acdream/docs/research/2026-08-09-chat-retail-window-shell.md

# Campaign CH slice CH6 — retail chat-window SHELL research

**Date:** 2026-08-09
**Scope:** RESEARCH ONLY. How retail Asheron's Call constructs, shows, sizes,
persists and fades the chat windows — the *shell*, not the text pipeline.
The per-window text FILTER model is already decoded in
[`2026-08-09-chat-retail-color-table.md`](2026-08-09-chat-retail-color-table.md)
§4 and is not re-derived here.

**Primary sources**
- `docs/research/named-retail/acclient_2013_pseudo_c.txt` (Sept 2013 EoR build,
  Binary Ninja pseudo-C, PDB-named)
- `docs/research/named-retail/acclient.h` (verbatim retail struct/enum defs)
- `docs/research/named-retail/symbols.json`
- `docs/research/named-retail/retail-default.keymap.txt`
- `docs/research/2026-06-25-retail-ui-layout-dump.json` — **an existing acdream
  dump of the retail gameplay-UI top-level windows.** It carries the LayoutDesc
  id, element ids, widget kinds and authored rects for every window below, and
  is the geometric oracle used throughout §2.
- `references/ACE/Source/...` (server cross-check)

**Binary-Ninja field-name caveat (applies throughout).** BN's struct-field
attribution in `gm*ChatUI::PostInit` is shifted by one slot relative to the true
member order (the classic artifact class in
`claude-memory/feedback_bn_decomp_field_names.md`). Every id↔role binding in §2
was therefore re-derived from the **authored rectangles in the layout dump**,
which agree with the decomp's *call order* exactly once the one-slot shift is
undone. Where a claim rests on BN naming alone it is marked UNVERIFIED.

---

## 1. Window lifecycle — how windows 1–4 get created and shown

### 1.1 They are not created on demand. They are authored, always-resident, and toggled.

`gmGamePlayUI::SetupChildren @0x004E9EC0` builds the whole gameplay UI once:

```
004e9ed5  m_pGameplayUI = CreateAndAddRootElement(0x10000006, 0x10000495)
004e9f40  var_b0 = 0x10000505    // floating chat window 1
004e9f48  var_ac = 0x1000050E    // floating chat window 2
004e9f50  var_a8 = 0x1000050F    // floating chat window 3
004e9f58  var_a4 = 0x10000510    // floating chat window 4
004e9f6a  hash("ID_Chat_Chat1_DefaultTitle") … "ID_Chat_Chat4_DefaultTitle"
004e9ff5  for i in 0..3:
004e9faf      StringInfo::SetStringIDandTableEnum(&title, ids[i], 0x10000001)
004e9fbf      el = UIElement::GetChildRecursive(m_pGameplayUI, windowIds[i])
004e9fd1      chat = el->DynamicCast(0x10000040)      // gmFloatyChatUI
004e9fe9      chat->vtable[0x2AC](&title)             // SetWindowTitle
```

So all four extra chat windows exist as **authored children of the gameplay-UI
root from the moment the game UI is built**. There is no window-manager `Create`
call, no per-window allocation at toggle time, and no floating-window registry.
Opening one is `SetVisible(true)`; closing one is `SetVisible(false)`.

Confirming class identity: `gmFloatyChatUI::GetUIElementType @0x004CE2B0` returns
`0x10000040`, and `gmFloatyChatUI::Register @0x004CE3D0` registers that element
class with the LayoutDesc element factory. `gmFloatyChatUI` derives from
`ChatInterface` and adds **no members at all** (`acclient.h:55144`), which is why
`m_eWindowID` (the first `ChatInterface` field, `acclient.h:54900`) is the only
per-window identity it carries.

### 1.2 Window identity comes from a LayoutDesc attribute, not from code

`ChatInterface::PostInit @0x004F3DD0`:

```
004f3de8  UIElement::GetAttribute_Enum(this, 0x1000007E, &this->m_eWindowID)
004f3df9  switch (m_eWindowID - 1) { … seed m_llTextTypeFilter defaults … }
```

**Attribute `0x1000007E` on the authored element IS the window id.** The main
chat window is id `0`; the floaties are `1..4` (the `PostInit` switch also has
arms for 5 and 8 — see the colour-table doc §4 for the seeded filter values).

### 1.3 The toggle is a KEYBIND, not a button

`docs/research/named-retail/retail-default.keymap.txt:150-153`, inside the
`ToggleWindows` action group:

```
ToggleFloatingChatWindow1 [ "" [ 0 DIK_1 ] 0x00000004 ]
ToggleFloatingChatWindow2 [ "" [ 0 DIK_2 ] 0x00000004 ]
ToggleFloatingChatWindow3 [ "" [ 0 DIK_3 ] 0x00000004 ]
ToggleFloatingChatWindow4 [ "" [ 0 DIK_4 ] 0x00000004 ]
```

`0x00000004` is the modifier mask (the same mask the `UseQuickSlot_14..18` rows
use for `Alt`+digit, versus `0x00000002` for the CTRL-digit quickslot rows —
corrected below; the original filing mislabeled `0x00000002` as "shift").
So retail's default is **Alt+1 … Alt+4**, not bare 1–4.
RESOLVED at Campaign CH slice CH6b: `retail-default.keymap.txt`'s own
`MetaKeys` legend (not a guess — the keymap file's literal index table) reads
```
MetaKeys
[
  1 [ 0 DIK_LSHIFT ]
  2 [ 0 DIK_LCONTROL ]
  2 [ 0 DIK_RCONTROL ]
  3 [ 0 DIK_LMENU ]
  3 [ 0 DIK_RALT ]
  4 [ 0 DIK_LWIN ]
  4 [ 0 DIK_RWIN ]
]
```
so the modifier-mask bit for MetaKeys index *N* is `1 << (N-1)`: **index 1 =
Shift → `0x1`, index 2 = Ctrl → `0x2`, index 3 = Alt → `0x4`, index 4 = Win →
`0x8`**. `0x00000004` on the `ToggleFloatingChatWindow1..4` rows is therefore
unambiguously **Alt**, cross-checked against the same file's own Alt+A/D
strafe and Alt+Enter/Tab/F4 rows (all `0x00000004`). `KeyBindings` already
carried `ModifierMask.Alt` for these four actions since Phase K.1c — no code
change was needed, only removing this hedge.

The dispatch chain is fully generic — there is no chat-specific code in it:

```
UIElementManager::KeyPressEvent @0x0045C300
  0045c347   DoVisibilityToggleAction(this, actionId)

UIElementManager::DoVisibilityToggleAction @0x0045B660
  0045b680   look up actionId in m_elementInputActionListenerTable
  0045b6c1   for each registered element:
                 BroadcastElementMessage(element, 0x31, actionId, 0)
```

and the registration side is a plain authored property:

```
UIElement::OnSetAttribute  (property switch) @0x00462D80
  004631d7   case 0x24:                       // "input action" property
  004631fb       UIElementManager::RegisterElementForInputAction(
                     UIElementManager::s_pInstance, enumValue, this)
```

**So: LayoutDesc property `0x24` on the window element declares the input action
that toggles it; the manager broadcasts element message `0x31` ("toggle
visibility") to every element registered for that action.** The same mechanism
serves `ToggleInventoryPanel`, `ToggleSpellbookPanel`, etc.

