded body
+size, and that read as the download volume. ESPN gzips every scoreboard, so
+it overstated what crossed the network about 14x: a college football
+Saturday's scoreboard is 865 KB decoded and 63 KB on the wire, and ledpi's
+"643 MB in 6 hours" of football was ~47 MB of actual traffic. Every counter
+set (totals, per plugin, per host) now has `wire_bytes` too, read from
+urllib3's count of the raw bytes it took off the socket. A response with no
+urllib3 response behind it is counted at its decoded size. `bytes` keeps its
+meaning.
+
 ### Cheap per-frame and per-fetch savings
 
 - `BaseOddsManager.get_odds()` no longer pretty-prints every odds response
@@ -505,6 +567,54 @@ policies are unchanged.
   stored `ttl` was stretched the same way. A memory hit is now also checked against
   the record's own timestamp, and a stale one falls through to disk, which
   returns a newer write if there is one.
+- An on-demand request that names a `*_live` mode now shows that mode. On
+  ledpi, `{"plugin_id": "football-scoreboard", "mode": "ncaa_fb_live"}` with
+  15 college games on answered 200 and showed `nfl_recent`. The session's
+  mode list kept a live mode only when the plugin's `has_live_content()`
+  said so. That method answers the live-priority question, and the sports
+  plugins answer it for favourite teams only. A mode the request names
+  (not one resolved from a bare plugin id) now leads the session, with the
+  plugin's other modes after it. If it has nothing to draw, the session
+  moves on to the next of those modes, like any empty on-demand mode. The
+  name is saved with the session (`named_mode` in
+  `display_on_demand_config`), so a restart resumes on it.
+- A restart during an on-demand session whose plugin then fails to load no
+  longer leaves a session with no modes. On ledpi, `clock-simple` failed
+  config validation after a crash. The display logged `No valid display
+  modes found for on-demand plugin 'clock-simple' after restoration` and
+  kept reporting the session as active until its first pass ended it as
+  `idle`. The cached request stayed behind for the next restart. The session
+  now ends at startup with status `error` and error `restore-failed`, which
+  `/display/on-demand/status` reports, and the cached request is dropped. The
+  same applies when the plugin system itself fails to start.
+- `POST /api/v3/config/schedule` and `/config/dim-schedule` accept a
+  disabled per-day schedule with every day off. That is the shape
+  `config.template.json` ships, so posting back what GET returned on a fresh
+  install answered 400 "At least one day must be enabled". An enabled per-day
+  schedule still needs a day on. A day that is off now keeps the times it
+  was posted with (the schedule picker sends them). Before, saving dropped
+  them, so turning the day back on showed the defaults.
+- `POST /api/v3/config/main` answers `restart_required: true` only when the
+  save changed a setting the running display does not apply by itself.
+  Brightness (`brightness.set` and the config watcher), the per-mode
+  durations and plugin sections are applied live. A brightness-only save,
+  such as the MQTT bridge's slider, or a save that changed nothing, no longer
+  shows the restart banner. Hardware, rotation order, timezone and every
+  other setting still ask for the restart.
+- `GET /api/v3/health` reports `degraded` when the display service is
+  stopped. Before, only the sub-checks changed, and the overall status stayed
+  `healthy` for as long as the last preview frame was under 60 s old.
+  `checks.display_loop.status` is now `stopped` when three things agree:
+  systemd says the service is not active, the control socket does not
+  answer, and there is no live heartbeat. Where the platform has no socket
+  (Windows) or it is switched off, nothing changes.
+- `GET /api/v3/display/current-status` no longer reports the stopped
+  display's last state (`is_display_active: true`) from the cache for up to
+  120 s. When the control socket does not answer and the render loop's
+  heartbeat is absent, stale, or from a process that is gone (#726's rules),
+  the answer is unknown, with every field `null`. A display that still beats
+  without a socket, Windows and a socket switched off read the cache as
+  before. New `web_interface.display_state.display_gone()`.
 - The garbage-collection timer (`GcMonitor`, above) no longer prints
   `Exception ignored while calling GC callback ... 'NoneType' object has no
   attribute 'perf_counter'` when the display service or a test run exits.
@@ -885,25 +995,6 @@ policies are unchanged.
   a runtime publisher that stops still goes `stale`, and a subscription that
   goes quiet still falls back to the cache. The cache path's 120 s rule is
   unchanged.
-- A plugin that pauses the Vegas scroll gets its pause when its display
-  duration is not a plain number. Several plugins (clock-simple, calendar,
-  countdown) return `display_duration` as it is in config.json, so a value
-  saved as `"20"` or `null` (the raw config editor, a hand edit) reached the
-  pause as a string or None; comparing it with the clock raised, and the
-  plugin flashed up and the scroll went straight on, at every one of its
-  turns. `inf` held the pause until something interrupted it, and 0, a
-  negative number or NaN ended it at once. The pause now reads the duration
-  as the rotation does (`finite_seconds()` in `base_plugin`): a numeric
-  string counts, anything else that is not a finite number (or a
-  `get_display_duration()` that raises) pauses for 30 s, and a number at or
-  below zero for 15 s, with one warning per plugin.
-- Reinstalling Weather, Music, Stocks or Leaderboard from the Plugin Store
-  while it is enabled asks for a display restart, as reinstalling any other
-  enabled plugin does. `POST /api/v3/plugins/install` looked for the
-  plugin's `enabled` flag under the store id (`weather`), but its config
-  section is under the id its manifest declares (`ledmatrix-weather`), so
-  `restart_required` was always false and the display kept running the
-  copy it had loaded. The check now uses the installed id.
 
 ### Scrolling
 
@@ -1350,6 +1441,17 @@ read any of them:
 
 ### Fixes
 
+- Updating a plugin from the store no longer deletes the files it wrote
+  beside itself. A monorepo update replaces the plugin directory with the
+  fresh download and deletes the old copy, so calendar's Google OAuth files
+  (`token.pickle`, `credentials.json`) were lost on every update and the
+  calendar stopped until they were restored by hand. Before the old copy is
+  removed, the update now copies over anything the plugin's `.gitignore`
+  excludes plus known secret/state files (`*.pickle`, `token.json`,
+  `credentials.json`, `config_secrets.json`, `.pkce_code_verifier`); files the
+  new release ships are never overwritten, and byte code is not carried. A
+  plugin updated with `git pull` no longer sweeps an untracked token into the
+  auto-stash, which is never popped (`src/plugin_system/plugin_local_files.py`).
 - Quieter routine logging. Every rotation logged each mode twice
   ("Switching to mode", then "Processing mode"), and a mode with nothing to
   show added "display() returned False" and "No content to display". Those

docs/REST_API_REFERENCE.md

@@ -159,14 +159,16 @@ there an unchecked checkbox — which the browser omits — is saved as
 }
 ```
 
-`restart_required` is always true here: display hardware, rotation,
-durations and general settings take effect when the display restarts, and
-the web UI shows its restart banner on the flag. (Plugin sections saved
-through this route reach the running plugin live, like
-`POST /plugins/config`.)
+`restart_required` is true when the save changed a setting that takes
+effect when the display restarts: display hardware, rotation order,
+timezone, general settings and the rest. The web UI shows its restart banner
+on the flag. It is false when the save changed only what the running display
+applies by itself, or nothing: `brightness`, the per-mode durations
+(`duration__<mode>`, `display.display_durations`) and plugin sections, which
+reach the running plugin live, like `POST /plugins/config`.
 
-A saved `brightness` is the exception: it reaches the panel without a
-restart. The route also sends it to the running display over the control
+A saved `brightness` reaches the panel without a restart. The route also
+sends it to the running display over the control
 socket (`brightness.set`), which puts it on the panel at once, and the
 response adds `"brightness_transport": "socket"`. Otherwise it is
 `"config"`, with `brightness_socket_error` giving the reason, and the
@@ -246,7 +248,10 @@ Replace the schedule configuration.
 ```
 
 A day whose `<day>_enabled` key is absent counts as enabled, with default
-times `07:00`-`23:00`. At least one day must be enabled.
+times `07:00`-`23:00`. An enabled schedule needs at least one day enabled; a
+disabled one (`"enabled": false`) may have every day off, as
+`config.template.json` ships it. A day that is off keeps the times sent for
+it, when they are valid `HH:MM`.
 
 **Response**:
 ```json
@@ -343,7 +348,11 @@ control socket ([IPC_CONTROL_SOCKET.md](IPC_CONTROL_SOCKET.md)), and `cache`
 when it came from the `display_current_state` cache key (no socket: the
 display is stopped or older, or this is Windows). A display whose render
 loop has not refreshed its state for 120 seconds is reported with every
-field `null`, either way.
+field `null`, either way. So is a stopped display: when the socket does not
+answer and the render loop's heartbeat
+(`/run/ledmatrix/display-heartbeat.json`) is absent, stale or from a process
+that is gone, the cache's last entry is not used. A display still beating
+without a socket, Windows, or a socket switched off reads the cache.
 
 ### List Display Modes
 
@@ -2281,7 +2290,11 @@ display snapshot. `data.status` is `healthy` or `degraded`, with
 (with `heartbeat_age_seconds`), `stalled` (no heartbeat for 60s: the panel is
 frozen even if the service is active; the status turns `degraded`), or
 `not_reported` when the display writes none (not started yet, the dev server,
-Windows), which does not affect the status. Its `source` is `socket` when the
+Windows), which does not affect the status, or `stopped` (with `source:
+"service"`) when the display service is not active, the control socket does
+not answer and there is no live heartbeat; the status then turns
+`degraded`. A platform with no control socket (Windows) or a socket switched
+off never reports `stopped`. Its `source` is `socket` when the
 age came from the display's state stream over the control socket (measured
 in memory by the display) and `heartbeat_file` when it came from
 `/run/ledmatrix/display-heartbeat.json`.
@@ -2344,8 +2357,9 @@ Replace the dim schedule. `dim_brightness` is 0-100 (default 30). In
 `per-day` mode the days can be sent either as the `days` object that GET
 returns, or as the web form's flat fields (`monday_enabled`,
 `monday_start`, `monday_end`, ...). A day that is not sent counts as
-enabled with default times `20:00`-`07:00`; at least one day must be
-enabled.
+enabled with default times `20:00`-`07:00`. As for the schedule above, an
+enabled dim schedule needs at least one day enabled and a disabled one may
+have every day off.
 
 ---
 

docs/RUN_LOOP_REDESIGN.md

@@ -181,13 +181,14 @@ that the harness patches in today.
   - dynamic duration (cycle complete, plugin cap, global cap)
   - live priority taking over and handing back; live round-robin
   - on-demand start/stop/expiry; pinned on-demand; a session resumed after
-    a restart
+    a restart, and one that cannot resume (its plugin did not load); a
+    request naming a live mode the plugin's live check would drop
   - schedule off and dim, with an on-demand override during downtime
   - WiFi notice; sync follower
   - Vegas, with and without `live_in_ticker`
 - Each trace row is `[start, mode, duration, exit_reason, frames,
   force_clear]`. The exit reason is the event that decided what came next.
-- All 16 tests run in under a second. The goldens were generated from
+- All 18 tests run in under a second. The goldens were generated from
   main's `run()` before any code moved.
 - Vegas uses `FakeVegas`, which implements only the contract the controller
   depends on: `run_iteration()` returns True after its duration and False

src/common/README.md

@@ -109,8 +109,11 @@ and the plugin test harness all use it. Most plugins get BDF text through
 [`espn_dates.py`](espn_dates.py). ESPN's site API rejects `dates=` ranges
 and truncates results when `limit` is above 500. `fetch_espn_scoreboard()`
 splits a range into month and day requests ESPN accepts and merges the
-results; `espn_date_chunks()`, `fetch_espn_date_chunks()`,
-`clamp_espn_limit()` and `merge_scoreboard_payloads()` are the pieces.
+results; `espn_date_chunks()`, `espn_request_chunks()`,
+`fetch_espn_date_chunks()`, `clamp_espn_limit()` and
+`merge_scoreboard_payloads()` are the pieces. A window's partial edge months
+are asked whole and trimmed to its days (US Eastern), and chunk requests share
+one process-wide cap of `ESPN_CHUNK_WORKERS` in flight.
 Every request goes through [`fetch_service`](#fetch_service), the chunks
 counted against the plugin that asked. Scoreboard plugins also bundle a copy
 for older cores.

src/common/espn_dates.py

@@ -26,10 +26,30 @@ A month can hold more than 500 events (college baseball's March does), and
 ESPN answers that with exactly ``limit`` events and no hint that more exist. A
 month chunk that comes back full is therefore re-asked day by day.
 
+A window's *partial* edge months are asked for whole, too, once the window
+covers ``ESPN_MONTH_COVER_MIN_DAYS`` or more of their days, and the answer is
+trimmed back to the window's days. A scoreboard's default fortnight either side
+of today (29 days, two partial months) was 29 day requests per league; it is
+now 2. Trimming needs ESPN's "game day", which is the event's start in US
+Eastern time -- checked against the live API on 2026-10-03: 417 of 417 soccer
+events across five leagues and three months (one of them spanning the end of
+daylight saving) came back from exactly the day query their Eastern date
+names. A short window (a live poll's one or two days) stays day by day, so it
+never downloads a whole month to read a day of it.
+
+Chunk requests share one process-wide budget of ``ESPN_CHUNK_WORKERS`` in
+flight, however many windows are being fetched at once. Each window used to get
+its own six, so a scoreboard starting eight leagues -- each with a recent and
+an upcoming manager -- had ~40 requests in flight, every one beyond a session's
+pool a new connection and a new DNS lookup. On a Pi whose resolver could not
+keep up, that was ~90 ``NameResolutionError`` lines within a minute of every
+start.
+
 Once a range has been rejected, later ranges skip straight to chunks for
 ``RANGE_RETRY_SECONDS`` instead of spending a doomed request first -- live
 scoreboards ask every 30 seconds. After that the range is tried again, so the
-workaround retires itself if ESPN reverts.
+workaround retires itself if ESPN reverts. A process starts inside that
+period, as if a range had just been rejected.
 
 ONE CACHE KEY PER SCOREBOARD
 ----------------------------
@@ -55,7 +75,7 @@ import re
 import threading
 import time
 from concurrent.futures import ThreadPoolExecutor
-from datetime import date, datetime, timedelta
+from datetime import date, datetime, timedelta, tzinfo
 from functools import partial
 from typing import Any, Callable, Dict, Iterable, List, Optional, Tuple, cast
 
@@ -100,16 +120,57 @@ RANGE_RETRY_SECONDS = 6 * 60 * 60
 # pool_maxsize of 10 so the shared Session never has to discard connections.
 ESPN_CHUNK_WORKERS = 6
 
+#: An edge month the window covers at least this many days of is asked for
+#: whole and trimmed, instead of one request per day (see module docstring).
+#: Below it the days are cheaper than the month: a whole month is two to
+#: three times the bytes of the half of it a fortnight window holds.
+ESPN_MONTH_COVER_MIN_DAYS = 7
+
+# Every chunk request in the process holds one of these while it is in flight
+# -- the cap is per process, not per window (see module docstring).
+_chunk_slots = threading.BoundedSemaphore(ESPN_CHUNK_WORKERS)
+
+
+def _eastern_zone() -> Optional[tzinfo]:
+    """US Eastern, the zone ESPN's ``dates=YYYYMMDD`` means, or None when
+    this Python has no time zone data (no edge month is trimmed then)."""
+    zone: Optional[tzinfo] = None
+    try:
+        from zoneinfo import ZoneInfo
+        zone = ZoneInfo("America/New_York")
+    except Exception:  # noqa: BLE001 - no zoneinfo module or no tz database
+        zone = None
+    if zone is not None:
+        return zone
+    try:
+        import pytz
+        return cast(tzinfo, pytz.timezone("America/New_York"))
+    except Exception:  # noqa: BLE001
+        return None
+
+
+_EASTERN = _eastern_zone()
+
+# What _fetch_one_chunk returns for a month that came back at the cap.
+_CAPPED: Any = object()
+
 _range_lock = threading.Lock()
-_ranges_rejected_until = 0.0
+# A process starts out assuming ranges are still rejected, as they have been
+# since 2026-09-15, and tries one again RANGE_RETRY_SECONDS in. Starting
+# from "unknown" cost one doomed range request per window at every start --
+# eleven 400s at once from a soccer board, each fetching before any had
+# answered -- to learn what every start learns.
+_ranges_rejected_until = time.monotonic() + RANGE_RETRY_SECONDS
 
 __all__ = [
     "ESPN_MAX_LIMIT",
     "ESPN_CHUNK_WORKERS",
+    "ESPN_MONTH_COVER_MIN_DAYS",
     "RANGE_RETRY_SECONDS",
     "clamp_espn_limit",
     "parse_espn_date_range",
     "espn_date_chunks",
+    "espn_request_chunks",
     "merge_scoreboard_payloads",
     "fetch_espn_date_chunks",
     "fetch_espn_scoreboard",
@@ -220,6 +281,79 @@ def espn_date_chunks(start: date, end: date) -> List[str]:
     return chunks
 
 
+def espn_request_chunks(
+    start: date,
+    end: date,
+    month_cover_min_days: Optional[int] = None,
+) -> List[Tuple[str, Optional[Tuple[date, date]]]]:
+    """The requests that fetch ``[start, end]``, as ``(dates, trim)`` pairs.
+
+    :func:`espn_date_chunks`, except that a partial edge month with
+    ``month_cover_min_days`` (default ``ESPN_MONTH_COVER_MIN_DAYS``) or more
+    of its days in the window becomes one ``YYYYMM`` request whose ``trim``
+    is the first and last of those days: its events that start outside them
+    (US Eastern) are dropped. ``trim`` is None for every other request.
+    Without time zone data nothing can be trimmed, so the edge days stay day
+    requests.
+    """
+    if month_cover_min_days is None:
+        month_cover_min_days = ESPN_MONTH_COVER_MIN_DAYS
+    planned: List[Tuple[str, Optional[Tuple[date, date]]]] = []
+    run: List[str] = []
+
+    def flush() -> None:
+        if (_EASTERN is not None and month_cover_min_days > 0
+                and len(run) >= month_cover_min_days):
+            planned.append((run[0][:6], (_parse_day(run[0]), _parse_day(run[-1]))))
+        else:
+            planned.extend((day, None) for day in run)
+        run.clear()
+
+    for chunk in espn_date_chunks(start, end):
+        if run and (len(chunk) != 8 or chunk[:6] != run[0][:6]):
+            flush()
+        if len(chunk) == 8:
+            run.append(chunk)
+        else:
+            planned.append((chunk, None))
+    flush()
+    return planned
+
+
+def _parse_day(text: str) -> date:
+    return date(int(text[:4]), int(text[4:6]), int(text[6:8]))
+
+
+def _eastern_day(stamp: Any) -> Optional[date]:
+    """The US Eastern date of an ESPN event ``date`` ("2026-10-10T11:30Z"),
+    or None when it cannot be read."""
+    if not isinstance(stamp, str) or _EASTERN is None:
+        return None
+    try:
+        moment = datetime.fromisoformat(stamp.strip().replace("Z", "+00:00"))
+    except ValueError:
+        return None
+    if moment.tzinfo is None:
+        return None
+    return moment.astimezone(_EASTERN).date()
+
+
+def _trim_to_days(payload: Any, first: date, last: date) -> Any:
+    """Drop the events of a month payload that start outside ``[first, last]``
+    (US Eastern). An event whose date cannot be read is kept: its day query
+    might well have returned it, and a game is never dropped on a guess.
+    """
+    if not isinstance(payload, dict) or not isinstance(payload.get("events"), list):
+        return payload
+    kept = []
+    for event in payload["events"]:
+        day = _eastern_day(event.get("date")) if isinstance(event, dict) else None
+        if day is None or first <= day <= last:
+            kept.append(event)
+    payload["events"] = kept
+    return payload
+
+
 def merge_scoreboard_payloads(payloads: List[Any]) -> Dict[str, Any]:
     """Fold chunk responses into one scoreboard payload.
