dialogs come to the front, and the red line names what it needs

- dialogs carry the reference's toplevel bit, so the stack-split window opens above Bagnon instead of behind it
- the failure line fills its blank — "Requires Blacksmith Hammer", not "Requires" — and Opening casts stop printing a label the spell itself marks "no text"
- the pet bar's Attack/Follow/Stay tooltips name their keybinding, and typed text reaches the focused frame through the client's own delivery walk
- AtlasLoot parses again: our Lua keeps 5.0's compatibility semicolon

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

crates/benilla-app/assets/ui/ActionBar.xml

@@ -363,15 +363,27 @@
         -- tooltip with its rank column, an ITEM slot the shared item renderer; ref
         -- ActionButton_SetTooltip). The re-enter keeps it live across page flips.
         -- Anchor: ref ActionButton_SetTooltip (ActionButton.lua l.366-372) branches on the
-        -- UberTooltips CVar — "1" → the default corner, else RIGHT/LEFT by bar. UberTooltips
-        -- registers with default "1" (byte-read: WoW.exe 0x48fdd9, default string 0x82e748;
-        -- sibling rows BlockTrades→"0", UnitNameRenderMode→"2" confirm the layout), and benilla
-        -- ships no CVar state, so the branch collapses to the stock-client one: action-bar
-        -- tooltips live at the screen's bottom-right, above the bags.
+        -- UberTooltips CVar — "1" → the default corner, else beside the button, LEFT for the bars
+        -- that sit at the screen edge and RIGHT for the rest. This branch was collapsed to the "1"
+        -- leg until B230, on the reason that benilla shipped no CVar state for anything to move
+        -- it; 0954 ended that and cvars.rs now registers the key (default "1", byte-read — the
+        -- evidence lives on that row), so the fork is live and the stock look is unchanged.
+        --
+        -- The ref's LEFT set is {MultiBarBottomRight, MultiBarRight, MultiBarLeft}. Only the first
+        -- exists here — the two VERTICAL bars are deliberately not shipped (see the header of
+        -- MultiBars.xml) — so it is the whole test, and the day they land they join it.
         function BenillaActionButton_OnEnter(button)
             BenillaActionButton_Adopt(button)
             if not GameTooltip then return end -- engine tests load file subsets
-            GameTooltip_SetDefaultAnchor(GameTooltip, button)
+            if GetCVar("UberTooltips") == "0" then
+                if button:GetParent() == MultiBarBottomRight then
+                    GameTooltip:SetOwner(button, "ANCHOR_LEFT")
+                else
+                    GameTooltip:SetOwner(button, "ANCHOR_RIGHT")
+                end
+            else
+                GameTooltip_SetDefaultAnchor(GameTooltip, button)
+            end
             GameTooltip:SetAction(BenillaActionButton_Action(button))
         end
 

crates/benilla-app/assets/ui/GameMenuFrame.xml

@@ -207,7 +207,10 @@
          MainMenuFrameButtonTemplate (Classic/Frame/MainMenuFrameTemplates.xml l.11-16) is that same
          button — same art, same trio, same size — so the header's ART BYTES note is unchanged. -->
 
-    <Frame name="GameMenuFrame" frameStrata="DIALOG" enableMouse="true" hidden="true">
+    <!-- `toplevel` is the ref's own (GameMenuFrame.xml l.3) and was dropped here: without the bit
+         a DIALOG-strata window opened over another one stays behind it (the B180/Bagnon class,
+         decision 1318). Restored alone; the ref's `movable`/`parent` stay off deliberately. -->
+    <Frame name="GameMenuFrame" frameStrata="DIALOG" toplevel="true" enableMouse="true" hidden="true">
         <!-- The era layout engine's own numbers (header SCOPE note): 28+144+28 wide, 32-top /
              28-bottom padding around our seven 21-unit rungs with three 20-unit section gaps. -->
         <Size><AbsDimension x="200" y="267"/></Size>

crates/benilla-app/assets/ui/ItemRef.xml

@@ -91,7 +91,9 @@
         end
     ]]></Script>
 
-    <GameTooltip name="ItemRefTooltip" inherits="GameTooltipTemplate" frameStrata="HIGH" hidden="true">
+    <!-- `toplevel` is the ref's (ItemRef.xml l.3) — a linked-item tooltip is a window that must
+         land above the HIGH-strata frames it is opened over (decision 1318). -->
+    <GameTooltip name="ItemRefTooltip" inherits="GameTooltipTemplate" frameStrata="HIGH" toplevel="true" hidden="true">
         <Size><AbsDimension x="128" y="64"/></Size>
         <Anchors>
             <Anchor point="BOTTOM"><Offset><AbsDimension x="0" y="80"/></Offset></Anchor>

crates/benilla-app/assets/ui/OptionsFrame.xml

@@ -1475,7 +1475,10 @@
         </Scripts>
     </Frame>
 
-    <Frame name="OptionsFrame" frameStrata="HIGH" enableMouse="true" hidden="true">
+    <!-- `toplevel` restored (ref OptionsFrame.xml l.3) — same class as the B180/Bagnon dialog
+         (decision 1318). NOTE our strata is HIGH where the ref is DIALOG; left as-is, flagged in
+         the record rather than silently changed. -->
+    <Frame name="OptionsFrame" frameStrata="HIGH" toplevel="true" enableMouse="true" hidden="true">
         <Size><AbsDimension x="920" y="724"/></Size>
         <Anchors><Anchor point="CENTER"/></Anchors>
         <!-- The 1.14 System-window dialog, verbatim (BACKDROP_DIALOG_32_32 — wow-ui-source

crates/benilla-app/assets/ui/PetActionBar.xml

@@ -98,13 +98,18 @@
         --
         --   4 bones, pivoting at the corners of a 0.02 x 0.02 unit square, each keyed with the
         --   SAME 5-key 2.000 s translation loop walking the square edge by edge (0.5 s per edge) —
-        --   the four phase-shifted by one corner each, so they chase one another round it.
+        --   the four phase-shifted by one corner each, so they chase one another round it. The
+        --   tracks are `interp == 1`, LINEAR (read at the bone record's M2Track word, not assumed):
+        --   constant speed along each edge, no easing into the corners, which is what the walk
+        --   below reproduces.
         --   4 particle emitters, one per bone: 300/s, life 1.000 s, Add blend, unlit, texture
-        --   Interface\Buttons\GlowStar.blp, size half-extent 0.005 -> 0.002 -> 0.001 over life,
-        --   colour (0.976,0.875,0.192,1) -> (0.996,0.945,0.745,1) -> (1,1,1,0). Its spline has 4
-        --   control points ALL AT THE ORIGIN, so a particle never travels: it is born where the
-        --   bone is and shrinks away there. The effect is therefore four comet trails running
-        --   round the button's rim, not a rotating ring.
+        --   Interface\Buttons\GlowStar.blp, size half-extent 0.005 -> 0.0015 -> 0.001 over life,
+        --   colour (0.976,0.875,0.192,1) -> (0.996,0.945,0.745,1) -> (1,1,1,0). Its spline is one
+        --   cubic Bezier segment whose 4 control points are ALL AT THE ORIGIN — checked, all four,
+        --   because the two INTERIOR handles are what would bend the path and reading only the
+        --   endpoints proves nothing — so a particle never travels: it is born where the bone is
+        --   and shrinks away there. The effect is therefore four comet trails running round the
+        --   button's rim, not a rotating ring.
         --
         -- Model units -> pixels is the same 1200 px/unit calibration BagFrame.xml derived from the
         -- Cooldown model (its own header carries the derivation), and it CROSS-CHECKS here: the
@@ -114,22 +119,44 @@
         -- number rather than eyeball a size.
         --
         -- OUR ONE DEPARTURE is the particle COUNT: 300 live per emitter is 12 000 quads across a
-        -- full bar, for a trail that reads identically at a fraction of that. We draw
-        -- BENILLA_PET_AUTOCAST_TRAIL per emitter, evenly spaced in AGE, which samples the same
-        -- position/size/colour ramps at the same times — the trail's shape, speed and colour are
-        -- the file's; only its density is ours.
+        -- full bar. We draw a SAMPLE of the same trail — each sample sits where the bone was `age`
+        -- ago and reads the same position/size/colour ramps at that age — so the trail's shape,
+        -- speed and colour are the file's; only its density is ours.
+        --
+        -- **How many samples is DERIVED, not chosen** (B228). It was 8, evenly spaced in age, on
+        -- the claim that the trail "reads identically at a fraction" of 300. Nobody had checked
+        -- that claim, and the arithmetic says it is false. The rim is 4 x 28.8 = 115.2 px round and
+        -- the lap is 2.0 s, so the bone travels 57.6 px/s and a particle's 1.0 s life spans 57.6 px
+        -- of rim — two whole edges. Eight even samples therefore sit 7.2 px apart, while the stars
+        -- behind the head are 4.3 px and then 2.9 px across: the back half of every trail was a row
+        -- of SEPARATED DOTS where the file draws one continuous streak. Marching dots are legible
+        -- motion in a way a smooth streak is not, which is the shape of thing a reporter reads as
+        -- "too fast" (nazriel, B228) even at the right period — and the period IS right, measured
+        -- through this file's own OnUpdate in `pet_bar_tests`.
+        --
+        -- So the samples are spaced by the file's own SIZE ramp rather than evenly: each step
+        -- advances one star-width / OVERLAP of arc. That spends samples in the tail where the
+        -- stars are small and saves them at the head where a 14.4 px star covers everything —
+        -- continuous everywhere, and cheaper than a uniform spacing fine enough for the tail.
         BENILLA_PET_AUTOCAST_PX_PER_UNIT = 1200
         BENILLA_PET_AUTOCAST_SCALE   = 1.2                                   -- the ref Model's
         BENILLA_PET_AUTOCAST_PERIOD  = 2.0                                   -- seq0 loop, seconds
         BENILLA_PET_AUTOCAST_LIFE    = 1.0                                   -- particle lifespan
-        BENILLA_PET_AUTOCAST_TRAIL   = 8                                     -- ours (see above)
         BENILLA_PET_AUTOCAST_EMITTERS = 4
+        -- Star-widths of cover on every point of the trail. 1 would put neighbours exactly edge to
+        -- edge, and a GlowStar fades out toward its own edge, so the seam would show; 2 (centres
+        -- half a width apart) is the cheapest value with no seam.
+        BENILLA_PET_AUTOCAST_OVERLAP = 2
         -- 0.02 units square, in pixels.
         BENILLA_PET_AUTOCAST_SIZE = 0.02 * BENILLA_PET_AUTOCAST_PX_PER_UNIT * BENILLA_PET_AUTOCAST_SCALE
-        -- Half-extent -> full width, in pixels: {0.005, 0.002, 0.001} * 2 * K * scale.
+        -- Half-extent -> full width, in pixels: {0.005, 0.0015, 0.001} * 2 * K * scale.
+        -- The MID key is 0.0015, not the 0.002 this table carried until B228: `m2part` printed the
+        -- ramp at 3 decimals and rounded it up, and the transcription took the printout at its
+        -- word. (Fixed at the instrument too — that column prints 4 decimals now.) 33% of the
+        -- mid-life star, so the trail was reading thicker than the file authors it.
         BENILLA_PET_AUTOCAST_STAR = {
             0.010 * BENILLA_PET_AUTOCAST_PX_PER_UNIT * BENILLA_PET_AUTOCAST_SCALE,
-            0.004 * BENILLA_PET_AUTOCAST_PX_PER_UNIT * BENILLA_PET_AUTOCAST_SCALE,
+            0.003 * BENILLA_PET_AUTOCAST_PX_PER_UNIT * BENILLA_PET_AUTOCAST_SCALE,
             0.002 * BENILLA_PET_AUTOCAST_PX_PER_UNIT * BENILLA_PET_AUTOCAST_SCALE,
         }
         -- The over-life colour ramp's three keys (the file's), as {r, g, b, a}.
@@ -167,15 +194,60 @@
             return out
         end
 
+        -- The trail's sample AGES, walked once off the size ramp above (the header has the why).
+        -- Each step advances by one local star-width / OVERLAP of rim, converted to seconds at the
+        -- bone's own rim speed — so the spacing tightens as the stars shrink and the streak stays
+        -- unbroken end to end. The count falls out of the file's numbers; nothing here picks it.
+        function BenillaPetAutocast_TrailAges()
+            local speed = 4 * BENILLA_PET_AUTOCAST_SIZE / BENILLA_PET_AUTOCAST_PERIOD  -- px/s of rim
+            local ages, age = {}, 0
+            while age < BENILLA_PET_AUTOCAST_LIFE do
+                table.insert(ages, age)
+                local w = BenillaPetAutocast_Ramp(BENILLA_PET_AUTOCAST_STAR, age / BENILLA_PET_AUTOCAST_LIFE)
+                -- The step is only positive while the ramp is. This runs at FILE LOAD, so a future
+                -- edit that put a zero in the size table would hang the whole UI before a frame
+                -- ever drew — the failure mode decision 1306 exists for. Bail instead.
+                if w <= 0 then break end
+                age = age + w / BENILLA_PET_AUTOCAST_OVERLAP / speed
+            end
+            return ages
+        end
+        BENILLA_PET_AUTOCAST_AGES  = BenillaPetAutocast_TrailAges()
+        BENILLA_PET_AUTOCAST_TRAIL = table.getn(BENILLA_PET_AUTOCAST_AGES)
+
         -- Build one button's sparkle pool: EMITTERS x TRAIL textures, hidden until autocast is on.
+        --
+        -- Everything about a sample EXCEPT where it is right now is fixed by its age — size,
+        -- colour and alpha are reads of the over-life ramps at an age that never changes for a
+        -- given slot — so all of it is set HERE, once. The per-frame pass moves the texture and
+        -- touches nothing else, which is why tripling the sample count still costs less per frame
+        -- than the eight-sample version did (it re-derived all five values for every spark, every
+        -- frame). Each spark also carries its fixed place in the loop: its emitter's quarter-lap
+        -- offset, less how far the bone has moved on since this star was born.
+        --
+        -- Built LAZILY, on the first frame the button actually autocasts. Most never do, and a
+        -- pool that is only ever hidden is pure cost — 22 buttons' worth of it across the bar and
+        -- the spell book.
         function BenillaPetAutocast_Create(button)
             button.sparks = {}
-            for i = 1, BENILLA_PET_AUTOCAST_EMITTERS * BENILLA_PET_AUTOCAST_TRAIL do
-                local t = button:CreateTexture(nil, "OVERLAY")
-                t:SetTexture("Interface\\Buttons\\GlowStar")
-                t:SetBlendMode("ADD")
-                t:Hide()
-                table.insert(button.sparks, t)
+            for e = 0, BENILLA_PET_AUTOCAST_EMITTERS - 1 do
+                for k = 1, BENILLA_PET_AUTOCAST_TRAIL do
+                    local age = BENILLA_PET_AUTOCAST_AGES[k]
+                    local life = age / BENILLA_PET_AUTOCAST_LIFE
+                    local size = BenillaPetAutocast_Ramp(BENILLA_PET_AUTOCAST_STAR, life)
+                    local c = BenillaPetAutocast_Ramp(BENILLA_PET_AUTOCAST_COLOR, life)
+                    local spark = button:CreateTexture(nil, "OVERLAY")
+                    spark:SetTexture("Interface\\Buttons\\GlowStar")
+                    spark:SetBlendMode("ADD")
+                    spark:SetWidth(size)
+                    spark:SetHeight(size)
+                    spark:SetVertexColor(c[1], c[2], c[3])
+                    spark:SetAlpha(c[4])
+                    spark:Hide()
+                    spark.phase = e / BENILLA_PET_AUTOCAST_EMITTERS
+                        - age / BENILLA_PET_AUTOCAST_PERIOD
+                    table.insert(button.sparks, spark)
+                end
             end
         end
 
@@ -183,34 +255,30 @@
         -- emitters are one model, and every autocasting button plays the same loop in step, which
         -- is what the reference's per-button Models do too (each starts at sequence time 0 and
         -- they never drift, the loop being exact).
+        --
+        -- Show and hide are EDGE work, not per-frame work: this runs for all ten buttons every
+        -- frame and all but one or two of them are dark, so the old unconditional Hide loop was
+        -- spending most of the bar's frame budget re-hiding what was already hidden.
         function BenillaPetAutocast_Update(button, clock)
-            if not button.sparks then return end
             if not button.autocasting then
-                for i = 1, table.getn(button.sparks) do button.sparks[i]:Hide() end
+                if button.sparksLit then
+                    for i = 1, table.getn(button.sparks) do button.sparks[i]:Hide() end
+                    button.sparksLit = nil
+                end
                 return
             end
-            local n = 1
-            for e = 0, BENILLA_PET_AUTOCAST_EMITTERS - 1 do
-                -- The four bones are the same lap, a quarter apart.
-                local phase = clock / BENILLA_PET_AUTOCAST_PERIOD + e / BENILLA_PET_AUTOCAST_EMITTERS
-                for k = 0, BENILLA_PET_AUTOCAST_TRAIL - 1 do
-                    -- Evenly spaced in AGE: this star was emitted `age` seconds ago, so it sits
-                    -- where the bone was then and has aged that far along the ramps.
-                    local age = (k / BENILLA_PET_AUTOCAST_TRAIL) * BENILLA_PET_AUTOCAST_LIFE
-                    local x, y = BenillaPetAutocast_Point(phase - age / BENILLA_PET_AUTOCAST_PERIOD)
-                    local life = age / BENILLA_PET_AUTOCAST_LIFE
-                    local size = BenillaPetAutocast_Ramp(BENILLA_PET_AUTOCAST_STAR, life)
-                    local c = BenillaPetAutocast_Ramp(BENILLA_PET_AUTOCAST_COLOR, life)
-                    local spark = button.sparks[n]
-                    n = n + 1
-                    spark:SetWidth(size)
-                    spark:SetHeight(size)
-                    spark:ClearAllPoints()
-                    spark:SetPoint("CENTER", button, "CENTER", x, y)
-                    spark:SetVertexColor(c[1], c[2], c[3])
-                    spark:SetAlpha(c[4])
-                    spark:Show()
-                end
+            if not button.sparks then BenillaPetAutocast_Create(button) end
+            local sparks = button.sparks
+            local lap = clock / BENILLA_PET_AUTOCAST_PERIOD
+            for i = 1, table.getn(sparks) do
+                local spark = sparks[i]
+                -- SetPoint on the SAME point replaces it, so no ClearAllPoints is needed.
+                local x, y = BenillaPetAutocast_Point(lap + spark.phase)
+                spark:SetPoint("CENTER", button, "CENTER", x, y)
+            end
+            if not button.sparksLit then
+                for i = 1, table.getn(sparks) do sparks[i]:Show() end
+                button.sparksLit = 1
             end
         end
 
@@ -233,7 +301,8 @@
             button.icon = getglobal(button:GetName() .. "Icon")
             button.cooldown = getglobal(button:GetName() .. "Cooldown")
             button.autocastable = getglobal(button:GetName() .. "AutoCastable")
-            BenillaPetAutocast_Create(button)
+            -- No sparkle pool here: BenillaPetAutocast_Update builds it on the first frame this
+            -- button actually autocasts, and most buttons never do.
         end
 
         -- Ref PetActionButton_OnClick. The CheckButton widget toggles itself BEFORE OnClick fires,
@@ -311,21 +380,49 @@
             PetActionBar_Update()
         end
 
-        -- Ref PetActionButton_OnEnter. A token slot has no spell to look up, so its tooltip is
-        -- built from the two strings the info call already returned (the ref's own `isToken` fork);
-        -- a spell slot goes through the engine's tooltip channel.
+        -- Ref PetActionButton_OnEnter (PetActionBarFrame.lua l.285-305). ONE predicate,
+        -- `isToken or UberTooltips == "0"`, decides two things at once:
+        --
+        --   * WHICH TOOLTIP. A token slot has no spell to look up, so its tooltip is built here out
+        --     of the two strings the info call already returned — and the BINDING (B230). Anything
+        --     else is a real pet SPELL and goes through the engine's tooltip channel, which appends
+        --     nothing. So the parenthesised key belongs to Attack/Follow/Stay and the reactions, and
+        --     NOT to the pet's own spells: putting it on every slot would be the wrong fix.
+        --   * WHERE IT SITS. Only the UberTooltips-off leg moves the tooltip beside the button
+        --     (ANCHOR_RIGHT); a token with the CVar on, and every spell, take the default corner.
+        --     The CVar registers "1" (cvars.rs carries the byte-read), so the off leg is the
+        --     player's to reach and not what ships.
+        --
+        -- The suffix is the ref's own expression and it is UNCONDITIONAL — no `if the key exists`
+        -- guard, unlike MicroButtonTooltipText's version of the same idea (MicroMenu.xml). That is
+        -- visible, not academic: GetBindingText answers "" for a nil key (UIParent.xml, transcribed
+        -- from the ref's own l.1819-1821), so 1.12 really does render "Attack ()" on a token slot
+        -- whose row is unbound. Stock, none are: BONUSACTIONBUTTON1-10 default to CTRL-1..CTRL-0
+        -- (bindings/commands.rs, byte-real from the install's bindings-cache.wtf).
+        --
+        -- Those same ten rows are the binding B218 found the CORNER hotkey glyph can never show
+        -- (the header's ref-divergence list has the `bindingSuffix == this:GetID()` string-vs-number
+        -- reason). Corner empty, tooltip parenthesised — both faithful, at once.
         function PetActionButton_OnEnter()
             local button = this
             if not GameTooltip then return end -- engine tests load file subsets
             if not button.tooltipName then return end
-            GameTooltip_SetDefaultAnchor(GameTooltip, button)
-            if button.isToken then
-                GameTooltip:SetText(button.tooltipName, 1.0, 1.0, 1.0)
+            local uber = GetCVar("UberTooltips")
+            if button.isToken or uber == "0" then
+                if uber == "0" then
+                    GameTooltip:SetOwner(button, "ANCHOR_RIGHT")
+                else
+                    GameTooltip_SetDefaultAnchor(GameTooltip, button)
+                end
+                local key = GetBindingText(GetBindingKey("BONUSACTIONBUTTON" .. button.index), "KEY_")
+                GameTooltip:SetText(button.tooltipName .. NORMAL_FONT_COLOR_CODE .. " (" .. key .. ")"
+                    .. FONT_COLOR_CODE_CLOSE, 1.0, 1.0, 1.0)
                 if button.tooltipSubtext then
                     GameTooltip:AddLine(button.tooltipSubtext, "", 0.5, 0.5, 0.5)
                 end
                 GameTooltip:Show()
             else
+                GameTooltip_SetDefaultAnchor(GameTooltip, button)
                 GameTooltip:SetPetAction(button.index)
             end
         end

crates/benilla-app/assets/ui/SpellBookFrame.xml

@@ -292,8 +292,8 @@
             -- The ref's `$parentAutoCast` is a <Model> playing UI-AutoCastButton.mdx; we have no
             -- Model widget, so the pet bar's byte-derived sparkle emitter stands in for it
             -- (decision 0982 — PetActionBar.xml owns the numbers and both surfaces share them).
-            -- Guarded because several engine test harnesses load this file without the pet bar.
-            if BenillaPetAutocast_Create then BenillaPetAutocast_Create(button) end
+            -- Nothing to build here: that emitter pools itself lazily, on the first frame a
+            -- button autocasts, which on this page is a pet-book spell and never a player one.
             BenillaSpellButton_UpdateButton(button)
         end
 

crates/benilla-app/assets/ui/StackSplit.xml

@@ -19,11 +19,15 @@
         for GossipFrame/QuestFrame/QuestLogFrame ("our engine ships no UIPanelButtonTemplate, so we
         draw it") — this is one more independent draw of the same real art.
 