UNVERIFIED: the numeric action-enum values behind
`ToggleFloatingChatWindow1..4` (the keymap file stores names, and the property
`0x24` value lives in the LayoutDesc, not in code). Cheapest resolution: dump
property `0x24` for elements `0x10000505/0x1000050E/0x1000050F/0x10000510` from
LayoutDesc `0x21000070`/gameplay root — one `ui-studio --dump` pass. Not needed
to implement CH6: acdream binds its own actions.

### 1.4 The 1/2/3/4 buttons in the main window are STATE MIRRORS, not the toggle

They exist and they are authored (§2.1, elements `0x10000522`–`0x10000525`), but
the code path is one-directional — window visibility drives the buttons:

```
gmGamePlayUI::ListenToElementMessage @0x004E9CE0
  004e9cee   if (idMessage == 0x18)                     // visibility changed
  004e9d29     if (idElement == 0x10000505 ||
                   (idElement > 0x1000050D && idElement <= 0x10000510))
  004e9d4c        CM_UI::SendNotice_SetPanelVisibility(idElement, elementVisibleBit)

gmMainChatUI::RecvNotice_SetPanelVisibility @0x004CCD80
  004ccd9c   0x10000505 -> child 0x10000522
  004ccda3   0x1000050E -> child 0x10000523
  004ccdaa   0x1000050F -> child 0x10000524
  004ccdb1   0x10000510 -> child 0x10000525
  004ccdbf   child = GetChildRecursive(this, mapped)
  004ccdcd     visible ? child->SetState(6) : child->SetState(1)
```

State 6 = "on/depressed", state 1 = "normal". No handler anywhere in the binary
switches on `0x10000522..0x10000525` as a *source* of a click — `grep` over the
whole pseudo-C returns only `gmMainChatUI::RecvNotice_SetPanelVisibility`.

**RESOLVED 2026-08-10 at Campaign CH slice CH6b; wording corrected at the
CH6a/b REJECT-review (NIT 4) — the original "ONLY function in the whole 2013
binary that branches on `idMessage == 1`" superlative was false**
(`gmFloatyChatUI::ListenToElementMessage @0x004CE330`, the floaty windows'
close-button handler, also branches on `idMessage == 1`; §1.5). **The
substantive claim stands: within `gmMainChatUI::ListenToElementMessage`
specifically, there is no case for `0x10000522`-`0x10000525`.** The prior
UNVERIFIED paragraph's hedge ("safe to wire both") is superseded by a direct
read of `gmMainChatUI::ListenToElementMessage @0x004CDA80`. It handles
exactly two element ids: `0x1000046f` (max/min, dispatching
`HandleMaximizeButton`) and the talk-focus menu's selection message
(`idMessage == 7`, checked against `this->m_pCCS` / a `0x1000000b` attribute
read). There is no case, anywhere in that function or its base-class fallback
(`ChatInterface::ListenToElementMessage`, called unconditionally at the
function's tail), for `0x10000522`–`0x10000525`. **This citation is still a
true statement about `gmMainChatUI::ListenToElementMessage` — see the round-4
correction below for why it is the wrong function to have grepped.**

---

**ROUND-4 CORRECTION (2026-08-10) — the user's own retail memory ("clicking
opens/closes the window") overruled the CONCLUSION above, and re-attacking the
question with that as the starting axiom (per CLAUDE.md: the user's retail
memory is the axiom, not a hypothesis to be argued down) found the exact
mechanism the CH6b pass missed.**

The CH6b pass's mistake was scope, not accuracy: it proved
`gmMainChatUI::ListenToElementMessage` has no click case for these ids, then
reasoned "a button's click message… routes to its LISTENING PARENT" and
stopped there. **That's wrong — `UIElement_Button` (Type 1, the class every
one of these four indicators actually is) overrides its OWN click handling
and never asks its parent window first:**

```
UIElement_Button::HandleButtonClick @0x00471E50
  00471e65   if (UIElement::GetAttribute_Enum(this, 0x12, &actionId))   // reads its OWN property
  00471e72     if (actionId != 1)
  00471e95       build an InputEvent(actionId)
  00471eb2       ICIDM::GetActionMap()… dispatch through the action map
```

Property `0x12` here is the SAME kind of "input action" enum as `0x24` — not
the parent-window dispatch §1.4's original text assumed didn't apply to
clicks. This IS the generic mechanism a click uses to reach
`UIElementManager::DoVisibilityToggleAction @0x0045B660` (§1.3's own citation,
previously assumed keybind-only) → `BroadcastElementMessage(target, 0x31,
actionId, 0)` for every element registered under that action id via
`RegisterElementForInputAction` (§1.3's property-`0x24` registration —
confirmed as the ONLY call site of `RegisterElementForInputAction` in the
whole binary) → the RECEIVING element's generic base-class handler:

```
UIElement::ListenToElementMessage @0x00462340
  00462447   case 8:                          // idMessage - 0x29 == 8, i.e. raw idMessage 0x31
  0046244f     GetAttribute_Enum(this, 0x58, &mode)   // reads the RECEIVER's OWN property
  00462459     if (mode == 1) SetVisible(!currentlyVisible)   // toggle
  0046245c     else if (mode == 2) SetVisible(1)               // force-show
  0046245f     else if (mode == 3) SetVisible(0)               // force-hide
```

Neither `gmMainChatUI`, `gmFloatyChatUI`, nor `ChatInterface` overrides
`ListenToElementMessage` for raw idMessage `0x31` (confirmed by reading all
three switches directly — none has a case landing on it), so EVERY window
falls through to this base-class handler unconditionally. This is a complete,
generic, working "click toggles a registered listener's visibility" system —
exactly the "authored button behavior" this reconciliation task hypothesized
— and it is NOT gated on keyboard input the way the original §1.3 text
assumed; `UIElementManager::KeyPressEvent`'s call to `DoVisibilityToggleAction`
is simply ONE caller among the several that can reach it (the button's own
`HandleButtonClick` is another).

**What the authored DATA shows, checked directly against the committed
fixtures:**

- `chat_2100006f.json` — all four indicator elements (`0x10000522`-`0x10000525`)
  DO author an Enum-kind property `0x12` (confirmed by `Kind: 0` = Enum in the
  fixture's own property dump, matching `LayoutImporter.ConvertProperty`'s
  `EnumBaseProperty → UiPropertyKind.Enum` mapping exactly), with values
  `0x10000114`-`0x10000117` in element-id order (re-verified directly against
  the committed fixture's own `UnsignedValue` fields — an earlier pass here
  misread these as `0x10000514`-`0x10000517`). This is a real, present,
  correctly-typed action id — the button-click half of the generic mechanism
  is genuinely armed.
- `chat_floaty_2100005b.json` — the floating chat window's own fixture
  authors **NO Enum-kind property `0x24` anywhere** (its only hit on property
  number 36 decimal is `Kind: 4` = Integer, an unrelated attribute — not the
  registration property) and **no property `0x58` at all**. Since
  `RegisterElementForInputAction` has exactly one call site in the entire
  binary (the property-`0x24` handler in `UIElement::Initialize`; no class
  anywhere calls it directly from code), nothing in the shipped floating chat
  window LayoutDesc ever registers it as a listener for ANY action id.
  `DoVisibilityToggleAction` would look up the button's action id, find zero
  registered listeners, and silently return — the click would fire a real
  message with no receiver.
- The four action-id VALUES (`0x10000114`-`0x10000117`) are also not
  chat-specific: the first two, `0x10000114` and `0x10000115`, are
  `m_prevButton`/`m_nextButton` child-element ids for an UNRELATED
  pagination widget elsewhere in the decomp (`UIElement::GetChildRecursive(this,
  0x10000115)` / `(this, 0x10000114)`,
  `acclient_2013_pseudo_c.txt:194343-194344`, confirmed by direct read) — a
  coincidence of Turbine's global per-dat-file asset-id allocator (ids are
  assigned client-wide, not scoped per panel), not a cross-reference.
  (**Correction:** an earlier pass here misread the fixture's Enum values as
  `0x10000514`-`0x10000517` and built a since-retracted claim that they
  matched `gmFriendsUI::PostInit`'s own Add/Remove/Tell child ids — that
  match does not exist at the correct `0x1000011x` values and `gmFriendsUI`
  is not part of this finding.)

