fix(chat): CH4 re-review fixes — dialog-queue reentrancy, settings option-bit chokepoint

Should-fix 1: RetailDialogFactory.CloseDialog's queued branch removed the
active DialogInfo, ran DialogDone (whose callback can synchronously open a
new dialog under the SAME queue key — the two-stage house-abandon
confirmation does exactly this), then called OpenNextDialog, which did an
unconditional Dictionary.Add on a key the reentrant dialog had already
re-occupied. Retail's HashTable::add tolerates the duplicate; Dictionary
throws. OpenNextDialog now returns early when the queue key is already
active — the reentrant dialog's own eventual close drains the queue.

Should-fix 2: @join/@leave wrote the local RuntimeCharacterOptionsState bit
before sending, but the Settings Chat toggles reached a second binding
(SendSingleCharacterOption) that only sent the wire message, leaving the
Turbine membership gate stale until the next PlayerDescription.
LiveSessionRuntimeFactory.CreateCommandBindings now has one shared local
function for both entrances.

Should-fix 3: corrected TS-68/#360 wording again — retail's DoAllegiance
dispatcher table EXECUTES boot/ban/officer/title/motd/name/lock/house/
chat/broadcast locally through their own handlers; acdream shows the
unrecognized-subcommand refusal for all nine pending the #360 port. What
matches retail is the ownership rule (the verb never reaches
DoChannelCommand/the server), not the subcommand behavior itself. Removed
the inaccurate "matching retail, not merely harmless" / "now matches
this" claims from both the register row and the issue.

Nits: corrected the HouseAbandonDialogCallback_First citation (0x00580E1A
is DoHouse's load site for the callback pointer, not the function entry —
the entry is 0x00580240, with the stage-2 confirmation string built at
0x005802D8) in both ClientCommandController.cs and the mirrored test
comment; added an InlineData case pinning "@clist allegiance" to
RequestChannelList(0x02000000); converted RetailClientCommandCatalog.
KnownVerbs from a plain array to a FrozenSet<string> with
StringComparer.OrdinalIgnoreCase, matching the file's other lookup tables.

Suite: 12,221 passed / 4 skipped / 0 failed (Release), up from CH4's
12,216/4/0 — net +5 tests, no removals.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

