Operator, 2026-08-29, after the menu dismissal work landed (#17850) and while the separator/disabled leaf (#17851) was in review:
"if a scroll changes alignments, it would not be a deal-breaker, if that has no animations. cherry on top. we have global css, and the main thread side could even add css selectors which indicate spawn directions, if not already there. then our menu list css could use them, in case the new config is set."
The refinement that shaped it: a submenu spawning to the right should slide in from the left while fading up — directional, not a scale — because the movement reads as the child emerging from its parent.
Almost every piece already exists. This ticket is mostly wiring, and the design work was establishing that.
Live latest-open sweep: checked the latest 20 open issues at 2026-08-29T19:16Z; no equivalent found. Closed-issue sweep across menu animation / spawn animation / useAnimations menu / menu transition: no declined predecessor — but see #5008 below, which the closed sweep surfaced and which changes this ticket's naming. A2A in-flight claim sweep (12 messages, all read-states): no overlapping claim.
The Problem
A floating menu appears instantly at its resolved position. That is correct and fast, and it should stay the default — but it gives the user no cue about where the menu came from, which matters most in a cascade, where a submenu's relationship to its parent item is the entire information the user needs.
The naive implementation fails in three specific ways that are not obvious until you read the main thread, and each one is a real trap rather than a style preference:
Animating the root's transform fights the aligner.main/DomAccess.mjs:277-280 positions every aligned node with {top: 0, left: 0, transform: translate(x, y)}. A spawn translateX on the same node either clobbers placement or is clobbered by the next align.
Animating size feeds a re-align loop.addAligned() (DomAccess.mjs:147-153) attaches a ResizeObserver to the subject itself with syncAligns as the callback. A scale/height animation changes the observed box → re-align during the animation → which re-measures a moving rect, since getBoundingClientRect() returns post-transform visual geometry (documented at DomAccess.mjs:486-491).
Keying the animation on mount animates before the direction is known.component/Base.mjs:622 fires me.alignTo() on mount without awaiting it — so the node enters the DOM first and the zone is resolved a worker→main round-trip later. There is no anti-flash gate over that window.
The Architectural Reality
The direction signal already exists, is already correct, and is already consumed.
util/Rectangle.mjs runs a zone search — candidate zones are tried until one constrains (:530-560) — and stamps the resolved side onto the result:
It is correctly lifecycle-managed — the previous position is removed before a re-align (:243) and on teardown (:1287) — and resources/scss/src/component/Base.scss already consumes all four classes to place the box shadow at the correct edge. Directional spawn is a second consumer of a maintained signal, not a new mechanism. Because the aligner flips when space is short, the animation direction follows the flip for free.
The spawn-vs-resync distinction also already exists.addAligned() branches on if (!aligns.has(id)) (:151) and marks the id seen at :189; syncAligns() runs _aligns.delete(align.id) once the subject is detached. So first-align, re-align and dismissal are already three distinct, correct moments — and a reused menu instance (the pattern examples/menu/context/ deliberately uses) passes through the delete on dismissal and re-arms as a genuine first-align on reopen.
Durations are already a governed vocabulary.resources/scss/_motion.scss declares itself "the ONLY sanctioned duration/easing source" and states "a call-site duration/easing literal is a contract violation". It supplies --motion-base: 200ms, --motion-panel: 280ms, --ease-out-soft, --motion-stagger: 24ms, and collapses every duration to 0ms under prefers-reduced-motionat the vocabulary level — so consuming the tokens inherits the accessibility contract instead of re-implementing it.
Navigation is unaffected by an opacity/transform entrance.addon/Navigator.mjs:272 filters with node.offsetParent && node.matches(selector); an element at opacity: 0 keeps its offsetParent, so it stays navigable throughout. (An entrance built on display/visibility would break arrow-key navigation — recorded as a trap below.)
Naming — and the prior ruling the closed sweep surfaced
#5008 (CLOSED/COMPLETED, "dialog.Base: maximise is no longer animated") carries an operator design note:
"the dialog should get a new config called animate_. animateTargetId is still needed for spawn animations, but a more generic config to enable / disable all animations seems reasonable. afterSetAnimate() should add a new top level CSS rule like neo-animated."