 
@@ -250,37 +384,54 @@ def merge_scoreboard_payloads(payloads: List[Any]) -> Dict[str, Any]:
 def _fetch_one_chunk(
     session, url: str, params: Dict[str, Any], headers, timeout, logger, chunk: str,
     cache_max_age: Optional[float] = None,
-) -> Optional[Dict[str, Any]]:
+    trims: Optional[Dict[str, Tuple[date, date]]] = None,
+) -> Any:
     """GET a single ``dates=`` chunk, or None when it failed.
 
     One bad chunk must not sink the rest of the season, so every error is
     logged and swallowed here rather than raised to the gather below.
+
+    A month that comes back at the cap is truncated: it returns ``_CAPPED``,
+    its payload dropped here before it is ever held beside the others. A
+    month in ``trims`` loses its events outside the days given there.
+
+    The request holds one of the process-wide ``_chunk_slots`` while it runs.
     """
     try:
-        response = fetch_get(
-            session,
-            url,
-            params=dict(params, dates=chunk, limit=ESPN_MAX_LIMIT),
-            headers=headers,
-            timeout=timeout,
-            **_memo_kwargs(cache_max_age),
-        )
-        response.raise_for_status()
-        return cast(Optional[Dict[str, Any]], response_json(response))
+        with _chunk_slots:
+            response = fetch_get(
+                session,
+                url,
+                params=dict(params, dates=chunk, limit=ESPN_MAX_LIMIT),
+                headers=headers,
+                timeout=timeout,
+                **_memo_kwargs(cache_max_age),
+            )
+            response.raise_for_status()
+            payload = response_json(response)
     except Exception as exc:  # noqa: BLE001 - see docstring
         if logger:
             logger.warning("ESPN chunk %s failed, skipping it: %s", chunk, exc)
         return None
+    if len(chunk) == 6 and isinstance(payload, dict):
+        if len(payload.get("events") or []) >= ESPN_MAX_LIMIT:
+            return _CAPPED
+        trim = (trims or {}).get(chunk)
+        if trim is not None:
+            payload = _trim_to_days(payload, *trim)
+    return payload
 
 
 def _fetch_chunks(
     session, url: str, params: Dict[str, Any], headers, timeout, logger,
     chunks: List[str], cache_max_age: Optional[float] = None,
-) -> List[Optional[Dict[str, Any]]]:
+    trims: Optional[Dict[str, Tuple[date, date]]] = None,
+) -> List[Any]:
     """Fetch every chunk, returning payloads positionally aligned with ``chunks``.
 
     Requests go out ``ESPN_CHUNK_WORKERS`` at a time because a cold season is
-    over a hundred of them. The order they come back in is not significant --
+    over a hundred of them -- and no more than that across every window the
+    process is fetching, which ``_fetch_one_chunk``'s slot enforces. The order they come back in is not significant --
     callers keep ``chunks`` order from the returned list -- but it does mean
     the session is shared across threads, which is why this only ever issues
     GETs and never touches session state.
@@ -293,7 +444,7 @@ def _fetch_chunks(
         return []
     fetch = partial(
         _fetch_one_chunk, session, url, params, headers, timeout, logger,
-        cache_max_age=cache_max_age,
+        cache_max_age=cache_max_age, trims=trims,
     )
     if len(chunks) == 1:
         return [fetch(chunks[0])]
@@ -340,7 +491,9 @@ def fetch_espn_date_chunks(
     if span is None:
         return None
 
-    chunks = espn_date_chunks(*span)
+    planned = espn_request_chunks(*span)
+    chunks = [chunk for chunk, _ in planned]
+    trims = {chunk: trim for chunk, trim in planned if trim is not None}
     if logger:
         logger.debug(
             "Fetching ESPN date range %s as %d month/day chunks",
@@ -349,32 +502,31 @@ def fetch_espn_date_chunks(
 
     results = _fetch_chunks(
         session, url, params, headers, timeout, logger, chunks, cache_max_age,
+        trims,
     )
     attempted = len(chunks)
 
-    # A month that came back at the cap is truncated; its days replace it in
-    # place, so merged events stay in chunk order however the requests raced.
+    # A month that came back at the cap is truncated; its days (only the
+    # window's, for a trimmed edge month) replace it in place, so merged
+    # events stay in chunk order however the requests raced. Its payload was
+    # already dropped in the worker: a capped college-baseball month is ~2MB
+    # of parsed JSON, and holding four of them through ~120 day requests added
+    # ~25MB to the peak -- more than the concurrency itself. Low-memory boards
+    # (docs/LOW_MEMORY_BOARDS.md) have under 200MB of headroom.
     slots: List[Any] = results
     capped: Dict[int, List[str]] = {}
     for index, chunk in enumerate(chunks):
-        payload = slots[index]
-        if payload is None or len(chunk) != 6:
+        if slots[index] is not _CAPPED:
             continue
-        events = payload.get("events") if isinstance(payload, dict) else None
-        if len(events or []) >= ESPN_MAX_LIMIT:
-            if logger:
-                logger.info(
-                    "ESPN month %s hit the %d-event cap; re-asking it day by day",
-                    chunk, ESPN_MAX_LIMIT,
-                )
-            capped[index] = _days_of_month(chunk)
-            # Drop the truncated month now rather than after its days arrive:
-            # a capped college-baseball month is ~2MB of parsed JSON, and
-            # holding four of them through ~120 day requests added ~25MB to
-            # the peak -- more than the concurrency itself. Low-memory boards
-            # (docs/LOW_MEMORY_BOARDS.md) have under 200MB of headroom.
-            slots[index] = None
-    payload = events = None
+        if logger:
+            logger.info(
+                "ESPN month %s hit the %d-event cap; re-asking it day by day",
+                chunk, ESPN_MAX_LIMIT,
+            )
+        trim = trims.get(chunk)
+        capped[index] = (_days_of_month(chunk) if trim is None
+                         else espn_date_chunks(*trim))
+        slots[index] = None
 
     if capped:
         days = [day for index in sorted(capped) for day in capped[index]]

src/common/fetch_service.py

@@ -59,7 +59,10 @@ says how old with ``cache_max_age`` (``fetch_get(..., cache_max_age=ttl)``;
 Identical means what the validator store keys on: URL, query, effective
 headers and, for a session with cookies or auth, the session.
 
-**Counters.** Requests, merged requests, bytes, 304s, errors, HTTP errors,
+**Counters.** Requests, merged requests, bytes (``bytes`` decoded, as the
+caller reads them; ``wire_bytes`` as they crossed the network, which is
+what a metered connection pays for -- ESPN gzips, so the two differ ~14x),
+304s, errors, HTTP errors,
 adapter retries, throttled requests and seconds waited, plus requests
 answered without the network: ``memo_hits`` (the response cache) and
 ``cache_hits`` / ``legacy_cache_hits`` (a shared ESPN scoreboard cache entry,
@@ -201,6 +204,7 @@ _COUNTER_FIELDS = (
     "throttled",     # requests that waited for a host budget
     "overruns",      # requests that went after max_wait_seconds anyway
     "bytes",         # decoded response body bytes received
+    "wire_bytes",    # body bytes as they came off the socket (still compressed)
     "wait_seconds",  # time spent waiting for host budgets
     "memo_hits",     # answered from the response cache (max-age); nothing sent
     "cache_hits",    # scoreboard fetches answered from a shared ESPN cache entry
@@ -616,6 +620,30 @@ def _body_of(response: Any) -> Optional[bytes]:
     return content if isinstance(content, bytes) else None
 
 
+def _wire_bytes_of(response: Any, body: Optional[bytes]) -> int:
+    """How many body bytes came off the socket for ``response``: the
+    compressed size when the server sent gzip, which ESPN does for every
+    scoreboard (63 KB on the wire for an 865 KB college football Saturday).
+
+    urllib3's ``HTTPResponse.tell()`` counts the raw bytes read before
+    decoding. A response without one (a test double, an adapter that is not
+    urllib3) or one whose body was not read is counted at its decoded size,
+    or as 0, so the counter never claims less than it can prove.
+    """
+    if body is None:
+        return 0
+    raw = getattr(response, "raw", None)
+    tell = getattr(raw, "tell", None)
+    if callable(tell):
+        try:
+            read = tell()
+        except Exception:
+            read = None
+        if isinstance(read, int) and not isinstance(read, bool) and read > 0:
+            return read
+    return len(body)
+
+
 def _retries_of(response: Any) -> int:
     raw = getattr(response, "raw", None)
     retries = getattr(raw, "retries", None)
@@ -1117,6 +1145,7 @@ class FetchService:
                         http_errors=int(status is not None and status >= 400),
                         retries=_retries_of(response),
                         bytes=len(body) if body is not None else 0,
+                        wire_bytes=_wire_bytes_of(response, body),
                         throttled=int(waited > 0), overruns=int(overrun),
                         wait_seconds=waited)
         except Exception:

src/display_controller.py

@@ -25,6 +25,7 @@ import os
 import inspect
 import signal
 import json
+import math
 import threading
 import types
 from collections import deque
@@ -56,7 +57,6 @@ from src.ipc.contract import (
     PluginReloadResult,
 )
 from src.ipc.server import ControlServer, QueuedCommand, StateHub, start_control_server
-from src.plugin_system.base_plugin import finite_seconds
 from src.vegas_mode.render_pipeline import SYNC_SEND_INTERVAL
 
 # Get logger with consistent configuration
@@ -90,6 +90,19 @@ _MIN_INITIAL_UPDATE_TIMEOUT_SECONDS = 2.0
 DEFAULT_DYNAMIC_DURATION_CAP = 180.0
 
 
+def _finite_seconds(value: Any) -> Optional[float]:
+    """``value`` as seconds when it is a finite number or a numeric string,
+    else None. A bool is not a number here, though it is an int: True would
+    read as a one-second screen."""
+    if isinstance(value, bool):
+        return None
+    try:
+        seconds = float(value)
+    except (TypeError, ValueError, OverflowError):
+        return None
+    return seconds if math.isfinite(seconds) else None
+
+
 class _PluginReloadJob:
     """A ``plugin.reload`` whose slow half runs off the render thread.
 
@@ -355,6 +368,10 @@ class DisplayController:
         self.on_demand_last_error: Optional[str] = None
         self.on_demand_last_event: Optional[str] = None
         self.on_demand_schedule_override = False
+        # The mode the request named, when it named one (not a mode resolved
+        # from a bare plugin id). Shown even when the plugin's live checks
+        # would leave it out of the session (_on_demand_modes_for_plugin).
+        self._on_demand_named_mode: Optional[str] = None
         # Plugins that are disabled in config and loaded only because an
         # on-demand request named them. The main loop unloads each one once
         # on-demand has moved off it (_release_on_demand_plugins).
@@ -548,6 +565,17 @@ class DisplayController:
         except Exception:  # pylint: disable=broad-except
             logger.exception("Plugin system initialization failed")
             self.plugin_manager = None
+            # A restored session has no plugin to resume on. It may have been
+            # read already (on_demand_active) or not yet, if initialization
+            # failed before the restore ran; either way, end it visibly.
+            try:
+                cached_session = self.cache_manager.get('display_on_demand_config',
+                                                        max_age=3600)
+            except Exception:  # pylint: disable=broad-except
+                cached_session = None
+            if self.on_demand_active or cached_session:
+                self.cache_manager.clear_cache('display_on_demand_config')
+                self._set_on_demand_error('restore-failed')
             # Its state machine no longer describes what runs; let the last
             # snapshot go stale (readers then say unknown) rather than keep
             # refreshing it.
@@ -1136,6 +1164,55 @@ class DisplayController:
         except Exception:  # pylint: disable=broad-except
             logger.exception("Error running scheduled plugin updates")
 
+    #: A run of frames skipped because a plugin's update() held its lock is
+    #: reported once it has lasted this long.
+    DISPLAY_HOLD_REPORT_SECONDS = 0.25
+
+    #: (plugin_id, monotonic start) of the current run of skipped frames.
+    _display_hold: Optional[Tuple[str, float]] = None
+
+    def _note_display_hold(self, plugin_id: str, held: bool) -> None:
+        """Report how long a plugin's update() kept its display() from drawing.
+
+        While update() runs on the worker it holds the plugin's lock, and every
+        frame of that plugin's screen is skipped: the panel keeps showing the
+        last frame, which on a scroller is a frozen strip. Nothing said so --
+        the frames are not failures, and a scroll freeze of 5 s or more is a
+        gap to the frame stats, not a freeze. This times each such run and,
+        when it ends after DISPLAY_HOLD_REPORT_SECONDS or more, logs it
+        (rate-limited per plugin) and records it on the plugin's health as a
+        busy skip, which never touches the circuit breaker. Only frames of the
+        high-FPS loop are timed (see _display_once's ``report_hold``).
+        """
+        # The clock is read only when a run starts or ends: on a frame that
+        # draws with no run open, this is one attribute check.
+        current = self._display_hold
+        if held:
+            if current is None or current[0] != plugin_id:
+                self._display_hold = (plugin_id, time.monotonic())
+            return
+        if current is None:
+            return
+        self._display_hold = None
+        if current[0] != plugin_id:
+            return
+        seconds = time.monotonic() - current[1]
+        if seconds < self.DISPLAY_HOLD_REPORT_SECONDS:
+            return
+        pm = self.plugin_manager
+        warn = getattr(pm, '_warn_rate_limited', None)
+        if warn is not None:
+            warn(f"display-hold:{plugin_id}",
+                 "Display of %s held %.0f ms by its update()",
+                 plugin_id, seconds * 1000.0)
+        tracker = getattr(pm, 'health_tracker', None)
+        record = getattr(tracker, 'record_busy_skip', None)
+        if record is not None:
+            try:
+                record(plugin_id, "display hold", seconds)
+            except Exception:  # pylint: disable=broad-except
+                logger.debug("Could not record a display hold", exc_info=True)
+
     @contextmanager
     def _display_lock_or_skip(self, plugin_id):
         """Try-lock guard keeping a plugin's display() off its in-flight update().
@@ -1161,7 +1238,7 @@ class DisplayController:
             lock.release()
 
     def _display_once(self, plugin, mode: str, accepts_display_mode: bool,
-                      force_clear: bool = False):
+                      force_clear: bool = False, report_hold: bool = False):
         """Call ``plugin.display()`` directly for one frame of a render loop.
 
         Frames after a screen's first dispatch come through here rather than
@@ -1177,6 +1254,12 @@ class DisplayController:
                 ``display_mode`` so plugins with several modes stay on it.
             accepts_display_mode: Whether display() takes ``display_mode``.
             force_clear: Passed through to display().
+            report_hold: Time runs of frames skipped because update() holds
+                the plugin's lock (see _note_display_hold). Only the high-FPS
+                loop asks: its frames are ~8 ms apart, so a run measures the
+                hold, and a held scroller is a frozen strip. The 1 Hz loop's
+                frames are a second apart, so one skipped frame there would
+                read as a 1 s hold of a screen that did not visibly change.
 
         Each call is timed (two monotonic reads) and handed to
         PluginManager.note_display_duration, which logs and records slow
@@ -1191,6 +1274,12 @@ class DisplayController:
         display_watchdog.watchdog.beat()
         plugin_id = getattr(plugin, 'plugin_id', None)
         with self._display_lock_or_skip(plugin_id) as can_display:
+            if report_hold and plugin_id:
+                self._note_display_hold(plugin_id, held=not can_display)
+            elif can_display and self._display_hold is not None:
+                # A drawn frame outside the high-FPS loop: whatever run was
+                # open is over, unreported.
+                self._display_hold = None
             if not can_display:
                 return True
             started = time.monotonic()
@@ -1202,6 +1291,13 @@ class DisplayController:
                 note = getattr(self.plugin_manager, 'note_display_duration', None)
                 if note is not None and plugin_id:
                     note(plugin_id, time.monotonic() - started)
+                # A screen that drew once and holds makes no more
+                # update_display() calls, so a frame the preview throttle
+                # skipped would otherwise never reach the snapshot.
+                write_owed = getattr(getattr(self, 'display_manager', None),
+                                     'write_owed_snapshot', None)
+                if write_owed is not None:
+                    write_owed()
 
     def _health_tracker(self):
         """The plugin circuit breaker, or None when it is not enabled."""
@@ -1379,7 +1475,7 @@ class DisplayController:
         except Exception as err:  # pylint: disable=broad-except
             problem = f"get_display_duration() raised {type(err).__name__}: {err}"
         else:
-            seconds = finite_seconds(value)
+            seconds = _finite_seconds(value)
             if seconds is not None:
                 return seconds
             problem = f"display duration {value!r} is not a number"
@@ -1652,6 +1748,7 @@ class DisplayController:
         self.on_demand_expires_at = None
         self.on_demand_pinned = False
         self.on_demand_schedule_override = False
+        self._on_demand_named_mode = None
         # While the session ran, _evaluate_schedule may have forced
         # is_display_active on over a scheduled-off answer. Drop the minute
         # gate so the next _check_schedule recomputes it; otherwise the panel
@@ -1818,6 +1915,7 @@ class DisplayController:
         self.on_demand_pinned = on_demand_config.get('pinned', False)
         self.on_demand_requested_at = on_demand_config.get('requested_at')
         self.on_demand_expires_at = on_demand_config.get('expires_at')
+        self._on_demand_named_mode = on_demand_config.get('named_mode')
         self.on_demand_status = 'active'
         self.on_demand_schedule_override = True
         logger.info("On-demand mode detected during initialization: resuming on plugin '%s'; "
@@ -2304,13 +2402,25 @@ class DisplayController:
                 return modes[0]
         return plugin_id
 
-    def _on_demand_modes_for_plugin(self, plugin_id: str) -> List[str]:
+    def _on_demand_modes_for_plugin(self, plugin_id: str,
+                                    named_mode: Optional[str] = None) -> List[str]:
         """Every loaded display mode belonging to `plugin_id`, in rotation order.
 
         Live modes that actually have content lead, then the rest, then live
         modes with nothing to show -- so an on-demand request for a sports
         plugin opens on a game in progress rather than an empty live screen.
         Returns an empty list when the plugin has no loaded modes.
+
+        `named_mode` is a mode the request asked for by name. It is always
+        in the list, first when the checks below would have dropped it.
+        Those checks ask has_live_content(), which is the live-priority
+        question -- "should this plugin take the panel from the rotation?"
+        -- and the sports plugins answer it for favourite teams only. Asking
+        for ncaa_fb_live with fifteen games on and no favourite playing got
+        a 200 and nfl_recent on the panel. The plugin's display() is what
+        knows whether the mode has anything to draw; when it has not, the
+        session moves to the plugin's next mode like any empty on-demand
+        mode.
         """
         plugin_modes = self.plugin_display_modes.get(plugin_id, [])
         if not plugin_modes:
@@ -2351,6 +2461,15 @@ class DisplayController:
             # Only live modes available but no content - use them anyway
             ordered_modes = live_modes
 
+        if named_mode and named_mode in available_plugin_modes:
+            # The named mode leads whether or not the live check kept it: a
+            # second live mode with content is already in the list, but behind
+            # the first, so the session would rotate away before reaching it.
+            if named_mode not in ordered_modes:
+                logger.info("On-demand: showing %s as requested; plugin '%s' reports no "
+                            "live-priority content for it", named_mode, plugin_id)
+            ordered_modes = [named_mode] + [m for m in ordered_modes if m != named_mode]
+
         return ordered_modes
 
     def _apply_on_demand_pin(self, ordered_modes: List[str], resolved_mode: Optional[str],
@@ -2379,10 +2498,20 @@ class DisplayController:
 
         plugin_id = self.on_demand_plugin_id
 
-        ordered_modes = self._on_demand_modes_for_plugin(plugin_id)
+        ordered_modes = self._on_demand_modes_for_plugin(plugin_id, self._on_demand_named_mode)
         if not ordered_modes:
-            logger.warning("No valid display modes found for on-demand plugin '%s' after restoration", plugin_id)
-            self.on_demand_modes = []
+            # The plugin did not load this time (seen on a rig: its config
+            # failed validation after the crash that caused the restart), so
+            # there is nothing to resume. Leaving the session active with no
+            # modes published it as active for a plugin that was not running
+            # until the first pass ended it as an ordinary 'idle', and kept
+            # the cached request for the next restart to trip over. End it
+            # as a failure the status endpoint reports, and drop the cache.