-        NAMED DEFERRALS (both cited on the functions below, not just here):
-        - Typed-digit entry (`StackSplitFrame_OnChar`/`_OnKeyDown` in the ref): this engine has no
-          plain-frame OnChar/OnKeyDown keyboard capture, only EditBox focus (grepped; no such hook
-          exists anywhere in benilla-ui). The Left/Right spinner buttons are the only way to change
-          the split count.
+        TYPED-DIGIT ENTRY IS LIVE (decision 1319). It used to be this file's headline deferral
+        ("this engine has no plain-frame OnChar/OnKeyDown capture, only EditBox focus"), and the
+        engine now has the delivery walk that unblocks it (`script::keyboard`, off wow-re's
+        byte-verified `frame-key-script-delivery.md`). The handlers below are OUR OWN spinner
+        logic, written from the widget's behaviour, not a transcription of the reference's Lua
+        (their text is never copied into this repo). `enableKeyboard` on the frame is
+        what puts it in the walk's bucket at all.
+
+        NAMED DEFERRAL:
         - The ref's shift-click "insert the item link into the chat editbox" (StackSplitFrame isn't
           even opened in that branch — ContainerFrame.lua:568-569 diverts to
           `ChatFrameEditBox:Insert` instead of `OpenStackSplitFrame`) never triggers here: the
@@ -51,6 +55,10 @@
             frame.owner = owner
             owner.hasStackSplit = true
             frame.split = 1
+            -- `typed` distinguishes "1 because nothing has been typed yet" from "1 because the
+            -- player typed it": the first digit REPLACES the seed, later digits append. Cleared on
+            -- every open, so a dialog reopened after typing starts fresh.
+            frame.typed = nil
             StackSplitText:SetText(frame.split)
             StackSplitLeftButton:Disable()
             StackSplitRightButton:Enable()
@@ -92,26 +100,15 @@
             end
         end
 
+        -- The arrows go through the shared setter (below) rather than nudging the count and the
+        -- two buttons themselves: with typed entry live there are now three writers of
+        -- `frame.split`, and three copies of the enable/disable rule is how they drift apart.
         function BenillaStackSplitLeft_Click()
-            local frame = StackSplitFrame
-            if frame.split == 1 then return end
-            frame.split = frame.split - 1
-            StackSplitText:SetText(frame.split)
-            if frame.split == 1 then
-                StackSplitLeftButton:Disable()
-            end
-            StackSplitRightButton:Enable()
+            BenillaStackSplitSet(StackSplitFrame.split - 1)
         end
 
         function BenillaStackSplitRight_Click()
-            local frame = StackSplitFrame
-            if frame.split == frame.maxStack then return end
-            frame.split = frame.split + 1
-            StackSplitText:SetText(frame.split)
-            if frame.split == frame.maxStack then
-                StackSplitRightButton:Disable()
-            end
-            StackSplitLeftButton:Enable()
+            BenillaStackSplitSet(StackSplitFrame.split + 1)
         end
 
         function BenillaStackSplitOkay_Click()
@@ -126,6 +123,81 @@
             StackSplitFrame:Hide()
         end
 
+        -- The one place the count is written, so the buttons' enable state and the label can
+        -- never disagree with `frame.split` (the arrow handlers below call it too). Clamped to
+        -- [1, maxStack] because every caller can overshoot: typing 9 into a 5-stack, or the
+        -- arrows at their ends.
+        function BenillaStackSplitSet(n)
+            local frame = StackSplitFrame
+            if n < 1 then n = 1 end
+            if n > frame.maxStack then n = frame.maxStack end
+            frame.split = n
+            StackSplitText:SetText(n)
+            if n <= 1 then
+                StackSplitLeftButton:Disable()
+            else
+                StackSplitLeftButton:Enable()
+            end
+            if n >= frame.maxStack then
+                StackSplitRightButton:Disable()
+            else
+                StackSplitRightButton:Enable()
+            end
+        end
+
+        -- Typed digits (decision 1319). The engine hands OnChar the literal character, so this is
+        -- a digit test and an append: the first keystroke replaces the seeded 1, each later one
+        -- shifts a decimal place in. A digit that would overshoot the stack is CLAMPED rather than
+        -- rejected -- typing 9 against a 5-stack leaves 5, which is what the player meant -- and
+        -- clamping keeps `typed` set, so a following digit still appends against the clamp.
+        -- Anything that is not a digit is ignored here; the frame consumes the key regardless
+        -- (the engine's existence gate decides that, not this handler's return).
+        function StackSplitFrame_OnChar()
+            local c = arg1
+            if c < "0" or c > "9" then return end
+            local frame = StackSplitFrame
+            local digit = tonumber(c)
+            local n
+            if frame.typed then
+                n = frame.split * 10 + digit
+            else
+                n = digit
+            end
+            -- A leading 0 types nothing: the count can never be 0, so it stays 1 and unlatched,
+            -- and the next digit still reads as the first one.
+            if n < 1 then return end
+            frame.typed = true
+            BenillaStackSplitSet(n)
+        end
+
+        -- The editing/commit keys for typed entry. BACKSPACE drops the last digit (and unlatches
+        -- at the end, so the next digit starts over rather than appending to a stale 1); ENTER
+        -- commits exactly as Okay does and ESCAPE cancels, which is what a player expects of a
+        -- dialog they are typing into. The arrow keys drive the spinner. Every other key is
+        -- swallowed by the frame -- the reference's own consequence of having a key script bound
+        -- at all, and the reason this dialog must not stay open invisibly.
+        function StackSplitFrame_OnKeyDown()
+            local key = arg1
+            local frame = StackSplitFrame
+            if key == "BACKSPACE" then
+                local n = floor(frame.split / 10)
+                if n < 1 then
+                    frame.typed = nil
+                    BenillaStackSplitSet(1)
+                else
+                    BenillaStackSplitSet(n)
+                end
+            elseif key == "ENTER" then
+                BenillaStackSplitOkay_Click()
+            elseif key == "ESCAPE" then
+                BenillaStackSplitCancel_Click()
+            elseif key == "LEFT" then
+                BenillaStackSplitLeft_Click()
+            elseif key == "RIGHT" then
+                BenillaStackSplitRight_Click()
+            end
+        end
+
         -- OnHide (ref l.158-162): ANY hide — Okay, Cancel, or ToggleGameMenu's chain hook below —
         -- clears the owner's `hasStackSplit` flag (ref ItemButtonTemplate's OnHide, ContainerFrame.xml
         -- l.35-39, checks `this.hasStackSplit` to auto-close the spinner if the SLOT itself goes away
@@ -152,7 +224,14 @@
         <HighlightTexture file="Interface\Buttons\UI-Panel-Button-Highlight" alphaMode="ADD"><TexCoords left="0" right="0.625" top="0" bottom="0.6875"/></HighlightTexture>
     </Button>
 
-    <Frame name="StackSplitFrame" frameStrata="HIGH" hidden="true">
+    <!-- The frame attrs are the ref's own (l.4), and `toplevel` is load-bearing: Bagnon's bag
+         windows sit in the SAME HIGH stratum, and without the bit the dialog's Show raises
+         nothing — it loses the level tie to the later-shown bag window and the plate draws
+         UNDERNEATH the slots while the child buttons (level+1) poke through (the director's
+         08-14 Bagnon report; decision 1318). With the bit, Show runs the verified raise (compact,
+         then top-occupied+1). `enableKeyboard` alone is left off: its only consumer is the
+         typed-digit deferral in the file header. -->
+    <Frame name="StackSplitFrame" frameStrata="HIGH" parent="UIParent" toplevel="true" enableMouse="true" enableKeyboard="true" hidden="true">
         <!-- ref StackSplitFrame.xml l.4-7: 172×96. The CENTER anchor here is only the pre-first-open
              placeholder — OpenStackSplitFrame always repositions before the first Show. -->
         <Size><AbsDimension x="172" y="96"/></Size>
@@ -228,6 +307,8 @@
         </Frames>
         <Scripts>
             <OnHide>StackSplitFrame_OnHide()</OnHide>
+            <OnChar>StackSplitFrame_OnChar()</OnChar>
+            <OnKeyDown>StackSplitFrame_OnKeyDown()</OnKeyDown>
         </Scripts>
     </Frame>
 </Ui>

crates/benilla-app/assets/ui/StanceBar.xml

@@ -62,11 +62,16 @@
         -- Hover: the form's spell tooltip through the engine channel (ref ShapeshiftBar's
         -- GameTooltip:SetShapeshift; decision 0274 P2). Anchor: the ref's ShapeshiftButton
         -- OnEnter (BonusActionBarFrame.xml l.40-45) takes the default corner when UberTooltips
-        -- is "1" — its byte-verified stock default (ActionBar.xml has the read) — so the branch
-        -- collapses to the default corner here too.
+        -- is "1" — its byte-verified stock default — and sits beside the button otherwise. Like
+        -- the action bar's, that fork was collapsed to the "1" leg until B230 registered the CVar
+        -- (cvars.rs carries the byte-read); the stock look is unchanged, the off leg is reachable.
         function BenillaShapeshiftButton_OnEnter(button)
             if not GameTooltip then return end -- engine tests load file subsets
-            GameTooltip_SetDefaultAnchor(GameTooltip, button)
+            if GetCVar("UberTooltips") == "0" then
+                GameTooltip:SetOwner(button, "ANCHOR_RIGHT")
+            else
+                GameTooltip_SetDefaultAnchor(GameTooltip, button)
+            end
             GameTooltip:SetShapeshift(button.index)
         end
 

crates/benilla-app/assets/ui/UIDropDownMenu.xml

@@ -1119,7 +1119,8 @@
            it is ONE library file replicated, and only that one of its two sites is unguarded (the
            other is `info.textHeight or UIDROPDOWNMENU_DEFAULT_TEXT_HEIGHT or 10`). Writing a
            literal here would have been a silent fidelity divergence, not a fix. -->
-    <Button name="DropDownList1" inherits="UIDropDownListTemplate" id="1" parent="UIParent">
+    <!-- `toplevel` is authored on the ref's INSTANCES, not its template (decision 1318). -->
+    <Button name="DropDownList1" inherits="UIDropDownListTemplate" id="1" toplevel="true" parent="UIParent">
         <Scripts>
             <OnLoad>
                 local fontName, fontHeight, fontFlags = getglobal("DropDownList1Button1NormalText"):GetFont();
@@ -1131,7 +1132,7 @@
             </OnHide>
         </Scripts>
     </Button>
-    <Button name="DropDownList2" inherits="UIDropDownListTemplate" id="2" parent="UIParent">
+    <Button name="DropDownList2" inherits="UIDropDownListTemplate" id="2" toplevel="true" parent="UIParent">
         <Scripts>
             <OnHide>
                 CloseDropDownMenus(3);

crates/benilla-app/assets/ui/UiPanels.xml

@@ -1712,7 +1712,9 @@
          the click's hide so a second instance must be free; the ref's own StaticPopup2 case).
          Extend toward 4 the day a third coexists. StaticPopup2 seats below 1 (the ref's
          StaticPopup_SetUpPosition stack, simplified to a static anchor). -->
-    <Frame name="StaticPopupTemplate" virtual="true" enableMouse="true" frameStrata="DIALOG" hidden="true">
+    <!-- `toplevel` is the ref's (StaticPopup.xml l.3): a confirm popup must come to the front over
+         whatever DIALOG-strata window raised it (decision 1318). -->
+    <Frame name="StaticPopupTemplate" virtual="true" toplevel="true" enableMouse="true" frameStrata="DIALOG" hidden="true">
         <Size><AbsDimension x="320" y="72"/></Size>
         <Backdrop bgFile="Interface\DialogFrame\UI-DialogBox-Background" edgeFile="Interface\DialogFrame\UI-DialogBox-Border" tile="true">
             <BackgroundInsets><AbsInset left="11" right="12" top="12" bottom="11"/></BackgroundInsets>

crates/benilla-app/assets/ui/benilla.toc

@@ -106,8 +106,10 @@ StanceBar.xml
 # i.e. until the server sends `SMSG_PET_SPELLS` for a live pet or charm — so a class that
 # never has one never sees it. Right after the stance bar, whose row it shares (either
 # one showing raises UIParent.xml's managed "pet" delta), and after Cooldown.xml
-# (CooldownFrame_SetTimer), ActionBar.xml (the `MainMenuBar` anchor target) and
-# GameTooltip.xml (its hover). It is also the pet ARC's file, so decision 1066's three
+# (CooldownFrame_SetTimer), ActionBar.xml (the `MainMenuBar` anchor target),
+# GameTooltip.xml (its hover), Fonts.xml (NORMAL_FONT_COLOR_CODE, which wraps the binding
+# a token slot's tooltip appends) and UIParent.xml (GetBindingText, which renders it).
+# It is also the pet ARC's file, so decision 1066's three
 # right-click-menu dialogs (ABANDON_PET / RENAME_PET / PETRENAMECONFIRM) register here —
 # which adds UiPanels.xml to its dependency set, already far above.
 PetActionBar.xml

crates/benilla-app/src/cvars.rs

@@ -111,6 +111,16 @@ pub(crate) const REGISTERED: &[(&str, &str)] = &[
     // out-of-box look is hover-only numerals. That default is BEHAVIOUR-derived, not byte-read —
     // 1.12's registrar value for this var is not pinned in wow-re yet.
     ("statusBarText", "0"),
+    // Enhanced Tooltips (B230): 1.12's `UberTooltips`, the *Enhanced Tooltips* checkbox
+    // (`UIOptionsFrame.lua:15`, `USE_UBERTOOLTIPS`). **No host knob** — its consumers are Lua, and
+    // there are three: PetActionBar.xml forks the whole tooltip on it (a token's own text with the
+    // binding appended, vs the engine's pet-spell channel), ActionBar.xml and StanceBar.xml fork
+    // their anchor. Registered "1" — byte-read, not behaviour-derived: WoW.exe `0x48fdd9`, default
+    // string `0x82e748`, with the sibling rows `BlockTrades`→"0" and `UnitNameRenderMode`→"2"
+    // confirming the layout. Those three Lua sites each carried the reference's fork in prose and
+    // then collapsed it to this default, on the stated premise that benilla shipped no CVar state
+    // for anything to move. That premise expired with 0954, and this row is what un-collapses them.
+    ("UberTooltips", "1"),
     // The two chat-bubble switches (1139): 1.12's own registrar CVars over the bubble gate,
     // which held them as `const bool` from 0598 until this window had a page for them.
     // `ChatBubbles` is the reference's registered "1"; `ChatBubblesParty` is ON where the binary
@@ -236,9 +246,9 @@ fn apply_to_knobs(name: &str, value: &str, knobs: &mut Knobs) -> bool {
         "unitnameplayer" => knobs.names.player = v != 0.0,
         "unitnamenpc" => knobs.names.npc = v != 0.0,
         "unitnameown" => knobs.names.own = v != 0.0,
-        // A CVar with no HOST knob, because its consumer is Lua (1140). Known — so the caller
-        // dirties the config and the value persists — with nothing to apply on this side.
-        "statusbartext" => {}
+        // Two CVars with no HOST knob, because their consumers are Lua (1140, B230). Known — so
+        // the caller dirties the config and the value persists — with nothing to apply this side.
+        "statusbartext" | "ubertooltips" => {}
         // The two bubble switches (1139) — flags, like every other pair here.
         "chatbubbles" => knobs.bubbles.all = v != 0.0,
         "chatbubblesparty" => knobs.bubbles.party = v != 0.0,

crates/benilla-app/src/net/apply.rs

@@ -714,10 +714,12 @@ pub(super) fn apply_net_updates(
                 spell_id,
                 success,
                 reason,
+                arg,
             } => cast_result(
                 spell_id,
                 success,
                 reason,
+                arg,
                 &mut commands,
                 &self_guid,
                 &index,

crates/benilla-app/src/net/apply/pet.rs

@@ -160,7 +160,7 @@ fn pet_feedback_key(reason: u8) -> Option<&'static str> {
 pub(super) fn pet_cast_failed(spell_id: u32, reason: Option<u8>, errors: &mut CastErrors) {
     debug!("net: pet cast failed — spell {spell_id} reason {reason:?}");
     if let Some(reason) = reason {
-        errors.0.push((spell_id, reason));
+        errors.push_local(spell_id, reason);
     }
 }
 

crates/benilla-app/src/net/apply/spells.rs

@@ -77,6 +77,7 @@ pub(super) fn cast_result(
     spell_id: u32,
     success: bool,
     reason: Option<u8>,
+    arg: Option<u32>,
     commands: &mut Commands,
     self_guid: &SelfGuid,
     index: &GuidIndex,
@@ -153,7 +154,11 @@ pub(super) fn cast_result(
         // still shows "out of range"/"line of sight" even though no bar is (or should be) up.
         if let Some(reason) = reason {
             if !dont_report {
-                cast_errors.0.push((spell_id, reason));
+                cast_errors.0.push(crate::ui_action::CastFail {
+                    spell_id,
+                    reason,
+                    arg,
+                });
             }
         }
         // The cast bar's red "Failed" — only the showing cast's own failure turns it red.
@@ -1336,6 +1341,7 @@ mod tests {
                             75,
                             false,
                             Some(reason),
+                            None,
                             &mut commands,
                             &self_guid,
                             &index,

crates/benilla-app/src/target/click.rs

@@ -487,7 +487,7 @@ enum KeyFact {
 fn resolve_go_action(
     guid: u64,
     inputs: &mut GoLockInputs,
-    known: &std::collections::HashSet<u32>,
+    known: &std::collections::BTreeSet<u32>,
     go: Option<(&ObjectStore, u32)>,
     me_store: Option<&ObjectStore>,
     net: &NetCommands,
@@ -516,7 +516,14 @@ fn resolve_go_action(
     );
     let key_entry = match outcome {
         super::lock::LockOutcome::Unlocked => return GoAction::Use,
-        super::lock::LockOutcome::OpenBySpell(spell_id) => return GoAction::OpenLock(spell_id),
+        super::lock::LockOutcome::OpenBySpell(spell_id) => {
+            // The retest instrument for B247 (decision 1312): which of the player's openers this
+            // lock resolved to, by id. The spell's Spell.dbc name is what the cast bar prints, so
+            // reading the id off a probe run is how "the bar says Opening - No Text" becomes a
+            // machine-checkable fact instead of a screenshot.
+            debug!("target: lock {} → open by spell {spell_id}", tmpl.lock_id);
+            return GoAction::OpenLock(spell_id);
+        }
         super::lock::LockOutcome::OpenByKey(entry) => entry,
         super::lock::LockOutcome::Unmet => {
             // Unopenable — §8.8's routing, which keys off Lock.dbc **slot 0** regardless of which

crates/benilla-app/src/target/lock.rs

@@ -27,7 +27,7 @@
 //! the skill block. That is not an approximation on the client's part: the DBC encodes the same
 //! number (Pick Lock at 60 → 300, a rogue's cap; Mining at 60 → 300).
 
-use std::collections::HashSet;
+use std::collections::BTreeSet;
 
 use benilla_formats::{LockSlot, LOCK_KEY_ITEM, LOCK_KEY_SKILL};
 use bevy::prelude::*;
@@ -129,13 +129,20 @@ pub(crate) const GO_FLAG_LOCKED: u32 = 0x2;
 /// `matched_spell` is the out-param the toast routing needs; it is written for a LockType match
 /// even when the value test then fails.
 ///
-/// One deliberate difference: the reference walks its known-spell **array** in index order, while
-/// `known` is a set. That only decides *which* of several equally-matching spells lands in
-/// `matched_spell` (a miner knows four Mining ranks, all `LockType 3`), never the outcome — the
-/// scan keeps going past an insufficient one, so the best-valued opener still wins.
+/// **The scan order is part of the answer, not an implementation detail** (decision 1312). The
+/// reference walks its known-spell **array** in index order and *returns on the first sufficient
+/// match*, so when two known spells both match the LockType and both clear the requirement, the
+/// order decides which one is cast — and that spell's name is what the cast bar reads. This was
+/// documented here for a long time as a harmless difference ("it only decides which spell lands in
+/// `matched_spell`, never the outcome"), which is false: every character knows both 6478 "Opening"
+/// and 22810 **"Opening - No Text"** — both `SPELL_EFFECT_OPEN_LOCK` on `LockType 13` (Open
+/// Kneeling), both trivially sufficient against the `Skill == 0` slots the ground containers carry
+/// — so iterating a `HashSet` put Blizzard's placeholder name on the cast bar at the hash's whim
+/// (B247). `known` is a [`BTreeSet`] for exactly that reason: ascending spell id is the reference
+/// array's own order after login, the server building `SMSG_INITIAL_SPELLS` out of a `std::map`.
 pub(crate) fn resolve_lock(
     slots: &[LockSlot],
-    known: &HashSet<u32>,
+    known: &BTreeSet<u32>,
     spells: Option<&crate::ui_action::Spells>,
     me: Option<&ObjectStore>,
     items: &crate::items::Items,
@@ -375,7 +382,7 @@ mod tests {
         let mut matched = None;
         let out = resolve_lock(
             &slots,
-            &HashSet::new(),
+            &BTreeSet::new(),
             None,
             None,
             &items,
@@ -461,7 +468,7 @@ mod tests {
         assert_eq!(
             resolve_lock(
                 &scholomance,
-                &HashSet::from([6247]),
+                &BTreeSet::from([6247]),
                 Some(&spells),
                 None,
                 &items,
@@ -480,7 +487,7 @@ mod tests {
         assert_eq!(
             resolve_lock(
                 &scholomance,
-                &HashSet::from([6247]),
+                &BTreeSet::from([6247]),
                 Some(&spells),
                 None,
                 &items,
@@ -500,7 +507,7 @@ mod tests {
         assert_eq!(
             resolve_lock(
                 &scholomance,
-                &HashSet::from([1804]),
+                &BTreeSet::from([1804]),
                 Some(&spells),
                 None,
                 &items,
@@ -528,7 +535,7 @@ mod tests {
         assert_eq!(
             resolve_lock(
                 &searing_gorge,
-                &HashSet::from([6247]),
+                &BTreeSet::from([6247]),
                 Some(&spells),
                 None,
                 &items,
@@ -559,7 +566,7 @@ mod tests {
         assert_eq!(
             resolve_lock(
                 &vein,
-                &HashSet::from([2575]),
+                &BTreeSet::from([2575]),
                 Some(&with_mining),
                 None,
                 &items,
@@ -574,6 +581,83 @@ mod tests {
         );
     }
 