Two consequences:
"spawn animation" is the codebase's own term for this concept — dialog/Base.mjs:37 still declares animateTargetId_ for exactly it. The proposed config name should use that word rather than inventing a synonym.
list.Base already has animate_ (src/list/Base.mjs:27, reactive, default false) — so menu.Listinherits it today. It must not be overloaded: afterSetAnimate() (:219) lazy-imports list/plugin/Animate.mjs, which drives item motion within the list (reordering/insertion) and marks the owner neo-animated-list. Entrance-of-a-floating-surface and motion-of-items-inside-it are different concepts — the latter is the store-re-sort transition seen on the Fleet Manager's 3x3 agent cards, where items glide to new positions in a list that is already on screen.
The nearest sibling (animate_) is reactive, which settles the reactivity question against the more distant useAnimations precedent in component/DateSelector, date/SelectorContainer and calendar/view/YearComponent (all non-reactive, all a different family).
The Fix
1. main/DomAccess.mjs — make an unchanged alignment idempotent. Swap the neo-aligned-* class only when the subject does not already carry the resolved one, guarded on the SUBJECT rather than the cached align record. A resync then leaves the class — and anything keyed on it — untouched, while a real flip still changes it. No new class, no new state, no new observer.
2. src/menu/List.mjs — the opt-in config.animateSpawn_: false, reactive, whose afterSet toggles a root class (proposed neo-animate-spawn). Inherited animate_ is untouched.
4. resources/scss/src/menu/List.scss — four keyframe rules keyed on the class conjunction, consuming --motion-* tokens only:
Two prescriptions in The Fix were wrong and are struck rather than silently rewritten — the reasoning that killed them is the useful part.
An inner wrapper node is not needed. The premise was that the animation could not share the root with the aligner's transform. It can: the individual translate property composes with transform rather than clobbering it. Measured — transform: translate(300px,200px) plus translate: -40px 0 resolves to x = 260 with the matrix intact, and the codebase already uses the individual property in 7 places. No structural change to the vdom.
⚠️ The paragraph below is itself superseded by Correction 2 — it names unalign(), which does not
exist, and rests on "a same-zone resync leaves animation-name untouched", which was falsified at
runtime. Kept only as the reasoning trail; nothing in it describes what shipped.
Remove the marker on the first syncAligns — that cancels the entrance.onDocumentMutation resyncs every aligned subject on ordinary body churn, and opening a submenu is enough to trigger it, so a per-align removal strips the marker milliseconds in and the animation dies. It is cleared by unalign() instead. Safe because a resync that keeps the same zone leaves animation-name untouched, and an unchanged name does not restart a running animation (measured: an animation kept advancing across unrelated class churn, 583ms → 1583ms, and reset to 0 only when the name changed). A genuine zone flip replays, which the operator accepted up front.
One addition the ticket did not name:showSubMenu() must propagate the config. Without it the cascade stays still while only the root animates — inverting the effect the feature exists for. Both this and the marker defect passed review-by-reading and were caught only in the browser.
Correction 2 — the marker is gone (PR #17861 review, RA-1/RA-3)
neo-align-initial does not exist and should not. The design above rested on "a resync that keeps the same zone leaves animation-name untouched, so a running animation is not restarted." @neo-gpt falsified it at runtime and I reproduced it: align()removes the zone class, reads layout, then re-adds it — currentTime 949ms → animation-name: none with zero animations → a differentAnimation object at 0. The final name is identical throughout, which is precisely why a name-only assertion passes while the entrance restarts.
The fix moved to the root:align() swaps the zone class only when the subject does not already carry it. An ordinary resync leaves the class — and the running animation, and the shadow repaint — untouched; a real flip still changes it and restarts the entrance mirrored. The transient marker is deleted, which also retires the claim that the default path was byte-unchanged: nothing is added to every aligned subject any more.
The guard is on the SUBJECT, not the cached align record. Keying off lastAlign looks equivalent and is not — a hidden menu leaves the DOM and remounts classless while _aligns still reports the same zone, so that version never re-added the class and a reused menu stayed unaligned permanently. Caught by the reuse witness one run after writing it.
unalign() does not exist. I named it as the cleanup authority in code comments, the PR body and this ticket. The cleanup lives in syncAligns(). Corrected everywhere.