**Conclusion: the generic UI action system is real, it exists, and the
buttons genuinely arm their half of it — but the authored DATA available to
us (both committed fixtures, generated from the installed DAT) does not wire
a target for it.** This is consistent with, not a refutation of, the original
CH6b grep of `gmMainChatUI::ListenToElementMessage` — that citation was
looking in the wrong function, but its NEGATIVE RESULT (retail's window-level
message handlers never claim these clicks) still holds; the generic
mechanism, if it does connect the dots in real retail, does so entirely
below the level either grep could see. **Per CLAUDE.md, the user's retail
memory is the axiom regardless: acdream now wires each indicator's click to
toggle its floating window through the SAME `ToggleFloatingChatWindow`
chokepoint the `Alt+1..4` keybinds use
(`ChatWindowController.BindIndicatorClicks`, called by `RetailUiRuntime`
right after mounting the main chat window) — explicitly as USER-DIRECTED
retail behavior, not a claim that the generic-action-system data path has
been proven end-to-end.** `SetIndicatorOpen` stays the ONLY writer of the
indicator's `Selected` mirror (`UiButton.SuppressSelfToggle` stays `true`);
the click drives the real toggle, and the mirror reports the outcome back —
so the visual stays consistent through the round trip even though the write
is now two-way at the FEATURE level.

---

acdream ports this exactly: the four indicator buttons
(`ChatWindowController._indicatorButtons`) now carry a real `OnClick`
(`ChatWindowController.BindIndicatorClicks`, round 4); the mirror half is
unchanged — `ChatWindowController.SetIndicatorOpen` is still the ONLY writer
of `Selected`, still called from `RetailUiRuntime.OnWindowVisibilityChanged`
in response to the floating window's own visibility changing, regardless of
what triggered it (click, keybind, or a restored layout).

### 1.5 Closing a floaty window from its own title bar

```
gmFloatyChatUI::ListenToElementMessage @0x004CE330
  004ce344   if (idMessage == 1 && idElement == 0x1000052A)
  004ce346       this->vtable[…]( 0 )        // SetVisible(false)
```

Element `0x1000052A` is the 14×14 close button at (230, 86) in the floaty
layout (§2.2). `idMessage == 1` is "clicked".

### 1.6 The lock-UI cosmetic swap

`gmFloatyMainChatUI::UpdateLockedStatus @0x004D23D0` (driven by global message
`0x0D` via `gmFloatyMainChatUI::ListenToGlobalMessage @0x004D2940`) reads
`PlayerModule::LockUI` and swaps the 8 live border/corner elements for their 8
`_Locked` cosmetic twins. When the UI is locked the *interactive* Resizebars are
hidden and the inert art is shown — that is how retail disables resizing without
touching the resize code.

---

## 2. LayoutDesc geometry

### 2.1 Main chat window — LayoutDesc `0x2100006F`

Window element `0x10000601` (this is the id `gmGamePlayUI` looks up and the id
`SaveScreenLayout @0x004EAD50` writes as `<CHAT>`), layout root element
`0x10000600`, **authored extent 410 × 100**.

Full authored tree (from `2026-06-25-retail-ui-layout-dump.json`; rects are
absolute in the dump's 640×480 space, i.e. root at 0,0):

| element | rect (x,y,w,h) | role |
|---|---|---|
| `0x10000600` | 0,0,410,100 | window root (Group) |
| `0x10000693` | 0,0,5,5 | TL corner — **Locked twin** |
| `0x10000694` | 5,0,400,5 | top edge — Locked twin |
| `0x10000695` | 405,0,5,5 | TR corner — Locked twin |
| `0x10000696` | 0,5,5,90 | left edge — Locked twin |
| `0x10000697` | 0,95,5,5 | BL corner — Locked twin |
| `0x10000698` | 5,95,400,5 | bottom edge — Locked twin |
| `0x10000699` | 405,95,5,5 | BR corner — Locked twin |
| `0x1000069A` | 405,5,5,90 | right edge — Locked twin |
| `0x1000069B` | 0,0,5,5 | **TL corner — live Resizebar** |
| `0x1000069C` | 5,0,400,5 | **top edge — live Resizebar** |
| `0x1000069D` | 405,0,5,5 | **TR corner — live Resizebar** |
| `0x1000069E` | 0,5,5,90 | **left edge — live Resizebar** |
| `0x1000069F` | 0,95,5,5 | **BL corner — live Resizebar** |
| `0x100006A0` | 5,95,400,5 | **bottom edge — live Resizebar** |
| `0x100006A1` | 405,95,5,5 | **BR corner — live Resizebar** |
| `0x100006A2` | 405,5,5,90 | **right edge — live Resizebar** |
| `0x10000010` | 5,5,400,73 | transcript panel (Sprite) |
| `0x10000011` | 21,5,368,73 | transcript text (Group) |
| `0x1000048C` | 21,62,16,16 | new-unseen-text indicator (Button) |
| `0x10000012` | 389,5,16,73 | scrollbar column |
| `0x10000364/65/66`, `0x10000071`, `0x10000072` | — | scrollbar thumb pieces + up/down buttons |
| `0x1000046F` | 368,5,16,16 | max/min toggle (Button) |
| `0x10000522` | 5,5,16,16 | **chat-window-1 indicator (Button)** |
| `0x10000523` | 5,22,16,16 | **chat-window-2 indicator** |
| `0x10000524` | 5,39,16,16 | **chat-window-3 indicator** |
| `0x10000525` | 5,56,16,16 | **chat-window-4 indicator** |
| `0x10000013` | 5,78,400,17 | input row (Sprite) |
| `0x10000014` | 5,78,46,17 | talk-focus menu button |
| `0x10000015` | 5,78,46,17 | talk-focus menu group |
| `0x10000016` | 51,78,306,17 | **chat entry field** |
| `0x10000017` / `0x10000018` | 1px | entry field left/right rails |
| `0x10000019` | 359,78,46,17 | Send button |

Cross-checks that pin this layout to the code, independent of the dump:
- `gmMainChatUI::HandleMaximizeButton @0x004CCE50` looks up `0x1000046F`.
- `gmMainChatUI::RecvNotice_SetPanelVisibility @0x004CCD80` writes
  `0x10000522`–`0x10000525`.
- `gmFloatyMainChatUI::PostInit @0x004D2670` binds exactly the 16 ids
  `0x10000693`–`0x100006A2`, and casts eight of them via `DynamicCast(9)` —
  element type **9 is `UIElement_Resizebar`**
  (`UIElement_Resizebar::Register @0x0046B920` → `RegisterElementClass(9, …)`).