+    /// **B247** (decision 1312): two known openers on one LockType, both sufficient — the LOWEST
+    /// spell id wins, deterministically, because that is the reference array's order and the scan
+    /// returns on its first sufficient match.
+    ///
+    /// The real values: lock 43 (Hyacinth Mushroom and every other ground container) carries one
+    /// SKILL slot, `LockType 13` "Open Kneeling", `Skill 0`, `Action 0`. Three shipped spells carry
+    /// `SPELL_EFFECT_OPEN_LOCK` on type 13 — 6478 "Opening", 17667 "Light On Fire", and 22810
+    /// **"Opening - No Text"** — and `playercreateinfo_spell` grants 6478 *and* 22810 to every
+    /// race/class. Whichever this scan returns is the spell the click casts, so its Spell.dbc name
+    /// is what the cast bar prints: with a `HashSet` here the bar read Blizzard's placeholder about
+    /// half the time.
+    #[test]
+    fn two_sufficient_openers_pick_the_lower_spell_id() {
+        use benilla_formats::{OpenLock, SpellCatalog, SpellDisplay};
+
+        let mut mushroom = [LockSlot::default(); 8];
+        mushroom[0] = skill_slot(13, 0, 0);
+
+        // Both "Opening" spells: OPEN_LOCK on type 13, flat value 100 (base 99 + dice 1), which
+        // clears the slot's `Skill 0`. Identical in every input the resolver reads — the pick is
+        // decided by visit order and nothing else, which is the whole point.
+        let kneeling_opener = || SpellDisplay {
+            open_lock: Some(OpenLock {
+                lock_type: 13,
+                effect: 0,
+            }),
+            effect_base_points: [99, 0, 0],
+            effect_base_dice: [1, 0, 0],
+            ..Default::default()
+        };
+        let spells = crate::ui_action::Spells {
+            catalog: SpellCatalog::from_displays(
+                [(6478, kneeling_opener()), (22810, kneeling_opener())]
+                    .into_iter()
+                    .collect(),
+            ),
+            ..crate::ui_action::Spells::empty_for_tests()
+        };
+        let items = crate::items::Items::default();
+        let unlocked = GoFacts {
+            state: GO_STATE_READY,
+            flag_locked: false,
+            level: 0,
+        };
+
+        let mut matched = None;
+        assert_eq!(
+            resolve_lock(
+                &mushroom,
+                &BTreeSet::from([6478, 22810]),
+                Some(&spells),
+                None,
+                &items,
+                unlocked,
+                &mut matched,
+            ),
+            LockOutcome::OpenBySpell(6478),
+            "6478 \"Opening\", never 22810 \"Opening - No Text\""
+        );
+        assert_eq!(matched, Some(6478), "the toast's out-param agrees");
+
+        // Insertion order cannot change it — the set is ordered, so the scan is too.
+        let mut matched = None;
+        assert_eq!(
+            resolve_lock(
+                &mushroom,
+                &BTreeSet::from([22810, 6478]),
+                Some(&spells),
+                None,
+                &items,
+                unlocked,
+                &mut matched,
+            ),
+            LockOutcome::OpenBySpell(6478),
+        );
+    }
+
     /// An all-empty row is not a lock at all — `CMSG_GAMEOBJ_USE`.
     #[test]
     fn an_empty_row_is_unlocked() {
@@ -583,7 +667,7 @@ mod tests {
         assert_eq!(
             resolve_lock(
                 &slots,
-                &HashSet::new(),
+                &BTreeSet::new(),
                 None,
                 None,
                 &items,

crates/benilla-app/src/ui_action/cast_fail.rs

@@ -10,18 +10,33 @@
 //!    screen shows "Spell is not ready yet." while `SPELL_FAILED_NOT_READY` reads
 //!    "Not yet recovered", and "Not enough rage" for a rage spell's NO_POWER.
 //!
+//! 3. A per-reason **argument arm** (the second dispatch `0x6e1d8e`, 13 targets) fills that
+//!    message's own `%s`/`%d` from a DBC or item name before it is displayed. The two arms whose
+//!    tables benilla already loads live in [`FailArgs::fill`]; the drain fills three more before
+//!    calling here (below), and the rest strip.
+//!
 //! Strings are never hardcoded here: every message resolves from the VM's loaded
 //! `GlobalStrings.lua` by key, so localization rides for free. Suppression is faithful on
 //! both mechanisms: reason `0x17` (DONT_REPORT) never reaches display (control-flow), and a
 //! key absent from GlobalStrings displays as nothing (data — 0x08/0x21/0x75, happiness
-//! NO_POWER). Known approximations, each marked below: the DBC-name `%s` fills of the
-//! argument-formatted reasons are stripped instead of filled — except `0x78` TOTEMS and
-//! `0x5c` REAGENTS, whose item-name fills (and query-then-redisplay cache-miss behavior) the
-//! drain models before calling here ("Requires Mining Pick", decisions 0545 + 0552) — and
-//! `0x0a`'s item-spell leg (`ERR_INVALID_ITEM_TARGET`) is unmodeled — the drain does not
-//! know item-ness.
+//! NO_POWER).
+//!
+//! **The argument arms, and what is still approximate.** Filled: `0x5e` REQUIRES_SPELL_FOCUS and
+//! `0x5d` REQUIRES_AREA here (decision 1313 — `0x6e1f62`/`0x6e1fad`), and `0x78` TOTEMS / `0x5c`
+//! REAGENTS / `0x19`–`0x1b` EQUIPPED_ITEM_CLASS\* in the drain, which owns them because their
+//! fills need the item caches and the query-then-redisplay cache-miss behavior ("Requires Mining
+//! Pick", decisions 0545 + 0552). Still stripped rather than filled — each needs a DBC we do not
+//! load yet: `0x56` ONLY_SHAPESHIFT (form-name list), `0x8d` PREVENTED_BY_MECHANIC
+//! (`SpellMechanic.dbc`), `0x90` MIN_SKILL (`SkillLine.dbc`), `0x31` NEED_EXOTIC_AMMO and `0x84`
+//! PROSPECT_NEED_MORE. Stripping is a **deliberate divergence**, now byte-confirmed as one: on a
+//! bad id or an absent word the reference jumps to the default arm with the pointer still on the
+//! *unfilled* template, so it displays a literal `Requires %s` (wow-re §WIRE-ARGS C3 — the fill
+//! path's own buffer swap sits after the printf and is skipped). We show the bare stem instead:
+//! "Requires" reads as terse, "Requires %s" reads as broken software (§7 — judge by the result).
+//! `0x0a`'s item-spell leg (`ERR_INVALID_ITEM_TARGET`) is unmodeled — the drain does not know
+//! item-ness.
 
-use benilla_formats::SpellDisplay;
+use benilla_formats::{AreaTableCatalog, SpellDisplay, SpellFocusCatalog};
 
 /// Wire reason → its `SPELL_FAILED_*` GlobalStrings key (the `0x6e23e0` table, byte-exact).
 pub(super) const CAST_FAIL_KEYS: [&str; 146] = [
@@ -194,12 +209,64 @@ fn is_food(spell: Option<&SpellDisplay>) -> bool {
     spell.is_some_and(|d| matches!(d.category, 0xA | 0xB))
 }
 
+/// The argument arms' inputs: the wire's reason-specific word ([`super::CastFail::arg`]) and the
+/// DBC name tables the arms read. Both catalogs are `Option` because a client without game data
+/// has neither — the arm then declines and the template strips, exactly as an unmodeled arm does.
+#[derive(Default, Clone, Copy)]
+pub(super) struct FailArgs<'a> {
+    pub(super) arg: Option<u32>,
+    /// `SpellFocusObject.dbc` (`0xc0d800`) — the `0x5e` arm's names ("Anvil", "Forge", and the
+    /// Teldrassil moonwells the Crown of the Earth phials name).
+    pub(super) focus: Option<&'a SpellFocusCatalog>,
+    /// `AreaTable.dbc` (`0xc0e048`) — the `0x5d` arm's `AreaName`.
+    pub(super) areas: Option<&'a AreaTableCatalog>,
+}
+
+impl FailArgs<'_> {
+    /// The `%s` fill for the argument-formatted reasons this module owns, or `None` to leave the
+    /// template to the strip fallback.
+    ///
+    /// Both arms are byte-verified and §5 cross-checked (wow-re `cast-fail-strings.md`
+    /// §WIRE-ARGS): each reads the **wire's** first argument word — `[ebx+8]`, the handler's own
+    /// stack slot, never a re-read of `Spell.dbc` — indexes its DBC store, and `SStrPrintf`s the
+    /// reason's `SPELL_FAILED_*` template. The errorId stays the default `0x2c` (`"%s"`) for both,
+    /// so what the player reads IS the filled template. A single-`%s` template is what makes a
+    /// plain `replace` faithful here; the reference runs a real two-pass printf, which would
+    /// matter for a multi-specifier format.
+    ///
+    /// **`0x5e` REQUIRES_SPELL_FOCUS** (`0x6e1f62`) → `SpellFocusObject.dbc` (`0xc0d800`)
+    /// `Name_Lang` at `row + 0x4 + locale*4`, so `"Requires %s"` reads "Requires Starbreeze
+    /// Village Moonwell". The failing spell's own `RequiresSpellFocus` column holds the same
+    /// number vmangos copied onto the wire, so we fall back to it when the word is absent — a
+    /// benilla-side robustness margin, not a transcription: the client itself never reads
+    /// `SpellRec+0x3c` on this path.
+    ///
+    /// **`0x5d` REQUIRES_AREA** (`0x6e1fad`) → `AreaTable.dbc` (`0xc0e048`) `AreaName_Lang` at
+    /// `row + 0x2c + locale*4`, so `"You need to be in %s"` names the zone. This one has **no**
+    /// client-side stand-in: the server derives the id from its own `spell_area` rows and nothing
+    /// in `Spell.dbc` holds it, so an absent word means an unfilled message.
+    fn fill(&self, reason: u8, spell: Option<&SpellDisplay>) -> Option<String> {
+        match reason {
+            0x5D => self.areas?.name(self.arg?).map(str::to_string),
+            0x5E => {
+                let id = self
+                    .arg
+                    .filter(|&id| id != 0)
+                    .or_else(|| Some(spell?.requires_spell_focus).filter(|&id| id != 0))?;
+                self.focus?.name(id).map(str::to_string)
+            }
+            _ => None,
+        }
+    }
+}
+
 /// The displayed text for a failed cast — `None` = the reference shows nothing. `get` is the
 /// VM's GlobalStrings lookup (an absent or empty key resolves to `None`, the data-suppression
 /// face). Reasons beyond the table print their code — our debug affordance, not a ref string.
 pub(super) fn cast_fail_text(
     reason: u8,
     spell: Option<&SpellDisplay>,
+    args: FailArgs<'_>,
     get: &dyn Fn(&str) -> Option<String>,
 ) -> Option<String> {
     let get_display = |key: &str| get(key).filter(|s| !s.is_empty());
@@ -248,9 +315,12 @@ pub(super) fn cast_fail_text(
         return Some(format!("Spell failed ({reason:#04x})"));
     };
     let text = get_display(key)?;
-    // The argument-formatted reasons fill %s/%d from DBC names (0x6e1d8e's arm table) —
-    // unmodeled: strip the tokens so the stem reads clean ("Missing reagent: %s" →
-    // "Missing reagent"), never a raw % on screen.
+    // The argument arms (`0x6e1d8e`): fill the template's `%s` from the reason's own DBC.
+    if let Some(name) = args.fill(reason, spell) {
+        return Some(text.replace("%s", &name));
+    }
+    // An arm we don't model (or one whose lookup missed) — strip the tokens so the stem reads
+    // clean ("Missing reagent: %s" → "Missing reagent"), never a raw % on screen.
     if text.contains('%') {
         let stripped = text
             .replace("%s", "")
@@ -321,27 +391,36 @@ mod tests {
     fn overrides_replace_and_passthrough_reads() {
         let m = gs();
         let g = getter(&m);
-        assert_eq!(cast_fail_text(0x43, None, &g).unwrap(), "Out of ammo");
-        assert_eq!(cast_fail_text(0x59, None, &g).unwrap(), "Out of range.");
-        assert_eq!(cast_fail_text(0x76, None, &g).unwrap(), "Target too close");
         assert_eq!(
-            cast_fail_text(0x09, None, &g).unwrap(),
+            cast_fail_text(0x43, None, FailArgs::default(), &g).unwrap(),
+            "Out of ammo"
+        );
+        assert_eq!(
+            cast_fail_text(0x59, None, FailArgs::default(), &g).unwrap(),
+            "Out of range."
+        );
+        assert_eq!(
+            cast_fail_text(0x76, None, FailArgs::default(), &g).unwrap(),
+            "Target too close"
+        );
+        assert_eq!(
+            cast_fail_text(0x09, None, FailArgs::default(), &g).unwrap(),
             "You have no target."
         );
         // 0x3c: plain spell → spell cooldown; Attr&0x10 → ability; potion category → potion.
         let plain = spell(0, 0, 0);
         assert_eq!(
-            cast_fail_text(0x3C, Some(&plain), &g).unwrap(),
+            cast_fail_text(0x3C, Some(&plain), FailArgs::default(), &g).unwrap(),
             "Spell is not ready yet."
         );
         let ability = spell(1, 0, 0x10);
         assert_eq!(
-            cast_fail_text(0x3C, Some(&ability), &g).unwrap(),
+            cast_fail_text(0x3C, Some(&ability), FailArgs::default(), &g).unwrap(),
             "Ability is not ready yet."
         );
         let potion = spell(0, 4, 0);
         assert_eq!(
-            cast_fail_text(0x3C, Some(&potion), &g).unwrap(),
+            cast_fail_text(0x3C, Some(&potion), FailArgs::default(), &g).unwrap(),
             "Item is not ready yet."
         );
     }
@@ -354,12 +433,12 @@ mod tests {
         let g = getter(&m);
         let rage = spell(1, 0, 0);
         assert_eq!(
-            cast_fail_text(0x4D, Some(&rage), &g).unwrap(),
+            cast_fail_text(0x4D, Some(&rage), FailArgs::default(), &g).unwrap(),
             "Not enough rage"
         );
         let mana = spell(0, 0, 0);
         assert_eq!(
-            cast_fail_text(0x4D, Some(&mana), &g).unwrap(),
+            cast_fail_text(0x4D, Some(&mana), FailArgs::default(), &g).unwrap(),
             "Not enough mana"
         );
     }
@@ -371,13 +450,16 @@ mod tests {
     fn suppression_hex_fallback_and_template_strip() {
         let m = gs();
         let g = getter(&m);
-        assert_eq!(cast_fail_text(0x17, None, &g), None);
-        assert_eq!(cast_fail_text(0x08, None, &g), None);
+        assert_eq!(cast_fail_text(0x17, None, FailArgs::default(), &g), None);
+        assert_eq!(cast_fail_text(0x08, None, FailArgs::default(), &g), None);
         assert_eq!(
-            cast_fail_text(0x92, None, &g).unwrap(),
+            cast_fail_text(0x92, None, FailArgs::default(), &g).unwrap(),
             "Spell failed (0x92)"
         );
-        assert_eq!(cast_fail_text(0x5C, None, &g).unwrap(), "Missing reagent");
+        assert_eq!(
+            cast_fail_text(0x5C, None, FailArgs::default(), &g).unwrap(),
+            "Missing reagent"
+        );
     }
 
     /// The RUNTIME leg, end to end on the real data: the shipped `GlobalStrings.lua` executed
@@ -396,30 +478,93 @@ mod tests {
         s.run(&String::from_utf8_lossy(&src)).expect("runs clean");
         let g = |key: &str| s.lua().globals().get::<String>(key).ok();
 
-        assert_eq!(cast_fail_text(0x43, None, &g).unwrap(), "Out of ammo");
-        assert_eq!(cast_fail_text(0x59, None, &g).unwrap(), "Out of range.");
         assert_eq!(
-            cast_fail_text(0x3C, None, &g).unwrap(),
+            cast_fail_text(0x43, None, FailArgs::default(), &g).unwrap(),
+            "Out of ammo"
+        );
+        assert_eq!(
+            cast_fail_text(0x59, None, FailArgs::default(), &g).unwrap(),
+            "Out of range."
+        );
+        assert_eq!(
+            cast_fail_text(0x3C, None, FailArgs::default(), &g).unwrap(),
             "Spell is not ready yet."
         );
         let rage = spell(1, 0, 0);
         assert_eq!(
-            cast_fail_text(0x4D, Some(&rage), &g).unwrap(),
+            cast_fail_text(0x4D, Some(&rage), FailArgs::default(), &g).unwrap(),
             "Not enough rage"
         );
         // The environment gate's pair (decision 1056) — both are plain passthroughs, so what the
         // player reads IS the GlobalStrings value. A typo'd key here would degrade a real refusal
         // to a dead-looking button, which is what this test exists to catch.
         assert_eq!(
-            cast_fail_text(0x50, None, &g).unwrap(),
+            cast_fail_text(0x50, None, FailArgs::default(), &g).unwrap(),
             "Cannot use while swimming"
         );
         assert_eq!(
-            cast_fail_text(0x58, None, &g).unwrap(),
+            cast_fail_text(0x58, None, FailArgs::default(), &g).unwrap(),
             "Can only use while swimming"
         );
         // The data-suppression face on the real file: the absent keys show nothing.
-        assert_eq!(cast_fail_text(0x08, None, &g), None);
-        assert_eq!(cast_fail_text(0x21, None, &g), None);
+        assert_eq!(cast_fail_text(0x08, None, FailArgs::default(), &g), None);
+        assert_eq!(cast_fail_text(0x21, None, FailArgs::default(), &g), None);
+
+        // B255, end to end on the real data: the argument arms against the real DBCs and the real
+        // GlobalStrings templates. Without the fill these read as the bare stems "Requires" and
+        // "You need to be in" — which is exactly what shipped.
+        let focus =
+            benilla_formats::load_spell_focus_catalog(&mut chain).expect("SpellFocusObject");
+        let areas = benilla_formats::load_area_table_catalog(&mut chain).expect("AreaTable");
+        let args = |arg: u32| FailArgs {
+            arg: Some(arg),
+            focus: Some(&focus),
+            areas: Some(&areas),
+        };
+        // 0x5e REQUIRES_SPELL_FOCUS: the Crown of the Earth phials' own refusal. Focus 12 is the
+        // Starbreeze Village moonwell — using the Jade Phial at any *other* pool is the report.
+        assert_eq!(
+            cast_fail_text(0x5E, None, args(12), &g).unwrap(),
+            "Requires Starbreeze Village Moonwell"
+        );
+        assert_eq!(
+            cast_fail_text(0x5E, None, args(1), &g).unwrap(),
+            "Requires Anvil"
+        );
+        // 0x5d REQUIRES_AREA: area 1657 is Darnassus.
+        assert_eq!(
+            cast_fail_text(0x5D, None, args(1657), &g).unwrap(),
+            "You need to be in Darnassus"
+        );
+        // The fallbacks. An id the DBC doesn't name, and a wire word the server never sent, both
+        // decline the arm and fall through to the strip — never a raw `%s` on screen.
+        assert_eq!(
+            cast_fail_text(0x5E, None, args(999_999), &g).unwrap(),
+            "Requires"
+        );
+        assert_eq!(
+            cast_fail_text(0x5D, None, FailArgs::default(), &g).unwrap(),
+            "You need to be in"
+        );
+        // 0x5e alone has a client-side stand-in: the failing spell's own `RequiresSpellFocus`
+        // column is the very number the server copied onto the wire, so an absent word still
+        // fills. (Spell 4976 "Filling" — the Crystal Phial's — carries focus 11.)
+        let filling = SpellDisplay {
+            requires_spell_focus: 11,
+            ..Default::default()
+        };
+        assert_eq!(
+            cast_fail_text(
+                0x5E,
+                Some(&filling),
+                FailArgs {
+                    focus: Some(&focus),
+                    ..FailArgs::default()
+                },
+                &g
+            )
+            .unwrap(),
+            "Requires Shadowglen Moonwell"
+        );
     }
 }

crates/benilla-app/src/ui_action/cast_send.rs

@@ -290,7 +290,7 @@ fn send_spell_cast(
         // `HandleCastFailed`) — the inflight rec here is always an ordinary cast, so even an
         // on-next-swing press is refused while it holds.
         if pending.current(now) != Some(spell_id) {
-            cast_errors.0.push((spell_id, 0x61));
+            cast_errors.push_local(spell_id, 0x61);
         }
         debug!("ui_action: cast {spell_id} suppressed — a cast is already in flight");
         return;
@@ -343,7 +343,7 @@ fn send_spell_cast(
                 debug!(
                     "ui_action: cast {spell_id} refused locally — unbindable target ({reason:#x})"
                 );
-                cast_errors.0.push((spell_id, reason));
+                cast_errors.push_local(spell_id, reason);
                 return;
             }
         },
@@ -371,7 +371,7 @@ fn send_spell_cast(
                 dist_sq,
             ) {
                 debug!("ui_action: cast {spell_id} refused locally — range ({reason:#x})");
-                cast_errors.0.push((spell_id, reason));
+                cast_errors.push_local(spell_id, reason);
                 return;
             }
         }
@@ -417,7 +417,7 @@ fn send_spell_cast(
                 }
             }
         }
-        cast_errors.0.push((spell_id, commit.not_ready_reason()));
+        cast_errors.push_local(spell_id, commit.not_ready_reason());
         return;
     }
     // Rung 2 — the power gate (`0x60962c`, SPELL presses only: the item fork's clear-query jump
@@ -430,7 +430,7 @@ fn send_spell_cast(
         if let (Some(d), Some(store)) = (def, ctx.rel.self_store) {
             if !super::usable::can_afford(d, store) {
                 debug!("ui_action: cast {spell_id} refused locally — not enough power (0x4d)");
-                cast_errors.0.push((spell_id, 0x4d));
+                cast_errors.push_local(spell_id, 0x4d);
                 return;
             }
         }
@@ -451,7 +451,7 @@ fn send_spell_cast(
         def,
     ) {
         debug!("ui_action: cast {spell_id} refused locally — mounted (0x39)");
-        cast_errors.0.push((spell_id, 0x39));
+        cast_errors.push_local(spell_id, 0x39);
         return;
     }
     // The ENVIRONMENT leg of the SAME requirement validator (`0x609d39–0x609d7b`, between the
@@ -463,7 +463,7 @@ fn send_spell_cast(
     // (ledger B176) and lets you eat mid-swim (B155's eating half). The condition is [`state`]'s.
     if let Some(reason) = state::cast_water_refusal(ctx.self_move_flags, def) {
         debug!("ui_action: cast {spell_id} refused locally — water side ({reason:#x})");
-        cast_errors.0.push((spell_id, reason));
+        cast_errors.push_local(spell_id, reason);
         return;
     }
     // The moving leg of the SAME requirement validator (`0x609de3`, after the mounted/posture/
@@ -482,7 +482,7 @@ fn send_spell_cast(
         let cast_time_ms = spells.map_or(0, |s| s.cast_time_ms(d, caster_level));
         if state::cast_moving_refusal(ctx.self_move_flags, cast_time_ms, def) {
             debug!("ui_action: cast {spell_id} refused locally — moving (0x2e)");
-            cast_errors.0.push((spell_id, 0x2e));
+            cast_errors.push_local(spell_id, 0x2e);
             return;
         }
     }
@@ -507,7 +507,7 @@ fn send_spell_cast(
         if let Some(refusal) = d.form_refusal(form, form_is_stance) {
             let reason = refusal.reason();
             debug!("ui_action: cast {spell_id} refused locally — the form gate ({reason:#x})");
-            cast_errors.0.push((spell_id, reason));
+            cast_errors.push_local(spell_id, reason);
             return;
         }
     }
@@ -651,6 +651,7 @@ fn send_spell_cast(
 mod tests {
     use super::*;
     use crate::net::Reputations;
+    use crate::ui_action::CastFail;
     use bevy::ecs::system::RunSystemOnce;
     use crossbeam_channel::Receiver;
 
@@ -934,7 +935,7 @@ mod tests {
         assert!(rx.try_recv().is_err());
         assert_eq!(
             world.resource::<CastErrors>().0,
-            vec![(MOUNT, 0x61)],
+            vec![CastFail::local(MOUNT, 0x61)],
             "a different spell mid-cast is 6e4d97's \"Another action is in progress\""
         );
     }
@@ -959,7 +960,10 @@ mod tests {
 
         send(&mut world, HEARTHSTONE, HEARTH_COMMIT);
         assert!(rx.try_recv().is_err(), "an item on cooldown never sends");
-        assert_eq!(world.resource::<CastErrors>().0, vec![(HEARTHSTONE, 0x28)]);
+        assert_eq!(
+            world.resource::<CastErrors>().0,
+            vec![CastFail::local(HEARTHSTONE, 0x28)]
+        );
 
