Context
list.plugin.Animate's measured mode (#18604: owner itemHeight: null + measureItemHeight: true) takes the tallest rendered item as the list's one row height. Tried on its first real consumer today — the Institution's fleet roster (neomjs/neo-agent-institution#128) — it splits in two:
- Homogeneous items (the calm roster, every card the same anatomy): 126 px fixed rows become 104 px, no dead reserve. The intended win.
- Mixed items (one card shows a control-status row, another a two-line lane): the pitch follows the tallest card in the whole list while every other card stays as tall as its own content. Short cards float in tall slots — the gaps read as cards again, the defect the consumer ticket set out to remove.
Operator design ruling on the captures (2026-09-18): cards in a grid have the same height.
The Problem
The mode measures items, so it cannot size them: createItem writes no inline height in measured mode on purpose (Animate.mjs, "no inline height: the item is as tall as its content, which is what gets measured"). A written height would be what the next measurement reads — in both directions: an element with an inline height answers a rect read with that height whatever its content does, so the row would freeze at its first measurement (measured while building the fix: unit and real-DOM mutation, the row stays at its first value). That guard is correct, and it is exactly what leaves the items ragged: uniform pitch, non-uniform items.
Fixed mode has no such split — style.height = owner.itemHeight makes item, slot and (through the consumer's height: 100%) surface one thing. Measured mode needs the same equality without that trap.
The Architectural Reality
src/list/plugin/Animate.mjs: createItem (the mode branch), measureItems → owner.getDomRect(ids) over the item ids, token-fenced; rowHeight getter; repositionItems.
src/main/DomAccess.mjs: getBoundingClientRect, getLayoutRect — neither answers "how tall would this element be if its inline height were released".
DomAccess.measure({id, value}) is the nearest sibling and the placement precedent (the main thread answering a size the worker cannot compute), but it answers a different question: it converts one CSS length to px for one node — % as a fraction of the offsetParent's height, any other unit through an off-screen measuring div — for component.Base#measure (the floating/align path, its only caller). It never reads an element's own natural size, it is single-node where this read must batch (release all → read all → restore all, or it thrashes layout per item), and its two branches already measure different axes. Overloading it (value: 'auto') would inherit that ambiguity; the new read sits beside it instead.
- Consumers size the surface off the item: the roster's
> .fm-agent-card {height: 100%}, the list example's cards.
The Fix — options, each with its falsifier
| Option |
Shape |
Falsifier |
| A — measure the content, size the slot |
measure each item's content element (component lists: the item component's root) and write height: rowHeight on the li |
the content must then NOT stretch to the li (height: 100% re-arms the trap), so an equal-height surface has to be painted by the li — every consumer restyles, selection and focus rings included |
| B — a natural-height read in one main-thread task |
a DomAccess read that releases the inline height of the given elements, reads their rects and restores it inside one task (release all → read all → restore all: one forced layout for any number of items, no painted intermediate frame); the plugin then writes height: rowHeight on every item, as fixed mode does |
cost at a few hundred items; an item whose own CSS transitions height |
C — CSS variable + min-height |
publish the row height on the list root, items take min-height: var(…) |
the measured element includes its min-height: grow-only by construction |
Recommendation: B. It keeps fixed mode's consumer contract (item = slot = surface) so a consumer switches modes by config alone, and the trap is closed where it lives — in the read.
Contract Ledger
| Target surface |
Source of authority |
Proposed behaviour |
Fallback |
Docs |
Evidence |
Neo.list.plugin.Animate measured mode — item height |
Animate.mjs#createItem, #measureItems (read at dev@70c2c94618) |
every item carries height: rowHeight once measured; before the first measurement it stays hidden and height-less, as today |
fixed mode unchanged |
the measureItemHeight_ member doc; examples/list/animate |
unit arms: equal item heights after a settle; a row that grew shrinks back when its tall item leaves |
Neo.main.DomAccess.getNaturalRect (+ the component.Base#getNaturalRect shortcut, beside getLayoutRect) — the LAYOUT box, so the answer can be written back as a CSS length |
DomAccess.mjs remote-method list (getBoundingClientRect is the sibling) |
for the given ids: release the named inline properties (properties, default ['height']), read their layout boxes, restore the values with their priority — one task |
callers that never pass it are untouched |
method JSDoc |
component-tier arm (a real DOM): the element's inline height and its priority are identical before and after, the returned height is its content's |
Behaviour change since #18604 (shipped 2026-09-14, one example consumer): measured items were content-height; they become row-height.
Acceptance Criteria
Out of Scope
Per-row heights (each row as tall as its own tallest item) — a different geometry (cumulative y), worth its own ticket if a consumer needs it. Variable item widths.
Avoided Traps
Writing the measured height back without changing the read (the row freezes at its first measurement). A hidden clone per item to measure (doubles the DOM and drifts from the real item's container queries — the roster card is container-type: inline-size).
Related
#18604 · #18670 · #18844 · neomjs/neo-agent-institution#128 (blocked on this)
Decision Record impact: none.