docs/architecture/retail-divergence-register.md

@@ -389,7 +389,7 @@ AP-94..AP-112 for the confirmed retail-UI completion gaps.
 | TS-64 | **Retail's sound-preference surface is only partly present.** Retail registers eight `[Sound]` keys in `SoundManager::InitPrefs` @ `0x005503F0`; two are unimplemented in acdream. (a) `s_bPlaySoundOnlyWhenActive` (default **1**) is checked against `Device::m_bIsActiveApp` in every entry point and in both `PlaySoundInternal` overloads, so an unfocused retail client is SILENT; acdream keeps playing when the window loses focus. (b) `s_SoundFeatures == 1` forces pan to dead centre; acdream's `RetailSoundMixer.Mix`/`GetPan` take a `panningEnabled` flag with conformance coverage, but no preference is wired behind it, so panning can never be turned off. The three enable bools (`Sound Disabled`, `Ambient Sound Disabled`, `Interface Sound Disabled`) also have no acdream counterpart — note retail's on-disk polarity is inverted relative to its backing variables, so a future reader must not assume the sense. | `src/AcDream.App/Audio/OpenAlAudioEngine.cs` (no focus gate); `src/AcDream.Core/Audio/RetailSoundMixer.cs` (`panningEnabled`, unwired) | Slice A2 kept its blast radius on the mixing model: window-focus state and a preference surface are host plumbing rather than mixing math, and the mixer parameter exists so wiring them later needs no math change. | Alt-tabbed acdream keeps making noise where retail goes quiet; users cannot disable panning or the individual sound classes. | `SoundManager::InitPrefs @ 0x005503F0`; `SoundManager::PlaySoundInternal @ 0x0054FEC0` and `@ 0x00550170`; `docs/research/2026-08-08-audio-retail-soundmanager-core.md` §1 |
 | TS-65 | **Volume-squared quirk applied on the ambient path only.** Retail multiplies its volume knob twice on several paths: `PlaySoundA(DataID, CPhysicsObj*)` passes `effect_sound_volume` as the `vol` argument and `GetAttenuation` then multiplies by `effect_sound_volume` again, and both `PlayAmbientSound*` entry points pre-multiply by `ambient_sound_volume` before that same second multiply — so those sliders are effectively squared. acdream's `RetailSoundMixer.TryGetAttenuation` applies the knob exactly once (which is what `GetAttenuation` itself does) and the animation-hook path does not pre-multiply. Slice A5 squares the ambient path, where two independent lanes byte-confirmed the double application. | `src/AcDream.Core/Audio/RetailSoundMixer.cs` (`TryGetAttenuation` remarks); `src/AcDream.App/Audio/OpenAlAudioEngine.cs` (`Play3DWave`) | Which `PlaySoundA` overload the animation-hook path reaches was not pinned by the lane-1 decode, and inventing a squaring on an unconfirmed overload would change every hook sound's loudness curve on a guess. Single-multiply is the conservative, decoded-function-exact choice; the open question is cheap to settle with a cdb breakpoint on the two overloads. | At a non-unity effect slider, hook sounds are louder than retail (slider 0.5 gives −6 dB where retail gives −12). At the default slider of 1.0 the two are identical, so this is inert until the user moves the slider. | `SoundManager::PlaySoundA @ 0x00550AF0`/`@ 0x00550B70`/`@ 0x005507A0`; `SoundManager::GetAttenuation @ 0x00550020`; `docs/research/2026-08-08-audio-retail-soundmanager-core.md` §3 D12 |
 | ~~TS-66~~ | **RETIRED 2026-08-08 (Campaign A listening-gate fix; user-reported).** `seen_outside` interiors now keep the OUTDOOR ambient set: the listener source resolves the per-cell `CEnvCell.seen_outside` bit through the physics cache's `CellPhysics` record (the same #107 field `AdjustPosition` reads) and converts the ENVCELL-local origin through the cell's `WorldTransform` into landblock coordinates before the 3×3 walk centres on it — an outdoor Position's origin is already landblock-local, an envcell's is not, and skipping the conversion would centre the walk on a wrong point by up to a landblock. A cell record not yet resident resolves to silence for that rebuild rather than a wrong walk. Sealed interiors (dungeons) remain silent, which is retail-correct. | retired | — | — | `Ambient` gate per `docs/research/2026-08-08-audio-retail-ambient-authoring.md` §6/§8; `CEnvCell::add_ambient_sounds` (folded `ret`); user listening gate 2026-08-08 ("in retail I get both outside ambient and the ambient from indoors") |
