houseplan-card/docs/USER-GUIDE.md

# House Plan — complete user guide

Current for **v1.73.0**. This guide describes the interface implemented by the
current source. [Русская версия](USER-GUIDE.ru.md).

House Plan adds a dedicated **House Plan** page to the Home Assistant sidebar.
That full-page panel is the primary way to view and edit the shared plan. The
integration also installs two optional Lovelace cards:

- `custom:houseplan-card` — the live plan, editors, state and actions;
- `custom:houseplan-space-card` — an inert rendering of one space with a link
  to the full plan.

Configuration, uploaded files and shared positions stay inside Home Assistant.
Only the explicit Help & feedback action can contact the House Plan support
relay, and exact plan geometry is attached only after you opt in and preview it.

> **Input support.** View and kiosk are supported on phones, tablets and wall
> touch panels. Create and maintain a plan on a desktop browser with a mouse
> and keyboard. Touch editing is best effort: an operation may be awkward,
> limited or absent, but it must not corrupt data or trigger an accidental
> Home Assistant action. The authority is [TOUCH-SUPPORT.md](TOUCH-SUPPORT.md).

## Contents

1. [Data model and terms](#1-data-model-and-terms)
2. [Installation and access](#2-installation-and-access)
3. [Adding a card](#3-adding-a-card)
4. [Quick start](#4-quick-start)
5. [Interface modes](#5-interface-modes)
6. [Navigation, zoom and input](#6-navigation-zoom-and-input)
7. [Spaces](#7-spaces)
8. [Rooms and walls](#8-rooms-and-walls)
9. [Doors, windows, gates and locks](#9-doors-windows-gates-and-locks)
10. [Devices](#10-devices)
11. [Tap actions](#11-tap-actions)
12. [Device visual states](#12-device-visual-states)
13. [Room fills and light](#13-room-fills-and-light)
14. [Background editor](#14-background-editor)
15. [Sun background and window rays](#15-sun-background-and-window-rays)
16. [Robot vacuums](#16-robot-vacuums)
17. [Kiosk](#17-kiosk)
18. [Static space card](#18-static-space-card)
19. [Plan maintenance](#19-plan-maintenance)
20. [Storage, multiple cards and backups](#20-storage-multiple-cards-and-backups)
21. [Current limitations](#21-current-limitations)
22. [Troubleshooting](#22-troubleshooting)
23. [Help and private feedback](#23-help-and-private-feedback)

<!-- docs-section: model -->

## 1. Data model and terms

Settings form a hierarchy. Global settings provide defaults; a space may
override them; a room may override its space; a marker may override its room.

| Level | Meaning | Stored data |
|---|---|---|
| Panel/card | Primary sidebar page or one optional dashboard instance | Last/initial space; dashboard cards may additionally set language, icon size, value/LQI display, kiosk and cycle |
| Global settings | Defaults for all spaces | Fill palette, background, Glow radius, north, sun, room-hover information and icon rules |
| Space | Floor, yard, garage or building | Plan image, scale, rooms, walls, openings, decor and display settings |
| Room | A closed outline | Name, optional HA area, temperature/humidity source and local fill |
| Wall | A room-contour or independent segment | Stable identity and thickness from 0 to 100 cm; zero-thickness appearance is selected per space |
| Opening | Door, window or gate on a wall | Size, orientation, contact and optional lock |
| Marker | A device shown on the plan | HA binding, room, action, presentation, light role and attachments |
| Background item | Visual context | Line, shape, text, furniture or plan-image transform |

A room is the spatial unit that owns area and an optional HA area. A partition
or column affects the physical rendering and light but does not create a room.

<!-- docs-section: installation -->

## 2. Installation and access

### Requirements

| Component | Requirement |
|---|---|
| Home Assistant | 2024.6.0 or newer |
| Installation | Any installation that supports custom integrations |
| Browser | Web Components, SVG and Pointer Events |
| Editing permission | Administrators by default |

### HACS

1. Search for **House Plan** in HACS and install it — the integration is in
   the HACS default catalog, no custom repository needed.
2. Restart Home Assistant.
3. Open **Settings → Devices & services → Add integration → House Plan**.
4. Keep “administrators only” enabled unless other users must edit the plan.

The integration registers both the sidebar panel and its Lovelace resource
automatically. After installing or updating House Plan, restart Home Assistant
and fully reload the page:
`Ctrl+F5` on Windows/Linux or `Cmd+Shift+R` on macOS.

#### Storage mode (Home Assistant default)

No YAML is normally needed. If automatic registration did not make the card
available, open **Settings → Dashboards → menu ⋮ → Resources → Add
resource**, enter `/houseplan_files/houseplan-card.js`, and select **JavaScript
module**.

#### YAML resources mode (Home Assistant 2026.2+)

To manage resources in `configuration.yaml` independently of the dashboard
mode, use:

```yaml
lovelace:
  resource_mode: yaml
  resources:
    - url: /houseplan_files/houseplan-card.js
      type: module
```

#### Legacy Home Assistant 2024.6–2026.1

Only for a full-YAML dashboard that is already managed in YAML, use:

```yaml
lovelace:
  mode: yaml
  resources:
    - url: /houseplan_files/houseplan-card.js
      type: module
```

`mode: yaml` changes the dashboard itself to YAML mode. Do not switch a storage
dashboard to legacy YAML just for House Plan; use the Storage mode instructions
above instead. Do not use
`/custom_components/houseplan/frontend/houseplan-card.js`; it is an on-disk
path, not the JavaScript URL served by Home Assistant.

### Manual installation

Copy the release folder to `config/custom_components/houseplan`, restart Home
Assistant, and add the integration. If the card is still unavailable, follow
the Storage or YAML resource instructions above for your Home Assistant
version and resource mode. Always copy the complete integration folder: the
stable resource URL remains one file, but that entry loads internal
content-hashed modules. A lone `houseplan-card.js` is not a supported install.

The ordinary View does not download editor code. The first opening of Plan,
Device or Background may therefore take a brief moment. If that internal module
cannot be loaded after one retry, the plan stays in View; no half-open editor is
kept. After a network failure the card invites you to check the connection and
press again — the next press starts a fresh download. Only when the tab holds
code from another build does the advice ask for a page refresh instead. A fully
stale proxy-cached `houseplan-card.js` no longer leaves an empty card after an
update: it shows a panel asking to reload the page.

### Permissions

Every signed-in user can view the plan. Home Assistant permissions still govern
device service calls. With the default integration option, only administrators
can edit configuration or upload files. Plan optimization and its undo always
require an administrator.

The **House Plan** sidebar item is visible to every signed-in user. A user who
cannot edit gets the complete View experience without editor controls. On an
empty installation that user sees a read-only explanation, not an Add space
button or an automatically opened setup dialog.

## 3. Adding a card

Normally there is nothing to add: open **House Plan** in the HA sidebar. It
occupies the available page area, keeps the existing space/editor controls, and
does not duplicate the product title inside the plan. Leaving the page ends an
editor session; returning keeps the last space and opens View.

The dashboard card remains available for layouts that intentionally embed the
plan. In a Sections view it requests full width by default; a size explicitly
chosen in Home Assistant remains authoritative.

Minimal configuration:

```yaml
type: custom:houseplan-card
title: House plan
```

| Field | Default | Purpose |
|---|---:|---|
| `title` | empty | Card title |
| `default_floor` | first/last opened | Initial/fallback space for an unpinned card |
| `floor` | not pinned | Keep this card on one stable space ID or zero-based YAML index |
| `language` | HA language | `auto`, `en`, `ru` or `de` |
| `icon_size` | `2.5` | Base marker size, 1–6% of plan width |
| `show_temperature` | `true` | Compact temperature and humidity values |
| `live_states` | `true` | Work/open/unavailable presentation and activity; alarms remain visible |
| `show_signal` | `true` | Base LQI display, overridable by a space |
| `kiosk` | `false` | Full plan without editors or header |
| `cycle` | `0` | Kiosk auto-cycle interval in seconds; `0` disables it |

The legacy card-level `tap_action` is ignored. Each marker owns its action.