**The 1/2/3/4 buttons are on the LEFT edge, stacked vertically** — not a tab row.

### 2.2 Floating chat windows 1–4 — LayoutDesc `0x2100005B`

All four windows instantiate **the same** LayoutDesc; only the window element id
and the `0x1000007E` window-id attribute differ. Layout root `0x100004F7`,
**authored extent 250 × 108**.

| element | rect (x,y,w,h) | role |
|---|---|---|
| `0x100004F7` | 0,80,250,108 | window root (Group) |
| `0x100004FC` | 0,80,5,5 | TL corner |
| `0x1000000F` | 5,80,240,5 | top edge |
| `0x100004FE` | 245,80,5,5 | TR corner |
| `0x100004D2` | 0,85,5,98 | left edge |
| `0x10000501` | 0,183,5,5 | BL corner |
| `0x100004D4` | 5,183,240,5 | bottom edge |
| `0x10000503` | 245,183,5,5 | BR corner |
| `0x100004D3` | 245,85,5,98 | right edge |
| `0x100004D9` | 5,85,240,16 | **title bar** |
| `0x10000528` | 5,100,240,5 | title/content divider |
| `0x10000529` | 5,85,240,20 | title-bar group (drag handle) |
| `0x1000052A` | 230,86,14,14 | **close button** |
| `0x10000010` | 5,105,224,60 | transcript panel |
| `0x10000011` | 5,105,224,60 | transcript text |
| `0x1000048C` | 5,169,16,16 | new-unseen-text indicator |
| `0x10000012` | 229,105,16,60 | scrollbar column (+ children) |
| `0x10000509` | 5,165,240,18 | input row |
| `0x10000016` | 5,165,202,18 | chat entry field |
| `0x1000052B`, `0x100002B5`, `0x10000209`, `0x1000020A`, `0x1000020B` | — | entry-field frame rails |
| `0x10000019` | 207,165,38,18 | Send button |

Independent code cross-check: `gmFloatyChatUI::SetWindowTitle @0x004CEAA0` does
`GetChildRecursive(this, 0x100004D9)` and `SetStringInfo` on it — exactly the
240×16 title strip above.

A floaty chat window has **no** talk-focus menu, **no** max/min button and **no**
1/2/3/4 indicators. It has a title bar (which the main window does not) and a
close button.

### 2.3 How retail encodes resizability — there are no "sizable edge" flags

Retail does **not** put a bitmask on the window. Each draggable edge and corner
is its own child element of type 9 (`UIElement_Resizebar`), and four authored
BOOL properties on that child say which border it drives:

`UIElement_Resizebar::StartMouseResizing @0x0046B7E0`

```
0046b7f8  GetAttribute_Bool(this, 0x2C, &bRight)
0046b806  GetAttribute_Bool(this, 0x2B, &bLeft)
0046b814  GetAttribute_Bool(this, 0x2D, &bTop)
0046b822  GetAttribute_Bool(this, 0x2A, &bBottom)

  if (bRight)      border = bTop ? BORDER_UR   : (bBottom ? BORDER_LR : BORDER_RIGHT)
  else if (bLeft)  border = bTop ? BORDER_UL   : (bBottom ? BORDER_LL : BORDER_LEFT)
  else if (bTop)   border = BORDER_TOP
  else if (bBottom)border = BORDER_BOTTOM

0046b88e  this->m_mousePressed = 1
0046b89f  UIElement::StartResizing(parent, border, pt.x, pt.y)
0046b8a4  this->SetState(3)                              // "pressed" art
```

`BorderLocation` (`acclient.h:4327`):
`BORDER_NONE=0, BORDER_UL=1, BORDER_TOP=2, BORDER_UR=3, BORDER_RIGHT=4,
BORDER_LR=5, BORDER_BOTTOM=6, BORDER_LL=7, BORDER_LEFT=8`.

So the property map is **`0x2A` = bottom, `0x2B` = left, `0x2C` = right,
`0x2D` = top**, and a corner grip simply sets two of them.

Drag lifecycle (`UIElement_Resizebar::ListenToElementMessage @0x0046B930`):
message `0x1C` (mouse down, `dwParam1 == 7` = left button) → `StartMouseResizing`
+ parent's vtable+0x60; `0x1E` (mouse move) → parent's vtable+0x4C
(`MouseResizeElement @0x00461130`); `0x1D` (mouse up) → parent
vtable+0x64 + `StopMouseResizing` → `UIElement::StopResizing @0x0045FD60`.

**Both chat layouts author all eight grips.** Retail is resizable from every
edge and every corner, including the top and the top corners, on both the main
and the floating chat windows.

**CORRECTED 2026-08-10 at Campaign CH slice CH6a implementation.** This claim
is wrong for the main window's plain TOP EDGE specifically, established by a
direct `ElementDesc.Type` dump of the installed DAT (ground truth, not a
decomp reading) of the 8 ids `0x1000069B`-`0x100006A2`: 7 of them are
Type 9 (`UIElement_Resizebar`) as claimed, but `0x1000069C` — the straight
top-edge strip between the two top corners — is Type **2**
(`UIElement_Dragbar`), not Type 9. So retail's main chat window is
resizable from every edge and corner EXCEPT the plain top strip, which is a
MOVE handle instead (there is no title bar, so the top strip does double
duty as the drag affordance). The two top CORNERS (`0x1000069B` UL,
`0x1000069D` UR) are still genuine Resizebar grips carrying the top bool, so
dragging from a corner still resizes the Y/top axis — only the straight
edge in between does not. This is fully consistent with, and explains, the
`UIElement_Resizebar::StartMouseResizing` cursor media dump: `0x1000069C`
carries cursor `0x06006119` (the four-arrow MOVE cursor, matching
`WindowMove`), not one of the two diagonal or the vertical resize cursor
ids the seven true grips carry. Not independently re-verified for the
floating-window layout `0x2100005B` (CH6b's scope); assume it needs the
same direct-dump check rather than trusting this row for that layout too.

### 2.4 Min/max extents

Retail reads them as ordinary integer attributes on the window element:
`gmMainChatUI::HandleMaximizeButton @0x004CCE50` uses
`GetAttribute_Int(this, 0x3C, &maxH)` and `GetAttribute_Int(this, 0x3E, &minH)`.
acdream already consumes the same family — `RetailWindowFrame.ResolveConstraint`
(`src/AcDream.App/UI/Layout/RetailWindowFrame.cs:160-169`) maps
`0x3C`=maxH, `0x3D`=maxW, `0x3E`=minH, `0x3F`=minW. No change needed.

---

## 3. Opacity — the two values

Named constants in the binary's read-only data at `0x007A8D50`:

```
007a8d50  uint32_t const Option_TextType_Property      = 0x1000007F
007a8d54  uint32_t const Option_DefaultOpacity_Property = 0x10000080
007a8d58  uint32_t const Option_ActiveOpacity_Property  = 0x10000081
```

| id | name | type | scope | meaning |
|---|---|---|---|---|
| `0x10000080` | `Option_DefaultOpacity_Property` | float | **global** (`InqOption`, not per-window) | opacity when the window's text entry does NOT have focus |
| `0x10000081` | `Option_ActiveOpacity_Property` | float | **global** | opacity while the window's text entry HAS focus |

Read path (`gmFloatyChatUI::UpdateFromPlayerModule @0x004CE3F0`,
identical in `gmFloatyMainChatUI::UpdateFromPlayerModule @0x004D2970`):

```
004ce42b  if (PlayerModule::InqOption(pm, 0x10000080, &prop) && prop->InqFloat(&v))
004ce445      ChatInterface::SetDefaultOpacity(this, v)
004ce465  if (PlayerModule::InqOption(pm, 0x10000081, &prop) && prop->InqFloat(&v))
004ce47f      ChatInterface::SetActiveOpacity(this, v)
```

Live-update path (`gmFloatyMainChatUI::RecvNotice_GameplayOptionChanged
@0x004D25A0`) switches on the same two ids and forwards anything else to
`ChatInterface::RecvNotice_GameplayOptionChanged @0x004F30E0` (which handles the
`0x1000007F` filter).