         // The byte law's other half (0948, correcting this test's pre-§5 shape): the record is
         // keyed (use-spell, item ENTRY), and a bare SPELL press queries (spell, 0) — the item
@@ -1075,7 +1079,10 @@ mod tests {
             })
             .expect("one-shot");
         assert!(rx.try_recv().is_err(), "an unaffordable press never sends");
-        assert_eq!(world.resource::<CastErrors>().0, vec![(HEARTHSTONE, 0x4d)]);
+        assert_eq!(
+            world.resource::<CastErrors>().0,
+            vec![CastFail::local(HEARTHSTONE, 0x4d)]
+        );
 
         world.resource_mut::<CastErrors>().0.clear();
         let base = ctx();
@@ -1122,7 +1129,7 @@ mod tests {
         assert!(rx.try_recv().is_err());
         assert_eq!(
             world.resource::<CastErrors>().0,
-            vec![(MOUNT, 0x61)],
+            vec![CastFail::local(MOUNT, 0x61)],
             "mid-cast outranks the cooldown rung (the ref's IsCasting precedes the validator)"
         );
     }

crates/benilla-app/src/ui_action/errors.rs

@@ -1,8 +1,8 @@
 //! The red error line's queues and resolvers — every route into `UIErrorsFrame`'s top line
 //! except the cast-fail two-layer display (its own module, [`super::cast_fail`]):
 //!
-//! - [`CastErrors`] — the wire `(spell_id, reason)` pairs (`SMSG_CAST_RESULT` + local cast
-//!   refusals), resolved by [`super::cast_fail`] at the drain.
+//! - [`CastErrors`] — the wire [`CastFail`]s (`SMSG_CAST_RESULT` + local cast refusals), resolved
+//!   by [`super::cast_fail`] at the drain.
 //! - [`MountErrors`] — the `SMSG_MOUNTRESULT`/`SMSG_DISMOUNTRESULT` code pairs, resolved by
 //!   [`mount_result_key`] (decision 0441 P2).
 //! - [`UiErrorKeys`] — client-LOCAL refusals straight by GlobalStrings key, the
@@ -21,12 +21,45 @@ use bevy::prelude::*;
 
 use crate::net::ObjectStore;
 
-/// Cast failures queued for the UI error line, as `(spell_id, reason)` — the wire pair from
-/// `SMSG_CAST_RESULT` and the local refusals alike. The spell id rides along because the
-/// display layer keys several messages on the failing spell's record ([`super::cast_fail`]:
-/// NO_POWER's power family, the 0x28/0x3c cooldown families).
+/// Cast failures queued for the UI error line — the wire triple from `SMSG_CAST_RESULT` and the
+/// local refusals alike. The spell id rides along because the display layer keys several messages
+/// on the failing spell's record ([`super::cast_fail`]: NO_POWER's power family, the 0x28/0x3c
+/// cooldown families).
 #[derive(Resource, Default)]
-pub(crate) struct CastErrors(pub Vec<(u32, u8)>);
+pub(crate) struct CastErrors(pub Vec<CastFail>);
+
+/// One queued cast failure: the failing spell, the wire reason, and the reason-specific argument
+/// word that fills the message's `%s` (the arm table at `0x6e1d8e` — a `SpellFocusObject.dbc` id
+/// for 0x5e, an `AreaTable.dbc` id for 0x5d).
+///
+/// `arg` is `None` for every **client-local** refusal, and that is faithful rather than a gap: the
+/// local refusals the reference raises itself go through `DisplayError` with no argText, and the
+/// two client-generated argument messages it *does* build (the lock toasts, MIN_SKILL) are
+/// [`UiError`]'s tenants, not this queue's.
+#[derive(Clone, Copy, Debug, PartialEq, Eq)]
+pub(crate) struct CastFail {
+    pub spell_id: u32,
+    pub reason: u8,
+    pub arg: Option<u32>,
+}
+
+impl CastFail {
+    /// A **client-local** refusal — no wire argument (see [`Self::arg`]).
+    pub(crate) const fn local(spell_id: u32, reason: u8) -> Self {
+        Self {
+            spell_id,
+            reason,
+            arg: None,
+        }
+    }
+}
+
+impl CastErrors {
+    /// Queue a [`CastFail::local`] refusal.
+    pub(crate) fn push_local(&mut self, spell_id: u32, reason: u8) {
+        self.0.push(CastFail::local(spell_id, reason));
+    }
+}
 
 /// (Dis)mount refusals queued for the UI error line, as the wire pair `(mount, code)` from
 /// `SMSG_MOUNTRESULT`/`SMSG_DISMOUNTRESULT` (decision 0441 P2) — resolved to text through the
@@ -200,7 +233,7 @@ pub(crate) fn reagent_totem_refusal(
         return false;
     };
     debug!("cast {spell_id} refused locally — missing totem/reagent ({reason:#04x})");
-    errors.0.push((spell_id, reason));
+    errors.push_local(spell_id, reason);
     true
 }
 
@@ -567,7 +600,7 @@ mod totem_reagent_tests {
             &items,
             &mut errors
         ));
-        assert_eq!(errors.0.as_slice(), &[(2575, 0x78)]);
+        assert_eq!(errors.0.as_slice(), &[CastFail::local(2575, 0x78)]);
 
         let mut reagents = [(0, 0); 8];
         reagents[0] = (17056, 1); // Slow Fall's Light Feather
@@ -580,7 +613,7 @@ mod totem_reagent_tests {
             &items,
             &mut errors
         ));
-        assert_eq!(errors.0.as_slice(), &[(130, 0x5c)]);
+        assert_eq!(errors.0.as_slice(), &[CastFail::local(130, 0x5c)]);
 
         let both = spell([2901, 0], reagents);
         let mut errors = CastErrors::default();
@@ -591,7 +624,7 @@ mod totem_reagent_tests {
             &items,
             &mut errors
         ));
-        assert_eq!(errors.0.as_slice(), &[(1, 0x78)]);
+        assert_eq!(errors.0.as_slice(), &[CastFail::local(1, 0x78)]);
 
         let plain = spell([0, 0], [(0, 0); 8]);
         let mut errors = CastErrors::default();

crates/benilla-app/src/ui_action/feed.rs

@@ -84,6 +84,10 @@ pub(super) fn feed_actions(
     mut items: ResMut<Items>,
     icons: Option<Res<ItemDisplays>>,
     sub_classes: Option<Res<crate::ui_items::ItemSubClasses>>,
+    // The two DBC name tables the cast-fail argument arms read (`FailArgs`): the crafting book's
+    // SpellFocusObject catalog and the map arc's AreaTable one, both already loaded.
+    spell_focus: Option<Res<crate::ui_tradeskill::SpellFocus>>,
+    areas: Option<Res<crate::area::AreaTableRes>>,
     commands: Res<NetCommands>,
     mut memory: Local<crate::ui_script::VmMemo<FeedMemory>>,
 ) {
@@ -102,12 +106,22 @@ pub(super) fn feed_actions(
     // item-cache miss the ref queries and shows nothing that frame, then its DBCACHECALLBACK
     // `0x6e29b0` REDISPLAYS when the answer lands — modeled by keeping the entry queued: the
     // ask-once query is away, and the frame the template answers, the fill succeeds and fires.
+    // The DBC-only arms (0x5d REQUIRES_AREA, 0x5e REQUIRES_SPELL_FOCUS) need no round trip and
+    // are filled inside [`cast_fail`] itself, off the wire's argument word.
     let self_store = self_q.iter().next();
-    let mut await_template: Vec<(u32, u8)> = Vec::new();
+    let mut await_template: Vec<crate::ui_action::CastFail> = Vec::new();
+    let fail_args = cast_fail::FailArgs {
+        arg: None,
+        focus: spell_focus.as_deref().map(|f| &f.catalog),
+        areas: areas.as_deref().map(|a| &a.0),
+    };
     let texts: Vec<String> = cast_errors
         .0
         .drain(..)
-        .filter_map(|(spell_id, reason)| {
+        .filter_map(|fail| {
+            let crate::ui_action::CastFail {
+                spell_id, reason, ..
+            } = fail;
             let d = spells.as_ref().and_then(|s| s.catalog.get(spell_id));
             let get = |key: &str| script.lua().globals().get::<String>(key).ok();
             // 0x19/0x1a/0x1b EQUIPPED_ITEM_CLASS* — the other argument-formatted family whose
@@ -154,7 +168,7 @@ pub(super) fn feed_actions(
                     // still pending → keep the entry queued for the redisplay.
                     None if items.template_answered_unknown(failing) => "UNKNOWN".to_string(),
                     None => {
-                        await_template.push((spell_id, reason));
+                        await_template.push(fail);
                         return None;
                     }
                 };
@@ -167,7 +181,25 @@ pub(super) fn feed_actions(
                     .filter(|s| !s.is_empty())
                     .map(|t| t.replace("%s", &name));
             }
-            cast_fail::cast_fail_text(reason, d, &get)
+            let text = cast_fail::cast_fail_text(
+                reason,
+                d,
+                cast_fail::FailArgs {
+                    arg: fail.arg,
+                    ..fail_args
+                },
+                &get,
+            );
+            // The retest instrument for this whole bug class (decision 1313). A red-line defect is
+            // reported as *seen* — B255 arrived as a screenshot of the word "Requires" — and until
+            // this line the only way to read what the client resolved was to look at the screen.
+            // Logging the reason, its wire argument and the resolved text makes an argument arm
+            // that silently declined (a missing word, an unnamed id) legible from a probe run.
+            debug!(
+                "ui_action: cast fail — spell {spell_id} reason {reason:#04x} arg {:?} → {:?}",
+                fail.arg, text
+            );
+            text
         })
         .collect();
     cast_errors.0.extend(await_template);

crates/benilla-app/src/ui_action/mod.rs

@@ -22,7 +22,7 @@
 //! [`errors`] (the red error line's two layers), [`usable`] (the castability walk the stance bar
 //! shares), [`weapon_icon`] (the auto-attack's borrowed weapon icon).
 
-use std::collections::{HashMap, HashSet};
+use std::collections::{BTreeSet, HashMap};
 
 use bevy::prelude::*;
 
@@ -65,8 +65,8 @@ pub(crate) struct CooldownEvents;
 // `cast_send`, so a second send path cannot be written by accident (decision 0914).
 pub(crate) use cast_send::{CastCommit, CastLadder};
 pub(crate) use errors::{
-    attack_actor_refusal, reagent_totem_refusal, ui_error_text, CastErrors, MountErrors, UiError,
-    UiErrorKeys,
+    attack_actor_refusal, reagent_totem_refusal, ui_error_text, CastErrors, CastFail, MountErrors,
+    UiError, UiErrorKeys,
 };
 // `pub(crate)`: the target chain registers the cursor pre-empt + the click commit, and the
 // spellbook/stance/craft drains thread the mode through the one cast-send path (decision 0792).
@@ -96,8 +96,17 @@ pub(crate) const SPELL_ATTACK: u32 = 6603;
 pub(crate) struct PlayerActions {
     /// Wire slot (0-based) → the slot's packed action. Lua action id = slot + 1.
     pub buttons: HashMap<u8, ActionButton>,
-    /// The spell book (`SMSG_INITIAL_SPELLS`), for gating/consumers to come.
-    pub spells: HashSet<u32>,
+    /// The spell book (`SMSG_INITIAL_SPELLS`).
+    ///
+    /// **Ordered, and that is load-bearing** (decision 1312). The reference keeps its known spells
+    /// in an ARRAY and the scans that hunt it — the GameObject lock resolver `0x5f83d0` above all —
+    /// stop at the first hit, so the visit order picks *which* of several equally-qualified spells
+    /// wins. A `HashSet` made that pick nondeterministic: every character knows both 6478
+    /// "Opening" and 22810 "Opening - No Text" (both `LockType 13`, both trivially sufficient), and
+    /// whichever the hash happened to reach first went on the cast bar (B247). Ascending spell id
+    /// is the array's own order after login — the server sends the initial batch out of a
+    /// `std::map`, so the wire arrives sorted.
+    pub spells: BTreeSet<u32>,
     /// Set on every book/bar arrival AND every local `action_sets` drain; cleared by the feed
     /// after re-resolving each slot's identity (icon/kind/action) and pushing. It is only ONE of
     /// the identity resolve's two triggers — a landed item template is the other (decision 0660,
@@ -226,8 +235,13 @@ fn track_learned_abilities(
     if !actions.is_changed() && !spells.is_changed() {
         return;
     }
-    let first_with = |effect: u32| {
-        actions.spells.iter().copied().find(|&id| {
+    // The **last** matching spell in ascending-id order, not the first: the reference latches at
+    // learn time and each new rank overwrites the global, so what it holds is the newest rank
+    // learned — and a rank chain is ascending by id (Skinning 8613 → 8617 → 8618 → 10768). Which
+    // end we take is only visible when a chain's ranks are known together, but "arbitrary" was
+    // never an answer: the set was a `HashSet` until 1312 and this picked at the hash's whim.
+    let last_with = |effect: u32| {
+        actions.spells.iter().copied().rfind(|&id| {
             spells
                 .catalog
                 .get(id)
@@ -235,9 +249,9 @@ fn track_learned_abilities(
         })
     };
     let (skinning, skin_player_corpse, feed_pet) = (
-        first_with(benilla_formats::SPELL_EFFECT_SKINNING),
-        first_with(SPELL_EFFECT_SKIN_PLAYER_CORPSE),
-        first_with(SPELL_EFFECT_FEED_PET),
+        last_with(benilla_formats::SPELL_EFFECT_SKINNING),
+        last_with(SPELL_EFFECT_SKIN_PLAYER_CORPSE),
+        last_with(SPELL_EFFECT_FEED_PET),
     );
     if (skinning, skin_player_corpse, feed_pet)
         != (

crates/benilla-app/src/ui_action/targeting/item.rs

@@ -260,7 +260,7 @@ pub(crate) fn commit_item_cast_on_pick(
                     "ui_action: cast {spell_id} refused at the item bind ({reason:#x}) — \
                      the cursor stays up"
                 );
-                ladder.cast_errors.0.push((spell_id, reason));
+                ladder.cast_errors.push_local(spell_id, reason);
             }
             ItemBind::ConfirmBind => {
                 debug!("ui_action: item bind confirm for {spell_id} on item {item_guid:#x}");

crates/benilla-app/src/ui_cast.rs

@@ -518,6 +518,30 @@ fn local_self_cancel(
     }
 }
 
+/// The casting bar's **label** for a spell — the `name` argument of `SPELLCAST_START` /
+/// `SPELLCAST_CHANNEL_START`, which the reference `CastingBarFrame` puts straight into
+/// `CastingBarText:SetText(...)`.
+///
+/// Almost always the spell's `Spell.dbc` name, with two ways to come back empty:
+///
+/// - **`AttributesEx3 & 0x4`** (`SPELL_ATTR_EX3_NO_CASTING_BAR_TEXT`,
+///   [`SpellDisplay::no_casting_bar_text`]) — the attribute that exists precisely so an internal
+///   spell with no player-facing name never puts one on the bar. Three shipped rows carry it, and
+///   one of them is *named after the bit*: **22810 "Opening - No Text"**, the opener the client
+///   casts at `LockType 13` ground containers. Printing the name unconditionally is what put that
+///   placeholder over a gathered Hyacinth Mushroom (B247, decision 1312).
+/// - **no catalog row** (an unknown id, or no client data) — the empty string this always
+///   returned in that case.
+///
+/// The bar still shows, fills and sweeps either way; only the text is suppressed.
+fn cast_bar_label(spells: Option<&crate::ui_action::Spells>, id: u32) -> String {
+    spells
+        .and_then(|s| s.catalog.get(id))
+        .filter(|d| !d.no_casting_bar_text())
+        .map(|d| d.name.clone())
+        .unwrap_or_default()
+}
+
 /// Drain the queue into FrameScript events — the reference `CastingBarFrame` contract
 /// (extracted 1.12 `CastingBarFrame.lua`): `SPELLCAST_START(name, ms)`,
 /// `SPELLCAST_CHANNEL_START(ms, name)`, `SPELLCAST_CHANNEL_UPDATE(remaining_ms)`, and the
@@ -561,13 +585,7 @@ fn feed_cast_bar(
     script.set_casting(
         auto_repeat.0.is_some() || inflight(&pending, &queued_melee, started, now).is_some(),
     );
-    let name = |id: u32| -> String {
-        spells
-            .as_ref()
-            .and_then(|s| s.catalog.get(id))
-            .map(|d| d.name.clone())
-            .unwrap_or_default()
-    };
+    let name = |id: u32| cast_bar_label(spells.as_deref(), id);
     for edge in feed.0.drain(..) {
         let (event, args): (&str, Vec<ScriptValue>) = match edge {
             CastBarEdge::Start {
@@ -636,6 +654,39 @@ mod tests {
 
     const FIREBALL: u32 = 133;
 
+    /// **B247** (decision 1312): the bar's label honours `SPELL_ATTR_EX3_NO_CASTING_BAR_TEXT`.
+    /// Spell 22810's Spell.dbc name — "Opening - No Text" — is Blizzard's own annotation of the
+    /// attribute, and it reached a player's screen because we printed the name regardless.
+    #[test]
+    fn the_no_casting_bar_text_attribute_blanks_the_label() {
+        use benilla_formats::{SpellCatalog, SpellDisplay};
+        let named = |name: &str, attributes_ex3: u32| SpellDisplay {
+            name: name.to_string(),
+            attributes_ex3,
+            ..Default::default()
+        };
+        let spells = crate::ui_action::Spells {
+            catalog: SpellCatalog::from_displays(
+                [
+                    (6478, named("Opening", 0)),
+                    (22810, named("Opening - No Text", 0x4)),
+                ]
+                .into_iter()
+                .collect(),
+            ),
+            ..crate::ui_action::Spells::empty_for_tests()
+        };
+        assert_eq!(cast_bar_label(Some(&spells), 6478), "Opening");
+        assert_eq!(
+            cast_bar_label(Some(&spells), 22810),
+            "",
+            "the bar fills and sweeps, but shows no name"
+        );
+        // The pre-existing empty cases are untouched: an unknown id, and no client data at all.
+        assert_eq!(cast_bar_label(Some(&spells), 133), "");
+        assert_eq!(cast_bar_label(None, 6478), "");
+    }
+
     #[test]
     fn a_fresh_guard_is_open() {
         let g = PendingCast::default();

crates/benilla-app/src/ui_script/bag_tests.rs

@@ -1006,10 +1006,9 @@ fn shift_click_on_a_stack_opens_the_split_frame() {
 /// B180 — the split dialog's parchment plate fills the whole 172×96 frame. The reference authors
 /// the plate 256×32 with NO anchors, a vestigial size the real client never renders (the TexCoords
 /// crop exactly 172×96 out of the 256×128 art — the frame's own size, a complete panel drawn 1:1
-/// over it). Our sized-anchor-less fallback draws CENTERED at the authored size instead, which
-/// squashed the dialog to a 32-tall bar: the count and arrows floated above it and Okay/Cancel
-/// straddled its bottom edge. The plate's authored size is dropped (StackSplit.xml says why); this
-/// pins the render so the vestigial size cannot come back.
+/// over it). 1308 first dodged this by dropping the size; 1310 then landed the byte-verified law
+/// (an anchor-less region gets an implicit SetAllPoints at creation, size unread under the two
+/// corners) and restored the ref's own text — this pins the render through the real mechanism.
 #[test]
 fn the_split_frame_plate_fills_the_dialog() {
     let mut s = UiScript::new().unwrap();
@@ -1057,6 +1056,132 @@ fn the_split_frame_plate_fills_the_dialog() {
     assert!(s.errors().is_empty(), "script errors: {:?}", s.errors());
 }
 
+/// B180's Bagnon follow-on (director, 08-14): the dialog opened UNDER an addon bag window —
+/// Bagnon's windows are `frameStrata="HIGH"`, the dialog's own stratum, and our re-expression had
+/// dropped the reference's `toplevel="true"`, so the dialog's Show raised nothing and lost the
+/// level tie to the later-shown window (its child buttons at level+1 poked through; the plate did
+/// not). With the ref's attrs restored, Show runs the verified raise (toplevel.rs: compact, then
+/// top-occupied-plus-one) and the whole dialog lands above. The synthetic window stands in for
+/// Bagnon: same stratum, shown after load, overlapping the dialog.
+#[test]
+fn the_split_frame_raises_over_a_same_stratum_window() {
+    let mut s = UiScript::new().unwrap();
+    s.set_screen_size(1024.0, 768.0);
+    let (x, y) = open_backpack_with_a_five_stack(&mut s);
+
+    s.run(
+        r#"
+        local w = CreateFrame("Frame", "FakeBagnon", UIParent)
+        w:SetFrameStrata("HIGH")
+        w:SetPoint("BOTTOMLEFT", 0, 0)
+        w:SetSize(1024, 768)
+        local bg = w:CreateTexture(nil, "BACKGROUND")
+        bg:SetTexture("Interface\\FakeBagnonBG")
+        bg:SetAllPoints()
+        w:Show()
+    "#,
+    )
+    .unwrap();
+    s.resolve();
+
+    s.set_modifiers(true, false, false);
+    s.mouse_button(x, y, "LeftButton", true);
+    s.mouse_button(x, y, "LeftButton", false);
+    s.set_modifiers(false, false, false);
+    s.resolve();
+
+    assert!(
+        s.eval::<bool>("return StackSplitFrame:IsShown()").unwrap(),
+        "the spinner opened"
+    );
+    let (dialog, bagnon) = s
+        .eval::<(i64, i64)>("return StackSplitFrame:GetFrameLevel(), FakeBagnon:GetFrameLevel()")
+        .unwrap();
+    assert!(
+        dialog > bagnon,
+        "Show must raise the toplevel dialog over the same-stratum window \
+         (dialog level {dialog}, window level {bagnon})"
+    );
+    // The symptom itself: the plate paints AFTER the window's background in draw order.
+    let order: Vec<String> = s
+        .extract()
+        .iter()
+        .filter_map(|q| match &q.content {
+            QuadContent::Texture { path: Some(p), .. }
+                if p.contains("FakeBagnonBG") || p.contains("UI-MoneyFrame") =>
+            {
+                Some(p.clone())
+            }
+            _ => None,
+        })
+        .collect();
+    assert_eq!(order.len(), 2, "both textures on screen: {order:?}");
+    assert!(
+        order[0].contains("FakeBagnonBG"),
+        "the plate must draw over the window, not under it: {order:?}"
+    );
+    assert!(s.errors().is_empty(), "script errors: {:?}", s.errors());
+}
+
+/// Typed-digit entry in the split spinner (decision 1319) — the director's ask, and the deferral
+/// this file's header carried since 0216. Pins the whole chain in one go: the dialog is in the
+/// keyboard walk (`enableKeyboard`), a digit reaches its `OnChar`, the first digit REPLACES the
+/// seeded 1 while later digits append, an over-max entry clamps instead of being rejected,
+/// BACKSPACE drops a digit, and ENTER commits exactly as Okay does.
+#[test]
+fn typing_a_number_into_the_split_spinner_sets_the_count() {
+    let mut s = UiScript::new().unwrap();
+    s.set_screen_size(1024.0, 768.0);
+    let (x, y) = open_backpack_with_a_five_stack(&mut s);
+
+    s.set_modifiers(true, false, false);
+    s.mouse_button(x, y, "LeftButton", true);
+    s.mouse_button(x, y, "LeftButton", false);
+    s.set_modifiers(false, false, false);
+    s.resolve();
+    assert_eq!(
+        s.eval::<i64>("return StackSplitFrame.split").unwrap(),
+        1,
+        "opens seeded at 1"
+    );
+
+    // A digit is consumed by the dialog — which is also what stops it firing action button 3.
+    assert!(s.char_input("3"), "the spinner consumed the digit");
+    assert_eq!(
+        s.eval::<i64>("return StackSplitFrame.split").unwrap(),
+        3,
+        "the first digit REPLACES the seed rather than appending to it"
+    );
+    assert_eq!(
+        s.eval::<String>("return StackSplitText:GetText()").unwrap(),
+        "3",
+        "and the label follows"
+    );
+
+    // A second digit appends, then clamps to the 5-stack rather than being thrown away.
+    assert!(s.char_input("7"));
+    assert_eq!(
+        s.eval::<i64>("return StackSplitFrame.split").unwrap(),
+        5,
+        "37 against a 5-stack clamps to 5"
+    );
+    // BACKSPACE drops a digit off the clamped value.
+    assert!(s.frame_key_input("BACKSPACE"), "the dialog took BACKSPACE");
+    assert_eq!(s.eval::<i64>("return StackSplitFrame.split").unwrap(), 1);
+
+    // Type a real value and commit with ENTER — the Okay path, so the carry is picked up.
+    assert!(s.char_input("4"));
+    assert_eq!(s.eval::<i64>("return StackSplitFrame.split").unwrap(), 4);
+    assert!(s.key_input("ENTER"), "ENTER is consumed by the dialog");
+    assert!(
+        !s.eval::<bool>("return StackSplitFrame:IsShown()").unwrap(),
+        "ENTER commits and closes, like Okay"
+    );
+    let held = s.cursor_item().expect("ENTER committed the split");
+    assert_eq!(held.count, Some(4), "the typed count is what got picked up");
+    assert!(s.errors().is_empty(), "script errors: {:?}", s.errors());
+}
+
 /// Okay in the split spinner only picks the split carry up (ref/cursor.rs `SplitContainerItem` —
 /// a pickup, not a self-contained move); a SUBSEQUENT placement is what actually queues the
 /// `ContainerMove` with `count: Some(n)`, drained the same way any other container move is.

