Known limitations

Performance.renderStats and repaintRects field names and semantics lack an authoritative list

DevTools-injected mouse events force a full frame each time, unlike real mouse input, inflating frame counts and style work measured during animations

The Frame Time History vertical axis cannot scale below 20ms when the entire time window is below that level

`devtools_ui.inl` still requires a manual generation step

Consumers cannot use the incremental-layout equivalence self-check (KUI_DEBUG_LAYOUT_EQUIVALENCE_CHECK)

Reduce the hard dependency of ext3d on the libc++ toolchain on Linux

The legend overlap can shrink to zero, whereas Chromium does not shrink it below the block-start border

A rendered legend stretches to the fieldset content width instead of shrinking to its own content width

Fragment-level compositing of ::first-line with 0 < opacity < 1

Percentage handling for fixed-table min-content/fit-content differs from subsequent spanning-cell intrinsic contributions

Zero-width cells after fixed percentage tracks exhaust the width budget include extra padding in their frames

The root strut in the flat IFC fallback still uses natural font height for line-height:normal, differing from text-item normal line-height

Flex items' automatic minimum size (min-width:auto) is rounded up to a whole pixel

Column flex layout does not transfer the natural ratio back to a replaced image height after cross-axis stretching

The Flex grid branch still rounds the root width up after Grid has sized fractional auto tracks

When a flex item is or contains a grid with fit-content() columns, or explicit columns plus implicit fr columns, its flex base still uses geometry from an infinite-width trial layout

Per-glyph coverage after missing webfont glyphs and the Windows generic sans fallback order

Complete native outline clipping, hit testing, shadows and DPI behavior for top-level `frameRadius` windows on Linux/macOS

Individual EXEs can fail transiently in the full ordinary test suite when other builds run on the same machine; failure state is not preserved for diagnosis

Generated-content `::first-letter` does not receive decoration lines declared on `::before` itself

Generic-family fallback glyphs still differ from Chrome in three cases

The 1.36em advance estimate for default geometric list markers differs from the used marker box in Chrome

Go/Rust canonical-example mirrors and manifests have no validation gate

Difference in advance at preserved-space break opportunities and line-start glyph-shaping boundaries in Grid

Fixed repeat is expanded in computed styles, so serialization of computed track values loses repeat notation

Horizontal percentage/ratio queries for fixed rows still lack stretch wrappers, fixed minmax, and a maximum used-width reference

Exact snapshots of adjacent C8 guardrail geometry unrelated to the element's own percentage basis

Temporary exact snapshots of adjacent A176 guardrail geometry must be removed as each root cause is resolved

Grid intrinsic widths and approximate final layout of character sequences with preserved tabs still differ from actual font tab stops

getComputedStyle for template tracks still returns computed values when a Grid box exists rather than used track lengths

KUI exact-point hit testing differs from browser hit regions at subpixel edges

Build and validate reduced ICU Data on macOS/Linux, then enable filtering only after establishing unit/WPT/pixel baselines for each platform

Linux/macOS idle-wake timers never change their interval after being scheduled

Consolidate the IFC baseline solver

Long words still break in the middle for several shapes in paragraphs with break boundaries, preserving the pre-fix behavior

The first/last baseline exported by an IFC can fall below the root strut baseline for mixed primary/fallback fonts and author-specified line heights

BrowserCompat applies the 20px broken-image fallback height to images with no src

Investigate the full set of inline atomic justify pixel failures and final resource states

`box-decoration-break:clone` affects painting of multiline inlines, but layout does not add padding and borders to both ends of each fragment

For an opacity inline effect owner wrapping only a float, visitInlineCBs clears the float overflow recorded by layoutBlockChildren back to zero during the same pass

Unify anonymous parent IFCs, IB-split start/end edges, and marker ownership for ordinary inline list-items

Roles, markers, and block-in-inline split geometry for ordinary display:inline list-item

InlineLayout collects runs differently for standalone inline roots and whole paragraphs: <br> is swallowed (the nested-relative-positioning case disappeared with B24)

Complete inline geometry for inline list-item and RTL inside markers

Inside text markers do not wrap when wider than the line and do not stretch during justification

Inside text list markers do not paint decoration lines propagated from li

IntrinsicSizer rounds the max-content width of joined Arabic words up and adds 1px, making the box about 1-2px wider than Chrome

IntrinsicSizer shapes and sums inline max-content one text node at a time, missing shaping interactions across nodes

IntrinsicSizer does not cover several shapes handled by the legacy keyword evaluator, preventing a full switch for block keyword widths; row flex item keyword widths in A59 already use it

Edge measurement for standalone lines of whitespace-preserving text is currently unified only under a Grid ancestor