They are also settable per-element from the LayoutDesc:
`ChatInterface::OnSetAttribute @0x004F3F60` accepts `0x10000080` / `0x10000081`
as element attributes and **falls back to `0x3F800000` (= 1.0f)** when the
property value cannot be read (`004f3fb2`, `004f3f84`). That 1.0f is the only
default the code itself carries.

The options UI wires them to two linked sliders — `gmChatOptionsUI::InitOptions
@0x0049FC60`:

```
0049fcc0  UIOption_Slider::SetGameplayOptionProperty(slider1, 0x10000080)
0049fd1a  UIOption_Slider::SetGameplayOptionProperty(slider2, 0x10000081)
0049fd5c  DualHash<UIOption_Slider*,UIOption_Slider*>::add(&m_hashSliderLinks, &slider1, &slider2)
```

The `DualHash` link is why the two sliders track each other in the retail
options panel, and it is mirrored in code:
`ChatInterface::SetDefaultOpacity @0x004F3BC0` calls `SetActiveOpacity` when
active < default, and `SetActiveOpacity @0x004F3C40` calls `SetDefaultOpacity`
when default > active. **Invariant: `activeOpacity >= defaultOpacity` always.**

Which one is applied right now (`SetDefaultOpacity @0x004F3BC0`,
`SetActiveOpacity @0x004F3C40`):

```
if (this window's root is the active element
    && GetFocusDescendant(activeElement) == m_chatEntry)
        m_fCurrentOpacity = m_fActiveOpacity
else    m_fCurrentOpacity = m_fDefaultOpacity
```

and application is one call on the window's own render surface —
`ChatInterface::SetOpacity @0x004F3120`:

```
004f3124  this->m_fCurrentOpacity = v
004f312a  obj = m_object ?: UIRegion::GetObjectA(m_parent)
004f314b  surface = obj->vtable[0x1C]()          // get render surface
004f3156  surface->vtable[0x48](v)               // set surface alpha
```

**Retail fades the whole composited window surface — chrome, backgrounds AND
text — with one alpha, not a per-widget background tint.** This is the shape the
future acdream user setting should take.

**RESOLVED at Campaign CH slice CH6c (2026-08-10) by static decomp, not cdb —
the values are constructor-literal, so no live attach was needed.** Retail's
shipped defaults are PER WINDOW CLASS, not one constant:

```
004f4550  ChatInterface::ChatInterface(this, arg2, arg3)      // BASE ctor
004f459f      this->m_fDefaultOpacity = 0.5f;
004f45a5      this->m_fCurrentOpacity = 0.5f;
004f45ab      this->m_fActiveOpacity  = 1f;

004cd0f0  gmMainChatUI::gmMainChatUI(this, arg2, arg3)         // derived, calls base first
004cd0ff      ChatInterface::ChatInterface(this, arg2, arg3);
004cd148      this->m_fDefaultOpacity = 1f;                     // OVERRIDES base
004cd14e      this->m_fCurrentOpacity = 1f;
                                                                 // m_fActiveOpacity left at base's 1f

004ce2c0  gmFloatyChatUI::Create(arg1, arg2)                   // the 4 floating windows
004ce2e0      ChatInterface::ChatInterface(eax, arg1, arg2);   // NO override — keeps base 0.5/1.0
```

`gmFloatyMainChatUI` (element class `0x10000050`, the concrete class actually
instantiated for the retail main chat window — its `DynamicCast` accepts both
`0x10000050` and `0x10000041`) calls `gmMainChatUI::gmMainChatUI` as its own
base constructor (`0x004D22B0`) and adds no opacity override of its own, so it
inherits `gmMainChatUI`'s 1.0/1.0.

**So: the main chat window is ALWAYS FULLY OPAQUE in both states (Default=1.0,
Active=1.0) unless a saved `GameplayOptions` value overrides it via
`UpdateFromPlayerModule`; the four floating windows default to
Default=0.5/Active=1.0 (translucent when idle, opaque once the chat entry has
focus).** These are IN-MEMORY CONSTRUCTED starting values for each window
INSTANCE's fields — they get overwritten the moment `UpdateFromPlayerModule`
successfully reads a persisted `0x10000080`/`0x10000081` value from
`PlayerModule::InqOption` (the SAME global option for every window instance),
which is why the two options are still correctly described as GLOBAL rather
than per-window: only a NEVER-SAVED option (a fresh character, nothing in the
`GameplayOptions` blob yet) lets the per-class constructed defaults show
through, and even then only until the user's first slider drag pushes one
shared value into every live window via `RecvNotice_GameplayOptionChanged`.

**CH6c review-fix round (2026-08-10): the shared default above was WRONG.**
Shipping the base `ChatInterface` value (0.5/1.0) as ONE shared global
default, combined with the scope extension to every registered window, faded
the WHOLE registered UI (radar, vitals, toolbar, main chat, ...) to 50%
opacity out of the box — including several windows that can never take
keyboard focus at all, so they were stuck at 0.5 permanently. acdream now
ships `gmMainChatUI`'s per-class 1.0/1.0 override (`0x004CD0F0`) as the
shared global default instead (register row AP-190 in
`docs/architecture/retail-divergence-register.md`), which is retail-identical
for the 11 non-chat windows and the main chat window and leaves only the four
floating chat windows diverging from retail's 0.5-while-idle fade — a
default-VALUE divergence the transparency slider still fully covers. The
linking invariant (active >= default, restored by dragging the OTHER value —
verified from `SetDefaultOpacity`/`SetActiveOpacity`'s own bodies, matching
the summary already recorded above) is ported exactly regardless of which
default seeds it.

### 3.1 Two more residuals found at the CH6c review (not yet ported)

Both are decomp-verified and both are recorded as new AP-190 clauses; neither
is implemented this round.

**(a) Retail eases opacity between endpoints; acdream snaps.**
`ChatInterface::ListenToGlobalMessage @0x004F3840` is the handler for global
message id `3`, armed (`UIListener::RegisterForGlobalMessage(this, 3)`) from
the element-focus messages `0x1A`/`0x1E`/`0x28`/`0x29`/`0x2E` inside the
window's `ListenToElementMessage` switch at `0x004F5275`. Once armed, every
tick nudges the live opacity toward whichever endpoint
`IsTextEntryFocused` currently selects by 5% of the endpoint delta
(`fabsl(target - current) * 0.05f`), and unregisters from the global message
once the value lands within FP-epsilon of the target. acdream's
`RetailWindowOpacityController.Apply` sets the target opacity directly on the
focus-change event — the START and END states are retail-exact, but the
transition is an instant snap instead of a roughly 20-tick fade. Porting the
lerp needs a UI frame-tick hook `RetailWindowOpacityController` does not have
today (it only reacts to `DescendantFocusChanged`); deferred.