`auto` recognizes the primary Home Assistant locale: `de`, `de-DE`, `de-AT`
and `de-CH` use German, while `fr`, `fr-FR`, `fr-CA`, `fr-BE` and `fr-CH` use
French (a community translation — thank you, @OUARZA). Lazy dictionaries
(German, French) are downloaded once on first use and shared by
all House Plan cards on the page. Until it is ready, a neutral busy surface is
shown instead of briefly flashing English; if both bounded download attempts
fail, the card becomes usable in English and says so with a "Could not load
the language pack" toast.

<!-- docs-section: first-run -->

## 4. Quick start

Use this order for a first working room:

1. Add a space with **+**, or import Home Assistant floors.
2. Upload SVG/PNG/JPG/WebP, reuse an uploaded plan, or choose no image.
3. Set the real size of a grid cell. It controls wall, opening, furniture and
   area measurements.
4. In Plan choose **Walls**, draw the boundary, then click its first point to
   close the first room face.
5. Name the room and bind it to an HA area, or use **No area**.
6. Add wall thickness and openings if needed.
7. In Device, position auto-discovered markers and choose their actions.
8. In Background, align the plan image and add labels, lines or furniture.
9. Return to View and confirm live values and actions.

![Creating the first space with synthetic data](images/03-space-create.png)

![Closing a new wall chain on its first point](images/04-room-contour-close.png)

Do all plan creation on desktop. Phone, tablet and wall-panel View remain full
product surfaces after setup.

<!-- docs-section: modes -->

## 5. Interface modes

View is the state with no editor open. Close the active editor to return to it.

| Mode | Devices | Geometry | Background/openings |
|---|---|---|---|
| View | Live and actionable | Read-only; room hover shows summary | Visible according to space settings |
| Plan | Hidden | Rooms, walls, columns and openings editable | Openings always visible |
| Device | Draggable; click opens settings | Read-only background | Read-only background |
| Background | Dimmed and inert | Dimmed | Background objects editable |
| Kiosk | Actionable as in View | Read-only | Read-only; no editors |
| Static card | Live states, values and alarms as in View (`live_states`), no tooltips or more-info | Render only | Render only; not interactive |

### Summary panel

The two-part control at the end of the View header opens **Summary panel
settings** with its left gear and shows or hides the panel with its right
sidebar icon. Only the right half is highlighted when the panel is on. In kiosk
the control floats above the plan. The panel is off on a new screen until you
explicitly show it. Its contents are shared for the whole House Plan
installation; the show/hide choice belongs only to your Home Assistant user
and this particular card or sidebar panel. A read-only user and kiosk can
change that local choice without being allowed to edit the shared contents.

An administrator can name the panel, add up to ten ordered blocks, show a block
on every space or one selected space, and add up to twenty values to a block.
A value can be the state of any entity available to that user, the number of
real HA devices represented on valid plans, the total clean room area across
all spaces, or the current date and time. Entity values use Home Assistant's
own formatting and units. Rows are informational only: pressing them never
opens more-info or calls a service.

Press a value's source field to open its picker. Only that row's picker is
shown: search by friendly name or exact entity ID, then choose one of the
matching entities or a built-in value. A broad search shows the first 100
matches and asks you to refine it; the search still covers every entity. Escape
or a press outside closes the picker without changing the draft. A missing old
source remains visible until you explicitly replace it.

The wide settings dialog separates general settings from the block cards.
Use a block's eye button to show or hide it, its grip or arrow buttons to
reorder it, and the dashed add buttons to add values or blocks. A source field
shows its friendly name above the entity ID. On a narrow screen or with
enlarged text the fields stack; Save and Cancel remain in the footer. There
are no icon/text size controls in this dialog; previously saved per-card
sizes are preserved.

**Display on mobile devices** is disabled while **Show panel on this card**
is off. Its previous value is kept, not reset. Both switches are drafts until
Save; Cancel, Escape or closing the dialog discards their changes. Leaving the
House Plan page, reconnecting under another user, or changing edit permission
closes an open summary draft rather than carrying it into the new context.

The compact panel floats over the plan without resizing it, with a separate
header and scrollable block cards. Showing and hiding it both use a brief
slide and fade; reduced-motion preferences are respected. It appears on the right
when the House Plan working area is at least as wide as it is tall and at the
bottom otherwise. It temporarily hides when the card cannot fit a readable
panel, and restores itself after the card grows. Turning off **Display on
mobile devices** also hides it whenever Home Assistant reports a narrow view;
the show/hide button remains pressed because the local choice was not erased.
The panel and its controls are absent from all three editors and from the
static space card.

The room highlight remains available in View and kiosk. To keep that highlight
but hide the floating room summary, turn off **General settings → Show the room
information window on hover**. The option is on by default and does not affect
device tooltips.

For an occasional Zigbee placement check, an administrator can enable
**General settings → Show Zigbee links when hovering over a device**. The option
is off by default. Load the provider snapshot there: **Read ZHA data** reads
ZHA's existing cache, while **Update map** starts an explicit Zigbee2MQTT raw
network-map scan for each entered base topic (default `zigbee2mqtt`). The latter
may take 10 seconds to 2 minutes and can temporarily slow the Zigbee network.

After data is loaded, moving a real mouse over a mapped Zigbee marker shows
only its observed direct neighbours. Links to markers on the current space are
lines; drawable neighbours on other spaces are summarized as a temporary
count. An arrow on a line shows the next step towards the coordinator: an
ordinary device points to its parent, while arrows pointing into a router show
devices whose path goes through it. Neighbour links outside the derived path
tree remain plain lines.

The active diagnostic layer is deliberately drawn above room names and devices
that are not part of the shown link, so a busy plan cannot hide the route. The
complete source and locally connected device markers remain above the lines.
An unknown-quality gray dashed link has a thin dark outline for contrast; this
does not change its meaning. The whole layer is pointer-transparent, so device
and room actions continue to work normally.

If the next step is in another space, a short bubble names that space. If the
needed router or coordinator is not placed on the plan, the bubble says so. If
the snapshot has no coordinator or the graph is disconnected, House Plan does
not invent a direction and leaves the link without an arrow. This is a stable
path approximation derived from the neighbour snapshot, not the route used by
every current packet. The layer does not appear on touch/pen, in kiosk, in
editors or in the static card, and hovering never starts a scan.

Each editor has a stable primary toolbar. Tool parameters and selected-object
actions appear in a context tray over the top of the canvas. On a narrow screen
the tray scrolls horizontally instead of shrinking the plan.

Where an editor offers a color sample, one click opens the House Plan color
picker. Hue, saturation, brightness, an exact HEX value and opacity (when the
setting supports it) are available together; there is no second browser color
dialog. The Hue track shows the full colour spectrum at a glance, with a
contrasting ring keeping its slider visible. Changes remain a draft until the
owning properties dialog is saved.

This is the same control everywhere: decor and custom fills, the global
light/temperature/LQI/Glow/wall palettes, global and per-space backgrounds,
room colour, marker Glow and the device activity ripple. Settings that already
have opacity show it in the picker; colour-only settings do not acquire one.
**Default** and **Inherited** background actions remain beside the colour
sample and do not save the displayed fallback unless a colour is changed.