+            logger.error("On-demand session for plugin '%s' cannot resume after the "
+                         "restart: the plugin has no loaded display modes (did it "
+                         "fail to load?); ending it", plugin_id)
+            self.cache_manager.clear_cache('display_on_demand_config')
+            self._set_on_demand_error('restore-failed')
             return
 
         # A restart must not silently un-pin: the pin is part of the request
@@ -2583,7 +2712,10 @@ class DisplayController:
         if resolved_mode in self.available_modes:
             self.current_mode_index = self.available_modes.index(resolved_mode)
 
-        ordered_modes = self._on_demand_modes_for_plugin(resolved_plugin_id)
+        # Named: the request gave this mode itself, rather than a plugin id
+        # (or a mode the plugin doesn't have) that resolved to a default.
+        named_mode = resolved_mode if mode == resolved_mode else None
+        ordered_modes = self._on_demand_modes_for_plugin(resolved_plugin_id, named_mode)
         if not ordered_modes:
             logger.error("No valid display modes found for plugin '%s'", resolved_plugin_id)
             self._set_on_demand_error("no-modes")
@@ -2600,6 +2732,7 @@ class DisplayController:
         self.on_demand_requested_at = now
         self.on_demand_expires_at = (now + duration) if duration else None
         self.on_demand_pinned = pinned
+        self._on_demand_named_mode = named_mode
         self.on_demand_status = 'active'
         self.on_demand_last_error = None
         self.on_demand_last_event = 'started'
@@ -2628,6 +2761,7 @@ class DisplayController:
             'mode': resolved_mode,
             'duration': duration,
             'pinned': pinned,
+            'named_mode': named_mode,
             'requested_at': now,
             'expires_at': self.on_demand_expires_at
         }
@@ -3914,7 +4048,8 @@ class DisplayController:
                             _frame_start = time.perf_counter()
                             try:
                                 result = self._display_once(
-                                    manager_to_display, active_mode, _accepts_display_mode)
+                                    manager_to_display, active_mode, _accepts_display_mode,
+                                    report_hold=True)
                                 if isinstance(result, bool) and not result:
                                     logger.debug("Display returned False, breaking early")
                                     break

src/display_manager.py

@@ -317,6 +317,11 @@ class DisplayManager:
         # is handed to the writer; this only once it has been saved, so an
         # mtime touch never vouches for a frame still waiting to be written.
         self._saved_snapshot_digest: Optional[int] = None
+        # A changed frame reached _write_snapshot_if_due() inside the write
+        # interval and was skipped. Nothing writes it unless update_display()
+        # runs again, and a screen that draws once and holds never calls it
+        # again -- see write_owed_snapshot().
+        self._snapshot_owed = False
         self._snapshot_dir_prepared = False
         # Background writer used mid-scroll; see _write_snapshot_if_due.
         self._snapshot_cond = threading.Condition()
@@ -1788,9 +1793,10 @@ class DisplayManager:
 
             if frame_checksum is not None:
                 digest = frame_checksum
+                frame_changed = digest != self._last_snapshot_digest
                 action = snapshot_policy.decide(
                     now, self._last_snapshot_ts, self._last_snapshot_touch_ts,
-                    viewer_fresh, digest != self._last_snapshot_digest)
+                    viewer_fresh, frame_changed)
             else:
                 # Ask as if the frame had changed before paying to find out.
                 # decide() is monotone in frame_changed -- a SKIP for a
@@ -1802,22 +1808,36 @@ class DisplayManager:
                     now, self._last_snapshot_ts, self._last_snapshot_touch_ts,
                     viewer_fresh, True)
                 if action is snapshot_policy.SnapshotAction.SKIP:
+                    # Not hashed, so not known to be unchanged: owed until a
+                    # later look finds it written or unchanged.
+                    self._snapshot_owed = True
                     return
                 digest = zlib.adler32(self.image.tobytes())
-                if digest == self._last_snapshot_digest:
+                frame_changed = digest != self._last_snapshot_digest
+                if not frame_changed:
                     # Unchanged after all: the decision an unchanged frame gets.
                     action = snapshot_policy.decide(
                         now, self._last_snapshot_ts,
                         self._last_snapshot_touch_ts, viewer_fresh, False)
             if action is snapshot_policy.SnapshotAction.SKIP:
+                # A changed frame inside the write interval stays owed: the
+                # next update_display() would write it, but a static screen
+                # may not make one -- write_owed_snapshot() covers that.
+                self._snapshot_owed = frame_changed
                 return
             if (action is snapshot_policy.SnapshotAction.TOUCH
                     and self._saved_snapshot_digest == digest):
                 # mtime bump only: keeps the health check (snapshot age)
                 # green without paying for a PNG encode of an unchanged frame
+                # (this frame is already on disk, so nothing is owed).
+                self._snapshot_owed = False
                 os.utime(self._snapshot_path, None)
                 self._last_snapshot_touch_ts = now
                 return
+            # Owed until the write below succeeds: if it raises, the frame
+            # stays owed and write_owed_snapshot() retries it, rather than a
+            # held screen leaving the preview stale after one failed write.
+            self._snapshot_owed = True
             # (A TOUCH for a frame that isn't on disk yet -- still queued, or
             # its write failed -- is written instead: touching would make the
             # older file on disk look current.)
@@ -1842,9 +1862,39 @@ class DisplayManager:
             self._last_snapshot_ts = now
             self._last_snapshot_touch_ts = now
             self._last_snapshot_digest = digest
+            self._snapshot_owed = False
         except Exception as e:
             self._log_snapshot_failure(e)
 
+    def write_owed_snapshot(self) -> None:
+        """Write a frame the snapshot throttle skipped, once it is due.
+
+        The preview snapshot is only ever written from update_display(), and
+        at most once per write interval (snapshot_policy). A frame pushed
+        inside that interval is skipped, and is written by the next
+        update_display() that comes after it -- but a screen that draws its
+        card once and then holds it makes no further call. Its frame was on
+        the panel and never in the preview: soccer's recent/upcoming cards
+        skip redundant redraws, and the first one after an on-demand start
+        (pushed a few milliseconds after the controller's clear) left
+        /api/v3/display/current and the web preview black for the whole
+        screen while the panel showed the card.
+
+        The render loop calls this after each frame. Cheap when nothing is
+        owed (one attribute read); otherwise the usual policy decides, so
+        the write still waits out the interval and an unchanged frame is
+        never re-encoded.
+        """
+        if not self._snapshot_owed:
+            return
+        try:
+            if self._writes_suppressed():
+                return
+            with self._update_lock:
+                self._write_snapshot_if_due()
+        except Exception as e:  # pylint: disable=broad-except
+            self._log_snapshot_failure(e)
+
     def _log_snapshot_failure(self, error: Exception) -> None:
         # Snapshot failures must never break display — but they must not
         # be silent either: the snapshot's mtime is the web UI's display

src/plugin_system/base_plugin.py

@@ -11,7 +11,6 @@ Stability: Stable - maintains backward compatibility
 from abc import ABC, abstractmethod
 from enum import Enum
 from typing import Dict, Any, Optional, List
-import math
 import os
 import sys
 from src.deprecation import deprecated, warn_deprecated
@@ -241,26 +240,6 @@ def resolve_vegas_participation(plugin: Any, plugin_id: Optional[str] = None) ->
     return legacy_vegas_participation(plugin)
 
 
-def finite_seconds(value: Any) -> Optional[float]:
-    """``value`` as seconds when it is a finite number or a numeric string,
-    else None. A bool is not a number here, though it is an int: True would
-    read as a one-second screen.
-
-    How the core reads a plugin's get_display_duration() -- the rotation
-    (DisplayController._get_display_duration) and the Vegas static pause --
-    which several plugins answer straight from config.json, so a value saved
-    as "20" or null arrives as a string or None. A number at or below zero is
-    returned as it is; each caller has its own rule for that.
-    """
-    if isinstance(value, bool):
-        return None
-    try:
-        seconds = float(value)
-    except (TypeError, ValueError, OverflowError):
-        return None
-    return seconds if math.isfinite(seconds) else None
-
-
 class BasePlugin(ABC):
     """
     Base class that all plugins must inherit from.

src/plugin_system/plugin_local_files.py

@@ -0,0 +1,204 @@
+"""
+Files a plugin writes beside itself at runtime, which an update must keep.
+
+A store update replaces a plugin's directory with a fresh download and then
+deletes the old copy. Anything the plugin created there -- OAuth tokens, a
+client-secrets file, a PKCE verifier, cached state -- is in no release, so the
+fresh download does not contain it and deleting the old copy destroys it. On
+2026-10-04 updating calendar 1.2.9 -> 1.2.12 that way deleted its
+``token.pickle`` and ``credentials.json``, and the calendar stopped until they
+were restored from a backup.
+
+What counts as "the plugin's own local file" is the union of:
+
+* :data:`KNOWN_STATE_PATTERNS` -- secret and state files plugins are known to
+  write, kept even when a plugin forgot to gitignore them; and
+* whatever the plugin's own ``.gitignore`` (old copy or new) excludes. A file
+  the author ignores is by definition not part of a release.
+
+A file the new release ships is never overwritten: tracked content wins. Byte
+code (``__pycache__``, ``*.pyc``) and ``.git`` are never carried, since they
+belong to the old code rather than to the user.
+"""
+
+from __future__ import annotations
+
+import fnmatch
+import os
+import re
+import shutil
+from pathlib import Path
+from typing import Iterable, List, Optional, Pattern, Tuple
+
+__all__ = [
+    'KNOWN_STATE_PATTERNS',
+    'carry_over_local_files',
+    'is_known_state_file',
+    'local_files_to_keep',
+]
+
+# Basename globs. Kept even when the plugin's .gitignore does not list them.
+KNOWN_STATE_PATTERNS: Tuple[str, ...] = (
+    'token.pickle',
+    '*.pickle',
+    'token.json',
+    'credentials.json',
+    'config_secrets.json',
+    '.pkce_code_verifier',
+)
+
+_NEVER_CARRY_DIRS = frozenset({'.git', '__pycache__'})
+_NEVER_CARRY_SUFFIXES = ('.pyc', '.pyo')
+
+
+def is_known_state_file(rel_path: str) -> bool:
+    """True when ``rel_path``'s basename is a known secret/state file."""
+    name = rel_path.replace('\\', '/').rsplit('/', 1)[-1]
+    return any(fnmatch.fnmatchcase(name, p) for p in KNOWN_STATE_PATTERNS)
+
+
+class _GitIgnore:
+    """The subset of gitignore semantics plugin .gitignore files use.
+
+    Supports comments, ``!`` negation (last match wins), a trailing ``/`` for
+    directory-only patterns, anchoring by a leading or embedded ``/``, ``*``,
+    ``?``, ``[...]`` and ``**``. As in git, a file under an ignored directory
+    is ignored regardless of later negations.
+    """
+
+    def __init__(self, lines: Iterable[str]):
+        self._rules: List[Tuple[Pattern[str], bool, bool]] = []
+        for raw in lines:
+            line = raw.rstrip('\n').rstrip()
+            if not line or line.startswith('#'):
+                continue
+            negate = line.startswith('!')
+            if negate:
+                line = line[1:]
+            elif line.startswith('\\'):
+                line = line[1:]
+            dir_only = line.endswith('/')
+            line = line.rstrip('/')
+            if not line:
+                continue
+            anchored = '/' in line
+            line = line.lstrip('/')
+            body = self._translate(line)
+            regex = body if anchored else r'(?:.*/)?' + body
+            self._rules.append((re.compile(r'\A' + regex + r'\Z'), negate, dir_only))
+
+    @staticmethod
+    def _translate(pattern: str) -> str:
+        out, i, n = [], 0, len(pattern)
+        while i < n:
+            if pattern.startswith('**/', i):
+                out.append(r'(?:.*/)?')
+                i += 3
+            elif pattern.startswith('/**', i) and i + 3 == n:
+                out.append(r'/.*')
+                i += 3
+            elif pattern.startswith('**', i):
+                out.append(r'.*')
+                i += 2
+            elif pattern[i] == '*':
+                out.append(r'[^/]*')
+                i += 1
+            elif pattern[i] == '?':
+                out.append(r'[^/]')
+                i += 1
+            elif pattern[i] == '[':
+                end = pattern.find(']', i + 1)
+                if end == -1:
+                    out.append(re.escape('['))
+                    i += 1
+                else:
+                    cls = pattern[i + 1:end]
+                    if cls.startswith('!'):
+                        cls = '^' + cls[1:]
+                    out.append('[' + cls.replace('\\', '\\\\') + ']')
+                    i = end + 1
+            else:
+                out.append(re.escape(pattern[i]))
+                i += 1
+        return ''.join(out)
+
+    def _decide(self, rel: str, is_dir: bool) -> Optional[bool]:
+        verdict = None
+        for regex, negate, dir_only in self._rules:
+            if dir_only and not is_dir:
+                continue
+            if regex.match(rel):
+                verdict = not negate
+        return verdict
+
+    def ignores(self, rel_path: str) -> bool:
+        if not self._rules:
+            return False
+        parts = rel_path.replace('\\', '/').split('/')
+        for depth in range(1, len(parts)):
+            if self._decide('/'.join(parts[:depth]), True):
+                return True
+        return bool(self._decide('/'.join(parts), False))
+
+
+def _read_gitignore(plugin_dir: Path) -> List[str]:
+    try:
+        return (plugin_dir / '.gitignore').read_text(
+            encoding='utf-8', errors='replace').splitlines()
+    except OSError:
+        return []
+
+
+def local_files_to_keep(old_dir: Path, new_dir: Path) -> List[str]:
+    """Relative paths (``/``-separated) in ``old_dir`` to copy into ``new_dir``.
+
+    Regular files only; symlinks and anything the new release already ships
+    are skipped.
+    """
+    old_dir, new_dir = Path(old_dir), Path(new_dir)
+    ignore = _GitIgnore(_read_gitignore(old_dir) + _read_gitignore(new_dir))
+    keep: List[str] = []
+    for root, dirs, files in os.walk(old_dir):
+        dirs[:] = sorted(d for d in dirs if d not in _NEVER_CARRY_DIRS
+                         and not os.path.islink(os.path.join(root, d)))
+        rel_root = os.path.relpath(root, old_dir)
+        for name in sorted(files):
+            if name.endswith(_NEVER_CARRY_SUFFIXES):
+                continue
+            full = os.path.join(root, name)
+            if os.path.islink(full) or not os.path.isfile(full):
+                continue
+            rel = name if rel_root == '.' else f"{rel_root}/{name}".replace('\\', '/')
+            if not (is_known_state_file(rel) or ignore.ignores(rel)):
+                continue
+            if os.path.lexists(new_dir / rel):
+                continue
+            keep.append(rel)
+    return keep
+
+
+def carry_over_local_files(
+    old_dir: Path, new_dir: Path
+) -> Tuple[List[str], List[Tuple[str, str]]]:
+    """Copy the plugin's local files from ``old_dir`` into ``new_dir``.
+
+    Copies rather than moves, so ``old_dir`` stays a complete copy until the
+    caller deletes it. Returns ``(copied, failed)`` where ``failed`` pairs a
+    relative path with the error; the caller should keep ``old_dir`` when
+    anything failed.
+    """
+    copied: List[str] = []
+    failed: List[Tuple[str, str]] = []
+    try:
+        candidates = local_files_to_keep(old_dir, new_dir)
+    except OSError as e:
+        return copied, [('.', str(e))]
+    for rel in candidates:
+        dest = Path(new_dir) / rel
+        try:
+            dest.parent.mkdir(parents=True, exist_ok=True)
+            shutil.copy2(Path(old_dir) / rel, dest)
+            copied.append(rel)
+        except OSError as e:
+            failed.append((rel, str(e)))
+    return copied, failed

src/plugin_system/store_install.py

@@ -22,6 +22,7 @@ from src.plugin_system.plugin_loader import (
     contained_plugin_dir, requirements_to_install,
 )
 from src.plugin_system.plugin_dirs import BACKUP_MARKER
+from src.plugin_system.plugin_local_files import carry_over_local_files
 from src.plugin_system.repo_urls import (
     USER_AGENT, github_api_headers, github_owner_repo, normalize_repo_url,
 )
@@ -92,7 +93,9 @@ class _InstallMixin:
                 raise
 
             if installed:
-                self._discard_backup(plugin_id, backup_path, "install")
+                self._discard_backup(
+                    plugin_id, backup_path, "install",
+                    new_path=self._existing_install(plugin_id) or plugin_path)
                 return True
 
             self._restore_backup(plugin_id, plugin_path, backup_path, "Install")
@@ -133,8 +136,33 @@ class _InstallMixin:
             return f"could not set aside {plugin_path}: {e}"
         return None
 
-    def _discard_backup(self, plugin_id: str, backup_path: Path, action: str) -> None:
-        """Remove the set-aside copy after a successful (re)install."""
+    def _discard_backup(
+        self, plugin_id: str, backup_path: Path, action: str,
+        new_path: Optional[Path] = None,
+    ) -> None:
+        """Remove the set-aside copy after a successful (re)install.
+
+        With ``new_path`` (where the new copy landed), first carries the
+        plugin's own runtime files -- OAuth tokens, client secrets, anything
+        its .gitignore excludes -- from the old copy into the new one: no
+        release contains them, so deleting the old copy would destroy them.
+        See src/plugin_system/plugin_local_files.py. If any could not be
+        copied the old copy is kept, so nothing is lost.
+        """
+        if new_path is not None and new_path.is_dir():
+            copied, failed = carry_over_local_files(backup_path, new_path)
+            if copied:
+                self.logger.info(
+                    "Kept %d local file(s) of %s across the %s: %s",
+                    len(copied), plugin_id, action, ", ".join(copied))
+            if failed:
+                self.logger.error(
+                    "Could not carry %s's local files into the new copy (%s); "
+                    "the previous copy is kept at %s -- copy them back by hand",
+                    plugin_id,
+                    "; ".join(f"{rel}: {err}" for rel, err in failed),
+                    backup_path)
+                return
         if not self._safe_remove_directory(backup_path):
             self.logger.warning(
                 "%s of %s succeeded but the previous copy at %s could not be "
@@ -542,7 +570,8 @@ class _InstallMixin:
                     raise
                 temp_dir = None  # Prevent cleanup since we moved it
                 if backup_path is not None:
-                    self._discard_backup(plugin_id, backup_path, "install")
+                    self._discard_backup(
+                        plugin_id, backup_path, "install", new_path=final_path)
 
             # Install dependencies
             self._install_dependencies(final_path)

src/plugin_system/store_update.py

@@ -10,6 +10,9 @@ import subprocess  # nosec B404 - list-form argv only, no shell  # nosemgrep
 from pathlib import Path
 from typing import Dict, Optional, Tuple
 from src.plugin_system.plugin_dirs import BACKUP_MARKER
+from src.plugin_system.plugin_local_files import (
+    KNOWN_STATE_PATTERNS, is_known_state_file,
+)
 from src.plugin_system.repo_urls import same_repo
 
 
@@ -302,7 +305,11 @@ class _UpdateMixin:
                 installed = False
 
             if installed:
-                self._discard_backup(plugin_id, backup_path, "update")
+                # install_plugin may land the new copy under the manifest id
+                # rather than the old directory name.