Contract Ledger Matrix
Target Surface
Source of Authority
Proposed Behavior
Fallback
Docs
Evidence
Neo.menu.List#animateSpawn_
new
Reactive Boolean, default false. afterSet toggles the root class that gates the entrance rules. Distinct from inherited list.Base#animate_, which loads list/plugin/Animate.mjs for item motion and is untouched.
list/Base.mjs:27 + :219 establish the reactive sibling; dialog/Base.mjs:37 establishes "spawn animation" as the term
neo-align-initial
WITHDRAWN
Removing and re-adding any class around a layout read destroys the running animation rather than preserving it, so a transient marker cannot carry this contract. Replaced by making the existing neo-aligned-* swap idempotent in align().
n/a
n/a
measured: a different Animation object with progress reset to 0
--menu-spawn-*
menu/List.scss
Offset distance for the directional slide, aliased from _motion.scss durations/easings
unset → geometry sheet fallback
SCSS comment
_motion.scss is the sanctioned source; a literal is a contract violation
neo-aligned-{top,right,bottom,left}
existing, DomAccess.mjs:292
Unchanged. Read as the direction input.
n/a
already documented
already consumed by component/Base.scss for shadow edge
Decision Record impact
none — aligned-with _motion.scss's stated vocabulary authority; no ADR interaction.
Acceptance Criteria
main/DomAccess.mjs leaves the neo-aligned-* class untouched when a resync resolves the same zone, and swaps it on a real flip — witnessed by WAAPI animation identity and monotonic progress, not by the resolved animation-name, which is identical across a restart.
menu.List#animateSpawn_ defaults false; with the default, the rendered DOM and computed styles are unchanged from today — asserted, not assumed.
With animateSpawn: true, a submenu resolving to zone right has computed animation-name of the from-the-left keyframes; zone left resolves to the mirrored keyframes. No wrapper node: the individual translate property composes with the aligner-owned transform.
The flip case: a submenu opened near the right viewport edge resolves neo-aligned-left and its animation mirrors accordingly — the case that is painful to hand-code and silent when broken.
A later animateSpawn change reaches already-cached submenus in BOTH directions — showSubMenu() copies the value at creation only, so a reused submenu would otherwise stay stale.
Under prefers-reduced-motion: reduce (Playwright emulateMedia), the resolved animation duration is 0s — exercising the _motion.scss collapse rather than trusting it.
No duration or easing literal appears in menu/List.scss; every value resolves through a --motion-* token.
Arrow-key navigation reaches every item during the entrance window (guards against a display/visibility implementation, which would drop items from Navigator's offsetParent filter).
Coverage lives in the component tier (test/playwright/component/menu/), which drives real pointer input in a real browser. Assertions are on the class/animation-name contract, never on a sampled visual mid-animation state, and the animateSpawn: false arm is verified able to fail.
Out of Scope
Exit / dismissal animation. Deliberate, and the reason is specific: dismissal now cascades unmount() through the whole tree from three entry points (leaf click, outside pointer, Escape), and that cascade was regressed as recently as #17850. Animating exits means deferring DOM removal until the animation ends, which opens a window where a menu is visibly present but logically dismissed — and would invalidate the [3, 0] level-count witness added in PR #17852. Its own ticket, after the cascade has mileage.
Any other floating surface (dialog, picker, tooltip). If this shape proves out, generalising it to component.Base is a follow-up — but a generic config is not justified by one consumer.
Per-item entrance stagger.--motion-stagger exists and menus are short enough to use it, but it is a separate visual decision.
Changing neo-aligned-* semantics or the zone search. Read-only inputs here.
Avoided Traps
Do not reuse list.Base#animate_. It is inherited and available, and it means something else entirely. Operator's example, 2026-08-29: "think of fleet manager agent cards => 3x3. we re-sort the store, and items transition to their new positions." That is items moving between positions inside a settled list; this ticket is a floating surface arriving. One flag for both would load list/plugin/Animate.mjs into every menu that only wanted a fade, and would make "animate this menu" ambiguous between two unrelated behaviours.
Do not animate the menu root's transform. It is the aligner's positioning channel (DomAccess.mjs:280).
Do not animate anything that changes the root's layout box. The ResizeObserver at DomAccess.mjs:153 observes the subject and re-aligns on change — this is the mechanical reason a slide is correct and a scale is not.
Do not key the entrance on mount. The zone is not known yet (component/Base.mjs:622 does not await alignTo).