<!-- docs-section: input -->

## 6. Navigation, zoom and input

The canvas grows with actual content. **Fit all** frames rooms and any distant
objects; each space keeps its local View viewport.

| Scenario | Mouse | Touch View | Touch editors | Keyboard |
|---|---|---|---|---|
| Zoom and pan | Wheel; drag empty space; `−`/`+`; double-click free background to Fit all | Pinch; drag; double-tap free background to Fit all | Available but precision is not guaranteed; no Fit-all double-tap | — |
| Room | A clean click fits the room with 10% margins | One tap fits the room; repeated room taps never become Fit all | — | `Enter`/`Space` on a visible room label |
| Change space | Click a tab | Tap; kiosk swipe at 1:1 | Tap a tab | — |
| Device | Click/double-click per mode | Tap; safe actions equal desktop | Drag/properties are best effort | `Esc` closes the top surface |
| Walls drawing or precise drag | Full contract | Not applicable | Best effort; use desktop for Resize and exact nodes | `Shift` changes magnet/angle; `Esc` finishes a Walls chain or cancels the current precise drag |
| Editor history | Undo/Redo controls | Not applicable | Controls may work; no gesture guarantee | `Ctrl/Cmd+Z`, `Ctrl/Cmd+Shift+Z`, `Ctrl+Y` |
| Kiosk sizing | — | Hold empty space for 3 seconds | — | — |

At zoom above 1:1 a horizontal kiosk gesture pans instead of changing space.
Any manual kiosk operation pauses auto-cycle for 60 seconds.

Wheel, `−`/`+`, **Fit all**, the return arrow and a free-background
double-click/tap in View or kiosk use a short smooth camera transition. Rapid
wheel input changes the current destination
instead of building a queue. Pinch and pan stay directly under the fingers.
With the operating system's reduced-motion preference enabled, every zoom
command is immediate.

### Cancel and undo

- `Esc` finishes an active Walls chain without deleting its accepted segments.
  In Split and other tools it cancels the unfinished path, current
  drag/resize/rotation, or the top dialog without undoing an already committed
  action.
- `Ctrl/Cmd+Z` undoes an editor command; redo is `Ctrl/Cmd+Shift+Z` or `Ctrl+Y`.
- Undo/Redo stores up to 50 named commands for Plan and Background.

<!-- docs-section: spaces -->

## 7. Spaces

A space may represent a floor, garden, garage or separate structure. It stores
its image, grid scale, geometry, Background layer and display overrides.

### Plan source

- upload SVG, PNG, JPG or WebP;
  Very large rasters (roughly 32 megapixels and up) are recognised from the
  file header before anything heavy happens: a dialog shows the exact
  resolution and memory numbers and offers a safe reduced copy. Your original
  file is never modified and the current plan stays untouched in every
  outcome. Images wider than 16384 px per side cannot be displayed by
  browsers at all — reduce those on a desktop first. SVG uploads as-is and is
  never rasterised.
- select a previously uploaded plan;
- choose no image and draw the geometry by hand.

The plan image keeps its proportions initially. Background can later move,
scale or rotate it. Detaching a plan never deletes its server file; deletion
requires an explicit user action.

### Grid scale

The scale field is the real size of one grid cell. A new metric space starts at
**1 cm per cell**; with an imperial Home Assistant unit system it starts at
**1 in per cell** and the field is shown in inches. Floor import uses the same
default for every new space.

Choose a finer cell when you need more precise snap points. It changes only the
number of grid points per metre, not how the finished plan looks: physically
equal rooms, walls, openings, labels and markers retain the same appearance.
Existing spaces keep their stored scale. A legacy space without a scale still
uses the 5 cm compatibility fallback and is not silently migrated.

### Tab order

Space tabs follow the order in which the spaces were created, and that order can
be changed: in any editor mode, grab a tab with the mouse and drag it to a new
position. The new order is saved immediately and applies everywhere — the tabs,
the kiosk swipe between floors and the carousel arrows. While dragging, a thin
divider shows the exact insertion side; release outside the tab strip to keep
the existing order.

Dragging works **with a mouse and in the editors only**. In ordinary View and on
touch screens a tab still does one thing: it switches the space. There it is the
primary way to navigate, and a gesture must not compete with a plain tap. Order
is changed on a computer, like the rest of the plan work.

If a card anywhere pins its floor **by number** (`floor: 0`), remember that the
number means a position: after a reorder such a card shows a different floor.
The card warns about this once. Pin the floor by space id instead of a number to
avoid it entirely.

### Display settings

A space can show/hide room borders, names, LQI, Background and openings. It can
override fill mode, Glow, day-cycle background, north, sun rays, room-card font
scale and which room metrics are visible.

Room cards are positioned and scaled on the plan. View renders only the metrics
enabled for that space.

In an existing space's settings, **Copy** creates a new space from its walls,
openings, columns, Background objects, image transform, scale and display
settings. The suggested name is the first free numbered copy, but duplicate
names are allowed. Rooms and device placements are deliberately not copied;
room walls become ordinary independent walls, ready for a different room
layout. House Plan opens the accepted copy in Plan with **Walls** selected. If
the saved plan must first be repaired, a separate warning explains that
**Optimize plans** will change the whole plan; cancelling that warning writes
nothing.

Deleting a space is blocked while any active device still points to the space,
one of its rooms or a saved position on it, provided another space remains.
Move or delete those devices first; then the confirmed delete removes the
space-owned layout. The sole remaining space can still be deleted after
confirmation: affected devices keep their bindings, icons, actions and settings
but become unplaced. Plan images and attachments are not deleted automatically.

![Room card with temperature, LQI and light state](images/08-room-card.png)

<!-- docs-section: plan-tools -->

## 8. Rooms and walls

### Create a room

Select **Walls** and draw one continuous chain. Every completed segment is
saved immediately as an ordinary independent wall. Changing tool, editor,
floor or leaving the card finishes the session-local chain: compatible straight
sections become one wall and a proven duplicate over room masonry is absorbed
immediately. Undo/Redo remains segment-by-segment but restores a canonical
result, so a normal current-version chain leaves no work for **Optimize plans**.
Reloading preserves the accepted walls but cannot finish the interrupted
session. When the
latest segment creates bounded endpoint/T/X faces, House Plan offers them from
smallest to largest. Save creates that room and consumes exactly coincident
chain walls, Keep as walls rejects only that candidate, and Cancel leaves all
accepted walls in place with no partial rooms.

While drawing an open chain, `Esc` finishes all accepted segments as ordinary
independent walls and keeps **Walls** selected; the next click starts a new
chain. `Ctrl/Cmd+Z` instead removes the last accepted point and segment. Pan,
pinch and `pointercancel` neither finish the chain nor add geometry.

Existing segment endpoints and lines appear above walls while drawing. An
endpoint grows when the next click will join it. A point on a line shows where
the click will create a valid junction.

A segment which is visually horizontal or vertical within `0.25°` is stored as
an exact axis: House Plan moves only the free endpoint, and preview already
shows the final result. A real diagonal remains unchanged. Older invisible
one-grid-step slopes are offered separately by **Optimize plans**, with the
number of walls and maximum movement shown before confirmation.

### Plan tools at a glance

| Tool | Result | Room area | Light and shadow | Main limit |
|---|---|---|---|---|
| Walls | Continuous wall chain; offers rooms when it closes faces and finishes open chains as independent walls | Only a confirmed room has area | Positive thickness blocks light; zero thickness follows the space's dashed/solid policy | Partial room overlap is rejected; there is no separate Partition or Boundary drawing tool |
| Column | Square or circular support | Does not change area | Blocks light inside its shape | One shape/size/rotation; not a wall or room |
| Opening | Door, window or gate | Does not change area | Door/gate passage follows state; window may cast sun | Must fit completely on a suitable wall segment |

Other operations edit existing geometry:

| Operation | Result |
|---|---|
| Merge | Joins adjacent rooms; a dialog chooses the surviving identity, name and area |
| Split | Cuts a room from one wall to another; the larger part keeps the original room |
| Resize | Moves one eligible horizontal/vertical wall without changing room topology. Live labels report the two changing **inner** side-wall dimensions, highlight those walls, and place each affected room's area beside its side of the moving wall |
| Thickness | Changes one span or every wall of a room, including zero-thickness walls |
| Delete room | Deletes the room after choosing whether its exclusive physical walls remain; shared walls always remain |

![Selected partition and its Plan context tray](images/05-plan-context-tray.png)

Deleting a room also clears that exact room from direct device assignments and
vacuum segment maps, including maps owned by a vacuum on another floor.
Merging rooms redirects the same references to the surviving room. The change
is part of the room command: Undo and Redo restore or reapply geometry and
references together without changing unrelated marker settings.