+                self._discard_backup(
+                    plugin_id, backup_path, "update",
+                    new_path=self._existing_install(plugin_id) or plugin_path)
                 return True
 
             # Bad network, registry error...: the user keeps a working plugin.
@@ -509,8 +516,12 @@ class _UpdateMixin:
                             for line in untracked_result.stdout.strip().split('\n'):
                                 if line.startswith('??'):
                                     # Untracked file
-                                    file_path = line[3:].strip()
-                                    untracked_files.append(file_path)
+                                    file_path = line[3:].strip().strip('"')
+                                    # Tokens and secrets stay out of the
+                                    # stash (see below), so they alone are
+                                    # not a reason to stash.
+                                    if not is_known_state_file(file_path):
+                                        untracked_files.append(file_path)
                         
                         # Check for tracked file changes
                         status_result = subprocess.run(
@@ -537,9 +548,17 @@ class _UpdateMixin:
                     if has_changes:
                         self.logger.info(f"Stashing local changes in {plugin_id} before update")
                         try:
-                            # Use -u to include untracked files in stash
+                            # Use -u to include untracked files in stash --
+                            # except the plugin's tokens and secrets, which a
+                            # repo may have forgotten to gitignore. The stash
+                            # is never popped, so a stashed token.pickle would
+                            # vanish from the plugin and break it.
+                            stash_cmd = (
+                                ['git', '-C', str(plugin_path), 'stash', 'push', '-u',
+                                 '-m', f'LEDMatrix auto-stash before update {plugin_id}', '--', '.']
+                                + [f':(exclude,glob)**/{p}' for p in KNOWN_STATE_PATTERNS])
                             stash_result = subprocess.run(
-                                ['git', '-C', str(plugin_path), 'stash', 'push', '-u', '-m', f'LEDMatrix auto-stash before update {plugin_id}'],
+                                stash_cmd,
                                 capture_output=True,
                                 text=True,
                                 timeout=30,

src/vegas_mode/coordinator.py

@@ -18,11 +18,10 @@ import math
 import sys
 import time
 import threading
-from typing import Optional, Dict, Any, FrozenSet, List, Callable, TYPE_CHECKING
+from typing import Optional, Dict, Any, List, Callable, TYPE_CHECKING
 
 from src import display_watchdog
 from src.common import render_gate
-from src.plugin_system.base_plugin import finite_seconds
 from src.vegas_mode.config import VegasModeConfig
 from src.vegas_mode.elements import LiveEpochs
 from src.vegas_mode.plugin_adapter import PluginAdapter
@@ -54,14 +53,6 @@ _FPS_HEARTBEAT_INTERVAL = 300.0
 #: every plugin. Game state doesn't change within a quarter second.
 _LIVE_PRIORITY_CHECK_INTERVAL = 0.25
 
-#: Seconds a static pause shows a plugin whose display duration can't be
-#: used, as long as the rotation shows it: 30 when get_display_duration()
-#: raises or answers something that is not a number
-#: (DisplayController._get_display_duration), 15 when it answers a number at
-#: or below zero (DisplayController._resolve_durations).
-_UNREADABLE_DURATION = 30.0
-_NOT_POSITIVE_DURATION = 15.0
-
 
 def _percentile(ordered: List[float], fraction: float) -> float:
     """Nearest-rank percentile of an already-sorted list.
@@ -101,9 +92,6 @@ class VegasModeCoordinator:
     _live_reason: Optional[str] = None
     # Set only while Vegas has changed the GIL switch interval; read with getattr.
     _saved_switch_interval: Optional[float]
-    #: Plugins already warned about a display duration the pause can't use,
-    #: so a bad setting logs once, not at every turn. Replaced, not mutated.
-    _duration_warned: FrozenSet[str] = frozenset()
 
     def __init__(
         self,
@@ -1022,7 +1010,7 @@ class VegasModeCoordinator:
             # Wait for the plugin's display duration. Monotonic, like the
             # iteration clock: an NTP step on an RTC-less Pi would otherwise
             # end the pause at once or stretch it by the correction.
-            duration = self._static_pause_duration(plugin)
+            duration = plugin.get_display_duration()
             start = time.monotonic()
 
             while time.monotonic() - start < duration:
@@ -1058,42 +1046,6 @@ class VegasModeCoordinator:
 
         return True
 
-    def _static_pause_duration(self, plugin: 'BasePlugin') -> float:
-        """Seconds a static pause shows ``plugin``: its display duration,
-        read the way the rotation reads it.
-
-        Several plugins return their display_duration setting straight from
-        config.json, so one saved as "20" or null came back as a string or
-        None; comparing it with the clock raised, and the pause's broad
-        except ended the pause at every one of the plugin's turns. inf
-        paused until something interrupted it, and NaN, False, 0 or a
-        negative number ended the pause at once. A numeric string counts
-        (finite_seconds); anything else, or a raise, gets
-        _UNREADABLE_DURATION, and a number at or below zero
-        _NOT_POSITIVE_DURATION, logged once per plugin.
-        """
-        try:
-            value = plugin.get_display_duration()
-        except Exception as err:  # pylint: disable=broad-except
-            problem = f"get_display_duration() raised {type(err).__name__}: {err}"
-            seconds = _UNREADABLE_DURATION
-        else:
-            seconds = finite_seconds(value)
-            if seconds is not None and seconds > 0:
-                return seconds
-            if seconds is None:
-                problem = f"display duration {value!r} is not a number"
-                seconds = _UNREADABLE_DURATION
-            else:
-                problem = f"display duration {value!r} is not above zero"
-                seconds = _NOT_POSITIVE_DURATION
-        plugin_id = plugin.plugin_id
-        if plugin_id not in self._duration_warned:
-            self._duration_warned = self._duration_warned | {plugin_id}
-            logger.warning("[%s] %s; its static pause lasts %.0fs (logged once)",
-                           plugin_id, problem, seconds)
-        return seconds
-
     def _end_static_pause(self) -> None:
         """End static pause and restore scroll state."""
         should_resume_scrolling = False

test/_run_loop_harness.py

@@ -724,11 +724,14 @@ class RunLoopHarness:
         self.clock.at(t, post)
 
     def restore_on_demand(self, plugin_id: str, mode: Optional[str] = None,
-                          duration: Optional[float] = None, pinned: bool = False):
+                          duration: Optional[float] = None, pinned: bool = False,
+                          named_mode: Optional[str] = None):
         """Start with an on-demand session resumed from the cache, as after
         a restart: the state _select_startup_plugins restores, then
-        _populate_on_demand_modes_from_plugin, as __init__ calls it."""
+        _populate_on_demand_modes_from_plugin, as __init__ calls it. A
+        session that cannot resume is logged as ``on-demand-error``."""
         dc = self.controller
+        dc._on_demand_named_mode = named_mode
         dc.on_demand_active = True
         dc.on_demand_plugin_id = plugin_id
         dc.on_demand_mode = mode
@@ -739,6 +742,8 @@ class RunLoopHarness:
         dc.on_demand_status = 'active'
         dc.on_demand_schedule_override = True
         dc._populate_on_demand_modes_from_plugin()
+        if dc.on_demand_status == 'error':
+            self.log("on-demand-error", dc.on_demand_last_error)
 
     def wifi_message(self, t: float, message: str, duration: float = 5):
         def write():

test/fixtures/run_loop_golden/on_demand_named_live.json

@@ -0,0 +1,29 @@
+{
+  "screens": [
+    [0.0, "clock", 5.0, "on-demand-start", 6, false],
+    [5.0, "sports_live", 15.0, "duration", 15, true],
+    [20.0, "sports_recent", 15.0, "duration", 15, true],
+    [35.0, "sports_upcoming", 5.0, "on-demand-requested-stop", 6, true],
+    [40.0, "clock", 20.0, "duration", 20, true],
+    [60.0, "sports_live", 15.0, "display-false", 11, true],
+    [75.0, "sports_recent", 15.0, "duration", 15, true],
+    [90.0, "sports_upcoming", 10.0, "on-demand-start", 11, true],
+    [100.0, "sports_live", 0.0, "empty", 1, true],
+    [100.0, "sports_recent", 15.0, "duration", 15, true],
+    [115.0, "sports_upcoming", 15.0, "duration", 15, true],
+    [130.0, "sports_live", 0.0, "empty", 1, true],
+    [130.0, "sports_recent", 10.0, "on-demand-requested-stop", 11, true],
+    [140.0, "sports_upcoming", 15.0, "duration", 15, true],
+    [155.0, "clock", 5.0, "horizon", 5, true]
+  ],
+  "events": [
+    [5.0, "request", "start:n1"],
+    [5.0, "on-demand-start", "sports"],
+    [40.0, "request", "stop:n2"],
+    [40.0, "on-demand-requested-stop"],
+    [100.0, "request", "start:n3"],
+    [100.0, "on-demand-start", "sports"],
+    [140.0, "request", "stop:n4"],
+    [140.0, "on-demand-requested-stop"]
+  ]
+}

test/fixtures/run_loop_golden/on_demand_restore_failed.json

@@ -0,0 +1,10 @@
+{
+  "screens": [
+    [0.0, "clock", 20.0, "duration", 20, false],
+    [20.0, "weather", 20.0, "duration", 20, true],
+    [40.0, "clock", 20.0, "horizon", 20, true]
+  ],
+  "events": [
+    [0.0, "on-demand-error", "restore-failed"]
+  ]
+}

test/test_api_v3_install_restart_installed_id.py

@@ -1,123 +0,0 @@
-"""POST /plugins/install asks for a restart by the id the plugin installed as.
-
-A store install needs a display restart when config.json already enables the
-plugin (a reinstall, or a config carried over): the display loads a plugin
-when its ``enabled`` flag changes, and this flag did not. The route read the
-flag under the registry id it was given. An aliased entry installs under
-another id -- ``weather`` installs a directory whose manifest declares
-``ledmatrix-weather``, and its config section is ``ledmatrix-weather`` -- so
-reinstalling an enabled Weather never reported that a restart was needed,
-and the display kept running the old copy.
-"""
-
-import json
-from unittest.mock import MagicMock
-
-import pytest
-
-from test._api_v3_test_helpers import api_v3_client, api_v3_module  # noqa: F401
-
-INSTALL = "/api/v3/plugins/install"
-
-
-@pytest.fixture
-def store(api_v3_module, tmp_path):
-    """The store installs registry entry ``weather`` as ``installed_id``."""
-    manager = api_v3_module.api_v3.plugin_store_manager
-    manager.install_plugin.return_value = True
-    manager.get_registry_info.return_value = None
-    manager._find_plugin_path.return_value = None
-
-    def installs_as(installed_id):
-        path = tmp_path / installed_id
-        path.mkdir()
-        (path / "manifest.json").write_text(json.dumps({"id": installed_id}),
-                                            encoding="utf-8")
-        manager._find_plugin_path.side_effect = (
-            lambda pid: path if pid == "weather" else None)
-
-    manager.installs_as = installs_as
-    return manager
-
-
-@pytest.fixture
-def config(api_v3_module):
-    """config.json with an ``enabled`` flag for each plugin id given."""
-    def sections(enabled):
-        api_v3_module.api_v3.config_manager.load_config.return_value = {
-            plugin_id: {"enabled": flag} for plugin_id, flag in enabled.items()}
-    return sections
-
-
-@pytest.fixture
-def queued(api_v3_module):
-    queue = MagicMock()
-
-    def enqueue(operation_type, plugin_id, operation_callback=None):
-        queue.callback_result = operation_callback(MagicMock())
-        return "op-1"
-
-    queue.enqueue_operation.side_effect = enqueue
-    api_v3_module.api_v3.operation_queue = queue
-    return queue
-
-
-def _direct(client):
-    return client.post(INSTALL, json={"plugin_id": "weather"}).get_json()
-
-
-def _queued(client, queue):
-    client.post(INSTALL, json={"plugin_id": "weather"})
-    return queue.callback_result
-
-
-class TestDirectInstall:
-    def test_an_aliased_install_enabled_under_its_installed_id_asks_for_a_restart(
-            self, api_v3_client, store, config):
-        store.installs_as("ledmatrix-weather")
-        config({"ledmatrix-weather": True})
-        body = _direct(api_v3_client)
-        assert body["status"] == "success"
-        assert body["restart_required"] is True
-        assert body["restart_message"]
-
-    def test_an_enabled_section_under_the_registry_id_alone_does_not(
-            self, api_v3_client, store, config):
-        """The display knows the plugin as ledmatrix-weather; nothing runs
-        under a section called weather."""
-        store.installs_as("ledmatrix-weather")
-        config({"weather": True})
-        assert _direct(api_v3_client)["restart_required"] is False
-
-    def test_an_aliased_install_that_is_not_enabled_needs_no_restart(
-            self, api_v3_client, store, config):
-        store.installs_as("ledmatrix-weather")
-        config({"ledmatrix-weather": False})
-        assert _direct(api_v3_client)["restart_required"] is False
-
-    def test_an_install_under_its_own_id_is_unchanged(self, api_v3_client, store, config):
-        store.installs_as("weather")
-        config({"weather": True})
-        assert _direct(api_v3_client)["restart_required"] is True
-
-    def test_an_install_that_cannot_be_found_uses_the_requested_id(
-            self, api_v3_client, store, config):
-        config({"weather": True})
-        assert _direct(api_v3_client)["restart_required"] is True
-
-
-class TestQueuedInstall:
-    def test_an_aliased_install_enabled_under_its_installed_id_asks_for_a_restart(
-            self, api_v3_client, store, config, queued):
-        store.installs_as("ledmatrix-weather")
-        config({"ledmatrix-weather": True})
-        result = _queued(api_v3_client, queued)
-        assert result["success"] is True
-        assert result["restart_required"] is True
-        assert result["restart_message"]
-
-    def test_an_enabled_section_under_the_registry_id_alone_does_not(
-            self, api_v3_client, store, config, queued):
-        store.installs_as("ledmatrix-weather")
-        config({"weather": True})
-        assert _queued(api_v3_client, queued)["restart_required"] is False

test/test_api_v3_status_accuracy.py

@@ -0,0 +1,326 @@
+"""Four web answers that disagreed with the rig they describe (found on ledpi).
+
+1. POST /config/schedule refused the schedule GET returns on a fresh install
+   (config.template.json: per-day, every day off, schedule disabled) with
+   "At least one day must be enabled", as did /config/dim-schedule. A
+   disabled schedule needs no enabled day.
+2. A brightness-only POST /config/main answered ``restart_required: true``,
+   though the display applies brightness live (brightness.set over the
+   socket, and the config watcher). The flag now says whether anything
+   changed that the running display does not pick up by itself.
+3. /health stayed "healthy" with the display service stopped: only the
+   sub-checks changed. Service inactive, no socket and no live heartbeat
+   is now ``display_loop: stopped`` and "degraded".
+4. /display/current-status kept answering ``is_display_active: true`` from
+   the cache for up to 120 s after the display stopped. With no socket and
+   no live heartbeat it is now unknown.
+"""
+
+import copy
+import json
+import os
+import sys
+import time
+from pathlib import Path
+from unittest.mock import patch
+
+import pytest
+
+sys.path.insert(0, str(Path(__file__).parent.parent))
+
+from test._api_v3_test_helpers import api_v3_client, api_v3_module  # noqa: F401,E402
+
+from src import display_watchdog  # noqa: E402
+from src.ipc import client as control_client  # noqa: E402
+from web_interface import display_state  # noqa: E402
+
+REPO = Path(__file__).resolve().parent.parent
+TEMPLATE = json.loads((REPO / 'config' / 'config.template.json').read_text(encoding='utf-8'))
+
+
+@pytest.fixture
+def store(api_v3_module, monkeypatch):
+    state = {'config': {}, 'saves': 0}
+    api_v3_module.api_v3.config_manager.load_config.side_effect = \
+        lambda *a, **k: copy.deepcopy(state['config'])
+
+    def fake_save(_manager, config, **_kwargs):
+        state['config'] = copy.deepcopy(config)
+        state['saves'] += 1
+        return True, ''
+
+    monkeypatch.setattr(api_v3_module, '_save_config_atomic', fake_save)
+    return state
+
+
+# --- 1. schedules ---------------------------------------------------------------
+
+SCHEDULE_ROUTES = [('/api/v3/config/schedule', 'schedule'),
+                   ('/api/v3/config/dim-schedule', 'dim_schedule')]
+
+
+@pytest.mark.parametrize('route,section', SCHEDULE_ROUTES)
+def test_the_templates_disabled_per_day_schedule_saves_back(api_v3_client, store,
+                                                            route, section):
+    stored = copy.deepcopy(TEMPLATE[section])
+    stored['mode'] = 'per-day'
+    assert stored['enabled'] is False
+    assert not any(day['enabled'] for day in stored['days'].values())
+    store['config'] = {section: copy.deepcopy(stored)}
+
+    read = api_v3_client.get(route).get_json()['data']
+    resp = api_v3_client.post(route, json=read)
+
+    assert resp.status_code == 200, resp.get_json()
+    saved = store['config'][section]
+    assert saved['enabled'] is False and saved['mode'] == 'per-day'
+    # The disabled days keep their times: switching one on finds them.
+    assert saved['days'] == stored['days']
+
+
+@pytest.mark.parametrize('route,section', SCHEDULE_ROUTES)
+def test_an_enabled_per_day_schedule_still_needs_a_day(api_v3_client, store, route, section):
+    body = copy.deepcopy(TEMPLATE[section])
+    body.update(enabled=True, mode='per-day')
+    resp = api_v3_client.post(route, json=body)
+    assert resp.status_code == 400
+    assert 'At least one day must be enabled' in resp.get_json()['message']
+    assert store['saves'] == 0
+
+
+@pytest.mark.parametrize('route', [r for r, _ in SCHEDULE_ROUTES])
+def test_the_pickers_form_post_with_every_day_off_saves(api_v3_client, store, route):
+    """What schedule-picker.js posts: flat hidden inputs, booleans as strings,
+    times for every day."""
+    body = {'enabled': 'false', 'mode': 'per_day', 'start_time': '07:00', 'end_time': '23:00'}
+    for day in ('monday', 'tuesday', 'wednesday', 'thursday', 'friday', 'saturday', 'sunday'):
+        body.update({f'{day}_enabled': 'false', f'{day}_start': '06:30', f'{day}_end': '22:15'})
+    resp = api_v3_client.post(route, json=body)
+    assert resp.status_code == 200, resp.get_json()
+
+
+def test_an_invalid_time_on_a_disabled_day_is_dropped_not_refused(api_v3_client, store):
+    body = {'enabled': False, 'mode': 'per-day',
+            'days': {'monday': {'enabled': False, 'start_time': 'soon', 'end_time': '22:00'}}}
+    resp = api_v3_client.post('/api/v3/config/schedule', json=body)
+    assert resp.status_code == 200, resp.get_json()
+    assert store['config']['schedule']['days']['monday'] == {'enabled': False,
+                                                             'end_time': '22:00'}
+
+
+# --- 2. restart_required on /config/main ------------------------------------------
+
+STORED_MAIN = {
+    'timezone': 'America/Chicago',
+    'display': {
+        'hardware': {'rows': 32, 'cols': 64, 'chain_length': 2, 'brightness': 90,
+                     'disable_hardware_pulsing': False, 'inverse_colors': False,
+                     'show_refresh_rate': False},
+        'runtime': {'gpio_slowdown': 4},
+        'display_durations': {'clock': 15},
+        'use_short_date_format': False,
+    },
+}
+
+
+@pytest.fixture
+def main_store(store):
+    store['config'] = copy.deepcopy(STORED_MAIN)
+    return store
+
+
+def _save_main(client, body):
+    with patch('web_interface.blueprints.api_v3.control_client.brightness_set',
+               side_effect=control_client.ControlError('no_socket', 'x')):
+        resp = client.post('/api/v3/config/main', data=json.dumps(body),
+                           content_type='application/json')
+    assert resp.status_code == 200, resp.get_json()
+    return resp.get_json()
+
+
+def test_a_brightness_only_save_needs_no_restart(api_v3_client, main_store):
+    body = _save_main(api_v3_client, {'brightness': 40})
+    assert main_store['config']['display']['hardware']['brightness'] == 40
+    assert body['restart_required'] is False
+
+
+def test_a_brightness_save_on_a_config_without_a_display_section(api_v3_client, store):
+    """The route creates display.hardware and display.runtime on the way;
+    empty sections are not a change."""