Live latest-open sweep: latest 20 open issues read at 2026-09-18T12:35Z + an all-state search (Animate measured row height equal stretch) — nothing equivalent. A2A sweep: latest messages carry no claim on list/plugin/Animate.mjs beyond my own merged #18844. Memory Core: my #18604 design turn recorded "variable per-row heights = stage two"; no decision covers equal-height items. Own open assignments: no overlap.
Origin Session ID: 1ef6c04a-10f6-4977-ad7d-0e2c7f343c59
Retrieval Hint: "Animate measured mode equal item heights natural-height read one task grow-only trap roster cards same height"
Context
list.plugin.Animate's measured mode (#18604: owneritemHeight: null+measureItemHeight: true) takes the tallest rendered item as the list's one row height. Tried on its first real consumer today — the Institution's fleet roster (neomjs/neo-agent-institution#128) — it splits in two:Operator design ruling on the captures (2026-09-18): cards in a grid have the same height.
The Problem
The mode measures items, so it cannot size them:
createItemwrites no inline height in measured mode on purpose (Animate.mjs, "no inline height: the item is as tall as its content, which is what gets measured"). A written height would be what the next measurement reads — in both directions: an element with an inlineheightanswers a rect read with that height whatever its content does, so the row would freeze at its first measurement (measured while building the fix: unit and real-DOM mutation, the row stays at its first value). That guard is correct, and it is exactly what leaves the items ragged: uniform pitch, non-uniform items.Fixed mode has no such split —
style.height = owner.itemHeightmakes item, slot and (through the consumer'sheight: 100%) surface one thing. Measured mode needs the same equality without that trap.The Architectural Reality
src/list/plugin/Animate.mjs:createItem(the mode branch),measureItems→owner.getDomRect(ids)over the item ids, token-fenced;rowHeightgetter;repositionItems.src/main/DomAccess.mjs:getBoundingClientRect,getLayoutRect— neither answers "how tall would this element be if its inline height were released".DomAccess.measure({id, value})is the nearest sibling and the placement precedent (the main thread answering a size the worker cannot compute), but it answers a different question: it converts one CSS length to px for one node —%as a fraction of theoffsetParent's height, any other unit through an off-screen measuringdiv— forcomponent.Base#measure(the floating/align path, its only caller). It never reads an element's own natural size, it is single-node where this read must batch (release all → read all → restore all, or it thrashes layout per item), and its two branches already measure different axes. Overloading it (value: 'auto') would inherit that ambiguity; the new read sits beside it instead.> .fm-agent-card {height: 100%}, the list example's cards.The Fix — options, each with its falsifier
height: rowHeighton thelili(height: 100%re-arms the trap), so an equal-height surface has to be painted by theli— every consumer restyles, selection and focus rings includedDomAccessread that releases the inline height of the given elements, reads their rects and restores it inside one task (release all → read all → restore all: one forced layout for any number of items, no painted intermediate frame); the plugin then writesheight: rowHeighton every item, as fixed mode doesheightmin-heightmin-height: var(…)min-height: grow-only by constructionRecommendation: B. It keeps fixed mode's consumer contract (item = slot = surface) so a consumer switches modes by config alone, and the trap is closed where it lives — in the read.
Contract Ledger
Neo.list.plugin.Animatemeasured mode — item heightAnimate.mjs#createItem,#measureItems(read atdev@70c2c94618)height: rowHeightonce measured; before the first measurement it stays hidden and height-less, as todaymeasureItemHeight_member doc;examples/list/animateNeo.main.DomAccess.getNaturalRect(+ thecomponent.Base#getNaturalRectshortcut, besidegetLayoutRect) — the LAYOUT box, so the answer can be written back as a CSS lengthDomAccess.mjsremote-method list (getBoundingClientRectis the sibling)properties, default['height']), read their layout boxes, restore the values with their priority — one taskBehaviour change since
#18604(shipped 2026-09-14, one example consumer): measured items were content-height; they become row-height.Acceptance Criteria
rowHeight, after each settle pass.#18604token fence still drops a stale read.examples/list/animate"Measured rows" renders equal cards in the browser, through filter and mode switches; mixed content is witnessed on a real DOM in the component tier (the example's cards share one natural height).Out of Scope
Per-row heights (each row as tall as its own tallest item) — a different geometry (cumulative
y), worth its own ticket if a consumer needs it. Variable item widths.Avoided Traps
Writing the measured height back without changing the read (the row freezes at its first measurement). A hidden clone per item to measure (doubles the DOM and drifts from the real item's container queries — the roster card is
container-type: inline-size).Related
#18604·#18670·#18844· neomjs/neo-agent-institution#128 (blocked on this)Decision Record impact: none.
Live latest-open sweep: latest 20 open issues read at 2026-09-18T12:35Z + an all-state search (
Animate measured row height equal stretch) — nothing equivalent. A2A sweep: latest messages carry no claim onlist/plugin/Animate.mjsbeyond my own merged#18844. Memory Core: my#18604design turn recorded "variable per-row heights = stage two"; no decision covers equal-height items. Own open assignments: no overlap.Origin Session ID: 1ef6c04a-10f6-4977-ad7d-0e2c7f343c59 Retrieval Hint: "Animate measured mode equal item heights natural-height read one task grow-only trap roster cards same height"