Wall thickness is stored in real units. A room may have different thicknesses
on different spans. Open wall branches and T-junctions are allowed; shared
geometry remains joined without painted end caps. At a saved T-junction both
physical half-walls stay solid: the bounded bevel removes only an excessive
projecting corner and never leaves a white triangular gap. A short arm stops at
its real endpoint; the join repair never draws a cap, shadow or light barrier
where no wall was saved. At junctions of three or more walls the removed bevel
remains connected to the surrounding room/background rather than becoming a
small enclosed hole. At a perpendicular T/X junction, the complete physical
width of every participating wall remains solid through the node, including
closely spaced neighbouring junctions. This is a rendering correction: a valid
plan may remain unchanged when **Optimize plans** is run.
Legacy near-axis walls are different: **Optimize plans** may explicitly
straighten them, counts a shared wall once across both rooms and leaves an
unsafe candidate unchanged. Cancel writes nothing and the confirmed batch has
the normal one-deep Undo.
Resize changes one room, or exactly two rooms when their shared wall coincides
endpoint-to-endpoint. The wall stops at the first corner, opening, foreign room
or other position that would change topology; no more than two rooms can change.
It also stops where extending or shortening an adjacent wall would turn shared
material into outer material (or the reverse), so one saved thickness never
silently serves both roles. If neither direction has even one safe grid step,
the handle explains that only part of a shared wall cannot be moved.
Resize also preserves every unrelated wall exactly: changing the length of a
neighbouring wall cannot shift a thickness boundary to an invented off-grid
point. An ambiguous candidate is rejected instead of damaging another wall.
Partial shared walls, diagonal walls and walls overlapped by an independent
partition/column keep a dimmed handle with an explanatory tooltip and
cannot start a drag. The former corner scale frame was removed. An ordinary
opening on the moving wall follows it once; a side-wall opening stops the
moving masonry at its physical jamb. Release creates one Undo step, while Esc
or an interrupted pointer writes nothing.

A wall thickness of **0 cm** is a real wall-axis record without a masonry body:
it does not create hatch, wall area, an opening tunnel or a valid opening host.
It is drawn and edited with the same Walls and Thickness tools as every other
wall. In Space settings choose whether all zero-thickness walls are
**Dashed** or **Solid**. Dashed zero walls let Glow and sun through; solid zero
walls are zero-area light barriers. Missing settings use Dashed. Changing the
style affects every `0 cm` wall in that space; there is no separate Boundary
tool or separate virtual-wall type.

During the drag, the moving wall itself has no redundant length badge. An outer
wall shows one area badge; a shared wall shows two on opposite sides, each with
a short leader. In a narrow room the area stays visible and may extend outside
the room rather than overlap another area or the room-settings button.

### Wall junction limits

To keep a plan physically meaningful, the editor refuses a write that would
create an impossible junction. The thresholds are absolute — they do not scale
with `cell_cm`:

| Rule | Threshold |
|---|---|
| Angle between neighbouring walls of one node | at least 15° |
| Walls meeting in one node | at most 6 |
| Wall length | at least 20 cm and never below its own thickness |
| Distance between non-incident nodes, and node to foreign wall | at least 5 cm |
| Room interior left after subtracting the masonry | at least 25 cm² |

Length is measured along the WALL, not along a single contour piece: a short
filler segment that compensates a thickness step is legal as long as the whole
wall is longer than 20 cm. A T-joint (a wall end landing on the middle of
another wall) is not forbidden by the distance rule.

If the check itself cannot run (an internal error), the change is not saved
either — a "The junction check could not run" toast appears: the editor
never waves a write through on faith.

The check runs on writes only. An already saved plan is never re-judged:
migration, import and backup restore are never blocked, and an edit that does
not touch the offending element passes as usual. The refusal appears where you
work: drawing and Thickness leave the value unapplied and raise a toast naming
the rule, while Resize stops the wall at the last allowed position and, once
per gesture, names the rule the next step would break.

### HA area binding

One HA area may be bound to one room. The binding drives automatic device
placement, room LQI, light and aggregate climate. A real temperature or
humidity sensor manually assigned to another House Plan room follows that
placement for the automatic room average instead of remaining in its registry
HA area. This also works for a room without an area; its manually assigned real
sensors provide the automatic average, while an explicit room measurement
source still takes priority.

## 9. Doors, windows, gates and locks

Choose Opening, select door/window/open passage/gate, and click a wall. Defaults
are 90 cm, 120 cm, 90 cm and 300 cm. The complete opening must fit on the wall.
Before the click, door, window and gate show their translucent architectural
symbol. An open passage instead shows the exact future wall cut as a translucent
wall-coloured segment with an orange boundary mark at each end. Its depth follows
the real wall thickness; after saving it has no standalone symbol.

The placement preview also draws a thin dimension line from each jamb to the
physical inner end of the wall. On a wall shared by two rooms, four values are
shown — two along each room's inner face — because their usable boundaries can
differ. On a finished independent wall, each value stops at the nearest
physical face of a connected wall; where no such face exists, it keeps the
distance to the independent wall's own endpoint. These richer dimensions apply
only before a new opening is placed; dragging an existing opening retains its
two established end-distance badges.

On a finished independent wall, a new or directly edited opening must leave a
jamb at each endpoint equal to at least half that wall's real thickness. The
same limit applies to placement, drag, rebind and length edits. Existing
near-end openings remain visible and are not moved until their geometry is
edited.

An opening may bind a contact; doors and gates may also bind a lock. View paints
the moving leaf and state. A lock badge is green when locked and red when
unlocked. A plan tap never toggles a lock. The opening
card provides a labelled
lock/unlock control, with confirmation before unlocking.

Openings may slide along joined wall corners. A double click in Plan opens
properties. Thick walls keep visible jambs and align the symbol to the correct
face.

<!-- docs-section: devices -->

## 10. Devices

House Plan reads Home Assistant device, entity and area registries. A device in
a bound HA area receives an automatic marker. Service-only records, bridges and
other non-spatial records are filtered; a light group may replace its members.
Newly discovered devices get a red dot until first opened in Device.

When the Area of a direct HA device or a separately placed entity changes,
House Plan moves its automatic marker to the room bound to the new Area. A
previous manual drag is layout, not a room override: it is discarded, the
ordinary room grid chooses the new position and the red attention dot appears.
Selecting a room explicitly in the marker settings overrides HA Area placement.
Ambiguous or unbound Areas never make House Plan guess a destination.

### Bindings

| Binding | Use |
|---|---|
| HA device | Uses its related entities and resolves a primary function |
| HA entity | Exact entity binding after enabling “Show entities” |
| Virtual device | Label/icon/description only, with no live active state |

The same binding cannot be used by two markers.

When an exact HA entity belongs to a device, placing that entity gives its
channel to the entity marker. The automatic parent marker, if needed, contains
only the remaining active, HA-visible and unplaced entities; it disappears
when that residual is empty. HA-hidden siblings alone do not keep an automatic
parent on the plan. To show both the exact entity and the complete device,
place `entity:X` and `device:D` explicitly — two explicit markers are treated
as an intentional configuration. Deleting an entity marker returns the entity
to automatic parent discovery; its binding tombstone does not remove registry
data from the live HA device.

After deleting a complete HA device, you can restore only one of its entities:
open **Devices → Available**, enable **Show entities**, and select that entity. House Plan
returns the selected marker with a fresh position while the complete device
and its other entities remain deleted. The complete device stays available in
**Available again** if you later decide to restore it explicitly as well.