+    store['config'] = {}
+    assert _save_main(api_v3_client, {'brightness': 40})['restart_required'] is False
+
+
+def test_the_display_form_with_only_brightness_changed_needs_no_restart(api_v3_client,
+                                                                       main_store):
+    hw = STORED_MAIN['display']['hardware']
+    body = {'__form_section': 'display', 'rows': 32, 'cols': 64, 'chain_length': 2,
+            'brightness': 55, 'gpio_slowdown': 4}
+    body.update({k: 'on' for k in ('disable_hardware_pulsing', 'inverse_colors',
+                                   'show_refresh_rate') if hw[k]})
+    assert _save_main(api_v3_client, body)['restart_required'] is False
+
+
+def test_a_mode_duration_needs_no_restart(api_v3_client, main_store):
+    body = _save_main(api_v3_client, {'duration__clock': 40})
+    assert main_store['config']['display']['display_durations']['clock'] == 40
+    assert body['restart_required'] is False
+
+
+@pytest.mark.parametrize('change', [{'rows': 64}, {'brightness': 40, 'chain_length': 3},
+                                    {'gpio_slowdown': 2}, {'timezone': 'UTC'}])
+def test_a_setting_the_display_reads_at_startup_still_needs_one(api_v3_client, main_store,
+                                                                change):
+    assert _save_main(api_v3_client, change)['restart_required'] is True
+
+
+def test_restart_needed_compares_leaves():
+    from web_interface.blueprints.api_v3.config import restart_needed
+    before = {'display': {'hardware': {'brightness': 90, 'rows': 32}}}
+    assert not restart_needed(before, copy.deepcopy(before))
+    assert not restart_needed(before, {'display': {'hardware': {'brightness': 10, 'rows': 32},
+                                                   'runtime': {}}})
+    assert restart_needed(before, {'display': {'hardware': {'brightness': 90}}})  # removed
+    assert not restart_needed({}, {'clock': {'enabled': True}}, live_paths=[('clock',)])
+    assert restart_needed({}, {'clockwork': {'enabled': True}}, live_paths=[('clock',)])
+
+
+# --- 3 and 4. a stopped display -----------------------------------------------------
+
+@pytest.fixture
+def no_display(monkeypatch, tmp_path):
+    """A Pi whose display service has stopped: the socket is expected here
+    but does not answer, and systemd took the heartbeat's directory away."""
+    monkeypatch.setattr(display_state, 'socket_supported', lambda: True)
+    monkeypatch.setattr(display_state, 'client_socket_paths', lambda: [str(tmp_path / 'gone')])
+    monkeypatch.setattr(display_state, 'read_state', lambda: None)
+    path = tmp_path / 'display-heartbeat.json'
+    monkeypatch.setattr(display_watchdog, 'HEARTBEAT_PATH', str(path))
+
+    def beat(age, pid=None):
+        path.write_text(json.dumps({'pid': os.getpid() if pid is None else pid,
+                                    'mono': time.monotonic() - age,
+                                    'wall': time.time() - age}))
+    return beat
+
+
+@pytest.fixture
+def service(monkeypatch):
+    status = {'active': False, 'returncode': 3, 'stdout': 'inactive', 'stderr': ''}
+    monkeypatch.setattr('web_interface.blueprints.api_v3.misc._get_display_service_status',
+                        lambda: dict(status))
+    return status
+
+
+@pytest.fixture
+def fresh_preview(tmp_path, monkeypatch):
+    """The preview frame the display left behind, under 60 s old: on its own
+    it kept the hardware check "connected"."""
+    from web_interface import display_preview
+    snapshot = tmp_path / 'preview.png'
+    snapshot.write_bytes(b'png')
+    monkeypatch.setattr(display_preview, 'SNAPSHOT_PATH', str(snapshot))
+
+
+def _health(client):
+    resp = client.get('/api/v3/health')
+    assert resp.status_code == 200, resp.get_json()
+    return resp.get_json()['data']
+
+
+class TestHealth:
+    def test_a_stopped_display_service_is_degraded(self, api_v3_client, no_display, service,
+                                                   fresh_preview):
+        data = _health(api_v3_client)
+        assert data['services']['display_service']['status'] == 'inactive'
+        assert data['checks']['display_loop']['status'] == 'stopped'
+        assert data['status'] == 'degraded'
+
+    def test_a_service_still_starting_is_not(self, api_v3_client, no_display, service,
+                                             fresh_preview):
+        """Active, before its socket and first heartbeat: not stopped."""
+        service.update(active=True, stdout='active', returncode=0)
+        data = _health(api_v3_client)
+        assert data['checks']['display_loop']['status'] == 'not_reported'
+        assert data['status'] == 'healthy'
+
+    def test_a_display_run_by_hand_is_not_stopped(self, api_v3_client, no_display, service,
+                                                  fresh_preview):
+        """The service is off but a display process beats (sudo python3 run.py)."""
+        no_display(age=2)
+        data = _health(api_v3_client)
+        assert data['checks']['display_loop']['status'] == 'running'
+        assert data['status'] == 'healthy'
+
+    @pytest.mark.parametrize('platform', ['no_unix_sockets', 'socket_off'])
+    def test_without_a_socket_to_expect_nothing_changes(self, api_v3_client, no_display,
+                                                        service, fresh_preview, monkeypatch,
+                                                        platform):
+        """Windows and the dev server (no systemd unit), or the socket
+        deliberately off: no heartbeat is no signal, as before."""
+        if platform == 'no_unix_sockets':
+            monkeypatch.setattr(display_state, 'socket_supported', lambda: False)
+        else:
+            monkeypatch.setattr(display_state, 'client_socket_paths', lambda: [])
+        service.update(returncode=-1, stdout='', stderr='systemctl not found')
+        data = _health(api_v3_client)
+        assert data['checks']['display_loop']['status'] == 'not_reported'
+        assert data['status'] == 'healthy'
+
+    def test_the_status_only_answer_says_degraded(self, api_v3_client, no_display, service,
+                                                  fresh_preview, monkeypatch):
+        monkeypatch.setattr('web_interface.blueprints.api_v3.misc.request_is_authenticated',
+                            lambda: False)
+        resp = api_v3_client.get('/api/v3/health')
+        assert resp.get_json()['data'] == {'status': 'degraded'}
+
+
+class TestCurrentStatus:
+    CACHED = {'mode': 'clock', 'plugin_id': 'clock', 'is_display_active': True,
+              'on_demand_active': False, 'last_updated': None}
+
+    @pytest.fixture
+    def cached(self, api_v3_module):
+        entry = dict(self.CACHED, last_updated=time.time() - 30)
+        cache = api_v3_module.api_v3.cache_manager
+        cache.get.side_effect = lambda key, *a, **kw: (
+            dict(entry) if key == 'display_current_state' else None)
+        return entry
+
+    def _status(self, client):
+        resp = client.get('/api/v3/display/current-status')
+        assert resp.status_code == 200
+        return resp.get_json()['data']
+
+    def test_a_stopped_display_is_not_reported_active(self, api_v3_client, no_display, cached):
+        data = self._status(api_v3_client)
+        assert not data.get('is_display_active')
+        assert data['mode'] is None and data['last_updated'] is None
+        assert data['source'] == 'cache'
+
+    def test_a_stale_heartbeat_is_not_active_either(self, api_v3_client, no_display, cached):
+        no_display(age=display_watchdog.HEARTBEAT_STALE_SECONDS + 5)
+        assert self._status(api_v3_client)['mode'] is None
+
+    @pytest.mark.skipif(os.name != 'posix', reason='process_exists answers only on POSIX')
+    def test_a_heartbeat_from_a_dead_process_is_not_active(self, api_v3_client, no_display,
+                                                           cached):
+        no_display(age=1, pid=2 ** 22 + 12345)
+        assert self._status(api_v3_client)['mode'] is None
+
+    def test_a_live_heartbeat_without_a_socket_reads_the_cache(self, api_v3_client,
+                                                               no_display, cached):
+        """An older display with no socket, still running."""
+        no_display(age=2)
+        data = self._status(api_v3_client)
+        assert data['mode'] == 'clock' and data['is_display_active'] is True
+
+    @pytest.mark.parametrize('platform', ['no_unix_sockets', 'socket_off'])
+    def test_without_a_socket_to_expect_the_cache_answers(self, api_v3_client, no_display,
+                                                          cached, monkeypatch, platform):
+        if platform == 'no_unix_sockets':
+            monkeypatch.setattr(display_state, 'socket_supported', lambda: False)
+        else:
+            monkeypatch.setattr(display_state, 'client_socket_paths', lambda: [])
+        data = self._status(api_v3_client)
+        assert data['mode'] == 'clock' and data['is_display_active'] is True

test/test_display_hold_report.py

@@ -0,0 +1,142 @@
+"""The report of a scrolling screen held by its plugin's update().
+
+While a plugin's update() runs it holds the plugin's lock, and its screen's
+frames are skipped -- on a scroller, a frozen strip -- with nothing logged.
+_note_display_hold times each such run and reports one of
+DISPLAY_HOLD_REPORT_SECONDS or more.
+"""
+
+import threading
+import types
+from unittest.mock import MagicMock
+
+import pytest
+
+
+class _Clock:
+    """display_controller's clock: moves only when run() sleeps or a test says."""
+
+    def __init__(self, start=10_000.0):
+        self.t = start
+
+    def now(self):
+        return self.t
+
+    def sleep(self, seconds):
+        self.t += max(seconds, 0.0005)
+
+    def module(self):
+        return types.SimpleNamespace(time=self.now, monotonic=self.now,
+                                     perf_counter=self.now, sleep=self.sleep)
+
+
+@pytest.fixture
+def clock(monkeypatch):
+    c = _Clock()
+    monkeypatch.setattr("src.display_controller.time", c.module())
+    return c
+
+
+class _Locks:
+    """get_plugin_lock for one plugin, whose lock the test can hold."""
+
+    def __init__(self):
+        self.lock = threading.Lock()
+
+    def __call__(self, plugin_id):
+        return self.lock
+
+
+@pytest.fixture
+def held(test_display_controller):
+    c = test_display_controller
+    locks = _Locks()
+    c.plugin_manager.get_plugin_lock = locks
+    c.plugin_manager._warn_rate_limited = MagicMock()
+    c.plugin_manager.health_tracker = MagicMock()
+    c._display_hold = None
+    return c, locks.lock
+
+
+def _plugin(plugin_id):
+    p = MagicMock()
+    p.plugin_id = plugin_id
+    p.display.return_value = True
+    return p
+
+
+class TestTheDisplayHoldReport:
+    def test_a_long_hold_is_reported_when_it_ends(self, held, clock):
+        c, lock = held
+        ticker = _plugin("ticker")
+        lock.acquire()                         # update() running
+        assert c._display_once(ticker, "ticker", False, report_hold=True) is True
+        clock.t += 0.2
+        assert c._display_once(ticker, "ticker", False, report_hold=True) is True
+        assert ticker.display.call_count == 0
+        c.plugin_manager._warn_rate_limited.assert_not_called()
+        clock.t += 0.2
+        lock.release()                         # update() done
+        c._display_once(ticker, "ticker", False, report_hold=True)
+        assert ticker.display.call_count == 1
+        key, message, plugin_id, ms = c.plugin_manager._warn_rate_limited.call_args[0]
+        assert key == "display-hold:ticker" and plugin_id == "ticker"
+        assert "held" in message and ms == pytest.approx(400.0)
+        c.plugin_manager.health_tracker.record_busy_skip.assert_called_once_with(
+            "ticker", "display hold", pytest.approx(0.4))
+
+    def test_a_short_hold_is_not(self, held, clock):
+        c, lock = held
+        ticker = _plugin("ticker")
+        lock.acquire()
+        c._display_once(ticker, "ticker", False, report_hold=True)
+        clock.t += 0.1
+        lock.release()
+        c._display_once(ticker, "ticker", False, report_hold=True)
+        c.plugin_manager._warn_rate_limited.assert_not_called()
+        c.plugin_manager.health_tracker.record_busy_skip.assert_not_called()
+
+    def test_a_hold_that_ends_on_another_plugins_screen_is_not_blamed_on_it(
+            self, held, clock):
+        c, lock = held
+        lock.acquire()
+        c._display_once(_plugin("ticker"), "ticker", False, report_hold=True)
+        clock.t += 1.0
+        lock.release()
+        c._display_once(_plugin("clock"), "clock", False, report_hold=True)
+        c.plugin_manager._warn_rate_limited.assert_not_called()
+        assert c._display_hold is None
+
+    def test_frames_that_draw_report_nothing(self, held, clock):
+        c, _lock = held
+        ticker = _plugin("ticker")
+        for _ in range(5):
+            c._display_once(ticker, "ticker", False, report_hold=True)
+            clock.t += 0.5
+        c.plugin_manager._warn_rate_limited.assert_not_called()
+
+    def test_the_1hz_loop_reports_no_holds(self, held, clock):
+        # A static screen's frames are a second apart: one skipped frame is
+        # not a measured hold, and nothing on the panel froze. (On ledpi the
+        # first version reported every such skip as "held 1000 ms".)
+        c, lock = held
+        board = _plugin("board")
+        lock.acquire()
+        c._display_once(board, "board", False)
+        clock.t += 1.0
+        lock.release()
+        c._display_once(board, "board", False)
+        c.plugin_manager._warn_rate_limited.assert_not_called()
+        c.plugin_manager.health_tracker.record_busy_skip.assert_not_called()
+
+    def test_a_run_left_open_is_dropped_by_a_1hz_frame(self, held, clock):
+        c, lock = held
+        ticker = _plugin("ticker")
+        lock.acquire()
+        c._display_once(ticker, "ticker", False, report_hold=True)
+        lock.release()
+        c._display_once(ticker, "ticker", False)        # the 1 Hz loop draws
+        assert c._display_hold is None
+        clock.t += 5.0
+        c._display_once(ticker, "ticker", False, report_hold=True)
+        c.plugin_manager._warn_rate_limited.assert_not_called()

test/test_espn_dates.py

@@ -489,3 +489,171 @@ class TestConcurrency:
 
         assert live["peak"] <= espn_dates.ESPN_CHUNK_WORKERS
         assert live["peak"] > 1, "chunks should actually overlap"
+
+
+class TestEdgeMonths:
+    """A window's partial edge months are asked whole and trimmed.
+
+    The default scoreboard window -- a fortnight either side of today -- spans
+    two partial months, so it used to cost 29 day requests per league. ESPN's
+    ``dates=YYYYMMDD`` means a US Eastern day (verified against the live API
+    on 2026-10-03, 417 of 417 soccer events), so a month answer trimmed to
+    the window's Eastern days is what the day requests returned.
+    """
+
+    def test_a_fortnight_either_side_is_two_requests(self):
+        planned = espn_dates.espn_request_chunks(date(2026, 9, 20), date(2026, 10, 18))
+        assert planned == [
+            ("202609", (date(2026, 9, 20), date(2026, 9, 30))),
+            ("202610", (date(2026, 10, 1), date(2026, 10, 18))),
+        ]
+
+    def test_a_live_polls_two_days_stay_two_days(self):
+        planned = espn_dates.espn_request_chunks(date(2026, 10, 2), date(2026, 10, 3))
+        assert planned == [("20261002", None), ("20261003", None)]
+
+    def test_the_threshold_is_inclusive(self):
+        n = espn_dates.ESPN_MONTH_COVER_MIN_DAYS
+        short = espn_dates.espn_request_chunks(date(2026, 10, 1), date(2026, 10, n - 1))
+        assert [chunk for chunk, _ in short] == [
+            "202610%02d" % day for day in range(1, n)]
+        enough = espn_dates.espn_request_chunks(date(2026, 10, 1), date(2026, 10, n))
+        assert enough == [("202610", (date(2026, 10, 1), date(2026, 10, n)))]
+
+    def test_whole_months_and_short_edges_are_unchanged(self):
+        planned = espn_dates.espn_request_chunks(date(2026, 8, 30), date(2026, 10, 2))
+        assert planned == [
+            ("20260830", None), ("20260831", None), ("202609", None),
+            ("20261001", None), ("20261002", None),
+        ]
+
+    def test_without_time_zone_data_edges_stay_days(self, monkeypatch):
+        monkeypatch.setattr(espn_dates, "_EASTERN", None)
+        planned = espn_dates.espn_request_chunks(date(2026, 9, 20), date(2026, 10, 18))
+        assert len(planned) == 29
+        assert all(trim is None for _, trim in planned)
+
+    @pytest.mark.parametrize("start,end", [
+        (date(2026, 9, 20), date(2026, 10, 18)),
+        (date(2026, 1, 25), date(2026, 3, 3)),
+        (date(2026, 12, 20), date(2027, 1, 9)),
+        (date(2026, 10, 5), date(2026, 10, 9)),
+    ])
+    def test_the_planned_requests_still_cover_every_day_exactly_once(self, start, end):
+        covered = []
+        for chunk, trim in espn_dates.espn_request_chunks(start, end):
+            if trim is None:
+                covered.extend(days_covered_by([chunk]))
+            else:
+                assert chunk == trim[0].strftime("%Y%m") == trim[1].strftime("%Y%m")
+                covered.extend(trim[0] + timedelta(days=offset)
+                               for offset in range((trim[1] - trim[0]).days + 1))
+        expected = [start + timedelta(days=offset) for offset in range((end - start).days + 1)]
+        assert covered == expected
+
+    def test_a_trimmed_month_keeps_only_the_windows_eastern_days(self):
+        september = [
+            # 03:30Z on the 20th is still the 19th in New York: outside.
+            {"id": "before", "date": "2026-09-20T03:30Z"},
+            {"id": "first", "date": "2026-09-20T14:00Z"},
+            {"id": "late", "date": "2026-09-30T23:30Z"},
+        ]
+        october = [
+            {"id": "oct1", "date": "2026-10-01T19:00Z"},
+            # 03:30Z on the 19th is the evening of the 18th in New York: inside.
+            {"id": "last", "date": "2026-10-19T03:30Z"},
+            {"id": "after", "date": "2026-10-19T14:00Z"},
+            {"id": "undated"},
+        ]
+        session = FakeSession({"202609": september, "202610": october})
+
+        data = fetch_espn_date_chunks(session, URL, params={"dates": "20260920-20261018"})
+
+        assert sorted(call["dates"] for call in session.calls) == ["202609", "202610"]
+        # An event with no readable date is kept, never dropped on a guess.
+        assert [e["id"] for e in data["events"]] == ["first", "late", "oct1", "last", "undated"]
+
+    def test_eastern_standard_time_is_honoured_after_the_clocks_change(self):
+        # 2026-11-01 ends daylight saving: Eastern is UTC-5 from then on.
+        november = [
+            {"id": "out", "date": "2026-11-15T04:30Z"},  # Nov 14, 23:30 EST
+            {"id": "in", "date": "2026-11-15T05:30Z"},   # Nov 15, 00:30 EST
+        ]
+        session = FakeSession({"202611": november})
+        data = fetch_espn_date_chunks(session, URL, params={"dates": "20261115-20261121"})
+        assert [e["id"] for e in data["events"]] == ["in"]
+
+    def test_a_capped_edge_month_re_asks_only_the_windows_days(self):
+        full = [{"id": "cap%d" % i, "date": "2026-10-05T18:00Z"} for i in range(ESPN_MAX_LIMIT)]
+        by_chunk = {"202610": full}
+        by_chunk.update({"202610%02d" % day: [{"id": "o%02d" % day}] for day in range(1, 32)})
+        session = FakeSession(by_chunk)
+
+        data = fetch_espn_date_chunks(session, URL, params={"dates": "20261001-20261010"})
+
+        sent = [call["dates"] for call in session.calls]
+        assert sent[0] == "202610"
+        assert sorted(sent[1:]) == ["202610%02d" % day for day in range(1, 11)]
+        assert [e["id"] for e in data["events"]] == ["o%02d" % day for day in range(1, 11)]
+
+
+class TestProcessWideChunkCap:
+    """The chunk cap holds across windows, not per window.