-| TS-68 | **Filed 2026-08-09 (Campaign CH slice CH4); corrected 2026-08-09 at the CH4 REJECT-review, Blocker 1.** `@allegiance`/`@all` and `@house`/`@hou` are real retail management-command dispatchers with 12 and 15 subcommands respectively (registry doc §2.5/§2.5b). acdream ports only the subset with simple parameterless/single-field wire shapes (allegiance `info`/`hometown`/`ho`; house `recall`/`re`/`mansion_recall`/`alleg_recall`/`ma`/`abandon`). For `@house`, every other subcommand (open, close, storage, remove, boot, boot_all, remove_all, guest, available, hooks, on, off) still falls through to ACE server-passthrough (which replies "Unknown command") — unchanged from the original filing. **The original filing was WRONG for `@allegiance`/`@all`: retail's own `DoAllegiance` never reaches DoChannelCommand/server-passthrough for an unrecognized subcommand** — it prints "Please see @help Allegiance for more information on how to use this command." locally (`label_57da4b`, 0x0057DA4B) and stays entirely client-side. acdream now matches this: every allegiance subcommand beyond `info`/`hometown`/`ho` (boot, ban, officer, title, name, lock, chat, broadcast, motd) shows the same retail refusal client-side instead of reaching ACE — closing a real bug where the unmatched subcommand text was instead broadcast to the Allegiance chat channel (0x02000000). The standalone `@motd` verb (reached directly, not via `@allegiance motd`) remains a separate, still-open gap. `RetailClientCommandCatalog.TryMatchHouse`/`TryMatchAllegiance` (`src/AcDream.UI.Abstractions/Panels/Chat/RetailClientCommandCatalog.cs`) | Retail would execute these locally (with its own usage/confirmation/refusal text). House's unported subcommands still reach ACE, which does not implement them as chat commands either — no functional loss on a real server, but a user typing e.g. `@house open` gets ACE's generic "Unknown command" instead of retail's real behavior. Allegiance's unported subcommands now correctly stay local with retail's own refusal text instead of reaching ACE at all — matching retail, not merely harmless. | `ClientCommunicationSystem::DoAllegiance @ 0x0057D5A0`; `DoHouse @ 0x00580860`; ACE `GameActionType` opcodes for each subcommand (all exist server-side) |
+| TS-68 | **Filed 2026-08-09 (Campaign CH slice CH4); corrected 2026-08-09 at the CH4 REJECT-review, Blocker 1.** `@allegiance`/`@all` and `@house`/`@hou` are real retail management-command dispatchers with 12 and 15 subcommands respectively (registry doc §2.5/§2.5b). acdream ports only the subset with simple parameterless/single-field wire shapes (allegiance `info`/`hometown`/`ho`; house `recall`/`re`/`mansion_recall`/`alleg_recall`/`ma`/`abandon`). For `@house`, every other subcommand (open, close, storage, remove, boot, boot_all, remove_all, guest, available, hooks, on, off) still falls through to ACE server-passthrough (which replies "Unknown command") — unchanged from the original filing. **The original filing was WRONG for `@allegiance`/`@all`: retail's own `DoAllegiance` never reaches DoChannelCommand/server-passthrough for an unrecognized subcommand** — it prints "Please see @help Allegiance for more information on how to use this command." locally (`label_57da4b`, 0x0057DA4B) and stays entirely client-side. **Corrected again 2026-08-09 at the CH4 re-review, SHOULD-FIX 3.** Retail does NOT refuse boot/ban/officer/title/motd/name/lock/house/chat/broadcast — `DoAllegiance`'s dispatcher table EXECUTES each one locally through its own handler (e.g. `DoAllegianceBoot @ 0x0057D646` is the dispatcher's call site into `ClientCommunicationSystem::DoAllegianceBoot`; `DoAllegianceBan`/`DoAllegianceOfficer`/`DoAllegianceOfficerTitle`/`DoMotd`/`DoAllegianceName`/`DoAllegianceLock`/`DoAllegianceHouse` are its siblings in the same table). acdream has none of those nine handlers ported (tracked by issue #360) and instead shows the SAME unrecognized-subcommand refusal ("Please see @help Allegiance...", `label_57da4b`, 0x0057DA4B) for every one of them, pending the #360 port. What matches retail here is the OWNERSHIP RULE — the verb never reaches `DoChannelCommand`/server-passthrough for `@allegiance`/`@all` regardless of subcommand — NOT the subcommand's actual behavior, which retail executes and acdream