### Device editor

- drag a marker to save its server-side position. One completed drag creates
  one position-history step; a cancelled, unchanged or failed drag creates
  none;
- persistent Undo/Redo buttons affect marker positions only. The same
  session-local history is available through `Ctrl/Cmd+Z`,
  `Ctrl/Cmd+Shift+Z` and `Ctrl+Y`, keeps up to 50 completed moves, and is not
  restored after reopening the card;
- click it to edit name, binding, room, tap action and presentation;
- **Add** opens the new-device dialog directly, without going through the
  device catalog;
- **Devices** opens one searchable lifecycle catalog. Its **On plan**,
  **Available**, **Hidden** and **Available again** tabs explain where every
  exact HA binding is and offer the next valid action;
- **Discovery filters** (#44) live on the **Available** tab: a switch that
  groups room lights into one marker (on by default) and the list of excluded
  integrations with search and a "Restore recommended" reset. Changes show
  appear/disappear counters before anything is written; Save stores the
  settings once. Filters only affect automatic candidates — a device you
  placed explicitly never disappears because of them, and an excluded
  candidate names its integration in the catalog;
- **Add virtual device** lives at the top of that catalog. Enable **Show
  entities** in **Available** to place an individual entity;
- **Show hidden on plan** is a local catalog switch. It reveals user-hidden
  and HA-disabled records as service ghosts only until you leave the Device
  editor; it never changes the saved Hidden flag;
- **Icon rules** edits the first-match regular-expression list.

An automatically discovered marker is already **On plan** even before it has
saved marker settings. The **New** badge is independent and remains until the
marker settings are opened. **Find on plan** centres and briefly selects the
marker without changing config or acknowledging that badge. Hide and Show are
reversible; Delete leaves an exact binding tombstone and moves an active HA
binding to **Available again**. A disabled or missing binding keeps its saved
category and receives a separate Home Assistant status instead of silently
moving to another tab.

The dialog shows binding provenance, exact next tap result, skipped targets and
a live presentation preview. A saved missing source is shown as missing rather
than silently replaced.

![Device editor with binding provenance and the exact action result](images/06-device-editor.png)

![Live preview of the selected device presentation](images/06-device-display-preview.png)

Hidden markers keep configuration and may still contribute to area aggregates.
An HA-disabled binding is excluded from rendering, state, actions, light and
aggregates until the same ID becomes active again.

### Presence radars

For a recognized presence radar, marker settings contain **Presence on the
plan**. Enable it, verify the exact Home Assistant sources and select the room
whose contour must contain the observations. Unknown/custom real devices can
use **Additional actions → This is a presence radar** and an explicit data
profile. The editor never guesses coordinate units, axis directions or a
bearing from entity names.

**Configure on plan** records the physical sensor position and direction,
independently of the decorative marker. Coordinate profiles can then use two
measured reference positions; an optional third point checks the result without
changing it. Use a desktop browser and stand alone/still at each reference.
**Check live data** distinguishes no target, stale or partial coordinates,
unavailable sources and presence without a usable position.

Live dots and range arcs do not intercept clicks and are clipped to the selected
room. Their short trail and smoothing exist only in the open browser session;
House Plan does not save raw radar samples or target history. Disable either
the radar itself or **General settings → Show live presence on the plan** to hide the
layer. See [Presence radars](RADAR.md) for supported profiles, setup and privacy.

## 11. Tap actions

| Gesture | View | Device editor |
|---|---|---|
| Short click/tap | Configured action | Open marker settings |
| Hold 600 ms | House Plan device card | No device control |
| Right click | Native HA more-info for the primary entity | Browser/editor context |

| Action | Behaviour | Safety |
|---|---|---|
| Device card | House Plan card with entities, model, description, links and files | No state change |
| HA more-info | Native dialog for the exact primary entity | No state change |
| Toggle state | Toggles the exact binding, supported device function or configured light-source group | Locks, alarm panels and protective garage/door/gate targets are no-op; confirmation is optional |
| Run | Runs an automation, script or scene | Explicit target; confirmation is optional |
| Do nothing | Short click/tap, Enter and Space are ignored | No dialog, command, confirmation, toast or press feedback |

When confirmation is enabled for **Toggle state**, the dialog shows the current
state and the exact expected result (`On`, `Off`, `Open`, `Closed` or `Stopped`).
A group shows the active/total count and lists unavailable targets separately;
the result describes only the targets that will receive the command. The text
is a snapshot, but Confirm re-resolves the live state and direction. If the
target set changed while the dialog was open, House Plan cancels the action and
asks you to try again.

A light defaults to Toggle; other devices default to the House Plan card. An
unsupported Toggle remains a visible no-op and is never changed into another
action behind the user's back. **Do nothing** is an explicit saved choice, not
the default: it keeps the marker's normal appearance, hover, long-press House
Plan card and right-click HA more-info while disabling only short activation.

If every explicitly configured `controls` target is unavailable, missing or
disabled in HA, a short tap sends no service call and the standard local House
Plan message names the target and explains that no action was performed. A
partially available group still operates only its available subset, so it does
not show the misleading no-action message.

When a device-bound marker has two or more own `light.*`/`switch.*` entities,
**Entity to toggle** appears below Toggle. It selects the exact own channel and
updates the target hint before Save. **Automatic** keeps the previous binding /
functional-role rules. A missing saved entity stays configured, shows a
warning and temporarily falls back; returning the same entity restores the
choice. This setting is independent from **Leading light entity**. With an
explicit external controls group, an explicitly selected own entity joins the
group; without a selection existing groups remain external-only.

![House Plan device card with state and safe actions](images/09-device-info.png)

<!-- docs-section: visual-states -->

## 12. Device visual states

Presentation uses one shared outer shell around three independent layers:
stable core, icon or value, and optional activity pulse. Visual priority is
**alarm → keyboard focus → selected → hover → semantic state → neutral**.

| State | Meaning | Examples |
|---|---|---|
| Red alarm | Critical condition, even with live states disabled | Smoke, gas, CO, leak, tamper/problem/safety, triggered alarm |
| Yellow | Device is doing its main job | Light/switch/fan on, active climate, vacuum cleaning, known appliance work |
| Red lock | Lock is unsecured | Unlocked/open lock |
| Green lock | Lock is secured | Locked lock |
| Orange | Physically open | Door/window contact, opening valve |
| Faded | Data unavailable | All relevant entities unknown, unavailable or absent |
| Neutral | No alarm, work or open condition | Off, closed, idle, standby, docked |

For a controller with `controls`, target work and controller availability are
independent. The controlled lights still decide whether the marker is yellow,
but only the controller's own active entities decide whether it fades. A live
battery, Zigbee LQI or update entity therefore keeps a wireless switch neutral
and opaque when all of its lamps are unavailable. If an active physical HA
device exposes no entities at all, House Plan also keeps its controller opaque:
missing telemetry alone is not evidence that the device is offline, so its
controlled target makes it yellow when working and neutral otherwise. If the
device does expose own entities but all of them are missing, `unknown` or
`unavailable`, the controller fades even if a target is on. A virtual controller
is always available.
This remains true when the same target was separately removed from the plan:
the removed marker is not restored, but it cannot make the controller look
offline or make its editor preview disagree with the plan.

For a composite appliance with a dedicated Power switch, Power=`on` alone
remains neutral. If Home Assistant also exposes a strict lifecycle entity such
as Status/Run state/Job state, active values (`start`, `running`, `washing`,
`rinse`, and similar work states) make the marker yellow; idle, paused and
terminal values remove it. Power=`off` or unavailable still fades the marker
even if the lifecycle value is stale. Mode, Program, Stage and remaining time
are not treated as independent proof of work, and an ordinary lone relay keeps
its existing yellow-on behaviour.

Activity may be a finite three-wave event, persistent presence, a travelling
transition, or persistent work. Continuous motion uses a 3.6 s cycle (green
presence, amber work, blue neutral transition), a short event lasts 3.3 s, and
the two-wave red alarm cycles in 2.4 s. Explicit saved pulse color/size remains
authoritative; the package size default is 1.5 diameters. `prefers-reduced-motion`
replaces ordinary motion with a compact colored indicator while the static red
alarm remains clear.