**(b) Retail's focus predicate is the chat entry field specifically; acdream's
is any focusable descendant.** `ChatInterface::IsTextEntryFocused @0x004F30A0`
tests `GetFocusDescendant(rootElement) == this->m_chatEntry` — literally the
window's text-entry element, not "some descendant of this window has focus."
acdream's `RetailWindowHandle.DescendantFocusChanged` (the event
`RetailWindowOpacityController` subscribes to) fires whenever ANY focusable
descendant of the window gains focus. For a window with exactly one
focusable child the two predicates coincide; for a window with several (a
settings panel's multiple controls, for example) acdream's broader predicate
holds ActiveOpacity while retail would already have faded back to
DefaultOpacity once focus left the specific text-entry widget.

**Pre-existing, unrelated: `UiMenu.cs:293`'s opacity bypass.** Popup menus
call `ctx.PushAlphaAbsolute(1f)` before drawing so a menu always reads solid
even when it is opened from a translucent (faded) window — this is a
deliberate acdream-only presentation choice (menus must stay legible
regardless of the host window's current fade state), not a divergence from
either of the two opacity mechanisms documented in this section, and it
predates the CH6c slice.

---

## 4. Persistence — how filters, geometry, visibility and title survive

### 4.1 The per-window option structure

Every per-window chat setting lives inside ONE global gameplay option, an
array indexed by `windowId - 1`. `PlayerModule::GetChatOptionStructure
@0x005D5300`:

```
005d5326  find option 0x1000008C in m_colGameplayOptions       // the ARRAY
005d5360  … if absent: SetPropertyName(&p, 0x1000008C); hash.add(p)
005d5473  if (arrayProp->type == 0x11 && (windowId-1) < arrayProp->capacity)
005d54a8      SetPropertyName(&elem, 0x1000008B)               // the ELEMENT
005d54b3      for i in currentCount..=(windowId-1): append elem
005d553b  return arrayProp->vtable[0x10C](windowId - 1)        // entry
```

- **`0x1000008C`** — the per-window option ARRAY (one entry per chat window).
- **`0x1000008B`** — the property name of each ARRAY ELEMENT (a nested bag).

`PlayerModule::InqChatWindowOption @0x005D5540` / `SetChatWindowOption
@0x005D5570` are thin `(windowId, propertyId)` accessors over that structure.

### 4.2 The complete `InqChatWindowOption` property family

| id | type | written by | read by |
|---|---|---|---|
| `0x1000007F` | bitfield64 | chat options UI (`gmChatOptionsUI::AddCheckboxBitfield64Option @0x0049EDA0`) | `ChatInterface::UpdateFromPlayerModule @0x004F3920`; live via `RecvNotice_GameplayOptionChanged @0x004F30E0` |
| `0x10000086` | int — window **X** | `gmFloatyChatUI::MoveTo @0x004CE840:004ce892`; `gmFloatyMainChatUI::MoveTo @0x004D2D10:004d2e06` | `UpdateFromPlayerModule` → `MoveTo(x,y)` |
| `0x10000087` | int — window **Y** | same `MoveTo` sites, `:004ce8da` / `:004d2e4e` | same |
| `0x10000088` | int — **width** | `gmFloatyChatUI::ResizeTo @0x004CE6D0:004ce722`; `gmFloatyMainChatUI::ResizeTo @0x004D2C00:004d2c5e` | `UpdateFromPlayerModule` → `ResizeTo(w,h)` |
| `0x10000089` | int — **height** | same `ResizeTo` sites, `:004ce770` / `:004d2cac` | same |
| `0x1000008A` | bool — **visible / open** | `gmFloatyChatUI::SetVisible @0x004CE9B0:004ce9ff` | `UpdateFromPlayerModule` → `SetVisible` |
| `0x1000008D` | StringInfo — **window title** | `gmFloatyChatUI::SetWindowTitle @0x004CEAA0:004ceb14` | `UpdateFromPlayerModule @0x004CE3F0:004ce63a` |

Two guards worth porting verbatim:

1. Every write is gated on `m_eWindowID != 0` — **the main chat window (id 0)
   never persists geometry, visibility, title or filter.** Only windows 1..4 do.
2. Position/size restore is skipped entirely when
   `CPlayerSystem::GetPlayerSystem()->m_layoutFromFile != 0`
   (`004ce4a8`, `004d2a25`) — i.e. a `LoadScreenLayout` file wins over the
   server-side option blob. Visibility and title still restore in that case.

`gmFloatyMainChatUI::MoveTo @0x004D2D10` additionally clamps the window inside
its parent before delegating (`004d2d53`–`004d2dbb`), so a saved position from a
larger resolution can never strand the window off-screen.

### 4.3 The local screen-layout file (a second, independent persistence path)

`gmGamePlayUI::SaveScreenLayout @0x004EAD50` / `LoadScreenLayout @0x004EA8F0`
write a plain-text file (path built by `CreateScreenLayoutPath @0x004EA690`)
with one line per window:

```
<SBOX> X:%d Y: %d W: %d H: %d       // 0x1000049A
<CHAT> X:%d Y: %d W: %d H: %d       // 0x10000601  ← main chat
<FCH1> X:%d Y: %d W: %d H: %d       // 0x10000505
<FCH2> …                            // 0x1000050E
<FCH3> …                            // 0x1000050F
<FCH4> …                            // 0x10000510
```

`LoadScreenLayout` matches the 4-character tags back to element ids
(`004eaa91`, `004eaab8`, `004eaadf`, `004eab06`). This is retail's
"save/load UI layout" feature and is the `m_layoutFromFile` flag's source.
Notably it is the ONLY path that persists the **main** chat window's geometry.

### 4.4 The wire

`m_colGameplayOptions` is packed whole into the character-options blob.
Server side, ACE treats it as **opaque bytes**:

- outbound: `references/ACE/Source/ACE.Server/Network/GameEvent/Events/GameEventPlayerDescription.cs:349-350, 393-394` —
  sets `CharacterOptionDataFlag.GameplayOptions` (`0x00000200`) and writes
  `Character.GameplayOptions` verbatim.
- inbound: `references/ACE/Source/ACE.Server/Network/GameAction/Actions/GameActionSetCharacterOptions.cs:183-190` —
  *"This is the last message... So it should be all that is left"* — reads the
  remaining bytes and calls `SetCharacterGameplayOptions(bytes)`.
- storage: `references/ACE/Source/ACE.Database/Models/Shard/Character.cs:49` —
  `public byte[] GameplayOptions`.

**ACE does not parse chat-window options at all.** It stores and echoes the
blob. Consequences for acdream:
- Anything acdream writes into that blob will round-trip through ACE unchanged.
- Nothing on the server validates it, so acdream owns the format entirely —
  which means acdream must match retail's `PropertyCollection` packing exactly
  if a retail client and acdream are ever to share a character.
- Until acdream can pack it, the cheapest correct behaviour is **local**
  persistence (the existing `SettingsStore` window-layout path), with the wire
  round-trip deferred.