crates/benilla-app/src/ui_script/input.rs

@@ -205,6 +205,30 @@ pub(super) fn feed_ui_input(
         // route law decides; unfocused they return unconsumed, and the camera/turn keys that
         // read arrows after UiInput see them exactly as before).
         let chord = keymap::chord(ev.key_code, mods, mac);
+        // A KEYBOARD FRAME gets these by NAME before their chord runs (decision 1319). They are
+        // the keys a focused EditBox receives as a semantic `EditAction` instead of a name, so
+        // they never reached `key_input` at all — and a dialog you type into needs its BACKSPACE.
+        // `frame_key_input` walks in the engine's own order and declines at a focused box, so the
+        // box's editing keys can never be stolen by a frame below it; a `true` here means a frame
+        // consumed the key, which suppresses BOTH the chord below and the key's binding (the
+        // reference's existence gate, wow-re `frame-key-script-delivery.md` §3).
+        let frame_named = match ev.key_code {
+            KeyCode::Backspace => Some("BACKSPACE"),
+            KeyCode::Delete => Some("DELETE"),
+            KeyCode::ArrowLeft => Some("LEFT"),
+            KeyCode::ArrowRight => Some("RIGHT"),
+            KeyCode::ArrowUp => Some("UP"),
+            KeyCode::ArrowDown => Some("DOWN"),
+            KeyCode::Home => Some("HOME"),
+            KeyCode::End => Some("END"),
+            _ => None,
+        };
+        if let Some(name) = frame_named {
+            if script.frame_key_input(name) {
+                capture.0 = true; // consumed by a frame: its binding must not also fire
+                continue;
+            }
+        }
         if let Some(name) = named {
             // The three box-event keys. An unconsumed press is not acted on here: every GAME
             // action of a key — ESCAPE's close/cancel ladder (TOGGLEGAMEMENU), TAB's targeting,
@@ -212,7 +236,12 @@ pub(super) fn feed_ui_input(
             // dispatch (decision 0997, `crate::bindings`), which runs right after this pass and
             // reads the capture gate written above, so a key a focused box consumed this frame
             // never also fires its binding — the real client's ESC precedence, table-wide.
-            script.key_input(name);
+            // Consumption suppresses the binding — a keyboard FRAME that took this key counts
+            // exactly as a focused box does (decision 1319; `capture.0` above already covers the
+            // box case, and this adds the frame one).
+            if script.key_input(name) {
+                capture.0 = true;
+            }
         } else if let Some(chord) = chord {
             match chord {
                 keymap::Chord::Edit(action) => {
@@ -244,7 +273,12 @@ pub(super) fn feed_ui_input(
             // plane European layouts type real text with (and macOS Option+letter comes through
             // with only `alt`, composing its special characters).
             if let Some(text) = &ev.text {
-                script.char_input(text);
+                // Same suppression as the named keys, and this is the case that bites: the number
+                // row IS a binding (the action buttons), so a digit typed into a keyboard frame —
+                // the stack-split spinner — must not also fire action button 3 (decision 1319).
+                if script.char_input(text) {
+                    capture.0 = true;
+                }
             }
         }
     }

crates/benilla-app/src/ui_script/pet_bar_tests.rs

@@ -184,10 +184,23 @@ fn the_shipped_pet_bar_drives_end_to_end() {
         1,
         "only Claw can autocast"
     );
+    // 4 emitters x the DERIVED trail length, on the one autocasting slot — and on that slot only,
+    // which is the lazy pool's own pin: the other nine buttons have no sparks at all, where they
+    // used to carry a full hidden set each.
+    let trail: usize = s.eval("return BENILLA_PET_AUTOCAST_TRAIL").unwrap();
+    assert_eq!(
+        trail, 23,
+        "the size ramp's own spacing (see PetActionBar.xml's header) comes out at 23 samples"
+    );
     assert_eq!(
         textures(&quads, "Interface\\Buttons\\GlowStar"),
-        32,
-        "4 emitters x 8 trail stars, on the one autocasting slot"
+        4 * trail,
+        "4 emitters x the trail, on the one autocasting slot"
+    );
+    assert!(
+        s.eval::<bool>("return PetActionButton1.sparks == nil")
+            .unwrap(),
+        "a button that never autocasts never pools a spark"
     );
 
     // Geometry, quoted from the ref: the bar's TOPLEFT is MainMenuBar's BOTTOMLEFT +(36,97),
@@ -631,3 +644,282 @@ fn the_keybind_pair_pushes_and_casts_without_the_clicks_forks() {
     );
     assert!(s.errors().is_empty(), "script errors: {:?}", s.errors());
 }
+
+/// **B228 — the autocast trail's motion, MEASURED off the shipped handler.**
+///
+/// nazriel reads the shine as "a bit too fast". The health pass answered that from the constants —
+/// the .m2 authors a 2.000 s loop and `BENILLA_PET_AUTOCAST_PERIOD` is 2.0 — which is a read, not a
+/// measurement: it says nothing about whether the OnUpdate wiring turns wall-clock seconds into
+/// that period, and an `elapsed` in the wrong unit would be invisible to it. So this drives the
+/// real `<OnUpdate>` with real seconds and reads the head spark's own SetPoint back.
+///
+/// What is pinned, all of it off the file (`m2bones` on `Interface\Buttons\UI-AutoCastButton.m2`):
+/// a 28.8 px square (0.02 units × 1200 px/unit × the ref Model's 1.2), one lap per 2.000 s, a
+/// corner every 0.500 s, LINEAR between them (`interp == 1` at the bone record's translation
+/// M2Track — so the half-edge sample sits at the half-time, no easing), and the four emitters one
+/// quarter-lap apart. If the reported overspeed is in the period, it fails here; if this holds and
+/// the report stands, the period is exonerated and the gap is in the trail's SHAPE.
+#[test]
+fn the_autocast_trail_laps_the_rim_once_every_two_seconds() {
+    let mut s = UiScript::new().unwrap();
+    s.set_screen_size(1024.0, 768.0);
+    load_pet_bar(&s);
+    declare_token_strings(&s);
+    s.set_pet_actions(true, true, true, hunter_slots());
+    s.fire_event("PET_BAR_UPDATE", vec![]);
+    s.resolve();
+
+    // Slot 4 is Claw, the one slot in the fixture with autocast RUNNING. `sparks[n]` is the pool
+    // `BenillaPetAutocast_Create` built: emitter-major, so emitter e's HEAD (age 0, the bone's own
+    // position) is index e·TRAIL + 1.
+    let head = |s: &UiScript, emitter: usize| -> (f32, f32) {
+        let expr = |axis: usize| {
+            format!(
+                "local n = {emitter} * BENILLA_PET_AUTOCAST_TRAIL + 1 \
+                 local a, b, c, x, y = PetActionButton4.sparks[n]:GetPoint() \
+                 return ({axis} == 0) and x or y"
+            )
+        };
+        (
+            s.eval::<f32>(&expr(0)).unwrap(),
+            s.eval::<f32>(&expr(1)).unwrap(),
+        )
+    };
+    let near = |got: (f32, f32), want: (f32, f32), what: &str| {
+        assert!(
+            (got.0 - want.0).abs() < 0.01 && (got.1 - want.1).abs() < 0.01,
+            "{what}: at {got:?}, expected {want:?}"
+        );
+    };
+
+    // 0.02 units × 1200 px/unit × 1.2 = 28.8 px across, so the corners sit ±14.4 from centre.
+    const H: f32 = 14.4;
+    let (bl, tl, tr, br) = ((-H, -H), (-H, H), (H, H), (H, -H));
+
+    // ── The clock WIRING: four half-second ticks of the shipped OnUpdate, one corner each. This
+    // is the half a constants read cannot reach — `arg1` really is seconds, and the bar's clock
+    // advances 1:1 with it.
+    let corner = |s: &mut UiScript, dt: f32, want: (f32, f32), what: &str| {
+        s.tick(dt);
+        near(head(s, 0), want, what);
+    };
+    corner(
+        &mut s,
+        0.5,
+        tl,
+        "0.5 s in: one edge, at the top-left corner",
+    );
+    corner(&mut s, 0.5, tr, "1.0 s: half a lap");
+    corner(&mut s, 0.5, br, "1.5 s: three quarters");
+    corner(
+        &mut s,
+        0.5,
+        bl,
+        "2.0 s: exactly one lap, back where it started",
+    );
+
+    // ── The position LAW, driven at exact clocks so no tick accumulation can blur it. The
+    // mid-edge sample is the `interp == 1` pin: linear means the half-time is the half-edge, and
+    // a spline-eased track would sit short of it.
+    let at = |s: &UiScript, clock: f32| -> (f32, f32) {
+        s.run(&format!(
+            "BenillaPetAutocast_Update(PetActionButton4, {clock})"
+        ))
+        .unwrap();
+        head(s, 0)
+    };
+    near(at(&s, 0.0), bl, "clock 0");
+    near(
+        at(&s, 0.25),
+        (-H, 0.0),
+        "quarter edge in: the LEFT edge's midpoint",
+    );
+    near(at(&s, 0.5), tl, "clock 0.5");
+    near(at(&s, 1.0), tr, "clock 1.0");
+    near(at(&s, 1.5), br, "clock 1.5");
+    near(at(&s, 2.0), bl, "clock 2.0 wraps onto clock 0");
+
+    // ── The trail is CONTINUOUS, which is what deriving the sample count buys (B228). No two
+    // neighbours in a trail sit further apart than the smaller of their two stars is wide, so the
+    // streak has no seam at any age. The 8-even-samples version failed this from mid-life back —
+    // 7.2 px apart with 4.3 px stars — and a row of separated dots is what reads as marching.
+    let worst_gap: f32 = s
+        .eval(
+            "local worst = 0 \
+             for k = 1, BENILLA_PET_AUTOCAST_TRAIL - 1 do \
+                 local a, b = PetActionButton4.sparks[k], PetActionButton4.sparks[k + 1] \
+                 local _, _, _, ax, ay = a:GetPoint() \
+                 local _, _, _, bx, by = b:GetPoint() \
+                 local d = math.sqrt((ax - bx) * (ax - bx) + (ay - by) * (ay - by)) \
+                 local w = math.min(a:GetWidth(), b:GetWidth()) \
+                 if d / w > worst then worst = d / w end \
+             end \
+             return worst",
+        )
+        .unwrap();
+    assert!(
+        worst_gap <= 1.0,
+        "the widest neighbour gap is {worst_gap:.2} star-widths — the streak has a seam"
+    );
+
+    // ── The four emitters are ONE LAP, a quarter apart — the file's four bones at the four
+    // corners, chasing each other (m2bones: pivots at (-0.001,0), (0.019,0), (0.019,0.02),
+    // (-0.001,0.02), same track, one corner of phase between neighbours).
+    s.run("BenillaPetAutocast_Update(PetActionButton4, 0.0)")
+        .unwrap();
+    for (e, want, name) in [
+        (0, bl, "emitter 0"),
+        (1, tl, "emitter 1"),
+        (2, tr, "emitter 2"),
+        (3, br, "emitter 3"),
+    ] {
+        near(head(&s, e), want, name);
+    }
+    assert!(s.errors().is_empty(), "script errors: {:?}", s.errors());
+}
+
+/// The pet bar plus the three files its TOOLTIP needs, in the TOC's own order: Fonts.xml (the
+/// colour codes that wrap the binding), GameTooltip.xml (the plate), and UIParent.xml — already a
+/// prerequisite for the managed stack — for `GetBindingText`, which renders the key.
+fn load_pet_bar_with_tooltip(s: &UiScript) {
+    for file in [
+        "Fonts.xml",
+        "UiPanels.xml",
+        "UIParent.xml",
+        "GameTooltip.xml",
+        "Cooldown.xml",
+        "ActionBar.xml",
+        "PetActionBar.xml",
+    ] {
+        load_xml(s, file);
+    }
+}
+
+/// A hunter bar hovered through the REAL gesture, with the real binding registry and the real CVar
+/// table behind it — `mouse_move` runs the shipped `<OnEnter>` with `this` bound, which is the only
+/// way this fork gets exercised the way nazriel exercised it.
+///
+/// Buttons are 30 px chained +8 from the bar's own origin: button 1 spans x[72,102] y[56,86] ⇒
+/// centre (87,71), button 4's left = 72 + 3·38 ⇒ centre (201,71) — the geometry
+/// `clicks_route_through_the_attack_toggle_fork` already leans on.
+const ATTACK_BUTTON: (f32, f32) = (87.0, 71.0);
+const CLAW_BUTTON: (f32, f32) = (201.0, 71.0);
+
+fn hovered_pet_bar() -> UiScript {
+    let mut s = UiScript::new().unwrap();
+    s.set_screen_size(1024.0, 768.0);
+    // The real command set and the real registered defaults, the way the app's own seed installs
+    // them — so `GetBindingKey("BONUSACTIONBUTTON1")` answers CTRL-1 (its byte-real default) and
+    // `GetCVar("UberTooltips")` answers "1" (its byte-read registrar value) rather than nil.
+    s.register_bindings(&crate::bindings::registry_commands());
+    s.register_cvars(crate::cvars::REGISTERED.iter().copied());
+    load_pet_bar_with_tooltip(&s);
+    declare_token_strings(&s);
+    s.set_pet_actions(true, true, true, hunter_slots());
+    s.fire_event("PET_BAR_UPDATE", vec![]);
+    s.resolve();
+    s
+}
+
+fn tooltip_line1(s: &UiScript) -> String {
+    s.eval::<String>("return GameTooltipTextLeft1:GetText() or \"\"")
+        .unwrap()
+}
+
+/// **B230 — Attack/Follow/Stay name their keybinding; the pet's own spells do not.**
+///
+/// nazriel: *"On 1.12 when you hover over Attack, Follow, Stay pet action buttons they show the
+/// keybinding - which is not the case with Benilla"*. The reference agrees with him and is
+/// narrower than "the pet bar": `PetActionButton_OnEnter` (PetActionBarFrame.lua l.285-305) forks
+/// on `isToken or UberTooltips == "0"`, and only THAT branch concatenates
+/// `NORMAL_FONT_COLOR_CODE.." ("..GetBindingText(GetBindingKey("BONUSACTIONBUTTON"..id), "KEY_")..")"`.
+/// The other branch is a bare `SetPetAction` and appends nothing — so a fix that suffixed every
+/// slot would be wrong, and the control below is what says ours doesn't.
+///
+/// The exact string is pinned, colour codes and all, because the ref's concatenation puts the
+/// colour BEFORE the space (`Attack|cffffd200 (CTRL-1)|r`, not `Attack |cffffd200(CTRL-1)|r` —
+/// which is what MicroMenu.xml's different-shaped version of the same idea produces).
+#[test]
+fn token_tooltips_name_their_binding_and_pet_spells_do_not() {
+    let mut s = hovered_pet_bar();
+
+    s.mouse_move(ATTACK_BUTTON.0, ATTACK_BUTTON.1);
+    s.resolve();
+    assert_eq!(
+        tooltip_line1(&s),
+        "Attack|cffffd200 (CTRL-1)|r",
+        "a command token carries BONUSACTIONBUTTON1's key in the normal-font colour"
+    );
+    // Uber on + a token ⇒ still the default corner; only the CVar-off leg owner-anchors.
+    assert!(
+        s.eval::<bool>("return GameTooltip.default ~= nil").unwrap(),
+        "a token's plate takes the default corner while UberTooltips is on"
+    );
+
+    // The CONTROL: slot 4 is Claw, a real pet spell, and it goes through the engine channel —
+    // its name, and nothing in parentheses.
+    s.mouse_move(CLAW_BUTTON.0, CLAW_BUTTON.1);
+    s.resolve();
+    let claw = tooltip_line1(&s);
+    assert_eq!(claw, "Claw", "a pet SPELL renders through SetPetAction");
+    assert!(
+        !claw.contains('('),
+        "…and appends no binding: {claw:?} (the ref's other branch)"
+    );
+    assert!(s.errors().is_empty(), "script errors: {:?}", s.errors());
+}
+
+/// The append is UNCONDITIONAL in that branch, so an unbound row really does render empty
+/// parentheses — `GetBindingText` answers `""` for a nil key (UIParent.xml, transcribed from the
+/// ref's own l.1819-1821), and nothing guards the `"("`. That is the reference's behaviour, not a
+/// slip of ours: MicroButtonTooltipText guards its own suffix and this one does not.
+///
+/// Stock it is unreachable — BONUSACTIONBUTTON1-10 default to CTRL-1..CTRL-0 — so it takes a
+/// deliberate unbind to see, which is exactly what this does.
+#[test]
+fn an_unbound_token_row_renders_the_references_empty_parentheses() {
+    let mut s = hovered_pet_bar();
+    s.run(r#"SetBinding("CTRL-1")"#).unwrap();
+    assert!(
+        s.eval::<Option<String>>(r#"return GetBindingKey("BONUSACTIONBUTTON1")"#)
+            .unwrap()
+            .is_none(),
+        "the row really is unbound now"
+    );
+
+    s.mouse_move(ATTACK_BUTTON.0, ATTACK_BUTTON.1);
+    s.resolve();
+    assert_eq!(
+        tooltip_line1(&s),
+        "Attack|cffffd200 ()|r",
+        "unbound: the parentheses stay, empty — the ref's own unguarded concatenation"
+    );
+    assert!(s.errors().is_empty(), "script errors: {:?}", s.errors());
+}
+
+/// The CVar is the fork's OTHER half, and it is not decoration: with `UberTooltips` off, a real pet
+/// SPELL stops going through the engine channel and takes the token branch too — same built text,
+/// same binding suffix — and the plate moves from the screen corner to beside the button
+/// (`SetOwner(this, "ANCHOR_RIGHT")`). Both are the reference's, in the same five lines.
+///
+/// This is also the row's reason to exist in `cvars::REGISTERED`: benilla registers `UberTooltips`
+/// at all because these Lua sites read it (the honest-tree rule's live-consumer half).
+#[test]
+fn ubertooltips_off_takes_the_spells_through_the_token_branch_and_moves_the_plate() {
+    let mut s = hovered_pet_bar();
+    s.set_cvar_engine("UberTooltips", "0");
+
+    s.mouse_move(CLAW_BUTTON.0, CLAW_BUTTON.1);
+    s.resolve();
+    assert_eq!(
+        tooltip_line1(&s),
+        "Claw|cffffd200 (CTRL-4)|r",
+        "with the CVar off a pet spell is built here too, binding and all"
+    );
+    assert!(
+        s.eval::<bool>("return GameTooltip:IsOwned(PetActionButton4)")
+            .unwrap(),
+        "…and the plate sits beside the button, not at the screen corner"
+    );
+    assert!(s.errors().is_empty(), "script errors: {:?}", s.errors());
+}

crates/benilla-app/src/ui_script/tooltip_anchor_tests.rs

@@ -192,11 +192,16 @@ fn your_own_portrait_explains_the_menu_instead_of_showing_your_health() {
 
 /// Action-bar hovers take the default corner too — ref ActionButton_SetTooltip l.366-372
 /// branches on the UberTooltips CVar, whose stock default is "1" (byte-read from WoW.exe
-/// 0x48fdd9 / default string 0x82e748; see ActionBar.xml). An empty slot renders nothing,
+/// 0x48fdd9 / default string 0x82e748; see `cvars::REGISTERED`). An empty slot renders nothing,
 /// but the anchor must already be seated — asserted through GetPoint, resolved rect or not.
+///
+/// Registered, not merely absent: before B230 the CVar was not in the table at all, so `GetCVar`
+/// answered nil and the "1" leg was reached by accident rather than by value. This seeds the real
+/// table so the pass means what it says.
 #[test]
 fn action_button_hover_takes_the_default_corner() {
-    let s = harness(&["Cooldown.xml", "ActionBar.xml"]);
+    let mut s = harness(&["Cooldown.xml", "ActionBar.xml"]);
+    s.register_cvars(crate::cvars::REGISTERED.iter().copied());
     s.run("BenillaActionButton_OnEnter(ActionButton3)").unwrap();
     assert!(s.errors().is_empty(), "hover errors: {:?}", s.errors());
     let ok: bool = s
@@ -212,6 +217,105 @@ fn action_button_hover_takes_the_default_corner() {
     );
 }
 
+/// The other leg of that same branch, live since B230 registered the CVar: with `UberTooltips`
+/// off, an action button's plate leaves the screen corner and seats BESIDE the button — LEFT for
+/// the bars the reference lists (`MultiBarBottomRight` is the only one benilla ships; the two
+/// vertical bars are not), RIGHT for everything else, including the main bar.
+///
+/// The two anchors are read off the resolved SetPoint pair, which is what `SetOwner` actually
+/// writes: ANCHOR_RIGHT = the plate's BOTTOMLEFT on the button's TOPRIGHT, ANCHOR_LEFT its mirror
+/// (`script/tooltip/verbs.rs`).
+#[test]
+fn ubertooltips_off_seats_action_bar_plates_beside_the_button() {
+    let mut s = harness(&["Cooldown.xml", "ActionBar.xml", "MultiBars.xml"]);
+    s.register_cvars(crate::cvars::REGISTERED.iter().copied());
+    s.set_cvar_engine("UberTooltips", "0");
+
+    let seat = |s: &UiScript| {
+        s.eval::<String>(
+            "local p, rel, rp = GameTooltip:GetPoint() \
+             return p .. \"/\" .. (rel and rel:GetName() or \"?\") .. \"/\" .. rp",
+        )
+        .unwrap()
+    };
+
+    s.run("BenillaActionButton_OnEnter(ActionButton3)").unwrap();
+    assert!(
+        s.eval::<bool>("return GameTooltip.default == nil").unwrap(),
+        "off: the main bar's plate is owner-anchored, not the default corner"
+    );
+    assert!(
+        s.eval::<bool>("return GameTooltip:IsOwned(ActionButton3)")
+            .unwrap(),
+        "…owned by the button it opened from"
+    );
+    assert_eq!(
+        seat(&s),
+        "BOTTOMLEFT/ActionButton3/TOPRIGHT",
+        "ANCHOR_RIGHT — the main bar is not in the ref's LEFT set"
+    );
+
+    // MultiBarBottomRight IS in that set — the one shipped member of it.
+    s.run("BenillaActionButton_OnEnter(MultiBarBottomRightButton1)")
+        .unwrap();
+    assert_eq!(
+        seat(&s),
+        "BOTTOMRIGHT/MultiBarBottomRightButton1/TOPLEFT",
+        "ANCHOR_LEFT — MultiBarBottomRight opens toward screen centre"
+    );
+
+    // And the CVar back on restores the corner — the fork is a fork, not a one-way door.
+    s.set_cvar_engine("UberTooltips", "1");
+    s.run("BenillaActionButton_OnEnter(MultiBarBottomRightButton1)")
+        .unwrap();
+    assert!(
+        s.eval::<bool>("return GameTooltip.default ~= nil").unwrap(),
+        "on: back to the screen corner"
+    );
+    assert!(s.errors().is_empty(), "hover errors: {:?}", s.errors());
+}
+
+/// The stance bar's own leg of the same branch (ref BonusActionBarFrame.xml l.40-45 — ANCHOR_RIGHT
+/// with no bar fork of its own), pinned separately because it is a different file's handler and a
+/// different tooltip verb; it was collapsed to the "1" leg alongside the action bar's and
+/// un-collapsed with it.
+#[test]
+fn ubertooltips_off_seats_stance_plates_beside_the_button() {
+    let mut s = harness(&["Cooldown.xml", "ActionBar.xml", "StanceBar.xml"]);
+    s.register_cvars(crate::cvars::REGISTERED.iter().copied());
+    s.set_shapeshift_forms(vec![benilla_ui::script::ShapeshiftFormView {
+        spell_id: 5487,
+        name: "Bear Form".into(),
+        texture: Some("Interface\\Icons\\Ability_Racial_BearForm".into()),
+        active: false,
+        castable: true,
+        cooldown: None,
+    }]);
+    s.fire_event("UPDATE_SHAPESHIFT_FORMS", vec![]);
+    s.resolve();
+
+    s.run("BenillaShapeshiftButton_OnEnter(ShapeshiftButton1)")
+        .unwrap();
+    assert!(
+        s.eval::<bool>("return GameTooltip.default ~= nil").unwrap(),
+        "on (the stock default): the screen corner"
+    );
+
+    s.set_cvar_engine("UberTooltips", "0");
+    s.run("BenillaShapeshiftButton_OnEnter(ShapeshiftButton1)")
+        .unwrap();
+    assert_eq!(
+        s.eval::<String>(
+            "local p, rel, rp = GameTooltip:GetPoint() \
+             return p .. \"/\" .. (rel and rel:GetName() or \"?\") .. \"/\" .. rp",
+        )
+        .unwrap(),
+        "BOTTOMLEFT/ShapeshiftButton1/TOPRIGHT",
+        "off: ANCHOR_RIGHT, beside the button"
+    );
+    assert!(s.errors().is_empty(), "hover errors: {:?}", s.errors());
+}
+
 /// Buff hovers hang BELOW the button — ref BuffFrame.xml l.37 is ANCHOR_BOTTOMLEFT (the buff
 /// row lives at the screen's top-right): the tooltip's TOPRIGHT seats on the button's
 /// BOTTOMLEFT.

crates/benilla-app/src/ui_spellbook.rs

@@ -31,7 +31,7 @@
 //! and the rank tie-break falling to the rank *string* only when the parsed number is 0 (the
 //! client's `SpellLevel` fallback for a digit-less rank is unmodeled — we have no `SpellLevel`).
 