The five display choices are icon + state; icon + state + activity; value +
state; always-static icon; and value + static icon. A separate value badge can
show an entity state,
useful attribute, average LQI or linked light state on any side of the marker.
**Value + static icon** combines the two: the marker content follows the same
rules as **value + state** — the same Value source, the same numbers and
localized states, the same icon fallback with its reason — while the colour
behaves like **always-static icon**. State, alarm, unavailability, live RGB
colour and activity never change the marker; there is no pulse at all, the
compact °/% and LQI readings stay hidden, and the separate value badge is
suppressed the same way (the setting is kept), because the value is already
inside the marker. A vacuum in this mode draws no live puck, trail or route
warning, and its saved history is not deleted.
For **value + state**, the Device editor also offers **Value source**. Keep
**Automatic (as before)** for the legacy choice, or select one of those same
readings—for example, cover position—to replace the icon with `42 %` rather
than `Open`. A temporarily unavailable saved source stays selected and shows
`—` until it recovers; it is not silently replaced. Changing this source never
changes what a click or tap does.
Text and adjacent values are sections of the same shell. They shrink to a
readable floor and then expand the shell; they are never ellipsized.
The complete visible value capsule is one hover and action target: clicking or
tapping its value section runs exactly the same configured action and safety
checks as the icon core.

Virtual devices use the ordinary neutral/hover background with a dashed outer
circle. An HA-less virtual device does not invent unavailable or activity;
a linked virtual light may still follow its real controller. Unavailable keeps the ordinary
presentation with the standard icon opacity reduction, no visual hover and no
motion; its existing click/tap still opens information or settings. Marker LQI
uses the same continuous red-to-green scale as before the package update; the
room fill gradient and the displayed number are unchanged.

Interactive View/kiosk and Device-editor markers have at least a 44×44 CSS px
target. Enter and Space reuse the exact current click and confirmation path;
Plan, Background, preview and the read-only static card add no tab stop.
In the full View, Tab focus also opens the same device tooltip as mouse hover;
moving to the next control closes or moves it. The active space is exposed as
the current item of the named space navigation. The static card remains
non-interactive.

## 13. Room fills and light

Space fill modes include user colour, temperature comfort range and LQI. Room
settings may override the space. A room has its own colour only while its fill
is set to its own **Custom color**; choosing **As the space** forgets that
colour and the room is painted like the rest of the space. Glow is independent
from the base fill.

When a room effectively uses the temperature fill, its settings show optional
lower and upper comfort bounds. Each blank field independently inherits the
matching space bound, and **As the space** clears both overrides. The range
changes only the room floor and opening-tunnel fill; room-card and tooltip
temperature values are unchanged.

A light source may come from automatic classification, an explicit Always
role, or a controlled source group. Walls, partitions and columns occlude Glow;
open passages transmit it. When a configured light source disappears or loses
its valid binding, its contribution is removed instead of keeping stale light.

Overlapping Glow pools add brightness and colour where browser SVG blending is
supported; otherwise House Plan uses a safe normal blend without changing the
saved setting.

<!-- docs-section: background -->

## 14. Background editor

The authoritative technical interaction contract is
[DECOR-EDITOR.md](DECOR-EDITOR.md).

While this mode is active, device markers and room labels do not intercept the
pointer (#362, #376): drawing works right through them.

The main-toolbar default colour and style for new objects is saved with the
plan (#377): it survives a page reload and is shared by everyone who edits
this plan.

| Tool | Create | Edit |
|---|---|---|
| Select | Select an item | Move, scale, rotate; double click properties; Delete/Backspace removes |
| Backdrop | Available when a plan image exists | Move, corner-scale, rotate; double click numeric size/angle |
| Line | Drag endpoints | Colour/opacity, physical thickness and solid/dashed style |
| Rectangle | Drag diagonal; Shift makes a square | Stroke plus independent fill, size and angle |
| Oval | Drag bounds; Shift makes a circle | Stroke plus independent fill, radii and angle |
| Text | Click to open dialog | Multiline text, HA tokens, colour, physical size and angle |
| Furniture | Pick a front-view category, pick a top-view variant, then click | Symbol, smooth size, horizontal/vertical mirror, colour, outline and wall magnet |
| Image | Upload PNG/JPEG/WebP/SVG or choose a previous upload, then click | Smooth size, horizontal/vertical mirror, opacity, angle and file replacement; no wall magnet |
| Erase | Click an item | Confirmed deletion, undoable |

Creation and ordinary decor transforms snap to the grid plus nearby
room/background anchors. Furniture resize is the exception: corners move
smoothly and preserve proportion (`Shift` changes axes independently), while
four middle handles change one axis. Crossing the opposite edge mirrors the
item. Rotation is smooth and `Shift` snaps it to 45°; signed size fields and
the two mirror checkboxes provide the same result numerically. Furniture is
selected within 10 physical centimetres of its drawn strokes, not throughout
its empty bounding box.
**Optimize plans** preserves the complete smooth position, size and rotation of
furniture and uploaded images; these authored transforms are not grid debt.
The plan image is interactive only with Backdrop selected. Undo/Redo shares the
50-command editor history.

The Furniture palette always uses two levels: categories first, then the
available plan variants. **All categories** returns to the first level and
disarms the current symbol. Existing placed furniture keeps its saved size and
position when the built-in artwork is updated.

The Image palette stores reusable files privately in House Plan. Each saved
canonical file is at most 2 MiB; PNG, JPEG, WebP and safe SVG are supported.
When a raster source exceeds that limit, the warning dialog offers to upload a
reduced copy while keeping the oversized original unavailable. Picking a file
arms one placement: the pointer preview shows the result, one click adds it at
100 cm wide (aspect-preserving, height capped at 200 cm), and the tool returns
to Select. Images use the same smooth handles, mirroring and `Shift`-45°
rotation as furniture, but never snap to a wall. Their complete rectangle is
selectable, including transparent pixels.

Deleting or replacing a placed image leaves the reusable file in the palette.
The palette deletes a file only after all placed copies in all spaces are gone.
If a file is missing or fails its integrity check, View hides it; Background
shows a crossed placeholder that can be selected and repaired with Replace.
Exports still keep that image object without embedding the absent file. A later
import shows the existing missing-content confirmation and, once confirmed,
keeps the same repairable placeholder instead of rejecting the whole plan.

![Selected line in the Background editor](images/07-background-editor.png)

Live text accepts `{sensor.entity}` and
`{climate.entity:current_temperature}` tokens. Missing, unavailable or complex
object values render as `—`; long values are shortened.

## 15. Sun background and window rays

These are independent features. **Follow the sun** changes the background from
day through golden hour to night and can fall back to browser time. Window rays
require `sun.sun`, a configured north direction and suitable exterior windows.
Weather cloud cover may reduce ray intensity.

For window rays, point the compass N arrow toward the place where true north
actually lies on the drawing. The value is the literal clockwise direction
from canvas-up to north, not an opposite correction for a rotated plan. If you
previously mirrored the compass to compensate for the old ray-direction bug,
return it to the real north after updating.

Rays remain visual only: they do not change Home Assistant state. Walls and
physical obstacles clip them; changing north or window geometry recalculates
the result.

## 16. Robot vacuums

The dock marker stays at its saved location while a live puck and path follow a
supported map source. Calibration maps source coordinates to plan coordinates.
Automatic room-name matching is a starting point; manual drag/stretch corrects
it. A diagnostic source picker reports missing or incompatible sources instead
of silently rebinding.

Multiple maps are represented as distinct sources/calibrations. A vacuum is
shown only in the space whose saved mapping currently matches the active map;
the dock remains in its configured space. Deleting a vacuum marker erases its
server-side trail immediately; adding the marker again starts the trail from
scratch. See [VACUUM.md](VACUUM.md) for the source dialect and calibration
contract.

## 17. Kiosk

Set `kiosk: true` on a card in a Panel view. Editors and the ordinary header are
removed. Pinch/drag navigate, double-tap on free background fits the whole plan, and a 1:1 horizontal swipe
changes space. Holding empty space for three seconds opens per-display icon and
text sizing. `cycle` enables automatic space changes; interaction pauses it for
60 seconds.

## 18. Static space card

`custom:houseplan-space-card` renders one configured space without live states,
hover, drag, more-info or actions. A footer button opens the full plan. Use it
for compact dashboard navigation, not for home control.