---

## 5. acdream inventory — what exists, what is missing

### 5.1 What we mount today

`RetailUiRuntime.MountChat()` — `src/AcDream.App/UI/RetailUiRuntime.cs:670-736`:

```
676   LayoutImporter.ImportInfos(dats, ChatWindowController.LayoutId)   // 0x21000006
691   ChatWindowController.Bind(info, layout, …)
708   RetailWindowFrame.Mount(Host.Root, root, …)
714       WindowName    = WindowNames.Chat
715       Chrome        = RetailWindowChrome.NineSlice
718       ContentWidth  = 490f
728       ResizableEdges = ResizeEdges.Left | ResizeEdges.Right | ResizeEdges.Bottom
729       Opacity       = 0.75f
```

`ChatWindowController` — `src/AcDream.App/UI/Layout/ChatWindowController.cs`:

- `:29` `LayoutId = 0x21000006`, `:33` `RootId = 0x1000000E`
- `:34` `ResizeBarId = 0x1000000F` — **dropped** at `:204-205`
- `:35-42` transcript panel / transcript / track / input bar / menu / input /
  send / max-min ids
- `:228`, `:237` `BackgroundColor = (0,0,0,0.35)` on transcript and input
- `:383-426` `ToggleMaximize` — a faithful port of
  `gmMainChatUI::HandleMaximizeButton @0x004CCE50`

### 5.2 Gap list

**G1 and G2 are CLOSED as of Campaign CH slice CH6a (2026-08-10).**
`ChatWindowController.LayoutId`/`RootId` now import `0x2100006F`/`0x10000600`;
the crop/rebase/orphan-pruning compensations are deleted; `LayoutImporter`
gained a Type-9 case (`UiResizeGrip`); `UiRoot` gives a directly-hit grip's
own edges priority over its generic proximity heuristic; the main window
mounts with `RetailWindowChrome.Imported` (0x2100006F's own border art is
its chrome). The two subsections below are kept verbatim as the historical
diagnosis — do not re-run this investigation.

**G1 — Wrong LayoutDesc for the main window.**
`ChatWindowController.cs:29` imports `0x21000006` with root `0x1000000E`.
Retail's EoR main chat window is **LayoutDesc `0x2100006F`**, window element
`0x10000601`, root `0x10000600`, 410 × 100 (§2.1), evidenced three independent
ways: the runtime layout dump, `gmGamePlayUI::SetupChildren`/`SaveScreenLayout`
element ids, and `gmMainChatUI`'s own `0x1000046F` / `0x10000522`–`0x10000525`
child lookups. `0x21000006` is a different (older/standalone) chat layout — its
root `0x1000000E` and its 800px-wide resize bar `0x1000000F` appear nowhere in
the EoR gameplay UI, and `ChatWindowController.cs:185-193` already documents
that it drags along stray unparented siblings (`0x1000001C/1D/1E`,
`0x10000526`) that had to be orphaned by hand. Every symptom in the user's
report (3) — stray geometry, hand-cropped 490px content width, a dropped
800px resize bar, a 9px hole patched by growing the transcript panel at
`:210-211` — is downstream of this single choice.
UNVERIFIED: the exact provenance of `0x21000006`. Cheapest resolution:
`AcDream.App ui-studio --layout 0x21000006 --dump` next to
`--layout 0x2100006F --dump` and diff the element sets — one command, no
connected session.

**G2 — Only three edges resize, and the resize model is not retail's.**
`RetailUiRuntime.cs:728` sets `ResizableEdges = Left | Right | Bottom`,
deliberately excluding `Top` because the authored resize bar was dropped (G1).
`UiRoot.HitEdges` (`src/AcDream.App/UI/UiRoot.cs:999-1010`) does support
corners — a corner is just two bits — so with `Top` masked out the top-left
and top-right corners are dead and only the two bottom corners work. That is
exactly the user's report (2). Retail authors **eight** grips (§2.3) with
per-grip bool properties `0x2A`/`0x2B`/`0x2C`/`0x2D`. Additionally,
`LayoutImporter` has no concept of element type 9 (`UIElement_Resizebar`) — grep
for `Resizebar` in `src/AcDream.App/UI` returns nothing — so authored grips
would be imported as inert sprites today.

**G3 — Window opacity is completely inert. CLOSED at Campaign CH slice CH6c
(2026-08-10).** `UiRenderContext.ApplyAlpha` already gated `DrawRect`/
`DrawFill`/`DrawSprite` before this slice (added back at `1da697ec`, well
before CH6 — the "zero consumers" framing below described the PUBLIC
`AlphaMod` property specifically, not the private `_alpha`/`ApplyAlpha` pair
those three draws already used); the actual gap was narrower than originally
scoped: (a) `DrawStringDat`/`DrawString` still passed `applyAlpha: false`, so
TEXT stayed sharp over a translucent window against retail's whole-surface
`SetOpacity` semantics — CH6c fixed both; (b) nothing ever SET a window's
`Opacity` below its 1f default, since `RetailUiRuntime.MountChat` deliberately
left it at 1f pending this slice. CH6c added
`RetailWindowOpacityController` (`src/AcDream.App/UI/RetailWindowOpacityController.cs`),
which drives every `RetailWindowManager`-registered window's live `Opacity`
from keyboard-focus state and the two retail-linked Default/Active floats,
now exposed as a Settings → Chat tab transparency slider pair
(`SettingsPanel.RenderChatTab`). See the verified defaults + linking
behavior above (§3) and register row AP-190. The original paragraph below is
kept verbatim as the historical record of what CH6a/b actually shipped —
do not re-run this investigation.

`RetailWindowFrame.cs:157` sets `outerFrame.Opacity`, and
`UiElement.DrawSelfAndChildren` (`src/AcDream.App/UI/UiElement.cs:465`) and
`DrawOverlays` (`:513`) push it onto `UiRenderContext`'s alpha stack. But
**`UiRenderContext.AlphaMod` (`src/AcDream.App/UI/UiRenderContext.cs:55`) has
zero consumers** — a repo-wide grep over `src/` and `tests/` returns only its
own declaration plus one `PushAlphaAbsolute` call in `UiMenu.cs:293`. No sprite,
rect or text draw ever multiplies by it. So `Opacity = 0.75f` at
`RetailUiRuntime.cs:729` changes nothing, and the only translucency the chat
window has is the two hard-coded `(0,0,0,0.35)` background tints at
`ChatWindowController.cs:228` and `:237` plus whatever alpha is baked into the
DAT chrome textures. That is the user's report (3). Retail applies ONE alpha to
the whole composited window surface including text (§3).

**G4 — No multi-window support of any kind.**
`WindowNames` has a single `Chat` entry; `RetailUiRuntime.MountChat()` mounts
exactly one window; there is no window-id concept, no `0x1000007E` attribute
read, no filter state, no `0x10000522`–`0x10000525` binding, and no
`ToggleFloatingChatWindow1..4` input action. Grep for `ToggleFloatingChat` or
`WindowId` under `src/AcDream.App/UI` returns nothing.

**G5 — No per-window filter state.** The 64-bit `m_llTextTypeFilter` model and
the `PostInit` seeded defaults are decoded (colour-table doc §4) but unbuilt.
`ChatWindowController` renders `vm.RecentLinesDetailed()` unfiltered
(`ChatWindowController.cs:477`).

**G6 — No gameplay-options wire.** `PlayerDescriptionParser.cs:433-443` slices
the inbound `GameplayOptions` blob **heuristically** (`TryHeuristicInventoryStart`)
and never parses it; `SocialActions.cs:43-49` records that the outbound
`SetCharacterOptions (0x01A1)` full-blob builder was **deleted** in CH3 as
malformed and callerless. So there is currently no way to read or write
`0x1000008B`/`0x1000008C`.