-use std::collections::{BTreeMap, HashSet};
+use std::collections::{BTreeMap, BTreeSet};
 
 use bevy::prelude::*;
 
@@ -231,7 +231,7 @@ fn feed_spellbook(
 /// Build the whole book (module doc): the app's own resolve — the engine holds no spell
 /// knowledge, only what's pushed here.
 fn build_book(
-    known: &HashSet<u32>,
+    known: &BTreeSet<u32>,
     catalog: &SpellCatalog,
     skill_lines: Option<&SkillLineCatalog>,
     race: u8,
@@ -458,7 +458,7 @@ mod tests {
         map.insert(668, spell("Language: Common", None, 0x80)); // DO_NOT_DISPLAY → hidden
         map.insert(818, spell("Cooking", None, 0x20)); // IS_TRADESKILL → hidden
         let catalog = SpellCatalog::from_displays(map);
-        let known: HashSet<u32> = [133, 145, 2136, 668, 818].into_iter().collect();
+        let known: BTreeSet<u32> = [133, 145, 2136, 668, 818].into_iter().collect();
 
         // No skill-line catalog → every shown spell lands in General (race/class irrelevant).
         let book = build_book(&known, &catalog, None, 1, 1, None, None);
@@ -495,7 +495,7 @@ mod tests {
         attack.icon = Some("Interface\\Icons\\Temp".into()); // the real DBC placeholder
         attack.effects[0] = 78; // SPELL_EFFECT_ATTACK — makes it the melee auto-attack
         let catalog = SpellCatalog::from_displays(HashMap::from([(ATTACK, attack)]));
-        let known: HashSet<u32> = [ATTACK].into_iter().collect();
+        let known: BTreeSet<u32> = [ATTACK].into_iter().collect();
 
         // The feed's resolved icon (armed → the weapon, unarmed → Spell-Reset) wins over Temp.
         for resolved in [
@@ -531,7 +531,7 @@ mod tests {
         throw.icon = Some("Interface\\Icons\\Ability_Throw".into());
         let catalog =
             SpellCatalog::from_displays(HashMap::from([(AUTO_SHOT, auto_shot), (THROW, throw)]));
-        let known: HashSet<u32> = [AUTO_SHOT, THROW].into_iter().collect();
+        let known: BTreeSet<u32> = [AUTO_SHOT, THROW].into_iter().collect();
 
         let bow = "Interface\\Icons\\INV_Weapon_Bow_02";
         let book = build_book(&known, &catalog, None, 1, 1, None, Some(bow.into()));

crates/benilla-app/src/ui_talent.rs

@@ -28,7 +28,7 @@
 //! composed here on a cp2 rise: this feed owns the points diff, and the chat arc's own kinds
 //! carry it as a SYSTEM line.
 
-use std::collections::HashSet;
+use std::collections::BTreeSet;
 
 use bevy::prelude::*;
 
@@ -151,7 +151,7 @@ fn feed_talents(
 }
 
 /// One talent's current rank: the highest rank whose spell is known (module doc).
-fn rank_of(t: &Talent, known: &HashSet<u32>) -> u32 {
+fn rank_of(t: &Talent, known: &BTreeSet<u32>) -> u32 {
     t.ranks
         .iter()
         .enumerate()
@@ -164,7 +164,7 @@ fn rank_of(t: &Talent, known: &HashSet<u32>) -> u32 {
 /// Build the pushed snapshot — the app's whole resolve (module doc).
 fn build_pages(
     catalog: &TalentCatalog,
-    known: &HashSet<u32>,
+    known: &BTreeSet<u32>,
     spells: &Spells,
     race: u8,
     class: u8,
@@ -341,11 +341,11 @@ mod tests {
     #[test]
     fn rank_is_the_highest_known_rank_spell() {
         let t = talent(1, 81, 0, 0, &[100, 101, 102]);
-        let known: HashSet<u32> = [100, 101].into_iter().collect();
+        let known: BTreeSet<u32> = [100, 101].into_iter().collect();
         assert_eq!(rank_of(&t, &known), 2);
-        assert_eq!(rank_of(&t, &HashSet::new()), 0);
+        assert_eq!(rank_of(&t, &BTreeSet::new()), 0);
         // A gap never happens on the live wire (learn-up-to), but the max() read stays honest.
-        let holey: HashSet<u32> = [102].into_iter().collect();
+        let holey: BTreeSet<u32> = [102].into_iter().collect();
         assert_eq!(rank_of(&t, &holey), 3);
     }
 }

crates/benilla-app/src/ui_trainer/mod.rs

@@ -22,7 +22,7 @@
 //! when the player walks out of the trainer's service range (or the trainer despawns) — the same
 //! `CloseTrainer` clear.
 
-use std::collections::HashSet;
+use std::collections::BTreeSet;
 
 use benilla_formats::{SkillLineCatalog, SpellCatalog};
 use benilla_protocol::messages::TrainerSpell;
@@ -157,7 +157,7 @@ fn resolve_service(
     trainer_type: u32,
     spells: &SpellCatalog,
     skill_lines: Option<&SkillLineCatalog>,
-    known: &HashSet<u32>,
+    known: &BTreeSet<u32>,
     icons: Option<&ItemDisplays>,
     items: &mut Items,
     commands: &NetCommands,
@@ -269,7 +269,7 @@ fn snapshot(
     open: &TrainerOpen,
     spells: &SpellCatalog,
     skill_lines: Option<&SkillLineCatalog>,
-    known: &HashSet<u32>,
+    known: &BTreeSet<u32>,
     icons: Option<&ItemDisplays>,
     items: &mut Items,
     commands: &NetCommands,

crates/benilla-app/src/ui_trainer/tests.rs

@@ -26,7 +26,7 @@ fn resolve_service(
     trainer_type: u32,
     spells: &SpellCatalog,
     skill_lines: Option<&SkillLineCatalog>,
-    known: &HashSet<u32>,
+    known: &BTreeSet<u32>,
 ) -> TrainerService {
     let mut deps = Deps::new();
     super::resolve_service(
@@ -62,7 +62,7 @@ fn snap(open: &TrainerOpen, spells: &SpellCatalog) -> Option<TrainerState> {
         open,
         spells,
         None,
-        &HashSet::new(),
+        &BTreeSet::new(),
         None,
         &mut deps.items,
         &deps.commands,
@@ -306,7 +306,7 @@ fn resolve_hops_the_learn_wrapper_to_the_taught_ability() {
         0,
         &spells,
         Some(&skills),
-        &HashSet::new(),
+        &BTreeSet::new(),
     );
     assert_eq!(svc.spell_id, 1605, "the buy id stays the wire wrapper");
     assert_eq!(
@@ -340,7 +340,7 @@ fn display_name_is_the_wire_spell_not_the_taught_one() {
         TRAINER_TYPE_TRADESKILL,
         &spells,
         Some(&skills),
-        &HashSet::new(),
+        &BTreeSet::new(),
     );
     assert_eq!(svc.name.as_deref(), Some("Apprentice Blacksmith"));
     assert_eq!(svc.subtext.as_deref(), None);
@@ -359,7 +359,7 @@ fn display_name_is_the_wire_spell_not_the_taught_one() {
         TRAINER_TYPE_TRADESKILL,
         &spells,
         Some(&skills),
-        &HashSet::new(),
+        &BTreeSet::new(),
     );
     assert_eq!(recipe.group_key, 2);
     assert_eq!(recipe.group_name, "Recipes");
@@ -392,7 +392,7 @@ fn resolve_reads_cost_state_and_gates_with_no_catalog() {
     let mut w = wire(2018, trainer_spell_state::RED, 1000, 20, 164);
     w.req_spells = [78, 0, 0];
     // Empty known set: the player knows nothing → the ability gate reads unmet on its own terms.
-    let svc = resolve_service(&w, TRAINER_TYPE_TRADESKILL, &spells, None, &HashSet::new());
+    let svc = resolve_service(&w, TRAINER_TYPE_TRADESKILL, &spells, None, &BTreeSet::new());
     assert_eq!(svc.spell_id, 2018);
     assert!(svc.name.is_none(), "no catalog → name in flight");
     assert_eq!(svc.cost, 1000);
@@ -429,7 +429,7 @@ fn resolve_reads_cost_state_and_gates_with_no_catalog() {
         0,
         &spells,
         None,
-        &HashSet::new(),
+        &BTreeSet::new(),
     );
     assert_eq!(plain.skill_req, None);
     assert!(plain.ability_reqs.is_empty());
@@ -447,12 +447,12 @@ fn ability_req_met_tracks_known_spells_not_the_service_category() {
     w.req_spells = [78, 0, 0]; // requires Heroic Strike (78)
 
     // Player doesn't know 78 → the prerequisite is unmet (red).
-    let unknown = resolve_service(&w, 0, &spells, None, &HashSet::new());
+    let unknown = resolve_service(&w, 0, &spells, None, &BTreeSet::new());
     assert_eq!(unknown.category, TrainerServiceCategory::Unavailable);
     assert!(!unknown.ability_reqs[0].met, "prereq unknown → unmet");
 
     // Player knows 78 → the prerequisite is met (white) EVEN THOUGH the service stays unavailable.
-    let known: HashSet<u32> = [78].into_iter().collect();
+    let known: BTreeSet<u32> = [78].into_iter().collect();
     let learned = resolve_service(&w, 0, &spells, None, &known);
     assert_eq!(learned.category, TrainerServiceCategory::Unavailable);
     assert!(
@@ -473,7 +473,7 @@ fn ability_req_shows_the_required_rank_on_real_data() {
     let mut w = wire(846, trainer_spell_state::RED, 100, 20, 0);
     w.req_spells = [78, 0, 0]; // Heroic Strike Rank 1
 
-    let known: HashSet<u32> = [78].into_iter().collect();
+    let known: BTreeSet<u32> = [78].into_iter().collect();
     let svc = resolve_service(&w, 0, &spells, None, &known);
     assert_eq!(
         svc.ability_reqs,
@@ -484,7 +484,7 @@ fn ability_req_shows_the_required_rank_on_real_data() {
         "the prereq shows its rank and reads met because the player knows it"
     );
     // Not known → same name, but unmet (red).
-    let svc = resolve_service(&w, 0, &spells, None, &HashSet::new());
+    let svc = resolve_service(&w, 0, &spells, None, &BTreeSet::new());
     assert!(!svc.ability_reqs[0].met);
     assert_eq!(svc.ability_reqs[0].name, "Heroic Strike (Rank 1)");
 }

crates/benilla-formats/src/bin/benilla-extract/m2dump.rs

@@ -1347,13 +1347,23 @@ pub fn m2part(chain: &mut Chain, internal_path: &str) -> Result<()> {
             println!("     recursion model (child emitters): {r}");
         }
         if let Some(s) = &e.spline {
-            let pt = |p: &[f32; 3]| format!("({:.2},{:.2},{:.2})", p[0], p[1], p[2]);
+            // EVERY control point, and at full precision. This used to print `first .. last`,
+            // which is actively misleading on a cubic Bézier: the path is `3K+1` points and the
+            // two INTERIOR handles of each segment are what bend it, so an emitter whose ends
+            // both sit at the origin can still fling its particles a long way — or not move them
+            // at all. Reading "(0,0,0) .. (0,0,0)" as "the particles never travel" is a guess
+            // the truncated line invites and cannot support (B228 nearly rested on it).
+            // Wrapped at 4 per line so a long path stays readable.
+            let pt = |p: &[f32; 3]| format!("({:.4},{:.4},{:.4})", p[0], p[1], p[2]);
             println!(
-                "     spline: {} control points, {} .. {}",
+                "     spline: {} control points ({} cubic segment(s)), model-local:",
                 s.points.len(),
-                s.points.first().map(pt).unwrap_or_default(),
-                s.points.last().map(pt).unwrap_or_default(),
+                s.points.len().saturating_sub(1) / 3,
             );
+            for (i, chunk) in s.points.chunks(4).enumerate() {
+                let row: Vec<String> = chunk.iter().map(pt).collect();
+                println!("       [{:>2}] {}", i * 4, row.join("  "));
+            }
         }
 
         // Emission timing: the rate/gate pair, per file sequence slot.
@@ -1454,8 +1464,12 @@ pub fn m2part(chain: &mut Chain, internal_path: &str) -> Result<()> {
                 c[0], c[1], c[2], c[3]
             );
         }
+        // 4 decimals, not 3: UI models author their sizes in thousandths, where `{:.3}` turns the
+        // autocast shine's 0.0015 mid-key into a printed "0.002" — a 33% error that a transcriber
+        // reads as authored truth (B228). The colour ramp above stays at 3 because 0..1 channels
+        // do not have that problem.
         println!(
-            "       size    {:.3} -> {:.3} -> {:.3} yd (half-extent)",
+            "       size    {:.4} -> {:.4} -> {:.4} yd (half-extent)",
             ol.scale[0], ol.scale[1], ol.scale[2]
         );
         println!(

crates/benilla-formats/src/skill_lines/mod.rs

@@ -337,7 +337,7 @@ impl SkillLineCatalog {
     pub fn highest_known_rank(
         &self,
         spell_id: u32,
-        known: &std::collections::HashSet<u32>,
+        known: &std::collections::BTreeSet<u32>,
     ) -> Option<u32> {
         let mut cur = self.chain_head(spell_id);
         let mut best = known.contains(&cur).then_some(cur);

crates/benilla-formats/src/skill_lines/tests.rs

@@ -361,12 +361,12 @@ fn real_rank_chains_resolve_the_highest_known_rank() {
 
     // From ANY rank, knowing only rank 8, the answer is rank 8 — the walk starts at the
     // chain head, so it works downward as well as upward.
-    let top: std::collections::HashSet<u32> = [11294].into_iter().collect();
+    let top: std::collections::BTreeSet<u32> = [11294].into_iter().collect();
     for id in SS {
         assert_eq!(cat.highest_known_rank(id, &top), Some(11294), "from {id}");
     }
     // Knowing an intermediate rank resolves to it, not to the top of the chain.
-    let mid: std::collections::HashSet<u32> = [1759].into_iter().collect();
+    let mid: std::collections::BTreeSet<u32> = [1759].into_iter().collect();
     assert_eq!(cat.highest_known_rank(1752, &mid), Some(1759));
     assert_eq!(cat.highest_known_rank(11294, &mid), Some(1759));
     // No rank of the chain known → no answer (the caller leaves the slot alone).
@@ -377,11 +377,11 @@ fn real_rank_chains_resolve_the_highest_known_rank() {
     // Frostbolt r1, Healing Touch r1, Renew r1, Immolate r1, Lightning Bolt r1.
     for id in [133u32, 143, 116, 5185, 139, 348, 403] {
         assert_eq!(cat.rank_successor(id), None, "spell {id} is not chained");
-        let known: std::collections::HashSet<u32> = [id].into_iter().collect();
+        let known: std::collections::BTreeSet<u32> = [id].into_iter().collect();
         assert_eq!(cat.highest_known_rank(id, &known), Some(id));
     }
     // A spell with no `SkillLineAbility` row at all (the auto-attack) is its own chain.
-    let attack: std::collections::HashSet<u32> = [6603].into_iter().collect();
+    let attack: std::collections::BTreeSet<u32> = [6603].into_iter().collect();
     assert_eq!(cat.rank_successor(6603), None);
     assert_eq!(cat.highest_known_rank(6603, &attack), Some(6603));
 }

crates/benilla-formats/src/spells/catalog_tests.rs

@@ -247,6 +247,43 @@ fn real_spell_catalog_reads_open_lock_types() {
         "Fireball opens no lock"
     );
 
+    // **B247's data fact** (decision 1312): `LockType 13` "Open Kneeling" — what lock 43 carries,
+    // and with it every ground container from the Hyacinth Mushroom up — has TWO shipped openers
+    // that `playercreateinfo_spell` grants to every race/class, and one of them wears Blizzard's
+    // own placeholder name. The lock resolver returns the FIRST sufficient match in the
+    // known-spell array's order, so which of these it reaches is what the cast bar prints; this
+    // pins the ambiguity to the data instead of leaving it a surprise.
+    assert_eq!(cat.get(6478).unwrap().open_lock_type(), Some(13));
+    assert_eq!(cat.get(6478).unwrap().name, "Opening");
+    assert_eq!(cat.get(22810).unwrap().open_lock_type(), Some(13));
+    assert_eq!(
+        cat.get(22810).unwrap().name,
+        "Opening - No Text",
+        "the placeholder is Blizzard's own Spell.dbc string, not a formatting bug of ours"
+    );
+
+    // ...and the attribute that says never to show it. `AttributesEx3 & 0x4`
+    // (`SPELL_ATTR_EX3_NO_CASTING_BAR_TEXT`) is what the cast bar must honour — 22810 carries it,
+    // 6478 does not, and that ONE bit is the only column separating two otherwise byte-identical
+    // rows. Exactly three rows in the shipped file carry it, so the whole set is pinned here: a
+    // fourth appearing means the data is not what this law was derived from.
+    assert!(cat.get(22810).unwrap().no_casting_bar_text());
+    assert!(!cat.get(6478).unwrap().no_casting_bar_text());
+    let silent: Vec<u32> = {
+        let mut v: Vec<u32> = cat
+            .iter()
+            .filter(|(_, d)| d.no_casting_bar_text())
+            .map(|(id, _)| id)
+            .collect();
+        v.sort_unstable();
+        v
+    };
+    assert_eq!(
+        silent,
+        vec![6477, 22810, 26380],
+        "6477 Opening / 22810 Opening - No Text / 26380 zzOLDSummon Mouth Tentacle Visual"
+    );
+
     // The totem (tool) and reagent columns the pre-send possession check reads (decision 0552;
     // the ref's `0x6e4000` at SpellRec+0xA0/+0xA8 = cols 40-41 / 42-49+50-57). A column slip
     // here silently breaks "Requires Mining Pick" / "Missing reagent: …".

crates/benilla-formats/src/spells/display.rs

@@ -407,6 +407,19 @@ impl SpellDisplay {
         self.attributes_ex3 & ATTR_EX3_NORMAL_RANGED_ATTACK != 0
     }
 
+    /// `AttributesEx3 & 0x4` — **the casting bar shows no name for this spell**
+    /// ([`ATTR_EX3_NO_CASTING_BAR_TEXT`], `SPELL_ATTR_EX3_NO_CASTING_BAR_TEXT`). The cast-bar feed
+    /// hands `SPELLCAST_START`/`SPELLCAST_CHANNEL_START` an empty name string when this is set, so
+    /// the bar fills and sweeps with a blank label.
+    ///
+    /// Only the *name* is suppressed, and only on the bar: the spell still casts, still animates,
+    /// and its failures still name it. Three shipped rows carry the bit; the one that reached a
+    /// player is 22810 **"Opening - No Text"**, the internal opener for `LockType 13` ground
+    /// containers, whose own name spells out what the attribute is for (decision 1312, B247).
+    pub fn no_casting_bar_text(&self) -> bool {
+        self.attributes_ex3 & ATTR_EX3_NO_CASTING_BAR_TEXT != 0
+    }
+
     /// The ranged-shot cooldown pad's gate (`0x6e2b60` at `0x6e2c2c`–`0x6e2c47`, byte-verified —
     /// wow-re `ranged-cooldown-sweep.md`): on the caster's own SPELL_GO self-insert, a
     /// `Attributes & 0x2` spell WITHOUT `AttributesEx2 & 0x20000`

crates/benilla-formats/src/spells/mod.rs

@@ -350,6 +350,15 @@ const COL_EFFECT_ITEM_TYPE_1: usize = 103;
 /// `AttributesEx3` bit `0x8000` — damage renders melee-white (`SPELL_ATTR3_NORMAL_RANGED_ATTACK`;
 /// the combat-text emitter's `B`-bit flip, wow-re `combattext-color-law.md`).
 const ATTR_EX3_NORMAL_RANGED_ATTACK: u32 = 0x8000;
+/// `AttributesEx3` bit `0x4` — **the cast bar shows no name for this spell**
+/// (`SPELL_ATTR_EX3_NO_CASTING_BAR_TEXT`, vmangos `Spells/SpellDefines.h:907`). Decision 1312.
+///
+/// Exactly three rows in the shipped 5875 file carry it, and one of them is *named after the bit*:
+/// 6477 "Opening", **22810 "Opening - No Text"**, 26380 "zzOLDSummon Mouth Tentacle Visual". That
+/// placeholder name is not a string benilla was ever meant to render — it is Blizzard's own note
+/// to themselves about what this attribute does to the spell, and it reached a player's screen
+/// because we printed the name unconditionally.
+const ATTR_EX3_NO_CASTING_BAR_TEXT: u32 = 0x4;
 /// `Attributes` bit `0x2` — the "uses the ranged slot" attribute (mangos `SPELL_ATTR_RANGED`).
 /// One half of the client's ranged-stance gate (module docs).
 const ATTR_RANGED: u32 = 0x2;

crates/benilla-protocol/examples/cooldown_probe.rs

@@ -120,6 +120,7 @@ fn main() -> Result<()> {
             spell_id,
             success,
             reason,
+            ..
         } => println!(
             "  pre: cast_result spell={spell_id} success={success} reason={reason:?} ({})",
             reason.map(fail_name).unwrap_or("-")
@@ -149,6 +150,7 @@ fn main() -> Result<()> {
                 spell_id,
                 success: false,
                 reason,
+                ..
             } if spell_id == CHARGE => {
                 println!(
                     "  refused: {:?} ({})",
@@ -183,6 +185,7 @@ fn main() -> Result<()> {
                     spell_id,
                     success,
                     reason,
+                    ..
                 } if spell_id == CHARGE => {
                     let name = reason.map(fail_name).unwrap_or("-");
                     println!(