does not yet. This still closes the real bug the original filing named (the unmatched subcommand text broadcast to the Allegiance chat channel, 0x02000000). The standalone `@motd` verb (reached directly, not via `@allegiance motd`) remains a separate, still-open gap. `RetailClientCommandCatalog.TryMatchHouse`/`TryMatchAllegiance` (`src/AcDream.UI.Abstractions/Panels/Chat/RetailClientCommandCatalog.cs`) | Retail would execute these locally (with its own usage/confirmation/refusal text). House's unported subcommands still reach ACE, which does not implement them as chat commands either — no functional loss on a real server, but a user typing e.g. `@house open` gets ACE's generic "Unknown command" instead of retail's real behavior. Allegiance's unported subcommands correctly stay local (never reach ACE) but show a generic refusal instead of retail's real per-subcommand execution — a user typing e.g. `@allegiance boot Name` gets "Please see @help Allegiance..." instead of retail's real boot confirmation/effect, until #360 ports the nine `DoAllegiance*`/`DoMotd`/`DoAllegianceHouse` handlers. | `ClientCommunicationSystem::DoAllegiance @ 0x0057D5A0`; `DoHouse @ 0x00580860`; ACE `GameActionType` opcodes for each subcommand (all exist server-side) |
 | TS-69 | **Filed 2026-08-09 (Campaign CH slice CH4).** `@day`, `@log`, and `@render` are registered retail verbs acdream recognizes only in the `/help <verb>` lookup table, not as executable client commands. `@day` needs a sky/time-of-day override hook the renderer doesn't expose; `@log` needs a safely-lifecycled chat-to-file writer (deferred to avoid an unaudited file-handle leak across reconnects); `@render` has no acdream equivalent to retail's `SmartBox::HandleRenderOption` render-option surface. All three fall through to server passthrough. `RetailCommandHelpTable` (`src/AcDream.UI.Abstractions/Panels/Chat/RetailCommandHelpTable.cs`) | A user typing `@day`/`@log`/`@render` gets ACE's "Unknown command" instead of retail's local toggle/file-copy/render-option behavior — cosmetic/QoL only, no gameplay impact | `ClientCommunicationSystem::DoDay @ 0x005706F0`; `DoSetOutput @ 0x0057E4F0`; `DoRenderOption @ 0x0057E120` |
 | TS-70 | **Filed 2026-08-09 (Campaign CH slice CH4).** The new `@index`/`@clist`/`@on`/`@off`/`@hslist`/`@allegiance info` outbound requests (`ClientCommandRequests.BuildIndexChannels`/`BuildListChannel`/`BuildOnChannel`/`BuildOffChannel`/`BuildListAvailableHouses`/`BuildAllegianceInfoRequest`) send the byte-correct retail wire request, but the corresponding inbound GameEvents (`ChannelIndex 0x0149`, `ChannelList 0x0148`, `AvailableHouses 0x0271`, `AllegianceInfoResponse 0x027C`) are registered in `GameEventType` but have no `GameEventWiring` handler — the server's reply is silently dropped rather than rendered. `src/AcDream.Core.Net/GameEventWiring.cs` | The request reaches ACE correctly (verifiable on the wire / server-side log) but the client shows nothing in response — looks like the command silently failed | ACE `GameEventChannelIndex`/`GameEventChannelList`/`GameEventHouseListAvailable`/`GameEventAllegianceInfoResponse` (`references/ACE/Source/ACE.Server/Network/GameEvent/Events/`) |
 | TS-67 | **Ambient contributions are computed in-plane.** Retail's `CLandBlock::add_ambient_sounds` @ `0x530310` positions each contributing land cell at its own SW terrain VERTEX, including that vertex's height, and `Ambient::CalcWeight` deliberately includes Z in its distance (where `CalcDir` deliberately excludes it — the two differ on purpose). acdream's gatherer supplies Z = 0 for the offset, so a cell's weight ignores the height difference between the listener and the terrain under that cell. | `src/AcDream.Core/Audio/AmbientSoundGatherer.cs` (`ContributeLandblock`) | Sampling the height needs the landblock's height table threaded into the walk alongside the terrain words; the walk already runs only on a 24 m crossing so the cost is not the obstacle, the extra plumbing at slice end was. The error is bounded by terrain relief inside 120 m and affects the crossfade weight only, never the direction. | On steep ground an ambient reads slightly louder than retail, because the true 3-D distance is longer than the planar one. | `CLandBlock::add_ambient_sounds @ 0x530310`; `Ambient::CalcWeight @ 0x550DD0` |