Do not write 200ms anywhere._motion.scss names that a contract violation, and the token carries the reduced-motion collapse.
Related
#17850 · PR #17852 (the dismissal cascade this must not disturb) · #17851 · PR #17856 (the sibling menu work)
#5008 (the prior animate_ / spawn-animation design note this ticket honours)
_motion.scss vocabulary · dialog/Base.mjs#animateTargetId_ (the existing spawn-animation surface)
Retrieval Hint: query_raw_memories("menu spawn animation neo-aligned zone direction DomAccess addAligned first align marker")
tobiu referenced in commit 8f466ba - "feat(menu): an opt-in entrance that follows the side the aligner chose (#17860) (#17861) on Aug 29, 2026, 11:28 PM
Context
Operator, 2026-08-29, after the menu dismissal work landed (#17850) and while the separator/disabled leaf (#17851) was in review:
The refinement that shaped it: a submenu spawning to the right should slide in from the left while fading up — directional, not a scale — because the movement reads as the child emerging from its parent.
Almost every piece already exists. This ticket is mostly wiring, and the design work was establishing that.
Live latest-open sweep: checked the latest 20 open issues at 2026-08-29T19:16Z; no equivalent found.Closed-issue sweep acrossmenu animation/spawn animation/useAnimations menu/menu transition: no declined predecessor — but see #5008 below, which the closed sweep surfaced and which changes this ticket's naming. A2A in-flight claim sweep (12 messages, all read-states): no overlapping claim.The Problem
A floating menu appears instantly at its resolved position. That is correct and fast, and it should stay the default — but it gives the user no cue about where the menu came from, which matters most in a cascade, where a submenu's relationship to its parent item is the entire information the user needs.
The naive implementation fails in three specific ways that are not obvious until you read the main thread, and each one is a real trap rather than a style preference:
transformfights the aligner.main/DomAccess.mjs:277-280positions every aligned node with{top: 0, left: 0, transform: translate(x, y)}. A spawntranslateXon the same node either clobbers placement or is clobbered by the next align.addAligned()(DomAccess.mjs:147-153) attaches aResizeObserverto the subject itself withsyncAlignsas the callback. A scale/height animation changes the observed box → re-align during the animation → which re-measures a moving rect, sincegetBoundingClientRect()returns post-transform visual geometry (documented atDomAccess.mjs:486-491).component/Base.mjs:622firesme.alignTo()on mount without awaiting it — so the node enters the DOM first and the zone is resolved a worker→main round-trip later. There is no anti-flash gate over that window.The Architectural Reality
The direction signal already exists, is already correct, and is already consumed.
util/Rectangle.mjsruns a zone search — candidate zones are tried until one constrains (:530-560) — and stamps the resolved side onto the result:solution.zone = zone; solution.position = zoneNames[zone]; // ['top','right','bottom','left']finishAlignResult(:73-85) exists specifically to preservezone/positionacross the offset move. Main then writes it to the DOM:subject.classList.add(`neo-aligned-${result.position}`); // DomAccess.mjs:292It is correctly lifecycle-managed — the previous position is removed before a re-align (
:243) and on teardown (:1287) — andresources/scss/src/component/Base.scssalready consumes all four classes to place the box shadow at the correct edge. Directional spawn is a second consumer of a maintained signal, not a new mechanism. Because the aligner flips when space is short, the animation direction follows the flip for free.The spawn-vs-resync distinction also already exists.
addAligned()branches onif (!aligns.has(id))(:151) and marks the id seen at:189;syncAligns()runs_aligns.delete(align.id)once the subject is detached. So first-align, re-align and dismissal are already three distinct, correct moments — and a reused menu instance (the patternexamples/menu/context/deliberately uses) passes through the delete on dismissal and re-arms as a genuine first-align on reopen.Durations are already a governed vocabulary.
resources/scss/_motion.scssdeclares itself "the ONLY sanctioned duration/easing source" and states "a call-site duration/easing literal is a contract violation". It supplies--motion-base: 200ms,--motion-panel: 280ms,--ease-out-soft,--motion-stagger: 24ms, and collapses every duration to0msunderprefers-reduced-motionat the vocabulary level — so consuming the tokens inherits the accessibility contract instead of re-implementing it.Navigation is unaffected by an opacity/transform entrance.