The compact card keeps radial light pools and wall shadows off by default, so
existing dashboards retain the cheapest render path. Set `light_pools: true`
to show the full plan's Glow: the same source colour, brightness and radius,
additive overlap, passages and shadows from walls, partitions and columns.
This is independent from `live_states`; disabling ordinary live marker dressing
does not disable light transport. The option is intentionally heavier than the
default because every visible source needs clipped floor visibility geometry.

```yaml
type: custom:houseplan-space-card
space: ground
fit: house
light_pools: true
```

`fit` controls only this compact card's frame:

- `content` (default) keeps the existing frame around all visible content and
  its 5% breathing room;
- `house` removes that intentional padding and fits every room, every visible
  opening symbol, and — when `show_borders` keeps them visible — every wall,
  partition and column (#384: hidden architecture does not widen the frame).
  A detached but valid wing stays in the frame. Backdrop, decor, room labels
  and device markers do not widen it.

Choose `house` when the building should occupy as much of the card as possible.
Auxiliary objects remain rendered, but an object outside the structural bounds
may be clipped. An image-only or empty space safely falls back to `content`.

By default the card uses the configured space name as its header. A non-empty
`title` replaces that text. Set `title: ""` explicitly when the surrounding
dashboard already provides the name: the header is omitted and the plan meets
the top of the stage, while the normal left, right and bottom breathing room is
kept. Omitting `title` is not the same as setting it to an empty string.
With `fit: house` all four intentional paddings are already zero, so
`title: ""` removes only the header and leaves the same structural frame.

## 19. Plan maintenance

### Save the current space as PDF

Administrators can press the printer button between **General settings** and
**Help & feedback** to download the current space as a clean one-page A4 plan.
The sheet always includes walls, partitions, columns and openings, but excludes
devices, live states, Glow, sunlight, room colours, vacuum trails and Zigbee
topology. The dialog can add dimensions and clean floor areas, room names,
Background-editor decor and the space backdrop. Its choices are remembered in
this browser.

House Plan lays out the complete selected content before choosing portrait or
landscape and a standard scale, then centres that complete scene on the sheet.
Physical walls use a grey base and architectural hatch. Measurements are
limited to horizontal and vertical walls, and a matching opposite pair is
printed once within its own room or connected outer contour. The footer
includes a scale bar, a vector compass when north is configured, the date and
version; the old architectural-symbol legend is no longer printed. The export
is read-only and always uses the flat plan, including while the card is in
isometric view. Its dialog remains usable without horizontal scrolling down to
a 320 px-wide View area. See [PDF export](PDF-EXPORT.md) for measurement,
image-limit and font details.

Rectangular facade steps keep a complete reconstructable dimension chain:
both adjacent exterior sections, the height of the step and one copy of its
depth remain visible without adding diagonal measurements.

Current plans give every stored wall segment a stable internal identity. This
keeps the wall's thickness and its door, window, gate or passage attached while
Resize, Split, Merge and other structural tools change surrounding geometry.
There is no new control and the plan is not rewritten merely by opening it.

An older plan is upgraded atomically on its first structural edit or when you
run **Optimize plans**. Names, colours and other presentation settings do not
trigger the upgrade. If old geometry is ambiguous, House Plan cancels the edit
without partially saving it and asks you to run **Optimize plans**. If the same
message remains, fix the reported conflicting wall geometry or attach that
space's export to a bug report.

Optimization compacts old off-grid geometry and repairs the plan's reference
graph while preserving rooms, bindings and supported settings. An exact
independent-import signature restores the copied space, room and positions. If
there is no exact copy, an active real device follows its unambiguous HA Area;
otherwise only its missing placement is detached, so the marker becomes
available on a valid plan without losing its settings.

Old plans can also contain invisible floating-point tails around ordinary grid
nodes. Optimize reports how many coordinate values it will canonicalize, the
maximum physical movement and only the affected spaces. This cleanup does not
pull intentional off-grid or diagonal geometry to a node. Current ordinary
edits apply the same invisible boundary automatically, so the noise cannot
return after a later room, opening, decor or marker-position save.

Equal neighbouring wall-thickness fragments are compacted only while they have
the same physical role: one outer room or the same pair of shared rooms. A
shared-to-outer transition or a change of shared-room pair stays as an exact
breakpoint even when the thickness is equal. Optimize may also
remove a different-thickness fragment shorter than half a grid step when equal
pieces of the same straight wall prove the replacement. This includes a
fragment touching exactly one room T-junction: the junction and perpendicular
wall do not move. A fragment between two room vertices or touching an opening
boundary is preserved. Ordinary opening, rendering, Save and editing never
perform this cleanup without explicit Optimize confirmation.

Before an editor stores a change to rooms, walls, boundaries, openings,
partitions or columns, House Plan builds the exact candidate with the
same physical-geometry engine used for display. If the result is unsafe, the
change is canceled before Undo history or server storage is touched and the
card reports that the wall geometry could not be built safely. Titles, colours,
markers and other non-geometry settings remain editable.

When an old plan contains an independent wall exactly on top of solid room
masonry, Optimize can absorb each proven covered section even when consecutive
room-wall intervals form the cover. Free or ambiguous residual sections remain
independent walls with stable identities. Doors, windows and gates stay in
place: each is reattached to the room wall or to the retained residual that
still hosts it. The resulting thickness is the wider original thickness, so
visible masonry does not shrink. Each independently stored section is absorbed
only when it is fully redundant; a free, partly covered or thicker partition
remains unchanged. The report counts absorbed independent-wall sections.

Every Plan editor tool draws room and independent-wall centre axes and endpoint
nodes through the same layer above wall bodies. An independent wall hidden
under other masonry additionally retains its source
diagnostic axis and nodes. These pointer-transparent layers do not change
snapping or selection and are absent outside the Plan editor. The diagnostic
disappears after Apply only when the corresponding independent geometry was
safely absorbed or removed.

Old positions are classified before Apply. A position whose room label, device
or light-group owner is proven absent is removed automatically and counted by a
plain-language category. A live owner in a deleted space is named and preserved
by default; **Remove old positions** explicitly adds only those entries to the
same Apply candidate. An owner that cannot be checked against a complete HA
registry is preserved without a destructive action. Raw IDs appear only inside
collapsed **Details**, and vacuum room mappings remain a separate warning for
manual review. Preview, the secondary option and Cancel do not write anything.
Plan images and attachments are never deleted merely because nothing currently
references them.

Optimization creates one server-side undo point which restores automatically
and explicitly removed positions with the rest of the previous layout. Any
later edit makes that undo stale, so create a Home Assistant backup before a
large maintenance operation.

If a temporary storage error interrupts Optimize or its server-side Undo,
House Plan does not let the next edit overwrite the unfinished half. The next
save first completes the recorded operation (or its safe rollback) and then
checks the edit against the fresh revisions. A stale browser may therefore ask
you to reload and retry. If storage is still unavailable, the save fails and
the recovery record remains for another attempt or a Home Assistant restart.

<!-- docs-section: multiple-cards -->

## 20. Storage, multiple cards and backups

### Portable JSON backup

**Global settings → Backup and transfer** exports either the complete House
Plan model or the current space. Import first shows a server-side preview with
type, versions, object counts, source and content-link state; nothing is written
until confirmation.

A full import replaces the model and creates one undo point. A space import
assigns new internal IDs and adds the copy without replacing global settings.
When that exact import map matches orphan references already present in the
target, the preview reports them and Apply restores them together with the
space.
Re-importing a copy still creates an independent space, but no longer grows
nested service prefixes in internal IDs. Preview and **Add space** use the same
prepared candidate. **Import reference details** reports links updated inside
the copy and in the existing plan. If more than one target is possible, House
Plan preserves the reference instead of guessing and recommends running
**Optimize plans** after the import.
Internal uploaded files are not embedded in JSON; an import to another HA
instance must explicitly detach those links.