crates/benilla-protocol/src/bin/benilla-world/world.rs

@@ -436,8 +436,11 @@ impl World {
                 spell_id,
                 success,
                 reason,
+                arg,
             } => {
-                println!("SMSG_CAST_RESULT: spell {spell_id} success={success} reason={reason:?}");
+                println!(
+                    "SMSG_CAST_RESULT: spell {spell_id} success={success} reason={reason:?} arg={arg:?}"
+                );
                 if self.dest_spell == Some(*spell_id) {
                     self.dest_verdict = Some((*spell_id, *success, *reason));
                 } else if self.cast_verdict.is_none() {

crates/benilla-protocol/src/events.rs

@@ -467,11 +467,15 @@ pub enum SessionEvent {
         new_spell_id: u32,
     },
     /// The server's verdict on our cast (`SMSG_CAST_RESULT`): `reason` is the failure code when
-    /// `success` is false (the client's error-text table keys on it; surfaced raw for now).
+    /// `success` is false (the client's error-text table keys on it), and `arg` the reason-specific
+    /// argument word that fills that message's `%s` — a `SpellFocusObject.dbc` id for
+    /// `REQUIRES_SPELL_FOCUS`, an `AreaTable.dbc` id for `REQUIRES_AREA`
+    /// ([`crate::messages::CastOutcome::Failed`]).
     CastResult {
         spell_id: u32,
         success: bool,
         reason: Option<u8>,
+        arg: Option<u32>,
     },
     /// The pet action bar's whole state (`SMSG_PET_SPELLS`, decision 0982) — or its **teardown**,
     /// which is the same event carrying a zero `pet_guid`. Server-authoritative: this is not a

crates/benilla-protocol/src/events/decode.rs

@@ -80,7 +80,11 @@ pub fn decode(packet: ServerPacket) -> Vec<SessionEvent> {
             success: outcome == CastOutcome::Ok,
             reason: match outcome {
                 CastOutcome::Ok => None,
-                CastOutcome::Failed { reason } => Some(reason),
+                CastOutcome::Failed { reason, .. } => Some(reason),
+            },
+            arg: match outcome {
+                CastOutcome::Ok => None,
+                CastOutcome::Failed { arg, .. } => arg,
             },
         }],
         ServerPacket::PetSpells(spells) => vec![SessionEvent::PetSpells(Box::new(spells))],
@@ -92,7 +96,7 @@ pub fn decode(packet: ServerPacket) -> Vec<SessionEvent> {
             spell_id,
             reason: match outcome {
                 CastOutcome::Ok => None,
-                CastOutcome::Failed { reason } => Some(reason),
+                CastOutcome::Failed { reason, .. } => Some(reason),
             },
         }],
         ServerPacket::ItemQueryResponse { entry, info } => {

crates/benilla-protocol/src/messages/pet.rs

@@ -380,12 +380,16 @@ pub(super) fn read_pet_action_feedback(r: &mut impl Read) -> io::Result<u8> {
 /// `SMSG_CAST_RESULT` — the same `SpellCastResult` vocabulary, decoded the same way (status `2` =
 /// fail) so the one message table renders both. vmangos only ever sends the fail status here, but
 /// reading it the shared way costs nothing and keeps the two decodes from drifting.
+///
+/// The body stops at the reason: unlike `CastResult`, `PetCastFailed::AppendBodyTo` writes **no
+/// argument words** at all, so a pet failure never carries a `%s` fill.
 pub(super) fn read_pet_cast_failed(r: &mut impl Read) -> io::Result<(u32, super::CastOutcome)> {
     let spell_id = read_u32_le(r)?;
     let status = read_u8(r)?;
     let outcome = if status == 2 {
         super::CastOutcome::Failed {
             reason: read_u8(r)?,
+            arg: None,
         }
     } else {
         super::CastOutcome::Ok

crates/benilla-protocol/src/messages/spells.rs

@@ -28,24 +28,52 @@ use crate::wire::{
 };
 
 /// `SMSG_CAST_RESULT`'s verdict: `u32 spellId, u8 status` — status `0` (`SPELL_RESULT_STATUS_OKAY`)
-/// ends the packet, status `2` (`SPELL_RESULT_STATUS_FAIL`) appends a `u8` failure reason
-/// (vmangos `CastResult::AppendBodyTo`; some reasons append extra arg words we skip).
+/// ends the packet, status `2` (`SPELL_RESULT_STATUS_FAIL`) appends a `u8` failure reason and then
+/// up to two **reason-specific argument words** (vmangos `CastResult::AppendBodyTo`).
 #[derive(Debug, Clone, Copy, PartialEq, Eq)]
 pub enum CastOutcome {
     Ok,
-    Failed { reason: u8 },
+    Failed {
+        reason: u8,
+        /// The failure's first argument word, when the server sent one — the `%s` source of the
+        /// client's argument-formatted messages (the arm table at `0x6e1d8e`): a
+        /// `SpellFocusObject.dbc` id for `REQUIRES_SPELL_FOCUS` (0x5e), an `AreaTable.dbc` id for
+        /// `REQUIRES_AREA` (0x5d), an `EquippedItemClass` for the 0x19–0x1b family, a
+        /// permanent-cooldown flag for `NOT_READY` (0x3c). `None` = the packet ended at the reason.
+        arg: Option<u32>,
+    },
 }
 
-/// Read `SMSG_CAST_RESULT` → `(spell_id, outcome)`. Reason-specific trailing args (a required
-/// spell-focus id, an equip class) are left unread — the slice ends with the packet, nothing
-/// follows in the stream.
-pub(super) fn read_cast_result(r: &mut impl Read) -> io::Result<(u32, CastOutcome)> {
+/// Read `SMSG_CAST_RESULT` → `(spell_id, outcome)`.
+///
+/// The trailing argument words are **positional, gated on the remaining body length** — and that
+/// is the reference's own decode, not merely a convenient one. `CastResultHandler 0x6e7330`
+/// (registered at `0x6e7155`) reads `u32 spellId`, `u8 status`, then on `status == 2` a
+/// `u8 reason`, then each word only `if cursor < size` (`0x6e737a` and `0x6e738c`, comparing
+/// CDataStore `+0x14` against `+0x10`) — never keyed off the reason value (wow-re
+/// `system/spell/scratch/cast-fail-strings.md` §WIRE-ARGS, §5 cross-checked). It has to be: vmangos
+/// writes `if (arg1 || arg2) << u32 arg1; if (arg2) << u32 arg2;`
+/// (`Server/Packets/Spell.cpp`, `CastResult::AppendBodyTo`), so a reason whose argument is
+/// legitimately 0 with no second argument behind it sends no word at all, and a reason-keyed decode
+/// would desync on exactly that case.
+///
+/// **The reference's absent value is `-1`, not zero** (`0x6e736e or eax,0xffffffff` initializes both
+/// slots before the gated reads). `None` here is that sentinel: every consumer treats it and an
+/// out-of-range id the same way — decline the fill — so the two encodings are behaviourally
+/// identical, and `Option` says which case it is at the type level.
+///
+/// The second word is read to keep the shape documented, then dropped: only the
+/// `EQUIPPED_ITEM_CLASS*` family (0x19–0x1b) sends one, and that arm's `%s` fill reads the failing
+/// spell's own `EquippedItemClass`/`EquippedItemSubClassMask` columns — the same two values the
+/// server copied out of `SpellEntry` to build the packet (`Spell::SendCastResult`).
+pub(super) fn read_cast_result(r: &mut &[u8]) -> io::Result<(u32, CastOutcome)> {
     let spell_id = read_u32_le(r)?;
     let status = read_u8(r)?;
     let outcome = if status == 2 {
-        CastOutcome::Failed {
-            reason: read_u8(r)?,
-        }
+        let reason = read_u8(r)?;
+        let arg = (r.len() >= 4).then(|| read_u32_le(r)).transpose()?;
+        let _arg2 = (r.len() >= 4).then(|| read_u32_le(r)).transpose()?;
+        CastOutcome::Failed { reason, arg }
     } else {
         CastOutcome::Ok
     };

crates/benilla-protocol/tests/spells.rs

@@ -142,7 +142,43 @@ fn spell_and_action_wire() {
             spell_id,
             success,
             reason,
-        } => assert_eq!((spell_id, success, reason), (6673, false, Some(0x55))),
+            arg,
+        } => assert_eq!(
+            (spell_id, success, reason, arg),
+            (6673, false, Some(0x55), None),
+            "a reason carrying no argument ends the body at the reason byte"
+        ),
+        _ => panic!("cast fail event"),
+    }
+    // The reason-specific argument words (`CastResult::AppendBodyTo`) are POSITIONAL — read by
+    // what is left in the body, never keyed off the reason. One word for 0x5e
+    // REQUIRES_SPELL_FOCUS (the `SpellFocusObject.dbc` id: 12 = "Starbreeze Village Moonwell",
+    // the `%s` of "Requires %s"); two for the 0x19 EQUIPPED_ITEM_CLASS family (class + subclass
+    // mask), whose second is read and dropped.
+    let focus = messages::parse_server(
+        messages::opcode::SMSG_CAST_RESULT,
+        &hx("111a0000025e0c000000"),
+    )
+    .unwrap();
+    match decode(focus).pop().unwrap() {
+        SessionEvent::CastResult { reason, arg, .. } => {
+            assert_eq!((reason, arg), (Some(0x5e), Some(12)));
+        }
+        _ => panic!("cast fail event"),
+    }
+    let equip = messages::parse_server(
+        messages::opcode::SMSG_CAST_RESULT,
+        &hx("111a0000021902000000f3a50200"),
+    )
+    .unwrap();
+    match decode(equip).pop().unwrap() {
+        SessionEvent::CastResult { reason, arg, .. } => {
+            assert_eq!(
+                (reason, arg),
+                (Some(0x19), Some(2)),
+                "arg2 read and dropped"
+            );
+        }
         _ => panic!("cast fail event"),
     }
 

crates/benilla-ui/src/loader/scripts.rs

@@ -39,10 +39,9 @@ impl Loader<'_> {
                 // is in the set, `OnDragStop`/`OnReceiveDrag` are NOT. This is XML-load-time
                 // ONLY: the Lua SetScript binding (`0x7748d0`) never auto-enables, so the law
                 // lives here and not in SetScript itself (a runtime-created frame still needs an
-                // explicit `EnableMouse(true)`, like the real client). The keyboard kinds
-                // (`OnChar` = kind 0, `OnKeyDown`/`OnKeyUp` = kind 1) and the wheel kind
-                // (`OnMouseWheel` = kind 3) have separate indexes in the real client that this
-                // engine doesn't model yet — un-modeled beside the SetScript keyboard gap above.
+                // explicit `EnableMouse(true)`, like the real client). The wheel kind
+                // (`OnMouseWheel` = kind 3) has a separate index this engine doesn't model yet;
+                // the KEYBOARD kinds are modelled and armed just below.
                 const MOUSE_KIND: [&str; 5] = [
                     "OnEnter",
                     "OnLeave",
@@ -53,6 +52,16 @@ impl Loader<'_> {
                 if MOUSE_KIND.iter().any(|k| name.eq_ignore_ascii_case(k)) {
                     self.call(wrapper, "EnableMouse", true, dbg);
                 }
+                // The KEYBOARD kinds, the same walker rule one index over (`OnChar` = kind 0,
+                // `OnKeyDown`/`OnKeyUp` = kind 1). Bucket membership is what the delivery walk
+                // iterates ([`crate::script::keyboard`]) and it is the FLAG, never the presence of
+                // a script — so an XML frame carrying a key handler but no `enableKeyboard`
+                // attribute would never be reached by the dispatcher at all, and its handler could
+                // not fire. XML-load-time only, like the mouse half.
+                const KEY_KIND: [&str; 3] = ["OnChar", "OnKeyDown", "OnKeyUp"];
+                if KEY_KIND.iter().any(|k| name.eq_ignore_ascii_case(k)) {
+                    self.call(wrapper, "EnableKeyboard", true, dbg);
+                }
                 if name.eq_ignore_ascii_case("OnLoad") {
                     onload = Some(func);
                 }

crates/benilla-ui/src/loader/tests.rs

@@ -461,18 +461,24 @@ mod loader_tests {
         assert!(s.eval::<bool>("return FineRan == true").unwrap());
     }
 
-    /// Unsupported handler names (e.g. the keyboard-focus handlers `OnKeyDown`/`OnChar`, not yet
-    /// modeled) are warn-once gaps, not hard errors, and don't stop the frame from building.
+    /// Unsupported handler names are warn-once gaps, not hard errors, and don't stop the frame
+    /// from building. The example is `OnCursorChanged` — caret geometry is host-side here, so its
+    /// four float args would all be zero, which is the silent-drop this warn exists to avoid.
+    /// (It used to be `OnKeyDown`: that one is *fired* since decision 1319 and now belongs to
+    /// `script::tests::keyboard`.)
     #[test]
     fn unsupported_script_name_is_a_warning() {
         let s = UiScript::new().unwrap();
         let doc = parse(
-            r#"<Ui><Frame name="Keyed"><Scripts><OnKeyDown>x = 1</OnKeyDown></Scripts></Frame></Ui>"#,
+            r#"<Ui><Frame name="Keyed"><Scripts><OnCursorChanged>x = 1</OnCursorChanged></Scripts></Frame></Ui>"#,
         );
         let report = load(&s, &doc, &no_files);
         assert_eq!(report.frames, 1);
         assert!(report.errors.is_empty(), "{:?}", report.errors);
-        assert!(report.warnings.iter().any(|w| w.contains("OnKeyDown")));
+        assert!(report
+            .warnings
+            .iter()
+            .any(|w| w.contains("OnCursorChanged")));
     }
 
     /// An unknown frame type is an error that drops that subtree but not the rest of the load.

crates/benilla-ui/src/script/editbox/tests.rs

@@ -459,13 +459,38 @@ fn get_number_parses_the_text() {
 
 #[test]
 fn key_and_char_paths_never_fire_generic_handlers() {
-    // OnKeyDown/OnChar aren't even in SCRIPT_KINDS, so SetScript rejects them — proving the runtime
-    // has no generic keyboard slots to fire (the C++ override replaced them).
-    let s = script();
-    let err = s
-        .run(r#"CreateFrame("EditBox", "E"):SetScript("OnChar", function() end)"#)
-        .unwrap_err();
-    assert!(err.to_string().contains("OnChar"), "{err}");
+    // The law (wow-re `frame-key-script-delivery.md` §1): `CSimpleEditBox` has its OWN input
+    // vtable (`0x77a900` char / `0x77b160` key-down) which handles the event and **never chains to
+    // the base implementation** — so a focused box's typing does not also run a generic `OnChar`
+    // bound on that same box.
+    //
+    // This used to be proved by `SetScript("OnChar", …)` *raising*, which it no longer does
+    // (decision 1319 made the three key names real). The invariant is unchanged and is now
+    // asserted directly: bind the generic handler, type into the focused box, and watch the text
+    // arrive with the handler never called.
+    let mut s = script();
+    s.run(
+        r#"
+        fired = 0
+        E = CreateFrame("EditBox", "E")
+        E:EnableKeyboard(true)
+        E:SetScript("OnChar", function() fired = fired + 1 end)
+        E:SetFocus()
+    "#,
+    )
+    .unwrap();
+    assert!(s.char_input("k"), "the focused box consumed the character");
+    assert_eq!(
+        s.eval::<String>("return E:GetText()").unwrap(),
+        "k",
+        "…by inserting it through its own path"
+    );
+    assert_eq!(
+        s.eval::<i64>("return fired").unwrap(),
+        0,
+        "the box's own vtable never chains to the generic OnChar"
+    );
+    assert!(s.errors().is_empty(), "{:?}", s.errors());
 }
 
 // ── history recall (`historyLines` / AddHistoryLine — decision 0288 P2) ──────────────────────

crates/benilla-ui/src/script/event.rs

@@ -162,6 +162,23 @@ pub(super) fn fire_visibility_changes(lua: &Lua, changed: Vec<FrameHandle>) {
     }
 }
 
+/// Is a handler bound under `script` for this frame id? — the **existence** question, asked
+/// without firing anything.
+///
+/// It exists for the keyboard walk ([`super::keyboard`]), where the reference's consumption gate is
+/// the presence of the slot rather than anything the handler does: `0x76b7d0` reads `[+0x188]`/
+/// `[+0x190]`, consumes on either, and fires only the first. A lookup failure reads as "absent" —
+/// the gate must never be able to raise out of the middle of a walk.
+pub(super) fn has_widget_handler(lua: &Lua, id: u32, script: &str) -> bool {
+    let Ok(scripts) = lua.named_registry_value::<Table>(REG_SCRIPTS) else {
+        return false;
+    };
+    match scripts.get::<Value>(id) {
+        Ok(Value::Table(per)) => matches!(per.get::<Value>(script), Ok(Value::Function(_))),
+        _ => false,
+    }
+}
+
 /// The one firing path. `event_name` is the `event` global + first modern positional (OnEvent only);
 /// `extra` are the `arg1..argN` globals + the trailing modern positionals. Globals are saved before
 /// and restored after (even on error) — nesting-safe. Holds no model borrow across the call.

crates/benilla-ui/src/script/keyboard.rs

@@ -0,0 +1,230 @@
+//! **Frame keyboard delivery** — the walk, the existence gate, and the key-name argument.
+//!
+//! Until this module, keys went straight to the focused EditBox and a plain frame could not receive
+//! a key at all: `OnChar`/`OnKeyDown`/`OnKeyUp` were the three script names
+//! [`super::object::events_regions`] deliberately *raised* on, because accepting a name this engine
+//! never fires is the silent-drop class 1203/1205/1211 each recorded. This is the machinery that
+//! note was waiting on, so those three names are now accepted there.
+//!
+//! ## The law (wow-re `system/ui/scratch/frame-key-script-delivery.md`, §5 trio — VERIFIED)
+//!
+//! An OS key event reaches **one dispatcher per channel**, and the dispatcher — not the frame — is
+//! what walks: `0x765f10` (key-down, kind-1 buckets, vtable `+0x60`) and `0x765df0` (char, kind-0
+//! buckets, `+0x5c`). Each walks **9 strata records, TOOLTIP (8) down to WORLD (0)**, and within a
+//! record the frame array **index 0 upward**, which the inserter `0x764aa0` keeps ordered by
+//! **level descending, ties oldest-registration-first**. Every frame in that order is *called*
+//! until one returns nonzero; that one **consumes** and both loops stop. So several frames may be
+//! called, at most one consumes.
+//!
+//! **The consumption gate is EXISTENCE, not handling** (`0x76b7d0`/`0x76bba0`/`0x76b760`, §3):
+//!
+//! | slot state | key-down returns | fires |
+//! |---|---|---|
+//! | `OnKeyDown` set | **consume** | `OnKeyDown` |
+//! | only `OnKeyUp` set | **consume** | *nothing* |
+//! | neither | decline | nothing |
+//!
+//! That asymmetry is real and is transcribed below: a frame carrying only an `OnKeyUp` swallows
+//! every key-down and runs no script. `OnChar` gates on its own slot alone; `OnKeyUp` on `+0x190`
+//! alone (no `OnKeyDown` fallback). **A 1.12 handler cannot signal "handled"** — the fire's return
+//! is discarded at all three call sites (§3.1) — so a Lua handler never influences consumption.
+//!
+//! **Bucket membership is the keyboard-enabled flag**, never the presence of a script (§3.2):
+//! `EnableKeyboard(true)` on a script-less frame puts it *in* the walk, where it is called and
+//! declines. XML `enableKeyboard` enables both kinds; an XML `<Scripts>` block auto-enables per
+//! handler; **Lua `SetScript` auto-enables nothing**, so a runtime-created frame needs an explicit
+//! `EnableKeyboard(true)` exactly as in the real client.
+//!
+//! ## Where the EditBox sits — it is a participant, not a separate stage
+//!
+//! The reference has no "editbox first" rule: `CSimpleEditBox` is a frame in the same buckets whose
+//! own vtable (`0x77b160` key-down / `0x77a900` char) consumes while focused and never chains to
+//! the base. Modelling it as a stage before or after the walk would get the order wrong whenever a
+//! keyboard frame and the focused box are in different strata — a TOOLTIP-strata frame really does
+//! pre-empt a focused HIGH-strata box, and a WORLD-strata one really does not. So [`walk`] visits
+//! the focused box **at its own strata/level**, handing it to the RF-0082 routing this crate
+//! already had ([`super::editbox`]) and taking a `true` as consumption.
+//!
+//! What is deliberately NOT re-derived here: the box's own focus acquisition (`autoFocus`
+//! self-acquire, click-to-focus, `SetFocus`) stays entirely in `editbox`, and an event no frame
+//! consumed still falls through to it — so every existing routing test holds unchanged.
+//!
+//! ## Not modelled, named
+//!
+//! The **sticky per-key-code key-up slot** (§2.1: `0x765fd0` routes key-up to `[root+code*4+0x84]`,
+//! the last frame that consumed *some* down for that code, and does **not** clear it on delivery)
+//! is not built: this engine's host feeds no key-up at all today. `OnKeyUp` is therefore accepted,
+//! stored, gated in the table above — and never fired. That is stated rather than hidden, because
+//! its one observable consequence is live: a frame with only an `OnKeyUp` still consumes key-downs.
+
+use mlua::Lua;
+
+use super::{event, Model};
+use crate::widget::FrameHandle;
+
+/// Which channel a dispatch is on — the two walks the reference registers separately.
+#[derive(Clone, Copy, PartialEq, Eq)]
+enum Channel {
+    /// `0x765df0` — kind-0 buckets, vtable `+0x5c`, fires `OnChar` with the literal character.
+    Char,
+    /// `0x765f10` — kind-1 buckets, vtable `+0x60`, fires `OnKeyDown` with the key NAME.
+    KeyDown,
+}
+
+/// The walk order (§2): every keyboard-enabled, effectively-visible frame, **strata descending,
+/// then level descending, then oldest registration first**.
+///
+/// Built as a sorted `Vec` per dispatch rather than kept as nine live buckets. The reference
+/// maintains the buckets incrementally because it is walking them on every key at 1.12 frame
+/// budgets; here a key press is a human-rate event and the candidate set is the *keyboard-enabled*
+/// frames only — single digits in any real interface — so an index that must be kept coherent with
+/// every Show/Hide/SetFrameLevel/SetFrameStrata would be pure drift surface for no measurable win.
+/// The ORDER is the law; the storage is ours.
+fn walk_order(model: &Model) -> Vec<FrameHandle> {
+    let mut candidates: Vec<(FrameHandle, u8, u16, u32)> = model
+        .arena
+        .iter_frames()
+        .filter(|(_, f)| f.effective_visible && f.keyboard_enabled)
+        .map(|(h, f)| (h, f.strata as u8, f.level, f.insertion_seq))
+        .collect();
+    // strata DESC, level DESC, insertion ASC — the inserter's own ordering (`0x764aa0`), where a
+    // frame goes before the first entry of strictly lower level and equal levels keep registration
+    // order.
+    candidates.sort_by(|a, b| b.1.cmp(&a.1).then(b.2.cmp(&a.2)).then(a.3.cmp(&b.3)));
+    candidates.into_iter().map(|(h, ..)| h).collect()
+}
+
+/// Does `frame` have a script bound under `name`?
+///
+/// The existence test the C++ gate performs on `+0x180`/`+0x188`/`+0x190` — asked of the script
+/// registry, which is where this engine keeps those slots.
+fn has_script(lua: &Lua, id: u32, name: &str) -> bool {
+    event::has_widget_handler(lua, id, name)
+}
+
+/// The dispatcher for one channel — [`Channel`]'s walk, gate and fire, in the reference's order.
+///
+/// Returns whether the event was **consumed**, which is what the caller must use to suppress the
+/// keybinding system: consumption is decided by the C++ existence gate alone, never by what the
+/// Lua handler did or returned (§3.1).
+fn walk(lua: &Lua, channel: Channel, arg: &str) -> bool {
+    let order = {
+        let model = lua.app_data_ref::<Model>().expect("model app_data");
+        walk_order(&model)
+    };
+    for h in order {
+        // The focused EditBox consumes through its OWN vtable, never the base one — so it is asked
+        // in its walk position and its answer is final for this event.
+        let is_focused_box = {
+            let model = lua.app_data_ref::<Model>().expect("model app_data");
+            model.focused_editbox == Some(h)
+        };
+        if is_focused_box {
+            let consumed = match channel {
+                Channel::Char => super::editbox::char_input(lua, arg),
+                Channel::KeyDown => super::editbox::key_input(lua, arg),
+            };
+            if consumed {
+                return true;
+            }
+            continue;
+        }
+        let id = {
+            let mut model = lua.app_data_mut::<Model>().expect("model app_data");
+            model.frame_id(h)
+        };
+        match channel {
+            // `0x76b760`: gates on the OnChar slot alone.
+            Channel::Char => {
+                if has_script(lua, id, "OnChar") {
+                    fire(lua, id, "OnChar", arg);
+                    return true;
+                }
+            }
+            // `0x76b7d0`: the OR gate — either key slot consumes, only `OnKeyDown` fires.
+            Channel::KeyDown => {
+                let down = has_script(lua, id, "OnKeyDown");
+                if down || has_script(lua, id, "OnKeyUp") {
+                    if down {
+                        fire(lua, id, "OnKeyDown", arg);
+                    }
+                    return true;
+                }
+            }
+        }
+    }
+    false
+}
+
+/// Fire one key handler with its single string argument (`0x7026f0(frame, &slot, "%s", str)` — the
+/// argument is a string on all three channels). Errors land in [`Model::errors`] like every other
+/// widget handler; a raising handler must not eat the key press.
+fn fire(lua: &Lua, id: u32, script: &str, arg: &str) {
+    let val = mlua::Value::String(match lua.create_string(arg) {
+        Ok(s) => s,
+        Err(e) => {
+            lua.app_data_mut::<Model>()
+                .expect("model app_data")
+                .errors
+                .push(e.to_string());
+            return;
+        }
+    });
+    if let Err(e) = event::fire_widget_handler(lua, id, script, vec![val]) {
+        lua.app_data_mut::<Model>()
+            .expect("model app_data")
+            .errors
+            .push(e.to_string());
+    }
+}
+
+/// The char channel (`0x765df0`) — `arg1` is the literal character, UTF-8 (§4.1: the code is
+/// UTF-8-encoded with no range guard, so a multi-byte codepoint arrives whole).
+pub(super) fn char_input(lua: &Lua, text: &str) -> bool {
+    walk(lua, Channel::Char, text)
+}
+
+/// The key-down channel (`0x765f10`) — `arg1` is the key NAME, with no modifier prefix, from the
+/// same table the keybinding chord uses (§4.2). The host speaks those names already
+/// (`crate::script::UiScript::key_input`'s contract), so no decode happens here.
+pub(super) fn key_input(lua: &Lua, key: &str) -> bool {
+    walk(lua, Channel::KeyDown, key)
+}
+
+/// The same key-down walk, for the keys this engine delivers to a **focused EditBox as a semantic
+/// [`crate::script::EditAction`] chord** rather than by name — BACKSPACE, DELETE, the arrows, HOME,
+/// END (decision 0301: the host's per-OS keymap owns which chord means what, so the box never sees
+/// these as names).
+///
+/// Those keys still have to reach a keyboard frame — a dialog you type into needs its BACKSPACE —
+/// and the ordering question that creates is real: whoever comes FIRST in the walk owns the key.
+/// So this runs the identical order and **declines at the focused box** rather than skipping past
+/// it: a frame above the box in walk order consumes and fires; a box above the frame ends the walk
+/// with `false`, and the caller then dispatches its chord exactly as before. The one thing this
+/// must never do is let a frame *below* the focused box steal the box's editing keys.
+pub(super) fn frame_key_input(lua: &Lua, key: &str) -> bool {
+    let order = {
+        let model = lua.app_data_ref::<Model>().expect("model app_data");
+        walk_order(&model)
+    };
+    for h in order {
+        {
+            let model = lua.app_data_ref::<Model>().expect("model app_data");
+            if model.focused_editbox == Some(h) {
+                return false; // the box owns this key; its chord path handles it
+            }
+        }
+        let id = {
+            let mut model = lua.app_data_mut::<Model>().expect("model app_data");
+            model.frame_id(h)
+        };
+        let down = has_script(lua, id, "OnKeyDown");
+        if down || has_script(lua, id, "OnKeyUp") {
+            if down {
+                fire(lua, id, "OnKeyDown", key);
+            }
+            return true;
+        }
+    }
+    false
+}