addon/Navigator.mjs:272filters withnode.offsetParent && node.matches(selector); an element atopacity: 0keeps itsoffsetParent, so it stays navigable throughout. (An entrance built ondisplay/visibilitywould break arrow-key navigation — recorded as a trap below.)Naming — and the prior ruling the closed sweep surfaced
#5008 (CLOSED/COMPLETED, "dialog.Base: maximise is no longer animated") carries an operator design note:
Two consequences:
dialog/Base.mjs:37still declaresanimateTargetId_for exactly it. The proposed config name should use that word rather than inventing a synonym.list.Basealready hasanimate_(src/list/Base.mjs:27, reactive, defaultfalse) — somenu.Listinherits it today. It must not be overloaded:afterSetAnimate()(:219) lazy-importslist/plugin/Animate.mjs, which drives item motion within the list (reordering/insertion) and marks the ownerneo-animated-list. Entrance-of-a-floating-surface and motion-of-items-inside-it are different concepts — the latter is the store-re-sort transition seen on the Fleet Manager's 3x3 agent cards, where items glide to new positions in a list that is already on screen.The nearest sibling (
animate_) is reactive, which settles the reactivity question against the more distantuseAnimationsprecedent incomponent/DateSelector,date/SelectorContainerandcalendar/view/YearComponent(all non-reactive, all a different family).The Fix
1.
main/DomAccess.mjs— make an unchanged alignment idempotent. Swap theneo-aligned-*class only when the subject does not already carry the resolved one, guarded on the SUBJECT rather than the cached align record. A resync then leaves the class — and anything keyed on it — untouched, while a real flip still changes it. No new class, no new state, no new observer.2.
src/menu/List.mjs— the opt-in config.animateSpawn_: false, reactive, whoseafterSettoggles a root class (proposedneo-animate-spawn). Inheritedanimate_is untouched.4.
resources/scss/src/menu/List.scss— four keyframe rules keyed on the class conjunction, consuming--motion-*tokens only:.neo-menu-list.neo-animate-spawn.neo-aligned-right { animation-name: neo-menu-spawn-from-left; }Nothing is added to the theme sheets: this is geometry and motion, and motion is mode-invariant by
_motion.scss's own design.Corrections from implementation (PR #17861)
Two prescriptions in The Fix were wrong and are struck rather than silently rewritten — the reasoning that killed them is the useful part.
An inner wrapper nodeis not needed. The premise was that the animation could not share the root with the aligner'stransform. It can: the individualtranslateproperty composes withtransformrather than clobbering it. Measured —transform: translate(300px,200px)plustranslate: -40px 0resolves tox = 260with the matrix intact, and the codebase already uses the individual property in 7 places. No structural change to the vdom.Remove the marker on the first— that cancels the entrance.syncAlignsonDocumentMutationresyncs every aligned subject on ordinary body churn, and opening a submenu is enough to trigger it, so a per-align removal strips the marker milliseconds in and the animation dies. It is cleared byunalign()instead. Safe because a resync that keeps the same zone leavesanimation-nameuntouched, and an unchanged name does not restart a running animation (measured: an animation kept advancing across unrelated class churn, 583ms → 1583ms, and reset to 0 only when the name changed). A genuine zone flip replays, which the operator accepted up front.One addition the ticket did not name:
showSubMenu()must propagate the config. Without it the cascade stays still while only the root animates — inverting the effect the feature exists for. Both this and the marker defect passed review-by-reading and were caught only in the browser.Correction 2 — the marker is gone (PR #17861 review, RA-1/RA-3)
does not exist and should not. The design above rested on "a resync that keeps the same zone leavesneo-align-initialanimation-nameuntouched, so a running animation is not restarted." @neo-gpt falsified it at runtime and I reproduced it:align()removes the zone class, reads layout, then re-adds it —currentTime949ms →animation-name: nonewith zero animations → a differentAnimationobject at 0. The final name is identical throughout, which is precisely why a name-only assertion passes while the entrance restarts.The fix moved to the root:
align()swaps the zone class only when the subject does not already carry it. An ordinary resync leaves the class — and the running animation, and the shadow repaint — untouched; a real flip still changes it and restarts the entrance mirrored. The transient marker is deleted, which also retires the claim that the default path was byte-unchanged: nothing is added to every aligned subject any more.The guard is on the SUBJECT, not the cached align record. Keying off