+
+    A soccer board starting eight leagues fetches sixteen windows at once.
+    With a pool of ``ESPN_CHUNK_WORKERS`` each, ~40 requests were in flight
+    and every one past a session's pool opened a connection -- and a DNS
+    lookup. On ledpi that was ~90 NameResolutionErrors per start.
+    """
+
+    def test_concurrent_windows_share_one_budget(self):
+        live = {"now": 0, "peak": 0}
+        guard = threading.Lock()
+
+        class CountingSession(FakeSession):
+            def get(self, url, params=None, headers=None, timeout=None):
+                with guard:
+                    live["now"] += 1
+                    live["peak"] = max(live["peak"], live["now"])
+                try:
+                    time.sleep(0.01)
+                    return super().get(url, params=params, headers=headers, timeout=timeout)
+                finally:
+                    with guard:
+                        live["now"] -= 1
+
+        sessions = [CountingSession() for _ in range(6)]
+        # Six leagues, so the fetch service cannot merge them into one, on a
+        # host with no token bucket: earlier tests may have spent ESPN's
+        # burst, and a bucket paced at 20/s would serialise these by itself.
+        threads = [
+            threading.Thread(target=fetch_espn_date_chunks,
+                             args=(session, "https://scores.example.test/league%d" % index),
+                             kwargs={"params": {"dates": "20260101-20261231"}})
+            for index, session in enumerate(sessions)
+        ]
+        for thread in threads:
+            thread.start()
+        for thread in threads:
+            thread.join(timeout=30)
+
+        assert all(len(session.calls) == 12 for session in sessions)
+        assert live["peak"] <= espn_dates.ESPN_CHUNK_WORKERS
+        assert live["peak"] > 1, "chunks should still overlap"
+
+
+def test_a_fresh_process_skips_the_doomed_range_request():
+    """Every start used to spend one 400 per window learning that ranges are
+    still rejected -- eleven at once from a soccer board. A new process now
+    starts inside the retry period instead."""
+    import subprocess
+    import sys
+    from pathlib import Path
+
+    out = subprocess.run(
+        [sys.executable, "-c",
+         "import src.common.espn_dates as e; print(e._ranges_known_rejected())"],
+        cwd=str(Path(__file__).resolve().parents[1]),
+        capture_output=True, text=True, timeout=60,
+    )
+    assert out.stdout.strip() == "True", out.stderr

test/test_fetch_service.py

@@ -576,6 +576,42 @@ class TestCounters:
         assert snap["hosts"]["site.api.espn.com"]["requests"] == 1
         assert snap["totals"]["bytes"] == 3 * len(b'{"ok": 1}')
 
+    def test_wire_bytes_are_the_compressed_size(self, service):
+        # Built the way requests builds a real response: a urllib3
+        # HTTPResponse carrying a gzip body, decoded when .content is read.
+        import gzip
+        import io
+
+        from requests.adapters import HTTPAdapter
+        from urllib3.response import HTTPResponse
+
+        decoded = json.dumps({"events": [{"id": str(i), "name": "x" * 200}
+                                         for i in range(50)]}).encode()
+        wire = gzip.compress(decoded)
+
+        def handler(url, kwargs):
+            raw = HTTPResponse(body=io.BytesIO(wire), status=200,
+                               headers={"Content-Encoding": "gzip",
+                                        "Content-Type": "application/json"},
+                               preload_content=False, decode_content=True)
+            request = requests.Request("GET", url).prepare()
+            response = HTTPAdapter().build_response(request, raw)
+            response.content  # what Session.get does for a non-streamed call
+            return response
+
+        response = service.get(FakeSession(handler), "https://site.api.espn.com/x")
+        assert response.content == decoded
+        totals = _counters(service)
+        assert totals["bytes"] == len(decoded)
+        assert totals["wire_bytes"] == len(wire) < len(decoded)
+
+    def test_wire_bytes_fall_back_to_the_decoded_size(self, service):
+        # No urllib3 response behind it (a test double, another adapter):
+        # count what is known rather than nothing.
+        service.get(FakeSession(), "https://api.test/x")
+        totals = _counters(service)
+        assert totals["wire_bytes"] == totals["bytes"] == len(b'{"ok": 1}')
+
     def test_errors_and_http_errors(self, service):
         def handler(url, kwargs):
             if url.endswith("/down"):
@@ -670,14 +706,14 @@ class TestCallerIdentity:
         assert _counters(global_service, plugin="football-scoreboard")["requests"] == 1
 
     def test_espn_chunks_on_worker_threads_count_against_the_caller(self, global_service):
-        from src.common.espn_dates import espn_date_chunks, fetch_espn_date_chunks, parse_espn_date_range
+        from src.common.espn_dates import espn_request_chunks, fetch_espn_date_chunks, parse_espn_date_range
 
         session = FakeSession(lambda url, kw: make_response(body=b'{"events": []}', url=url))
         dates = "20260801-20261015"
         with plugin_scope("baseball-scoreboard"):
             fetch_espn_date_chunks(session, "https://site.api.espn.com/s/scoreboard",
                                    params={"dates": dates})
-        chunks = len(espn_date_chunks(*parse_espn_date_range(dates)))
+        chunks = len(espn_request_chunks(*parse_espn_date_range(dates)))
         assert chunks > 1
         assert len(session.calls) == chunks
         assert _counters(global_service, plugin="baseball-scoreboard")["requests"] == chunks

test/test_on_demand_live_and_restore.py

@@ -0,0 +1,164 @@
+"""Two on-demand edges seen on a rig.
+
+  * A request naming a ``*_live`` mode got HTTP 200 and a different mode on
+    the panel. The session's mode list kept live modes only when the plugin's
+    has_live_content() said so, and that is the live-priority question,
+    which the sports plugins answer for favourite teams only: fifteen college
+    games on, no favourite playing, and ``ncaa_fb_live`` became
+    ``nfl_recent``.
+  * A restart during a session whose plugin then failed to load (its config
+    no longer validated) logged "No valid display modes found ... after
+    restoration" and left the session active with no modes: published as
+    active for a plugin that was not running, with its cached request kept
+    for the next restart.
+"""
+
+from unittest.mock import MagicMock
+
+import pytest
+
+SPORTS_MODES = ['nfl_live', 'nfl_recent', 'nfl_upcoming',
+                'ncaa_fb_live', 'ncaa_fb_recent', 'ncaa_fb_upcoming']
+
+
+def _sports_plugin(has_live_content=False):
+    plugin = MagicMock(spec=['display', 'has_live_content', 'has_live_priority',
+                             'get_live_modes'])
+    plugin.has_live_content.return_value = has_live_content
+    plugin.has_live_priority.return_value = True
+    plugin.get_live_modes.return_value = []
+    return plugin
+
+
+def _register(controller, plugin_id, modes, plugin):
+    controller.plugin_display_modes[plugin_id] = list(modes)
+    for mode in modes:
+        controller.plugin_modes[mode] = plugin
+        controller.mode_to_plugin_id[mode] = plugin_id
+        if mode not in controller.available_modes:
+            controller.available_modes.append(mode)
+
+
+@pytest.fixture
+def football(test_display_controller):
+    c = test_display_controller
+    _register(c, 'football-scoreboard', SPORTS_MODES, _sports_plugin())
+    return c
+
+
+class TestANamedLiveModeIsShown:
+
+    def test_it_is_the_first_screen(self, football):
+        football._activate_on_demand({'plugin_id': 'football-scoreboard',
+                                      'mode': 'ncaa_fb_live'})
+        assert football.on_demand_active
+        assert football.current_display_mode == 'ncaa_fb_live'
+        assert football.on_demand_mode == 'ncaa_fb_live'
+
+    def test_the_plugins_other_modes_follow_it(self, football):
+        football._activate_on_demand({'plugin_id': 'football-scoreboard',
+                                      'mode': 'ncaa_fb_live'})
+        assert football.on_demand_modes[0] == 'ncaa_fb_live'
+        assert set(football.on_demand_modes[1:]) == {
+            'nfl_recent', 'nfl_upcoming', 'ncaa_fb_recent', 'ncaa_fb_upcoming'}
+
+    def test_pinned_holds_it(self, football):
+        football._activate_on_demand({'plugin_id': 'football-scoreboard',
+                                      'mode': 'ncaa_fb_live', 'pinned': True})
+        assert football.on_demand_modes == ['ncaa_fb_live']
+
+    def test_a_bare_plugin_request_still_skips_quiet_live_modes(self, football):
+        """Only a mode asked for by name is kept: a plugin-only request
+        resolves to the plugin's first mode (nfl_live), and opening on an
+        empty live screen there is what the ordering exists to avoid."""
+        football._activate_on_demand({'plugin_id': 'football-scoreboard'})
+        assert not any(m.endswith('_live') for m in football.on_demand_modes)
+
+    def test_a_named_second_live_mode_with_content_leads(self, test_display_controller):
+        """With live content both live modes are kept, nfl_live first; a
+        request naming ncaa_fb_live must still open on it, not rotate away."""
+        c = test_display_controller
+        _register(c, 'football-scoreboard', SPORTS_MODES, _sports_plugin(has_live_content=True))
+        c._activate_on_demand({'plugin_id': 'football-scoreboard', 'mode': 'ncaa_fb_live'})
+        assert c.on_demand_modes[0] == 'ncaa_fb_live'
+        assert c.on_demand_modes.count('ncaa_fb_live') == 1
+        assert 'nfl_live' in c.on_demand_modes[1:]
+
+    def test_the_named_mode_survives_a_restart(self, football):
+        football._activate_on_demand({'plugin_id': 'football-scoreboard',
+                                      'mode': 'ncaa_fb_live'})
+        # The last on-demand config write, not the last write of any key: the
+        # font-usage publisher thread writes its own key at its own pace.
+        saved = [c for c in football.cache_manager.set.call_args_list
+                 if c.args and c.args[0] == 'display_on_demand_config'][-1]
+        config = saved.args[1]
+        assert config['named_mode'] == 'ncaa_fb_live'
+
+        football._reset_on_demand_fields()
+        football._select_startup_plugins(['football-scoreboard'], config)
+        football._populate_on_demand_modes_from_plugin()
+        assert football.on_demand_modes[football.on_demand_mode_index] == 'ncaa_fb_live'
+
+
+class TestARestoreWithNothingToResume:
+
+    @pytest.fixture
+    def restored(self, test_display_controller):
+        c = test_display_controller
+        c.config['clock-simple'] = {'enabled': True}
+        c._select_startup_plugins(['clock-simple'],
+                                  {'plugin_id': 'clock-simple', 'mode': 'clock-simple'})
+        assert c.on_demand_active
+        # The plugin's load then fails: nothing is registered for it.
+        c.cache_manager.clear_cache.reset_mock()
+        c._populate_on_demand_modes_from_plugin()
+        return c
+
+    def test_the_session_ends(self, restored):
+        assert not restored.on_demand_active
+        assert restored.on_demand_plugin_id is None
+        assert not restored.on_demand_schedule_override
+
+    def test_it_is_reported_as_an_error(self, restored):
+        assert restored.on_demand_status == 'error'
+        assert restored.on_demand_last_error == 'restore-failed'
+        # The last on-demand state write, not the last write of any key: the
+        # font-usage publisher thread writes its own key at its own pace.
+        published = [c for c in restored.cache_manager.set.call_args_list
+                     if c.args and c.args[0] == 'display_on_demand_state'][-1]
+        assert published.args[1]['status'] == 'error'
+        assert published.args[1]['error'] == 'restore-failed'
+
+    def test_the_cached_request_is_dropped(self, restored):
+        restored.cache_manager.clear_cache.assert_any_call('display_on_demand_config')
+
+
+def test_a_plugin_system_failure_ends_a_cached_session_not_yet_restored(
+        mock_config_manager, mock_display_manager, mock_cache_manager,
+        test_config_with_plugins, emulator_mode):
+    """Initialization can fail before the cached session is read, with
+    on_demand_active still False: the session must still end, visibly."""
+    from unittest.mock import patch
+    from src.display_controller import DisplayController
+
+    mock_config_manager.get_config.return_value = test_config_with_plugins
+    mock_config_manager.load_config.return_value = test_config_with_plugins
+    mock_cache_manager._memory_cache['display_on_demand_config'] = {
+        'plugin_id': 'clock-simple', 'mode': 'clock-simple'}
+    with patch('src.display_controller.ConfigManager', return_value=mock_config_manager), \
+         patch('src.display_controller.DisplayManager', return_value=mock_display_manager), \
+         patch('src.display_controller.CacheManager', return_value=mock_cache_manager), \
+         patch('src.display_controller.FontManager'), \
+         patch('src.plugin_system.PluginManager', side_effect=RuntimeError("boom")):
+        controller = DisplayController()
+        try:
+            assert controller.plugin_manager is None
+            assert not controller.on_demand_active
+            assert controller.on_demand_status == 'error'
+            assert controller.on_demand_last_error == 'restore-failed'
+            mock_cache_manager.clear_cache.assert_any_call('display_on_demand_config')
+        finally:
+            try:
+                controller.cleanup()
+            except Exception:
+                pass

test/test_run_loop_golden.py

@@ -133,6 +133,32 @@ def scenario_on_demand_restored(h: RunLoopHarness):
     h.restore_on_demand("sports", mode="sports_upcoming", duration=40)
 
 
+def scenario_on_demand_named_live(h: RunLoopHarness):
+    # Games are on until t=70, but none involves a favourite, so
+    # has_live_content() (the live-priority answer) stays False throughout.
+    # A request naming sports_live still opens on it (it opened on
+    # sports_recent); asked for again after the games end, it has nothing to
+    # draw and the session moves on to the plugin's next mode.
+    h.add_plugin(FakePlugin("clock", ["clock"], duration=20))
+    h.add_plugin(FakePlugin(
+        "sports", ["sports_live", "sports_recent", "sports_upcoming"], duration=15,
+        live_priority=True,
+        content=lambda t, mode: mode != "sports_live" or t < 70))
+    h.on_demand_request(5, "n1", plugin_id="sports", mode="sports_live")
+    h.on_demand_request(40, "n2", action="stop")
+    h.on_demand_request(100, "n3", plugin_id="sports", mode="sports_live")
+    h.on_demand_request(140, "n4", action="stop")
+
+
+def scenario_on_demand_restore_failed(h: RunLoopHarness):
+    # A restart during a session whose plugin then fails to load: the
+    # session ends as an error before the first screen, and the rotation
+    # runs normally from the top.
+    h.add_plugin(FakePlugin("clock", ["clock"], duration=20))
+    h.add_plugin(FakePlugin("weather", ["weather"], duration=20))
+    h.restore_on_demand("gone", mode="gone", duration=40)
+
+
 def scenario_schedule(h: RunLoopHarness):
     # The clock starts at 22:59:30. Off from 23:01 until 23:05 (the window
     # spans midnight); dimmed from 23:00 until 23:01.
@@ -194,6 +220,8 @@ SCENARIOS = {
     "on_demand": (scenario_on_demand, 240),
     "on_demand_pinned": (scenario_on_demand_pinned, 160),
     "on_demand_restored": (scenario_on_demand_restored, 100),
+    "on_demand_named_live": (scenario_on_demand_named_live, 160),
+    "on_demand_restore_failed": (scenario_on_demand_restore_failed, 60),
     "schedule": (scenario_schedule, 400),
     "wifi_notice": (scenario_wifi_notice, 150),
     "follower": (scenario_follower, 80),

test/test_snapshot_owed_frame.py

@@ -0,0 +1,183 @@
+"""A frame the preview throttle skipped still reaches the snapshot.
+
+The preview snapshot (/api/v3/display/current, the web UI's live preview) is
+only written from update_display(), at most once per write interval. A screen
+that draws its card once and then holds it -- soccer's recent/upcoming cards
+skip redundant redraws -- pushes exactly one frame. When that push lands inside
+the interval, e.g. a few milliseconds after the on-demand start's clear wrote a
+black frame, the throttle skips it and nothing ever writes it: on ledpi the
+preview stayed black for soccer's whole 15 s screen while the panel showed the
+card, and the next screen "rendered immediately".
+
+Runs the real DisplayManager on the emulator, like test_display_dirty_tracking.
+"""
+
+import os
+import sys
+import types
+
+os.environ["EMULATOR"] = "true"
+
+import pytest
+from PIL import Image
+
+sys.path.insert(0, os.path.join(os.path.dirname(__file__), ".."))
+
+
+@pytest.fixture(scope="module")
+def dm(tmp_path_factory):
+    from src.display_manager import DisplayManager
+    DisplayManager._instance = None
+    manager = DisplayManager({
+        "display": {
+            "hardware": {"rows": 32, "cols": 64, "chain_length": 2,
+                         "parallel": 1, "brightness": 90},
+            "runtime": {"gpio_slowdown": 0},
+        },
+    }, suppress_test_pattern=True)
+    manager._snapshot_path = str(
+        tmp_path_factory.mktemp("owed_snapshot") / "led_matrix_preview.png")
+    yield manager
+    DisplayManager._instance = None
+
+
+@pytest.fixture
+def viewer(dm, monkeypatch, tmp_path):
+    """A preview is open (1 s write interval); fresh snapshot bookkeeping."""
+    monkeypatch.setattr(dm, "_viewer_is_fresh", lambda now: True)
+    dm._viewer_was_fresh = True
+    dm._snapshot_path = str(tmp_path / "snap.png")
+    dm._last_snapshot_ts = 0.0
+    dm._last_snapshot_touch_ts = 0.0
+    dm._last_snapshot_digest = None
+    dm._saved_snapshot_digest = None
+    dm._snapshot_owed = False
+    dm.set_scrolling_state(False)
+    return dm
+
+
+def _lit(path):
+    with Image.open(path) as img:
+        return sum(1 for p in img.convert("RGB").getdata() if max(p) > 20)
+
+
+def _age_last_write(dm, seconds=2.0):
+    """As if `seconds` had passed since the last snapshot write."""
+    dm._last_snapshot_ts -= seconds
+    dm._last_snapshot_touch_ts -= seconds
+
+
+def _clear_then_draw_card(dm):
+    """The on-demand start's clear, then the card a few ms later."""
+    dm.clear()
+    dm.update_display()                       # black frame: written
+    assert _lit(dm._snapshot_path) == 0
+    dm.draw.rectangle([4, 4, 40, 20], fill=(255, 255, 0))
+    dm.update_display()                       # the card: inside the interval
+
+
+def _controller(dm):
+    from src import display_controller as dc_module
+    controller = dc_module.DisplayController.__new__(dc_module.DisplayController)
+    controller.plugin_manager = None
+    controller.display_manager = dm
+    return controller
+
+
+class _HoldingPlugin:
+    """Already showing its card: display() returns True and draws nothing."""
+
+    plugin_id = "holding"
+
+    def __init__(self):
+        self.calls = 0
+
+    def display(self, display_mode=None, force_clear=False):
+        self.calls += 1
+        return True
+
+
+def test_a_held_card_reaches_the_preview_on_the_next_frame(viewer):
+    dm = viewer
+    _clear_then_draw_card(dm)
+    assert _lit(dm._snapshot_path) == 0       # the throttle skipped the card
+
+    controller = _controller(dm)
+    plugin = _HoldingPlugin()
+    _age_last_write(dm)
+    # The render loop's next frame: the plugin draws nothing and makes no
+    # update_display() call, as soccer's switch cards do.