crates/benilla-ui/src/script/mod.rs

@@ -72,6 +72,7 @@ mod inspect;
 mod item_stats;
 mod item_text;
 pub mod keybind;
+mod keyboard;
 mod layout;
 mod loot;
 mod loot_roll;
@@ -353,7 +354,7 @@ pub const SCREEN: crate::layout::Handle = 0;
 /// corpus call sites across 91 addons), but it removes working behaviour from those 30 sites, so it
 /// is a change to make deliberately with FrameXML fixed first — not a side effect of widening this
 /// list.
-const SCRIPT_KINDS: [&str; 32] = [
+const SCRIPT_KINDS: [&str; 35] = [
     "OnLoad",
     "OnEvent",
     "OnUpdate",
@@ -404,6 +405,16 @@ const SCRIPT_KINDS: [&str; 32] = [
     // [`crate::layout::size_changed`]'s byte-verified epsilon test — see
     // [`UiScript::resolve_layout`]. `OnSizeChanged(self, width, height)`.
     "OnSizeChanged",
+    // The three KEY channels, unblocked by [`keyboard`]'s walk (wow-re
+    // `scratch/frame-key-script-delivery.md`, VERIFIED). They were the standing exception in
+    // [`object::events_regions`]'s note — accepted only once something fired them, which is that
+    // module's whole rule. `OnKeyUp` rides in with the other two deliberately: it is *gated* today
+    // (a frame carrying only an OnKeyUp consumes every key-down and runs nothing — the reference's
+    // own asymmetry) even though this engine's host feeds no key-up to fire it with, and accepting
+    // the name is what makes that consumption reachable.
+    "OnChar",
+    "OnKeyDown",
+    "OnKeyUp",
 ];
 
 // ─────────────────────────────────────────────────────────────────────────────────────────────
@@ -968,7 +979,11 @@ impl UiScript {
     /// focused box, inserts it (numeric/cap/password rules apply) or — for the Ctrl+A control code —
     /// selects all. Returns `true` if consumed (a focused box consumes every char).
     pub fn char_input(&mut self, text: &str) -> bool {
-        editbox::char_input(&self.lua, text)
+        // The frame walk first ([`keyboard`]): the focused box is a PARTICIPANT in it, at its own
+        // strata/level, so this is not "frames before boxes" — it is the reference's one dispatcher
+        // in the reference's order. An event no frame consumed still falls through to the box
+        // routing, which owns focus acquisition (`autoFocus` self-acquire and kin).
+        keyboard::char_input(&self.lua, text) || editbox::char_input(&self.lua, text)
     }
 
     /// Paste text from the host OS clipboard into the focused EditBox. The engine-free runtime can't
@@ -985,7 +1000,20 @@ impl UiScript {
     /// owns which chord means what). Routes per §1/§2; a focused box consumes the key even when
     /// it does nothing with it. Returns `true` if consumed.
     pub fn key_input(&mut self, key: &str) -> bool {
-        editbox::key_input(&self.lua, key)
+        // Same two-stage shape as `char_input` — see its note.
+        keyboard::key_input(&self.lua, key) || editbox::key_input(&self.lua, key)
+    }
+
+    /// A key the host delivers to a focused EditBox as an [`EditAction`] chord rather than by name
+    /// (BACKSPACE, DELETE, the arrows, HOME, END), offered to the **keyboard frames** first.
+    ///
+    /// Returns `true` if a frame consumed it, in which case the caller must NOT also dispatch the
+    /// chord — and the key's binding must not fire either (consumption suppresses it, wow-re §3).
+    /// A `false` means either nothing wanted it or the focused box owns it; the caller proceeds
+    /// exactly as it did before this entry point existed. See [`keyboard::frame_key_input`] for
+    /// why declining at the box is the faithful answer rather than skipping it.
+    pub fn frame_key_input(&mut self, key: &str) -> bool {
+        keyboard::frame_key_input(&self.lua, key)
     }
 
     /// One semantic text-editing operation on the focused EditBox — the output of the host's

crates/benilla-ui/src/script/tests/generic_for.rs

@@ -221,3 +221,65 @@ fn a_nested_long_string_parses_as_lua_50_does() {
         .expect("nested long string must parse");
     assert_eq!(v, "a [[b]] c");
 }
+
+/// **5.0's constructor compat-semicolon, restored** — the third dialect divergence (decision 1315).
+///
+/// Lua 5.0's `constructor()` opens its field loop with `testnext(ls, ';')  /* compatibility
+/// only */` — a Lua 4.0 leftover (4.0 separated a constructor's list part from its record part
+/// with `;`) that silently eats ONE extra `;` per separator position inside a table constructor.
+/// 5.1 deleted the line. AtlasLoot's `ButtonRegistry.lua` writes `Back_Title = AL["Factions"];;`
+/// twenty times; the real client loads it (director-verified on the live 1.12.1 client), stock
+/// 5.1 answered "unexpected symbol near ';'" and the addon's registry global never existed (B266).
+#[test]
+fn a_double_semicolon_inside_a_constructor_parses_as_lua_50_does() {
+    let s = script();
+
+    // The exact corpus shape: a record field, its `;` separator, and the compat extra.
+    let out: String = s
+        .eval(
+            r#"
+            local AL = { Factions = "Factions" }
+            local reg = {
+                ["REP1"] = { Title = "The Aldor"; Back_Page = "REPMENU"; Back_Title = AL["Factions"];; };
+            }
+            return reg["REP1"].Back_Title
+        "#,
+        )
+        .expect("the ButtonRegistry shape must parse");
+    assert_eq!(out, "Factions");
+
+    // The compat skip runs at loop TOP, so a lone `;` before `}` — legal 5.0, an empty record
+    // part in 4.0 terms — is eaten the same way.
+    assert_eq!(s.eval::<i64>("local t = {;} return 1").unwrap(), 1);
+
+    // List context too: the separator may be `;` and the extra one is still absorbed.
+    let n: i64 = s.eval("local t = { 1;; 2 } return t[1] + t[2]").unwrap();
+    assert_eq!(n, 3);
+}
+
+/// **The restoration is exactly one semicolon wide, and constructor-only.** Both halves of the
+/// boundary are the client's: 5.0 eats ONE extra `;` per position (a third dies in `listfield`),
+/// and statement-level `;;` is rejected by 5.0 and 5.1 alike — the half wow-5875-re byte-verified
+/// in the client's `luaL_loadbuffer` first (`system/ui/scratch/lua-dialect.md`).
+#[test]
+fn the_compat_semicolon_is_one_wide_and_statement_level_stays_rejected() {
+    let s = script();
+
+    let err = s
+        .eval::<i64>("local t = { 1;;; 2 } return t[1]")
+        .unwrap_err()
+        .to_string();
+    assert!(
+        err.contains("unexpected symbol"),
+        "a THIRD semicolon must still be a syntax error, as on 5.0: {err}"
+    );
+
+    let err = s
+        .eval::<i64>("local x = 1;; return x")
+        .unwrap_err()
+        .to_string();
+    assert!(
+        err.contains("unexpected symbol"),
+        "statement-level `;;` must stay rejected — that half is byte-verified: {err}"
+    );
+}

crates/benilla-ui/src/script/tests/input.rs

@@ -598,13 +598,10 @@ fn the_unfired_script_kinds_still_raise_rather_than_silently_accepting() {
     s.run(r#"Raiser = CreateFrame("EditBox", "RaiserBox")"#)
         .unwrap();
     for name in [
-        // The FLAG exists (`EnableKeyboard`/`IsKeyboardEnabled`, 1252); what is still absent is the
-        // keyboard INDEX and the strata walk that would deliver a key to a frame carrying it, so
-        // these three stay unfired (wow-re `scripts-auto-enable.md` kinds 0/1). Being enabled is
-        // not being a handler — the carve is explicit that the two are separable.
-        "OnKeyDown",
-        "OnKeyUp",
-        "OnChar",
+        // `OnKeyDown`/`OnKeyUp`/`OnChar` are NOT in this list any more: decision 1319 built the
+        // delivery walk, so they are fired and therefore accepted — the rule this test enforces,
+        // applied in the other direction. Their law lives in `script::tests::keyboard`.
+        //
         // Caret geometry is host-side; its four float args would all be zero.
         "OnCursorChanged",
         // 2.0's secure-frame system — no such slot exists in any 1.12 resolver.

crates/benilla-ui/src/script/tests/keyboard.rs

@@ -0,0 +1,186 @@
+//! The frame keyboard delivery law (decision 1319; wow-re
+//! `system/ui/scratch/frame-key-script-delivery.md`, §5 trio — VERIFIED).
+//!
+//! Each test pins one clause that a plausible-but-wrong implementation gets backwards.
+
+use super::common::script;
+
+/// The bucket gate (§3.2): membership is the **keyboard-enabled flag**, not the presence of a
+/// script. A Lua-created frame auto-enables nothing (the reference's `SetScript` doesn't either),
+/// so its handler is unreachable until `EnableKeyboard(true)` puts it in the walk.
+#[test]
+fn a_key_script_alone_does_not_put_a_frame_in_the_walk() {
+    let mut s = script();
+    s.run(
+        r#"
+        got = nil
+        f = CreateFrame("Frame", "KbUnenabled")
+        f:SetScript("OnChar", function() got = arg1 end)
+    "#,
+    )
+    .unwrap();
+    assert!(
+        !s.char_input("7"),
+        "an unenabled frame is not in the bucket"
+    );
+    assert_eq!(s.eval::<Option<String>>("return got").unwrap(), None);
+
+    s.run("f:EnableKeyboard(true)").unwrap();
+    assert!(
+        s.char_input("7"),
+        "enabled: now in the walk, and it consumes"
+    );
+    assert_eq!(
+        s.eval::<Option<String>>("return got").unwrap().as_deref(),
+        Some("7")
+    );
+    assert!(s.errors().is_empty(), "{:?}", s.errors());
+}
+
+/// The existence gate's asymmetry (§3): a frame carrying **only** an `OnKeyUp` consumes every
+/// key-down and runs nothing. This is the clause an implementation "tidies away" — and doing so
+/// silently un-suppresses that frame's keybindings.
+#[test]
+fn only_an_onkeyup_still_swallows_the_key_down() {
+    let mut s = script();
+    s.run(
+        r#"
+        downs = 0
+        f = CreateFrame("Frame", "KbUpOnly")
+        f:EnableKeyboard(true)
+        f:SetScript("OnKeyUp", function() downs = downs + 1 end)
+    "#,
+    )
+    .unwrap();
+    assert!(s.key_input("ESCAPE"), "the OnKeyUp slot alone consumes");
+    assert_eq!(
+        s.eval::<i64>("return downs").unwrap(),
+        0,
+        "…and fires nothing"
+    );
+
+    // Neither slot: declines outright.
+    s.run("f:SetScript(\"OnKeyUp\", nil)").unwrap();
+    assert!(!s.key_input("ESCAPE"), "no key slot at all: declines");
+    assert!(s.errors().is_empty(), "{:?}", s.errors());
+}
+
+/// The walk order (§2): **strata descending first**, and only then level. A LOW frame can never
+/// take a key from a HIGH one, however high its level — the same "a raise never changes stratum"
+/// shape the toplevel law has.
+#[test]
+fn the_walk_is_strata_then_level_then_registration() {
+    let mut s = script();
+    s.run(
+        r#"
+        winner = nil
+        function mk(name, strata, level)
+            local f = CreateFrame("Frame", name)
+            f:SetFrameStrata(strata)
+            f:SetFrameLevel(level)
+            f:EnableKeyboard(true)
+            f:SetScript("OnKeyDown", function() winner = name end)
+            return f
+        end
+        -- registered first, but the LOWEST stratum: must lose despite its huge level
+        mk("KbLow", "LOW", 99)
+        mk("KbHigh", "HIGH", 1)
+    "#,
+    )
+    .unwrap();
+    assert!(s.key_input("ESCAPE"));
+    assert_eq!(
+        s.eval::<String>("return winner").unwrap(),
+        "KbHigh",
+        "strata beats level"
+    );
+
+    // Within one stratum: higher level wins, and equal levels keep registration order.
+    s.run(
+        r#"
+        winner = nil
+        mk("KbHighTop", "HIGH", 5)
+        mk("KbHighTie", "HIGH", 5)
+    "#,
+    )
+    .unwrap();
+    assert!(s.key_input("ESCAPE"));
+    assert_eq!(
+        s.eval::<String>("return winner").unwrap(),
+        "KbHighTop",
+        "level 5 beats level 1; the earlier of the two 5s wins the tie"
+    );
+    assert!(s.errors().is_empty(), "{:?}", s.errors());
+}
+
+/// **At most one frame consumes** (§2, "Delivery"): the walk stops at the first consumer, so a
+/// second keyboard frame below it never sees the key.
+#[test]
+fn the_first_consumer_ends_the_walk() {
+    let mut s = script();
+    s.run(
+        r#"
+        seen = {}
+        function mk(name, strata)
+            local f = CreateFrame("Frame", name)
+            f:SetFrameStrata(strata)
+            f:EnableKeyboard(true)
+            f:SetScript("OnChar", function() table.insert(seen, name) end)
+        end
+        mk("KbTip", "TOOLTIP")
+        mk("KbDlg", "DIALOG")
+    "#,
+    )
+    .unwrap();
+    assert!(s.char_input("x"));
+    assert_eq!(s.eval::<i64>("return table.getn(seen)").unwrap(), 1);
+    assert_eq!(s.eval::<String>("return seen[1]").unwrap(), "KbTip");
+    assert!(s.errors().is_empty(), "{:?}", s.errors());
+}
+
+/// A hidden frame is in no bucket (the link gate the walk shares with the draw order), so it
+/// neither fires nor consumes — and the frame below it gets the key instead.
+#[test]
+fn a_hidden_frame_is_not_in_the_walk() {
+    let mut s = script();
+    s.run(
+        r#"
+        winner = nil
+        function mk(name, strata)
+            local f = CreateFrame("Frame", name)
+            f:SetFrameStrata(strata)
+            f:EnableKeyboard(true)
+            f:SetScript("OnChar", function() winner = name end)
+            return f
+        end
+        top = mk("KbHidden", "TOOLTIP")
+        mk("KbBelow", "DIALOG")
+        top:Hide()
+    "#,
+    )
+    .unwrap();
+    assert!(s.char_input("q"));
+    assert_eq!(s.eval::<String>("return winner").unwrap(), "KbBelow");
+    assert!(s.errors().is_empty(), "{:?}", s.errors());
+}
+
+/// A raising handler must not eat the key or abort the walk's caller: the error is recorded, the
+/// frame still consumed (consumption is the C++ gate's, never the handler's — §3.1).
+#[test]
+fn a_raising_handler_still_consumes_and_is_recorded() {
+    let mut s = script();
+    s.run(
+        r#"
+        f = CreateFrame("Frame", "KbBoom")
+        f:EnableKeyboard(true)
+        f:SetScript("OnChar", function() error("boom") end)
+    "#,
+    )
+    .unwrap();
+    assert!(s.char_input("z"), "the gate consumed regardless");
+    assert!(
+        s.errors().iter().any(|e| e.contains("boom")),
+        "the raise is recorded: {:?}",
+        s.errors()
+    );
+}

crates/benilla-ui/src/script/tests/mod.rs

@@ -17,6 +17,7 @@ mod font_object;
 mod frame_api;
 mod generic_for;
 mod input;
+mod keyboard;
 mod layout_gate;
 mod measure;
 mod minimap;

third_party/lua-src/BENILLA.md

@@ -35,7 +35,7 @@ downgrade, and it is the entire reason this is a one-hunk fork rather than a ven
 
 ## The delta, and the command that proves it
 
-**Current state: two hunks, in two files.** `src/lib.rs` additionally differs by having Lua
+**Current state: three hunks, in three files.** `src/lib.rs` additionally differs by having Lua
 5.2/5.3/5.4/5.5 stripped from the `Version` enum (with their source trees deleted) — benilla builds
 5.1 and only 5.1, and a fork that still *offered* the other four would answer a request for one with
 a missing directory at build time instead of a compile error.
@@ -44,6 +44,7 @@ a missing directory at build time instead of a compile error.
 |---|---|---|---|
 | `lvm.c` | 5.0's `OP_TFORPREP` table→`next` substitution, folded into the top of `OP_TFORLOOP` | `for k, v in someTable do` raised "attempt to call a table value"; it was the first session-start error for **60** corpus addons | 1215 |
 | `luaconf.h` | `LUA_COMPAT_LSTR` `1` → `2` | 5.1 kept 5.0's nesting machinery and put an advisory error in front of it; two corpus addons died on "nesting of `[[...]]` is deprecated" | — |
+| `lparser.c` | 5.0's compat-semicolon skip restored at the top of `constructor()`'s field loop — 5.0's own line, verbatim, comment included | one extra `;` after a field separator inside a table constructor (`Back_Title = AL["Factions"];;`, ×20 in AtlasLoot's `ButtonRegistry.lua`) parses on the client and died here with "unexpected symbol near `;`"; statement-level `;;` stays rejected by both dialects | 1315 |
 
 The ordering was deliberate and it held: the fork landed **inert**, byte-identical to upstream, so
 the build pipeline was proven transparent *before* any behavioural payload went into it — and when
@@ -59,7 +60,7 @@ diff -r third_party/lua-src/lua-5.1.5 \
   ~/.cargo/registry/src/*/lua-src-550.0.0/lua-5.1.5
 ```
 
-That must print **exactly** the two hunks in the table above and nothing else. If it ever prints
+That must print **exactly** the three hunks in the table above and nothing else. If it ever prints
 more, something crept in.
 
 ## Status of the payload

third_party/lua-src/lua-5.1.5/lparser.c

@@ -509,6 +509,23 @@ static void constructor (LexState *ls, expdesc *t) {
   checknext(ls, '{');
   do {
     lua_assert(cc.v.k == VVOID || cc.tostore > 0);
+    /* BENILLA: Lua 5.0's constructor compat-semicolon, restored (decision 1315;
+    ** see third_party/lua-src/BENILLA.md). 5.0 opened this loop with exactly the
+    ** line below -- its own comment read "compatibility only", a Lua 4.0 leftover
+    ** (4.0 separated a constructor's list part from its record part with `;`).
+    ** 5.1 deleted the line, so one extra `;` after a field separator -- the
+    ** `Back_Title = AL["Factions"];;` shape AtlasLoot's ButtonRegistry.lua writes
+    ** twenty times -- became "unexpected symbol near ';'" here while the 1.12.1
+    ** client (Lua 5.0) accepts it. Statement-level `;;` stays rejected by both,
+    ** which is the half wow-5875-re byte-verified first (system/ui/scratch/
+    ** lua-dialect.md 9.1), and byte-verified in the client itself rather than
+    ** inferred from stock 5.0: constructor is 0x6fd430, and its loop head
+    ** 0x6fd4a0 calls testnext(ls,';') WITHOUT testing EAX -- a skip, not a
+    ** separator check -- while the loop's continuation at 0x6fd4f5/0x6fd505 does
+    ** test it. Of the 16 call sites of testnext 0x6fccf0 image-wide, the only
+    ** other consumer of `;' is chunk 0x6fcccc, at STATEMENT level, with no such
+    ** head skip. This line restores the constructor context only. */
+    testnext(ls, ';');  /* compatibility only */
     if (ls->t.token == '}') break;
     closelistfield(fs, &cc);
     switch(ls->t.token) {