lastAlignlooks equivalent and is not — a hidden menu leaves the DOM and remounts classless while_alignsstill reports the same zone, so that version never re-added the class and a reused menu stayed unaligned permanently. Caught by the reuse witness one run after writing it.unalign()does not exist. I named it as the cleanup authority in code comments, the PR body and this ticket. The cleanup lives insyncAligns(). Corrected everywhere.Contract Ledger Matrix
Neo.menu.List#animateSpawn_false.afterSettoggles the root class that gates the entrance rules. Distinct from inheritedlist.Base#animate_, which loadslist/plugin/Animate.mjsfor item motion and is untouched.false→ today's instant spawn, byte-identicallist/Base.mjs:27+:219establish the reactive sibling;dialog/Base.mjs:37establishes "spawn animation" as the termneo-align-initialneo-aligned-*swap idempotent inalign().Animationobject with progress reset to 0--menu-spawn-*menu/List.scss_motion.scssdurations/easings_motion.scssis the sanctioned source; a literal is a contract violationneo-aligned-{top,right,bottom,left}DomAccess.mjs:292component/Base.scssfor shadow edgeDecision Record impact
none— aligned-with_motion.scss's stated vocabulary authority; no ADR interaction.Acceptance Criteria
main/DomAccess.mjsleaves theneo-aligned-*class untouched when a resync resolves the same zone, and swaps it on a real flip — witnessed by WAAPI animation identity and monotonic progress, not by the resolvedanimation-name, which is identical across a restart.menu.List#animateSpawn_defaultsfalse; with the default, the rendered DOM and computed styles are unchanged from today — asserted, not assumed.animateSpawn: true, a submenu resolving to zonerighthas computedanimation-nameof the from-the-left keyframes; zoneleftresolves to the mirrored keyframes. No wrapper node: the individualtranslateproperty composes with the aligner-ownedtransform.neo-aligned-leftand its animation mirrors accordingly — the case that is painful to hand-code and silent when broken.animateSpawnchange reaches already-cached submenus in BOTH directions —showSubMenu()copies the value at creation only, so a reused submenu would otherwise stay stale.prefers-reduced-motion: reduce(PlaywrightemulateMedia), the resolved animation duration is0s— exercising the_motion.scsscollapse rather than trusting it.menu/List.scss; every value resolves through a--motion-*token.display/visibilityimplementation, which would drop items fromNavigator'soffsetParentfilter).test/playwright/component/menu/), which drives real pointer input in a real browser. Assertions are on the class/animation-namecontract, never on a sampled visual mid-animation state, and theanimateSpawn: falsearm is verified able to fail.Out of Scope
unmount()through the whole tree from three entry points (leaf click, outside pointer, Escape), and that cascade was regressed as recently as #17850. Animating exits means deferring DOM removal until the animation ends, which opens a window where a menu is visibly present but logically dismissed — and would invalidate the[3, 0]level-count witness added in PR #17852. Its own ticket, after the cascade has mileage.component.Baseis a follow-up — but a generic config is not justified by one consumer.--motion-staggerexists and menus are short enough to use it, but it is a separate visual decision.neo-aligned-*semantics or the zone search. Read-only inputs here.Avoided Traps
list.Base#animate_. It is inherited and available, and it means something else entirely. Operator's example, 2026-08-29: "think of fleet manager agent cards => 3x3. we re-sort the store, and items transition to their new positions." That is items moving between positions inside a settled list; this ticket is a floating surface arriving. One flag for both would loadlist/plugin/Animate.mjsinto every menu that only wanted a fade, and would make "animate this menu" ambiguous between two unrelated behaviours.transform. It is the aligner's positioning channel (DomAccess.mjs:280).ResizeObserveratDomAccess.mjs:153observes the subject and re-aligns on change — this is the mechanical reason a slide is correct and a scale is not.component/Base.mjs:622does not awaitalignTo).200msanywhere._motion.scssnames that a contract violation, and the token carries the reduced-motion collapse.Related
animate_/ spawn-animation design note this ticket honours)_motion.scssvocabulary ·dialog/Base.mjs#animateTargetId_(the existing spawn-animation surface)Origin Session ID: 57d042dc-6295-4fea-8347-a79adb8135fc
Retrieval Hint:
query_raw_memories("menu spawn animation neo-aligned zone direction DomAccess addAligned first align marker")