**G7 — Local persistence has no opacity or window-id dimension.**
`RetailWindowLayoutPersistence.Capture` (`src/AcDream.App/UI/RetailWindowLayoutPersistence.cs:166-178`)
stores X/Y/W/H/Visible/Collapsed/Maximized only. It would persist four chat
windows correctly the moment they are registered under distinct `WindowNames`,
but it carries no opacity field and no filter field.

---

## 6. Recommended CH6 port shape

The single most valuable move is **G1**: import the layouts retail actually
uses. `0x2100006F` brings the eight resize grips, the 1/2/3/4 indicator buttons
and a coherent 410 × 100 root; `0x2100005B` is the floaty template. That
retires the hand-cropping, the dropped resize bar and the 9px patch in one
change, and it turns G2 from "add a resize model" into "import the one retail
authored".

### 6.1 Layering (matches the J4.1 communication-state pattern)

**Runtime — `RuntimeCommunicationState` extension** (it already owns the
transcript, reply/retell targets, rooms, friends and squelch, per
`docs/research/2026-07-26-slice-j4-1-communication-state.md`). Add a
presentation-free `ChatWindows` child owning, for ids 0..4:
- `ulong TextTypeFilter` per window, seeded by retail's `PostInit` switch
- `bool Open` per window (id 0 always open)
- `float DefaultOpacity` / `float ActiveOpacity` — **global**, not per-window,
  with retail's `active >= default` coupling enforced in the setter pair
  (`SetDefaultOpacity`/`SetActiveOpacity` semantics, §3)
- the routing predicate: display iff `windowId == m_eWindowID` **or**
  (`windowId == 0` && `TypeIsActive(type)`) — the rule already decoded in the
  colour-table doc §4, currently living nowhere.

Runtime owns this because no-window headless hosts already consume chat and
must not depend on presentation, and because the filter decides *routing*, not
appearance.

**Presentation — one `ChatWindowController` INSTANCE PER WINDOW.** The class is
already instance-based with no statics; give it a `WindowId` and a
`layoutId`/root-id pair so the same type binds both `0x2100006F` (main: adds
talk-focus menu, max/min, four indicator buttons) and `0x2100005B` (floaty:
adds title bar + close button). Register the four floaties as
`WindowNames.ChatWindow1..4` so `RetailWindowLayoutPersistence` picks them up
for free (G7).

**Input.** Four new actions `ToggleFloatingChatWindow1..4` in
`AcDream.UI.Abstractions/Input/`, defaulted to retail's modifier+digit chords
(§1.3), dispatched through the existing `InputDispatcher`. The four indicator
buttons call the same command; their pressed/normal state mirrors window
visibility exactly as `gmMainChatUI::RecvNotice_SetPanelVisibility` does.

**Settings.** Opacity becomes two `SettingsStore` floats plus two linked
sliders, matching `gmChatOptionsUI::InitOptions`'s `DualHash` linkage. Defer the
`0x1000008B`/`0x1000008C` wire (G6) — local persistence first, and file a
follow-up issue for the blob, because a malformed `0x01A1` is what CH3 already
had to delete once.

### 6.2 Slice sizing

| work | size | notes |
|---|---|---|
| **CH6a** — re-import main chat from `0x2100006F`; import + honour the eight Resizebar grips (element type 9, props `0x2A`–`0x2D`) in `LayoutImporter`; retire the 490px crop, the dropped resize bar and the 9px patch | **one slice**, but the biggest one | Fixes user reports (2) and (3)-artifacts. Visual gate required. |
| **CH6b** — make `Opacity` real: consume `UiRenderContext.AlphaMod` in every sprite/rect/text draw; delete the two hard-coded `(0,0,0,0.35)` tints | **one slice** | Fixes report (3)-transparency. Touches every widget draw path, so it wants its own slice and its own screenshot gate. Note the divergence: retail fades text too (§3); acdream's current comment at `UiRenderContext.cs:48-50` asserts the opposite as a deliberate choice — that row needs a decision and a register entry either way. |
| **CH6c** — Runtime `ChatWindows` state (filters + open flags + routing predicate) with no UI | **one slice** | Pure Runtime + tests, no visual gate. |
| **CH6d** — four floaty windows from `0x2100005B`, per-window controllers, toggle actions, indicator-button mirroring, close button, per-window local persistence | **one slice** | Depends on CH6a + CH6c. Fixes report (1). |
| **CH6e** — opacity settings UI + the two linked sliders | **small, deferrable** | |
| **CH6f** — `0x1000008B`/`0x1000008C` gameplay-options packing over `0x01A1` | **separate, later** | Do not bundle. CH3's deleted builder is the cautionary precedent. |

CH6a and CH6b both touch the render/import path and must not be parallelised
(`feedback_dont_parallelize_coupled_plan_slices`). CH6c is independent and can
run alongside either.

### 6.3 Divergence-register rows this work implies

- **Retires** any row asserting "chat window is not resizable from the top" once
  CH6a lands.
- **CLOSED at CH6c**: text now respects window alpha — `DrawStringDat`/
  `DrawString` route through `ApplyAlpha` exactly like `DrawSprite`/`DrawRect`/
  `DrawFill`, matching `ChatInterface::SetOpacity`'s whole-surface fade. No
  divergence row needed for this part.
- **New row AP-190** (CH6c): acdream applies the two opacity options to EVERY
  `RetailWindowManager` window (chat + floaties + vitals + toolbar +
  everything else), where retail's mechanism only ever runs from
  `ChatInterface`-derived windows; and ships ONE shared default (the base
  `ChatInterface` ctor's 0.5/1.0) rather than `gmMainChatUI`'s per-class
  1.0/1.0 override for the main window specifically.
- **New row** for local-only chat-window persistence until CH6f, since retail
  stores this server-side in `GameplayOptions`.

---

## Appendix — quick id reference

| id | meaning |
|---|---|
| `0x1000007E` | LayoutDesc attribute: chat window id (`m_eWindowID`) |
| `0x1000007F` | per-window text-type filter (bitfield64) |
| `0x10000080` | global default (unfocused) opacity (float) |
| `0x10000081` | global active (focused) opacity (float) |
| `0x10000086` / `0x10000087` | per-window X / Y |
| `0x10000088` / `0x10000089` | per-window width / height |
| `0x1000008A` | per-window visible flag |
| `0x1000008B` | per-window option-bag element name |
| `0x1000008C` | the per-window option ARRAY (global gameplay option) |
| `0x1000008D` | per-window title (StringInfo) |
| `0x2A` / `0x2B` / `0x2C` / `0x2D` | Resizebar bottom / left / right / top bools |
| `0x24` | LayoutDesc property: input action that toggles this element |
| `0x3C` / `0x3D` / `0x3E` / `0x3F` | max height / max width / min height / min width |
| `0x10000040` | element class `gmFloatyChatUI` |
| `0x10000041` | element class `gmMainChatUI` |
| `0x10000050` | element class `gmFloatyMainChatUI` |
| `9` | element class `UIElement_Resizebar` |
| `0x2100006F` | LayoutDesc — main chat window |
| `0x2100005B` | LayoutDesc — floating chat window (all four) |
| `0x10000601` | main chat window element (`<CHAT>`) |
| `0x10000505` / `0x1000050E` / `0x1000050F` / `0x10000510` | floating chat windows 1–4 (`<FCH1>`–`<FCH4>`) |
| `0x10000522`–`0x10000525` | main-window indicator buttons for windows 1–4 |
| `0x1000052A` | floaty-window close button |
| `0x100004D9` | floaty-window title bar |
| `0x1000046F` | main-window max/min button |