+    assert controller._display_once(plugin, "soccer_eng.1_recent", True) is True
+    assert plugin.calls == 1
+    assert _lit(dm._snapshot_path) > 0
+
+
+def test_the_owed_write_still_waits_out_the_interval(viewer, monkeypatch):
+    dm = viewer
+    _clear_then_draw_card(dm)
+    saves = []
+    monkeypatch.setattr(dm, "_save_snapshot", lambda image: saves.append(image))
+    dm.write_owed_snapshot()                  # still inside the interval
+    assert saves == []
+    _age_last_write(dm)
+    dm.write_owed_snapshot()
+    assert len(saves) == 1
+    # Written: nothing is owed, so later frames do no work and the unchanged
+    # frame is not encoded again.
+    assert dm._snapshot_owed is False
+    _age_last_write(dm)
+    dm.write_owed_snapshot()
+    assert len(saves) == 1
+
+
+def test_a_failed_owed_write_stays_owed_and_is_retried(viewer, monkeypatch):
+    dm = viewer
+    _clear_then_draw_card(dm)
+    _age_last_write(dm)
+    attempts = []
+
+    def failing_save(image):
+        attempts.append(image)
+        raise OSError("disk full")
+
+    monkeypatch.setattr(dm, "_save_snapshot", failing_save)
+    dm.write_owed_snapshot()                  # the write fails
+    assert len(attempts) == 1
+    assert dm._snapshot_owed is True          # still owed: a held screen
+    saves = []                                # makes no update_display()
+    monkeypatch.setattr(dm, "_save_snapshot", lambda image: saves.append(image))
+    dm.write_owed_snapshot()                  # retried on the next frame
+    assert len(saves) == 1
+    assert dm._snapshot_owed is False
+
+
+def test_nothing_owed_after_a_frame_that_was_written(viewer, monkeypatch):
+    dm = viewer
+    dm.draw.rectangle([0, 0, 8, 8], fill=(0, 255, 0))
+    dm.update_display()                       # due: written at once
+    assert dm._snapshot_owed is False
+    calls = []
+    monkeypatch.setattr(dm, "_write_snapshot_if_due",
+                        lambda *a, **k: calls.append(a))
+    dm.write_owed_snapshot()
+    assert calls == []
+
+
+def test_an_unchanged_frame_inside_the_interval_is_not_owed(viewer):
+    dm = viewer
+    dm.draw.rectangle([0, 0, 8, 8], fill=(0, 0, 255))
+    dm.update_display()
+    dm.update_display()                       # same frame, inside the interval
+    assert dm._snapshot_owed is False
+
+
+def test_a_controller_without_the_hook_still_draws():
+    """Controllers built without a display manager (tests) are unaffected."""
+    from src import display_controller as dc_module
+    controller = dc_module.DisplayController.__new__(dc_module.DisplayController)
+    controller.plugin_manager = None
+    plugin = _HoldingPlugin()
+    assert controller._display_once(plugin, "x", True) is True
+    controller.display_manager = types.SimpleNamespace()
+    assert controller._display_once(plugin, "x", True) is True
+    assert plugin.calls == 2

test/test_state_stream_readers.py

@@ -612,6 +612,12 @@ class TestEndToEnd:
         cached['display_current_state'] = {'mode': 'from-cache', 'last_updated': 1}
         path = str(tmp_path / 'control.sock')
         monkeypatch.setenv(c.SOCKET_PATH_ENV, path)
+        # The display is this process, and its render loop is beating: the
+        # cache is then still its answer once the socket goes.
+        heartbeat = tmp_path / 'display-heartbeat.json'
+        heartbeat.write_text(json.dumps({'pid': os.getpid(), 'mono': time.monotonic(),
+                                         'wall': time.time()}))
+        monkeypatch.setattr(display_watchdog, 'HEARTBEAT_PATH', str(heartbeat))
         hub = _hub_with_everything()
         server = ControlServer(path, state_hub=hub, keepalive=0.2)
         assert server.start()
@@ -638,3 +644,8 @@ class TestEndToEnd:
                 break
             time.sleep(0.05)
         assert (data['mode'], data['source']) == ('from-cache', 'cache')
+        # Stopped: systemd takes the heartbeat's directory with it, and the
+        # cache's last answer is no longer anyone's.
+        heartbeat.unlink()
+        data = _data(client, '/api/v3/display/current-status')
+        assert (data['mode'], data['source']) == (None, 'cache')

test/test_store_update_keeps_local_files.py

@@ -0,0 +1,244 @@
+"""A plugin update must keep the files the plugin wrote beside itself.
+
+Field incident, 2026-10-04: updating calendar 1.2.9 -> 1.2.12 from the web UI
+replaced plugin-repos/calendar/ with the fresh download and deleted the old
+copy -- and with it token.pickle and credentials.json, the plugin's Google
+OAuth files. No release contains them (the repo gitignores them), so the hot
+reload logged "Credentials file not found" and the calendar stayed broken
+until the files were restored by hand.
+
+Both update routes are covered: a monorepo plugin (registry ``plugin_path``),
+which is reinstalled into a fresh directory, and a plugin installed from its
+own git repository, which is updated with ``git pull`` after an auto-stash.
+"""
+
+import json
+import shutil
+import subprocess
+
+import pytest
+
+from src.plugin_system.plugin_local_files import (
+    is_known_state_file, local_files_to_keep,
+)
+from src.plugin_system.store_manager import PluginStoreManager
+
+PLUGIN_ID = "calendar"
+
+
+def _manifest(version):
+    return {"id": PLUGIN_ID, "name": "Calendar", "class_name": "CalendarPlugin",
+            "display_modes": ["calendar"], "version": version}
+
+
+def _write_release(target, version):
+    """What a download of ``version`` puts on disk."""
+    target.mkdir(parents=True, exist_ok=True)
+    (target / "manifest.json").write_text(json.dumps(_manifest(version)))
+    (target / "manager.py").write_text(f"VERSION = {version!r}\n")
+    (target / ".gitignore").write_text("credentials.json\ntoken.pickle\ncache/\n")
+
+
+def _drop_local_files(plugin_dir):
+    """What the plugin writes at runtime: OAuth files plus cached state."""
+    (plugin_dir / "token.pickle").write_bytes(b"\x80\x04oauth-token")
+    (plugin_dir / "credentials.json").write_text('{"installed": {}}')
+    (plugin_dir / "cache").mkdir()
+    (plugin_dir / "cache" / "events.json").write_text("[]")
+
+
+def _assert_local_files_kept(plugin_dir):
+    assert (plugin_dir / "token.pickle").read_bytes() == b"\x80\x04oauth-token"
+    assert (plugin_dir / "credentials.json").read_text() == '{"installed": {}}'
+    assert (plugin_dir / "cache" / "events.json").read_text() == "[]"
+
+
+def _leftover_backups(plugins_dir):
+    return [p.name for p in plugins_dir.iterdir() if "standalone-backup" in p.name]
+
+
+@pytest.fixture
+def store(tmp_path, monkeypatch):
+    mgr = PluginStoreManager(
+        plugins_dir=str(tmp_path / "plugin-repos"),
+        uninstalled_registry_path=str(tmp_path / "uninstalled.json"))
+    mgr.plugins_dir.mkdir(parents=True, exist_ok=True)
+    monkeypatch.setattr(mgr, "_install_dependencies", lambda *a, **k: True)
+    monkeypatch.setattr(mgr, "fetch_registry", lambda *a, **k: {"plugins": []})
+    return mgr
+
+
+class TestMonorepoUpdate:
+    @pytest.fixture
+    def installed(self, store, monkeypatch):
+        registry_entry = {
+            "id": PLUGIN_ID, "repo": "https://github.com/ChuckBuilds/ledmatrix-plugins",
+            "plugin_path": "plugins/calendar", "branch": "main",
+            "latest_version": "1.2.9",
+        }
+        monkeypatch.setattr(store, "get_plugin_info", lambda *a, **k: registry_entry)
+        release = {"version": "1.2.9"}
+
+        def fake_monorepo_download(download_url, plugin_subpath, target):
+            assert plugin_subpath == "plugins/calendar"
+            _write_release(target, release["version"])
+            return True
+
+        monkeypatch.setattr(store, "_install_from_monorepo", fake_monorepo_download)
+        assert store.install_plugin(PLUGIN_ID) is True
+
+        def publish(version):
+            registry_entry["latest_version"] = release["version"] = version
+        return store, store.plugins_dir / PLUGIN_ID, publish
+
+    def test_update_keeps_token_and_gitignored_files(self, installed):
+        store, plugin_dir, publish = installed
+        _drop_local_files(plugin_dir)
+
+        publish("1.2.12")
+        assert store.update_plugin(PLUGIN_ID) is True
+
+        assert json.loads((plugin_dir / "manifest.json").read_text())["version"] == "1.2.12"
+        _assert_local_files_kept(plugin_dir)
+        assert _leftover_backups(store.plugins_dir) == []
+
+    def test_token_is_kept_even_when_the_release_does_not_gitignore_it(self, installed):
+        store, plugin_dir, publish = installed
+        (plugin_dir / ".gitignore").unlink()
+        (plugin_dir / "token.pickle").write_bytes(b"tok")
+        (plugin_dir / "config_secrets.json").write_text("{}")
+
+        publish("1.2.12")
+        assert store.update_plugin(PLUGIN_ID) is True
+
+        assert (plugin_dir / "token.pickle").read_bytes() == b"tok"
+        assert (plugin_dir / "config_secrets.json").read_text() == "{}"
+
+    def test_release_content_wins_and_old_code_is_not_carried(self, installed):
+        store, plugin_dir, publish = installed
+        # A file the old copy had that the new release dropped, byte code, and
+        # an old copy of a file the new release also ships.
+        (plugin_dir / "removed_module.py").write_text("OLD = True\n")
+        (plugin_dir / "__pycache__").mkdir()
+        (plugin_dir / "__pycache__" / "manager.cpython-313.pyc").write_bytes(b"pyc")
+
+        publish("1.2.12")
+        assert store.update_plugin(PLUGIN_ID) is True
+
+        assert not (plugin_dir / "removed_module.py").exists()
+        assert not (plugin_dir / "__pycache__").exists()
+        assert "1.2.12" in (plugin_dir / "manager.py").read_text()
+
+    def test_reinstall_over_an_existing_copy_keeps_them_too(self, installed):
+        store, plugin_dir, publish = installed
+        _drop_local_files(plugin_dir)
+
+        assert store.install_plugin(PLUGIN_ID) is True
+
+        _assert_local_files_kept(plugin_dir)
+        assert _leftover_backups(store.plugins_dir) == []
+
+
+class TestInstallFromUrlReplace:
+    def test_replacing_an_installed_copy_keeps_the_token(self, store, monkeypatch):
+        plugin_dir = store.plugins_dir / PLUGIN_ID
+        _write_release(plugin_dir, "1.0.0")
+        _drop_local_files(plugin_dir)
+
+        def fake_clone(repo_url, target, branches):
+            _write_release(target, "2.0.0")
+            return "main"
+
+        monkeypatch.setattr(store, "_install_via_git", fake_clone)
+        result = store.install_from_url(
+            "https://github.com/example/ledmatrix-calendar", plugin_id=PLUGIN_ID)
+
+        assert result["success"] is True
+        assert json.loads((plugin_dir / "manifest.json").read_text())["version"] == "2.0.0"
+        _assert_local_files_kept(plugin_dir)
+
+
+def _git(*args, cwd):
+    subprocess.run(["git", "-c", "user.email=t@example.com", "-c", "user.name=t",
+                    "-c", "core.autocrlf=false", *args],
+                   cwd=cwd, check=True, capture_output=True)
+
+
+@pytest.mark.skipif(shutil.which("git") is None, reason="git not installed")
+class TestGitRepoUpdate:
+    @pytest.fixture
+    def cloned(self, store, tmp_path, monkeypatch):
+        monkeypatch.setattr(store, "get_plugin_info", lambda *a, **k: None)
+        upstream = tmp_path / "upstream"
+        _write_release(upstream, "1.0.0")
+        # This repo does NOT gitignore the token: an untracked, non-ignored
+        # file is exactly what `git stash push -u` used to sweep away.
+        (upstream / ".gitignore").write_text("cache/\n")
+        _git("init", "-q", "-b", "main", cwd=upstream)
+        _git("add", ".", cwd=upstream)
+        _git("commit", "-qm", "1.0.0", cwd=upstream)
+
+        plugin_dir = store.plugins_dir / PLUGIN_ID
+        _git("clone", "-q", str(upstream), str(plugin_dir), cwd=tmp_path)
+
+        def publish(version):
+            (upstream / "manifest.json").write_text(json.dumps(_manifest(version)))
+            _git("commit", "-qam", version, cwd=upstream)
+        return store, plugin_dir, publish
+
+    def test_pull_update_keeps_untracked_token(self, cloned):
+        store, plugin_dir, publish = cloned
+        _drop_local_files(plugin_dir)
+        # An unrelated untracked file, so the update really does stash.
+        (plugin_dir / "notes.txt").write_text("scratch")
+
+        publish("1.1.0")
+        assert store.update_plugin(PLUGIN_ID) is True
+
+        assert json.loads((plugin_dir / "manifest.json").read_text())["version"] == "1.1.0"
+        _assert_local_files_kept(plugin_dir)
+
+    def test_token_alone_does_not_trigger_a_stash(self, cloned):
+        store, plugin_dir, publish = cloned
+        (plugin_dir / "token.pickle").write_bytes(b"tok")
+
+        publish("1.1.0")
+        assert store.update_plugin(PLUGIN_ID) is True
+
+        assert (plugin_dir / "token.pickle").read_bytes() == b"tok"
+        stashes = subprocess.run(["git", "-C", str(plugin_dir), "stash", "list"],
+                                 capture_output=True, text=True, check=True)
+        assert stashes.stdout.strip() == ""
+
+
+class TestWhatIsKept:
+    @pytest.mark.parametrize("path,expected", [
+        ("token.pickle", True),
+        ("data/session.pickle", True),
+        ("credentials.json", True),
+        ("token.json", True),
+        ("config_secrets.json", True),
+        (".pkce_code_verifier", True),
+        ("manager.py", False),
+        ("config.json", False),
+    ])
+    def test_known_state_files(self, path, expected):
+        assert is_known_state_file(path) is expected
+
+    def test_gitignore_rules(self, tmp_path):
+        old, new = tmp_path / "old", tmp_path / "new"
+        new.mkdir()
+        for rel in ["a.log", "logs/x.txt", "sub/deep/b.log", "keep.log",
+                    "anchored.txt", "sub/anchored.txt", "assets/x/y_backup/z.png",
+                    "manager.py", "shipped.log"]:
+            (old / rel).parent.mkdir(parents=True, exist_ok=True)
+            (old / rel).write_text("x")
+        (new / "shipped.log").write_text("new")
+        (old / ".gitignore").write_text(
+            "# comment\n*.log\n!keep.log\nlogs/\n/anchored.txt\n"
+            "assets/**/*_backup/\n")
+
+        assert local_files_to_keep(old, new) == [
+            "a.log", "anchored.txt", "assets/x/y_backup/z.png",
+            "logs/x.txt", "sub/deep/b.log",
+        ]

test/test_vegas_static_mode.py

@@ -219,8 +219,7 @@ class TestCoordinatorStaticPause:
     def _plugin(self):
         plugin = MagicMock()
         plugin.plugin_id = 'clock'
-        # A moment: zero would pause 15 s, as the rotation shows it.
-        plugin.get_display_duration.return_value = 0.01
+        plugin.get_display_duration.return_value = 0
         return plugin
 
     def test_trigger_comes_from_the_pipeline(self):

test/test_vegas_static_pause_duration.py

@@ -1,197 +0,0 @@
-"""A Vegas static pause lasts as long as the rotation shows the plugin.
-
-The pause asked the plugin for get_display_duration() and compared the
-answer with the clock. Several plugins (clock-simple, calendar, countdown)
-return their display_duration setting as it is in config.json, so one saved
-as "20" or null -- the raw config editor, a hand edit -- reached that
-comparison as a string or None. The TypeError went to the pause's broad
-except, which ended the pause: the plugin flashed up and the scroll went on,
-at every one of its turns. inf paused until something interrupted it, and
-NaN, False, 0 or a negative number ended the pause at once.
-
-The pause now reads the answer the way the rotation does since #739, with
-the same helper (base_plugin.finite_seconds): a numeric string counts;
-anything else that is not a finite number, or a raise, gets the rotation's
-30 s; a number at or below zero gets its 15 s.
-"""
-
-import logging
-import os
-import threading
-from types import SimpleNamespace
-from unittest.mock import MagicMock
-
-os.environ.setdefault("EMULATOR", "true")
-
-import pytest
-
-from src.vegas_mode import coordinator
-
-NOT_NUMBERS = [None, '', 'twenty', True, False, float('nan'), float('inf'),
-               'inf', '1e400', [20], {'seconds': 20}]
-NOT_ABOVE_ZERO = [0, -5, '-5', '0']
-NUMBERS = [('20', 20.0), (' 7.5 ', 7.5), (12, 12.0), (12.5, 12.5)]
-
-
-class FakeClock:
-    """time.monotonic/time.sleep for the pause loop: sleeping moves the clock."""
-
-    #: A pause still going after this long never ends (inf did that).
-    LIMIT = 3600.0
-
-    def __init__(self):
-        self.now = 0.0
-
-    def monotonic(self):
-        return self.now
-
-    def sleep(self, seconds):
-        self.now += seconds
-        if self.now > self.LIMIT:
-            raise RuntimeError("the static pause never ended")
-
-
-@pytest.fixture
-def clock(monkeypatch):
-    fake = FakeClock()
-    monkeypatch.setattr(coordinator, 'time', fake)
-    return fake
-
-
-def _plugin(duration, plugin_id='clock-simple'):
-    plugin = MagicMock()
-    plugin.plugin_id = plugin_id
-    plugin.get_display_duration.return_value = duration
-    return plugin
-
-
-def _coord(*plugins):
-    coord = coordinator.VegasModeCoordinator.__new__(coordinator.VegasModeCoordinator)
-    coord.render_pipeline = MagicMock()
-    coord.render_pipeline.get_scroll_position.return_value = 0
-    coord.display_manager = MagicMock()
-    locks = {plugin.plugin_id: threading.Lock() for plugin in plugins}
-    coord.plugin_manager = SimpleNamespace(get_plugin_lock=locks.__getitem__)
-    coord._state_lock = threading.Lock()
-    coord._static_pause_active = False
-    coord._saved_scroll_position = None
-    coord._should_stop = False
-    coord._live_priority_active = False
-    coord._live_priority_check = None
-    coord._interrupt_check = None
-    coord.stats = {'static_pauses': 0}
-    return coord
-
-
-def _pause(coord, plugin, clock):
-    """One static pause: (whether it completed, how long it lasted)."""