For **Current space**, enable **Plan only** to transfer the architectural
template without devices or Home Assistant bindings. It keeps rooms, walls,
openings, decor, backdrop transforms and manually positioned room labels at
their chosen scale, but removes real and virtual markers, device positions,
Area assignments,
temperature/humidity sources and opening contacts/locks. Live values in text
labels become `—`; surrounding static text stays intact. The import preview
marks this file as plan-only and adds it through the normal space-import flow.

Plan-only is not full anonymisation: space and room names, static text, file
names, exact coordinates and external URLs remain in the JSON. Internal plan
files are still referenced rather than embedded and may need to be detached on
another Home Assistant instance.

### Importing a plan from Sweet Home 3D

If the home is already drawn in [Sweet Home 3D](https://www.sweethome3d.com/),
there is no need to draw it again: the converter at
[houseplan.tech/convert](https://houseplan.tech/convert) turns a `.sh3d` file
into import documents — one per level. From there it is the ordinary space
import under **Global settings → Backup and transfer**.

This is an **optional shortcut**, not a setup step: the normal paths — a
background image or drawing from scratch — stay unchanged. The conversion runs
in your browser; the file is never uploaded.

Carried over: levels, rooms with their names, walls with thickness, doors and
windows. Not carried over: furniture, materials, textures, lights, cameras —
and the binding of rooms to Home Assistant areas together with device
placement, because the file has neither; both are done in the editor. Curved
walls are straightened, walls thicker than 100 cm are clamped to the limit, and
a level with no drawn rooms cannot be converted at all — House Plan builds
geometry from rooms. The page lists every such case before you download
anything.

### Storage locations

| Data | Location |
|---|---|
| Spaces, geometry, Background and marker settings | HA storage `houseplan.config` |
| Marker and room-card positions | HA storage `houseplan.layout` |
| Plan files | `config/houseplan/plans` |
| Marker attachments | `config/houseplan/files` |
| Reusable custom decor images | `config/houseplan/assets` |
| Vacuum path | Separate House Plan HA storage |
| Cache, viewport and kiosk scale | That browser's `localStorage` |

### Several cards and clients

Use `floor` when separate card instances must stay on separate spaces. A stable
space ID is recommended:

```yaml
type: custom:houseplan-card
floor: ground
kiosk: true
cycle: 0
```

YAML also accepts an unquoted zero-based numeric index such as `floor: 1`.
Indexes follow the current server space order, so reordering spaces may change
which one is shown. A quoted value such as `floor: "1"` is a literal space ID.

A pinned card shows only its assigned space. It ignores the browser's shared
last-space record, `#space=` links, other floor tabs, swipe and kiosk cycling,
and it does not overwrite the shared last-space record. If the configured ID
or index is invalid, the card shows a configuration error instead of choosing
another space. Remove `floor` to restore normal navigation.

Unpinned `custom:houseplan-card` instances may still use different
`default_floor` values. That option is only the initial/fallback choice; the
last selected space, a valid `#space=` link or normal navigation may replace it.
If the saved id no longer exists, runtime still opens the first valid space and
the visual card editor shows the raw missing id with an inline warning until a
valid choice is made.

- configuration, rooms, Background and device layout are shared server data;
- `floor` is a permanent per-card navigation authority, while `default_floor`
  is only an initial/fallback choice for an unpinned card;
- current mode, selected space, zoom/pan and kiosk sizing are local;
- WebSocket broadcasts saved changes and revision checks reject a stale write;
- avoid editing the same object in two browsers: the second save may need a
  refresh and manual reapplication.

Every client that saves the shared configuration must return the revision from
`houseplan/config/get`. Omitting it is allowed only while the configuration
store is still empty; afterwards House Plan rejects the save as a conflict
instead of risking another client's work. If an old cached card repeatedly
reports conflicts, update House Plan and refresh the dashboard.

### Files and quotas

Plan files accept SVG/PNG/JPG/WebP up to 8 MB, with a 200-file/256-MB total.
Marker attachments accept PDF/PNG/JPG/WebP/TXT up to 50 MB, with a
1000-file/1-GB total and 50 links per marker. Writes are also refused below
512 MB free disk space. Detached files remain until explicitly deleted.

<!-- docs-section: limits -->

## 21. Current limitations

| Limitation | Practical effect |
|---|---|
| Rooms must be closed | An open path is walls, not a room with area |
| One HA area per room | To split one HA area visually, assign devices manually |
| Zero-thickness walls cannot host openings | Give the target wall a positive thickness before adding a door, window, gate or passage |
| Editors are desktop-first | Touch editing may be awkward, limited or absent |
| Room details use hover in View | A touch-only user may need an editor or another visible metric |
| Sun has no exterior 3D model | It cannot know shadows from trees, awnings or neighbouring structures |
| Icon rules use regular expressions | First matching rule wins and invalid expressions are rejected |

Storage guards allow up to 50 spaces, 400 rooms per space, 2,000 markers, 500
openings, 1,000 Background items and bounded wall/physical-object catalogues
per space. The configuration package is limited to 2 MB.

<!-- docs-section: diagnostics -->

## 22. Troubleshooting

### Card does not load

| Symptom | Check |
|---|---|
| `Custom element doesn't exist: houseplan-card` | Integration loaded; resource is `/houseplan_files/houseplan-card.js` with type `module`; restart HA, then hard-refresh with `Ctrl+F5` (Windows/Linux) or `Cmd+Shift+R` (macOS) |
| MIME `text/plain` | Replace `/custom_components/...` with `/houseplan_files/...` |
| Integration missing | Folder is exactly `custom_components/houseplan`; restart HA; inspect import errors |

### Devices or values are missing

- confirm the room's HA-area binding and the device/entity registry area;
- open Device → **Devices** and inspect **Hidden** / **Available again**;
- verify the selected source still exists and is available;
- remember that a virtual marker has no active state;
- inspect the dialog's exact target and skipped-target explanation.

### Geometry looks wrong

- confirm grid cell size before judging physical dimensions;
- use Fit all to include distant objects;
- check wall thickness and the space's zero-thickness wall style;
- use desktop for precise nodes, Split and Resize.
- if some masonry remains visible but Optimize or an edit reports unsafe wall
  geometry, export the affected space and attach it to a bug report. House Plan
  preserves known-valid wall components for inspection and does not repair or
  delete the ambiguous object during rendering.

### Vacuum does not move

- confirm the active map source and map identity;
- check calibration anchors and current source availability;
- verify that the active map is assigned to this space;
- open source diagnostics instead of deleting/recreating the marker.

When reporting a problem, include House Plan version, HA version, browser,
console/server errors and reproducible steps. Replace private entity IDs and
plans with synthetic equivalents.

<!-- docs-section: support -->

## 23. Help and private feedback

The **Help & feedback** button follows **General settings** in the card header.
It is available to users allowed to edit House Plan in View and all editors,
but is hidden in kiosk. The dialog contains the current card version, the
GitHub and Telegram links, and this guide. Russian UI opens the Russian guide;
all other languages open this English guide.

Enter a required message and, optionally, a contact such as an email, Telegram
username or WhatsApp number. The form is kept only in the open card instance:
it is not written to House Plan settings, local storage or Home Assistant.

The diagnostic attachment is **off by default**. When enabled, House Plan
builds an allowlisted package in the integration and shows its exact size,
SHA-256 and JSON before sending. The package excludes names, Home Assistant
entity/device/area IDs, live states, URLs, paths, files, message and contact.
It does include exact room/wall/opening geometry and home dimensions. Use
**Show data** to inspect the exact bytes and **Download JSON** to keep them.
The preview expires after ten minutes; refresh it before sending if required.

On success, keep the report ID shown by the dialog. A network or relay failure
does not close or clear the draft and never claims delivery: retry with the
same preview, or copy the message/download the JSON and continue through the
provided Telegram or GitHub links. The form is available when the card and
integration support the same feedback API; their release numbers may
temporarily differ during a normal HACS update. Update an old or incompatible
side. Retention and deletion details are in
[SUPPORT-PRIVACY.md](SUPPORT-PRIVACY.md).