Intrinsic width omits letter spacing after the final character

Calibrate the layout depth guard threshold

Chrome legacy `-webkit-line-clamp` clips height and exposes the next line top in bottom padding; KUI does not paint truncated lines

test_letter_spacing_per_line_start_offset pins three known Chrome differences in dev cases or skips

An inline crossing the line-clamp cutoff has no getClientRects() fragments for its discarded lines

For a legacy line-clamp host, only CSSOM scrollHeight includes discarded lines; scroll range, scrollWidth, and DevTools scroll information do not

Modern line-clamp shapes using SkParagraph native ellipses, including more than three lines or non-start alignment, still have trailing hanging-whitespace offsets or retain the first few letters of an extra word; F-WPT-FIX-950 covers only start-aligned LTR soft-wrap fallback with two or three lines

With overflow:visible, legacy line-clamp paints some text outside its container differently from Chrome

The `line-height:normal` cap of twice the font size differs from Chrome for fonts with unusually tall content bands, such as Javanese Text

Linux Canvas text measurement returns zero

It is unverified whether Linux select dropdowns and context menus close when the main window loses focus through Alt+Tab or clicking another application

Linux popup behavior without a compositor

Align Linux text rasterization settings with browsers and the desktop environment

Implement Linux TrayIcon

Align device-pixel snapping and antialiasing of curved markers under fractional transforms and zoom

Some test EXEs approach the run_tests.py per-EXE timeout of 300 s even in isolation and time out under load

Bind macOS menu keyboard shortcuts

Align macOS text rasterization settings with Chromium/CoreText behavior

Replace the deprecated macOS tray balloon API

The DOM.setOuterHTML name does not match its actual innerHTML operation

Remaining DevTools domains lack a unified typed parameter contract

The document-wide style invalidation cost of limited form constraints has not been quantified

Explicit form pattern validation lacks a provable complexity bound

A small baseline difference remains between the table bottom and Chrome

Restore projected damage clipping and caching during multicolumn fragment replay

The merged outline of a multiline inline element has no rounded corners

Remaining negative-z-index inline painting after glyphs moved into the negative layer: several decorations are still painted by the IFC root in the root's layer

Without a backend or SkParagraph, legacy anonymous-block IFC line-height differs from direct text boxes

`overflow-wrap` defers splitting a word to the next line only for authored spaces/tabs; it still splits on the current line after opportunities such as hyphens, slashes or ideographs, as does legacy segmented layout with floats/form controls

CSS 3D transforms (rotateY, perspective, transform-style, and others) have no effect

The fit-content(<length>) function produces a different result from Chrome: Chrome 400, KUI 218

Inline baseline subpixel phase differs from Chrome by one pixel row in 60 of 240 shapes

gap:<percentage> resolves to zero in containers with an indefinite inline size

A grid with width:max-content takes the maximum of inline siblings in a cell as if they were blocks rather than measuring the whole IFC together

Font loading is synchronous and local only: no remote @font-face src:url() download; unicode-range is parsed but not applied; font-display has no observable effect

scrollWidth of a nowrap box with text-overflow:ellipsis does not reflect its full untruncated width

tab-size is not implemented (measured width: 72 versus 118 in Chrome)

Negative text-indent does not produce a hanging indent

Line breaking after hyphens may not follow UAX #14

Complete paged multicol and specialized pagination semantics for Flex, Grid, and Table

The asynchronous image input history of performance-equivalence fixtures is not yet controlled

Experiment evidence paths have only 19 characters of headroom in the default output directory

Detect missing POSIX LSP executables

Unify anonymous IFCs and preserved line breaks around block siblings in <pre>, pre-wrap, and break-spaces

Keep justification opportunities, block height, and source mapping consistent for consecutive preserved spaces under pre-wrap + justify

The raw SkFont measurement path still maintains a third font-family table, with fallback glyphs differing from the paragraph path

Real-app corpus has three gaps: Lua-generated content lacks Chrome comparisons, performance pages include the loader navigation bar, and forge-ims is absent

The alternating previous-release comparison uses first-play maximum animation frame time as a blocking metric with thresholds narrower than same-build noise, causing false failures

Block-axis min/max intrinsic keyword constraints on replaced images do not transfer the natural ratio from a definite width

The main replaced-element sizing path is fixed; remaining work is limited to unifying unloaded-image paths and NativeGui iframe default-border conventions

When LTR content overflows a direction:rtl paragraph, text-overflow removes the logical tail instead of the part crossing the visual line-end edge

An inline wrapping text and preserved trailing spaces in an RTL `pre-wrap` paragraph lacks a fragment painted at the visual line end

When a native ellipsis in legacy RTL line-clamp lands at the truncation point of an inline with background/borders, its decoration area does not exclude the ellipsis

UTF-16-indexed placeholder rectangle queries in bidi-reordered lines return another atomic box or margin slot, affecting selection highlights and caret hit testing