-    start = clock.now
-    completed = coord._handle_static_pause(plugin)
-    return completed, clock.now - start
-
-
-class TestPauseLength:
-    @pytest.mark.parametrize('value, seconds', NUMBERS)
-    def test_numbers_and_numeric_strings_are_used(self, clock, value, seconds):
-        plugin = _plugin(value)
-        completed, lasted = _pause(_coord(plugin), plugin, clock)
-        assert completed is True
-        assert lasted == pytest.approx(seconds, abs=0.15)
-
-    @pytest.mark.parametrize('value', NOT_NUMBERS, ids=repr)
-    def test_anything_but_a_finite_number_pauses_for_30s(self, clock, value):
-        plugin = _plugin(value)
-        completed, lasted = _pause(_coord(plugin), plugin, clock)
-        assert completed is True
-        assert lasted == pytest.approx(30.0, abs=0.15)
-        plugin.display.assert_called_once_with(force_clear=True)
-
-    @pytest.mark.parametrize('value', NOT_ABOVE_ZERO, ids=repr)
-    def test_a_number_not_above_zero_pauses_for_15s(self, clock, value):
-        plugin = _plugin(value)
-        completed, lasted = _pause(_coord(plugin), plugin, clock)
-        assert completed is True
-        assert lasted == pytest.approx(15.0, abs=0.15)
-
-    def test_a_raising_get_display_duration_pauses_for_30s(self, clock):
-        plugin = _plugin(None)
-        plugin.get_display_duration.side_effect = KeyError('display_duration')
-        completed, lasted = _pause(_coord(plugin), plugin, clock)
-        assert completed is True
-        assert lasted == pytest.approx(30.0, abs=0.15)
-
-    def test_a_good_value_after_a_bad_one_is_used(self, clock):
-        plugin = _plugin(None)
-        coord = _coord(plugin)
-        assert _pause(coord, plugin, clock)[1] == pytest.approx(30.0, abs=0.15)
-        plugin.get_display_duration.return_value = 45
-        assert _pause(coord, plugin, clock)[1] == pytest.approx(45.0, abs=0.15)
-
-    def test_the_pause_can_still_be_interrupted(self, clock):
-        plugin = _plugin('twenty')
-        coord = _coord(plugin)
-        coord._interrupt_check = lambda: clock.now >= 5
-        completed, lasted = _pause(coord, plugin, clock)
-        assert completed is False
-        assert lasted == pytest.approx(5.0, abs=0.15)
-
-
-class TestWarning:
-    def test_logged_once_per_plugin(self, clock, caplog):
-        clock_plugin = _plugin('twenty')
-        calendar = _plugin(None, plugin_id='calendar')
-        coord = _coord(clock_plugin, calendar)
-        with caplog.at_level(logging.WARNING, logger='src.vegas_mode.coordinator'):
-            for _ in range(3):
-                for plugin in (clock_plugin, calendar):
-                    coord._handle_static_pause(plugin)
-        warnings = [r.getMessage() for r in caplog.records
-                    if 'display duration' in r.getMessage()]
-        assert len(warnings) == 2
-        assert any('clock-simple' in m and "'twenty'" in m for m in warnings)
-        assert any('calendar' in m and 'None' in m for m in warnings)
-
-
-class TestFiniteSeconds:
-    """The shared rule: what counts as a number of seconds."""
-
-    @pytest.mark.parametrize('value, seconds', NUMBERS + [(0, 0.0), ('-5', -5.0)])
-    def test_numbers_and_numeric_strings(self, value, seconds):
-        from src.plugin_system.base_plugin import finite_seconds
-        result = finite_seconds(value)
-        assert result == seconds and isinstance(result, float)
-
-    @pytest.mark.parametrize('value', NOT_NUMBERS + [pytest.param(10 ** 400, id='10**400')],
-                             ids=repr)
-    def test_anything_else_is_none(self, value):
-        from src.plugin_system.base_plugin import finite_seconds
-        assert finite_seconds(value) is None
-
-
-def _rotation_seconds(plugin):
-    """How long the rotation shows ``plugin`` (no dynamic duration, no
-    Rotation & Durations override): the two calls run() makes for a screen.
-    """
-    from src.display_controller import DisplayController
-    dc = object.__new__(DisplayController)
-    dc.config = {}
-    dc.plugin_modes = {'mode': plugin}
-    return dc._resolve_durations(plugin, 'mode', dc._get_display_duration('mode'), False)[1]
-
-
-class TestSameAsTheRotation:
-    """The pause and the rotation share finite_seconds; this pins their
-    fallbacks (30 s, 15 s) to each other too."""
-
-    @pytest.mark.parametrize('value', [value for value, _ in NUMBERS]
-                             + NOT_NUMBERS + NOT_ABOVE_ZERO, ids=repr)
-    def test_the_pause_lasts_as_long_as_the_rotation_shows_it(self, clock, value):
-        plugin = _plugin(value)
-        expected = _rotation_seconds(plugin)
-        assert _pause(_coord(plugin), plugin, clock)[1] == pytest.approx(expected, abs=0.15)
-
-    def test_a_raise_too(self, clock):
-        plugin = _plugin(None)
-        plugin.get_display_duration.side_effect = KeyError('display_duration')
-        expected = _rotation_seconds(plugin)
-        assert _pause(_coord(plugin), plugin, clock)[1] == pytest.approx(expected, abs=0.15)

test/web_interface/test_starlark_pixlet_routes.py

@@ -26,8 +26,15 @@ import pytest
 
 
 @pytest.fixture
-def client():
+def client(monkeypatch):
+    from web_interface import app as web_app
     from web_interface.app import app
+    # The captive-portal before_request hook shells out to systemctl/nmcli
+    # whenever its 30s cache is cold, so on a Linux host whether a request
+    # here runs subprocess depended on how long ago the previous one was --
+    # and several tests below stub subprocess. Pin it: no test in this file
+    # is about AP mode.
+    monkeypatch.setattr(web_app, 'is_ap_mode_active', lambda: False)
     app.config['TESTING'] = True
     with app.test_client() as c:
         yield c
@@ -1136,12 +1143,23 @@ class TestPixletEditorHostDefaultsButDoesNotOverride:
                 captured['env'] = env
             return FakeProcess()
 
+        # Swap the route module's own ``subprocess`` binding, not the shared
+        # ``subprocess.Popen``: patching the attribute on the real module is
+        # process-wide, and the app's before_request hook (the captive-portal
+        # check) runs ``subprocess.run`` -- ``with Popen(...)`` -- whenever its
+        # 30s AP-mode cache is cold on a host with systemctl. On the Linux CI
+        # runner that handed it this FakeProcess and 500'd the request, but
+        # only when the previous request was more than 30s earlier.
+        fake_subprocess = types.ModuleType('subprocess')
+        fake_subprocess.__dict__.update(mod.subprocess.__dict__)
+        fake_subprocess.Popen = fake_popen
+
         with patch.object(mod, '_validate_starlark_app_path',
                           return_value=(app_dir, None)), \
              patch.object(mod, '_PIXLET_EDITOR_SCRIPT', script), \
              patch.object(mod, '_PIXLET_EDITOR_STATE', state_file), \
              patch.object(mod, '_find_pixlet_binary', return_value='/usr/bin/pixlet'), \
-             patch.object(mod.subprocess, 'Popen', side_effect=fake_popen), \
+             patch.object(mod, 'subprocess', fake_subprocess), \
              patch.dict(os.environ):
             if operator_host is None:
                 os.environ.pop('PIXLET_EDITOR_HOST', None)

test/web_interface/test_web_process_runs_no_plugin_code.py

@@ -221,7 +221,9 @@ class TestTheWebProcessNeverRunsAPlugin:
     def test_saving_its_section_through_the_main_config(self, web):
         body = web.post("/api/v3/config/main", {PLUGIN_ID: {"message": "via main"}})
         assert web.stored()["message"] == "via main"
-        assert body["restart_required"] is True
+        # The display's config watcher hands the section to the running
+        # plugin (on_config_change), as for /plugins/config: no restart.
+        assert body["restart_required"] is False
         assert web.ran() == []
 
     def test_resetting_its_config(self, web):

web_interface/blueprints/api_v3/config.py

@@ -17,6 +17,8 @@ from src.pi5_matrix_support import is_raspberry_pi_5
 from web_interface.cache import invalidate_cache
 from web_interface.auth import SECTION as _WEB_AUTH_SECTION, strip_auth_section
 import web_interface.blueprints.api_v3 as _pkg
+import copy
+from typing import Any, Dict, Iterable, Tuple
 
 # Read through the module rather than bound by value: tests patch these
 # as module attributes, and a value binding would not see the patch.
@@ -33,6 +35,50 @@ FORM_SECTION_FIELD = '__form_section'
 GENERAL_FIELDS = ('timezone', 'city', 'state', 'country', 'web_display_autostart',
                   'plugins_directory', 'auto_update_enabled', 'auto_update_channel')
 
+#: Settings in config.json the running display applies without a restart,
+#: as key paths (a path covers everything under it). Brightness goes over the
+#: control socket (brightness.set) and the config watcher's refresh
+#: (DisplayController._refresh_config_cache); the per-mode durations are read
+#: from the live config each time a mode starts (_get_display_duration).
+#: Plugin sections are live as well (each plugin's on_config_change), and
+#: save_main_config adds the ones a request saves.
+LIVE_CONFIG_PATHS: Tuple[Tuple[str, ...], ...] = (
+    ('display', 'hardware', 'brightness'),
+    ('display', 'display_durations'),
+)
+
+
+_MISSING = object()
+
+
+def _config_leaves(config: Any, prefix: Tuple[str, ...] = ()) -> Dict[Tuple[str, ...], Any]:
+    """Every non-dict value in ``config``, by key path. An empty dict has none,
+    so a section created empty on the way to a field is not a change."""
+    if not isinstance(config, dict):
+        return {prefix: config}
+    leaves: Dict[Tuple[str, ...], Any] = {}
+    for key, value in config.items():
+        leaves.update(_config_leaves(value, prefix + (str(key),)))
+    return leaves
+
+
+def restart_needed(before: Dict[str, Any], after: Dict[str, Any],
+                   live_paths: Iterable[Tuple[str, ...]] = LIVE_CONFIG_PATHS) -> bool:
+    """Does going from config ``before`` to ``after`` need a display restart?
+
+    True when anything changed outside ``live_paths``. A save that changes
+    only live settings, or nothing at all, does not.
+    """
+    live = tuple(live_paths)
+    old, new = _config_leaves(before), _config_leaves(after)
+    for path in set(old) | set(new):
+        if old.get(path, _MISSING) == new.get(path, _MISSING):
+            continue
+        if not any(path[:len(prefix)] == prefix for prefix in live):
+            return True
+    return False
+
+
 #: Top-level fields save_main_config stores somewhere of its own (location,
 #: plugin_system, ...), never as a config key of the same name.
 _MAPPED_TOP_LEVEL_FIELDS = GENERAL_FIELDS + (
@@ -72,6 +118,22 @@ def _day_setting(data, day, flat_key, nested_key):
     return False, None
 
 
+def _disabled_day_times(data, day, start_key, end_key):
+    """The times posted for a day that is off, the valid ones.
+
+    Nothing reads them while the day is off, but the schedule picker posts
+    them and GET returns them, so keeping them means turning the day back on
+    finds what was there. An invalid one is dropped rather than refused, for
+    the same reason.
+    """
+    times = {}
+    for field, key in (('start_time', start_key), ('end_time', end_key)):
+        value = _day_setting(data, day, key, field)[1]
+        if value and _validate_time_format(value)[0]:
+            times[field] = value
+    return times
+
+
 @api_v3.route('/config/main', methods=['GET'])
 def get_main_config():
     """Get main configuration, with credentials redacted."""
@@ -257,11 +319,16 @@ def save_schedule_config():
 
                     day_config['start_time'] = start_time
                     day_config['end_time'] = end_time
+                else:
+                    day_config.update(_disabled_day_times(data, day, start_key, end_key))
 
                 schedule_config['days'][day] = day_config
 
-            # Validate that at least one day is enabled in per-day mode
-            if enabled_days_count == 0:
+            # An enabled per-day schedule needs a day to be on. A disabled
+            # one does not: every day off with the schedule off is what
+            # config.template.json ships, so refusing it meant a fresh
+            # install could not post back the schedule GET returned.
+            if enabled_days_count == 0 and enabled_value:
                 return error_response(
                     ErrorCode.VALIDATION_ERROR,
                     "At least one day must be enabled in per-day schedule mode",
@@ -465,11 +532,13 @@ def save_dim_schedule_config():
 
                     day_config['start_time'] = start_time
                     day_config['end_time'] = end_time
+                else:
+                    day_config.update(_disabled_day_times(data, day, start_key, end_key))
 
                 dim_schedule_config['days'][day] = day_config
 
-            # Validate that at least one day is enabled in per-day mode
-            if enabled_days_count == 0:
+            # As for the on/off schedule: only an enabled one needs a day on.
+            if enabled_days_count == 0 and enabled_value:
                 return error_response(
                     ErrorCode.VALIDATION_ERROR,
                     "At least one day must be enabled in per-day dim schedule mode",
@@ -557,6 +626,8 @@ def save_main_config():
 
         # Merge with existing config (similar to original implementation)
         current_config = api_v3.config_manager.load_config()
+        # What was stored, to tell which settings this save changed.
+        stored_config = copy.deepcopy(current_config)
         was_auto_update_enabled = bool((current_config.get('auto_update') or {}).get('enabled'))
 
         is_general_update = any(k in data for k in GENERAL_FIELDS)
@@ -1194,13 +1265,15 @@ def save_main_config():
                 message = f'{message}. {note}'
         except Exception:
             logger.warning("Automatic update setup could not be started", exc_info=True)
-        # Display hardware, rotation/durations and general settings take
-        # effect after a display restart; the UI shows its restart banner on
-        # this flag.
-        extra = {'restart_required': True}
-        # Brightness is the exception: the display applies a saved one
-        # without a restart. Over the control socket it lands at once,
-        # instead of when the config watcher next looks (up to ~2 s).
+        # Display hardware, rotation order and general settings take effect
+        # after a display restart; the UI shows its restart banner on this
+        # flag. Brightness, mode durations and plugin settings are applied by
+        # the running display (LIVE_CONFIG_PATHS), so a save that changed
+        # only those -- or nothing -- does not ask for one.
+        live_paths = LIVE_CONFIG_PATHS + tuple((plugin_id,) for plugin_id in plugin_keys_to_remove)
+        extra = {'restart_required': restart_needed(stored_config, current_config, live_paths)}
+        # Over the control socket a saved brightness lands at once, instead
+        # of when the config watcher next looks (up to ~2 s).
         if 'brightness' in data:
             saved = (current_config.get('display', {}).get('hardware', {}) or {}).get('brightness')
             if isinstance(saved, int) and not isinstance(saved, bool):

web_interface/blueprints/api_v3/display.py

@@ -386,15 +386,22 @@ def get_current_display_status():
     Read from the display's state stream over the control socket when it is
     available (``source: "socket"``). Otherwise from what the display
     publishes to the shared cache (display_controller._publish_current_mode_state)
-    when the active mode changes (``source: "cache"``).
+    when the active mode changes (``source: "cache"``). Unknown (every field
+    None) when the socket and the heartbeat both say the display is gone
+    (display_state.display_gone).
     """
-    state = display_state.current_status(display_state.read_state())
+    snapshot = display_state.read_state()
+    state = display_state.current_status(snapshot)
     source = 'socket'
     if state is None:
         source = 'cache'
-        cache = _cache_manager()
-        # memory_ttl=0: written by the display service; see get_on_demand_status.
-        state = cache.get('display_current_state', max_age=120, memory_ttl=0)
+        # A stopped display leaves its last answer in the cache, where it
+        # read as on (is_display_active: true) for the 120 s max_age. With
+        # no socket and no live heartbeat there is no display behind it.
+        if not display_state.display_gone(snapshot):
+            cache = _cache_manager()
+            # memory_ttl=0: written by the display service; see get_on_demand_status.
+            state = cache.get('display_current_state', max_age=120, memory_ttl=0)
     if state is None:
         state = {
             'mode': None,

web_interface/blueprints/api_v3/misc.py

@@ -102,6 +102,7 @@ def get_health():
         # the only signal, as it always was.
         # The display reports the same beat's age over the control socket's
         # state stream, measured in memory; the file is the fallback.
+        snapshot = None
         try:
             snapshot = display_state.read_state()
             if snapshot is not None:
@@ -132,6 +133,26 @@ def get_health():
                 'error': 'see logs for details'
             }
 
+        # A stopped display service. The heartbeat's absence alone says
+        # nothing (the dev server, the emulator and Windows write none), so
+        # the overall status stayed "healthy" with the display down until the
+        # last preview frame it left aged past 60 s (hardware: stale). Together
+        # the three signals are definite: systemd says the service is not
+        # active, the control socket does not answer, and there is no live
+        # heartbeat (display_state.display_gone, which is never true where
+        # the platform has no socket or it is switched off).
+        try:
+            if (not display_service_status.get('active')
+                    and display_state.display_gone(snapshot)):
+                health_status['checks']['display_loop'] = {
+                    'status': 'stopped',
+                    'note': 'The display service is not running',
+                    'source': 'service',
+                }
+        except Exception:
+            logger.warning("Health check could not tell whether the display is stopped",
+                           exc_info=True)
+
         # Check hardware connectivity (if display manager available)
         try:
             snapshot_path = display_preview.SNAPSHOT_PATH

web_interface/blueprints/api_v3/plugin_store.py

@@ -530,13 +530,11 @@ def install_plugin():
                     )
 
                 branch_msg = f" (branch: {branch})" if branch else ""
-                # plugin_id: the id to enable it by, and the id its config
-                # section is under (see _installed_plugin_id).
-                installed_id = _installed_plugin_id(plugin_id)
+                # plugin_id: the id to enable it by (see _installed_plugin_id).
                 return {'success': True,
                         'message': f'Plugin {plugin_id} installed successfully{branch_msg}',
-                        'plugin_id': installed_id,
-                        **_store_restart_fields('install', _plugin_enabled_in_config(installed_id))}
+                        'plugin_id': _installed_plugin_id(plugin_id),
+                        **_store_restart_fields('install', _plugin_enabled_in_config(plugin_id))}
             else:
                 error_msg = f'Failed to install plugin {plugin_id}'
                 if branch:
@@ -590,11 +588,10 @@ def install_plugin():
                 )
 
             branch_msg = f" (branch: {branch})" if branch else ""
-            installed_id = _installed_plugin_id(plugin_id)
             return success_response(
                 message=f'Plugin installed successfully{branch_msg}',
-                extra={'plugin_id': installed_id,
-                       **_store_restart_fields('install', _plugin_enabled_in_config(installed_id))})
+                extra={'plugin_id': _installed_plugin_id(plugin_id),
+                       **_store_restart_fields('install', _plugin_enabled_in_config(plugin_id))})
         else:
             error_msg = f'Failed to install plugin {plugin_id}'
             if branch:

web_interface/display_state.py

@@ -133,6 +133,39 @@ def on_demand_state(snapshot: Optional[Dict[str, Any]],
     return state
 
 
+def display_gone(snapshot: Optional[Dict[str, Any]]) -> bool:
+    """Is there positively no display behind a fallback to the cache?
+
+    True only when the socket should be there (this platform has one and it
+    is not switched off) but gave no ``snapshot``, and the render loop's
+    heartbeat file says nothing is running either: it is absent (systemd
+    removes its directory when the service stops), stale, or written by a
+    process that no longer exists -- #726's rules for the runtime snapshot.
+    Then what the display last left in the cache is a dead process's answer.
+
+    False whenever the answer is in doubt: a snapshot came in, the socket is
+    off or unsupported (Windows, the test suite, a deliberate ``off``), or a
+    live heartbeat says the display is running without a socket (an older
+    display). Those read the cache exactly as before.
+    """
+    if snapshot is not None:
+        return False
+    if not socket_supported() or not client_socket_paths():
+        return False
+    from src import display_watchdog
+    from src.plugin_system.plugin_runtime import process_exists
+    heartbeat = display_watchdog.read_heartbeat(display_watchdog.HEARTBEAT_PATH)
+    if heartbeat is None:
+        return True
+    age = display_watchdog.heartbeat_age(heartbeat)
+    if age is None or age >= display_watchdog.HEARTBEAT_STALE_SECONDS:
+        return True
+    pid = heartbeat.get('pid')
+    if isinstance(pid, int) and not isinstance(pid, bool) and process_exists(pid) is False:
+        return True
+    return False
+
+
 def loop_heartbeat_age(snapshot: Optional[Dict[str, Any]]) -> Optional[float]:
     """The render loop's heartbeat age now; None when the display has no
     beat to report yet (or there is no snapshot)."""
@@ -141,5 +174,5 @@ def loop_heartbeat_age(snapshot: Optional[Dict[str, Any]]) -> Optional[float]:
     return control_client.snapshot_loop_age(snapshot)
 
 
-__all__ = ['current_status', 'loop_heartbeat_age', 'on_demand_state', 'read_state',
-           'stop_subscription']
+__all__ = ['current_status', 'display_gone', 'loop_heartbeat_age', 'on_demand_state',
+           'read_state', 'stop_subscription']