Absolute static-position, self-alignment, and Flex/Grid alignment support under RTL and logical directions

KUI_DISABLE_MULTICOL, KUI_DISABLE_SHAPES, and KUI_DISABLE_SUBGRID are never read by code, although config.h claims NOT COMPILED IN and its code paths are gated out

Pagination fragmentation for Flex, Grid, and Table is not implemented

Reclassify compile-time-disabled multicol, CSS Shapes, and subgrid from out of scope to unimplemented and awaiting scheduling

Remaining differences between scroll anchoring and CSS Scroll Anchoring/Chrome

An overflow gate suppresses scroll-container first baselines, causing flex/grid baseline alignment to use the wrong line

Scroll offsets do not snap to device pixels: scrollTop/scrollLeft may be arbitrary fractions, and at the bottom of fractional layouts scrollTop + clientHeight falls short of scrollHeight by less than 1 px

Bring C SDK packaging into alignment across platforms

Decouple the SDK example gate from the legacy in-tree 3D switch

Extension and media-backend DLLs still lack VS_VERSION_INFO and a build ID

The block-stacking loop lays out inline content on either side of a float separately, placing preceding text and following inline elements on different lines

In segmented inline layout, top/bottom-aligned atomic boxes inside a span align to the span paragraph line rather than the outer line

Remaining selective-layout improvements

Line-end/block-end newline highlights on intermediate lines of a multiline selection differ from Chrome by about one space width

Changing the only `display:table-cell` child of a `display:inline-table` to `position:absolute` increases line height by 8px on the incremental path

Items do not yet preserve the unsafe modifier or propagate auto self values; non-abspos physical modifier combinations continue to use legacy values

Skia runtime CPU dispatch (`SkGraphics::Init()`) remains disabled on Linux/macOS pending local pixel validation

SkParagraph shapes each font run without surrounding context; joining text across fonts or formatting relies on an incomplete presentation-form compatibility layer

Trailing hanging-whitespace offset in native line-clamp ellipses, with the remaining boundary narrowed

SkParagraph assigns line-height rounding fractions to the ascent, letting atomic boxes taller than the text extend up to 0.5px above the line

Physically left-aligned paragraphs round soft-wrap widths up to whole pixels, fitting one more word than Chrome at fractional widths

The memcmp fast path in ComputedStyle::operator== can be misled by residual bytes in alignment padding

Nested @supports parentheses that contain neither a declaration nor a valid condition are treated as invalid, although the specification requires general-enclosed with a false result

A table cell first-baseline query does not descend into line boxes inside block children

The 45 known differing boxes in the C2 candidate guard temporarily use separate exact KUI snapshots

C2 unified candidate-content measurement still uses trial layout at non-atomic inline and nested CSS direct-cell boundaries

Spans laid out segment by segment beside floats lose br breaks and collapse multiline content into one line, including outside tables

Row flex container intrinsic contributions lose the natural minimum width of a table

The shrink-to-fit outer frame of an inline-block wrapping a table is still rounded up to whole pixels

`test_devtools_server` fails without retry when its chosen port is already in use

`VsyncFrameGateWindow.ImageLoadStartedWhileGatedShowsPromptly` intermittently fails its warmup assertion during parallel full-suite runs

`VsyncFrameGateWindow.LottieOnlyPostsAroundFrameChanges` intermittently exceeds p90 tolerance during parallel full-suite runs

Animation/performance tests sample wall-clock time; frame-count and duration assumptions fail in unoptimized Debug builds

`Win32WindowFeedback.RoundedShadowLayerCompositeUpdatesAllCorners` intermittently reports `layerRootReuses == 0`

Last-line detection, logical alignment, and justification for text-align-last

`text-decoration-skip-spaces` is unimplemented and the set of skipped whitespace differs from the specification

Shapes not covered by the text-indent glue carrier, including a break after the indent when the first word does not fit

Text list markers, including decimal numbering and ::marker content, still use paragraph painting phase and differ from Chrome when the first-line baseline is not on the strut

`text-underline-position` recognizes only `auto` and `from-font`; `under`, `left` and `right` are ignored

A transform-transition translate component switches discretely halfway when one endpoint is nonzero px and the other is a nonzero percentage

Establish an intrinsic-sizing matrix for BrowserCompat blockified/semi-replaced buttons

Establish an intrinsic-height, baseline, and line-box matrix for all BrowserCompat controls

Build a BrowserCompat native-control geometry matrix for Linux and macOS

Wrapped text spans with `vertical-align:top/bottom` differ from Chrome in position and line-box height

Running all ordinary tests overwrites tracked `tests/visual/output/typography_actual.png` and `typography_diff.png`

Consistency of trailing-space advances, wrap ownership, line boxes, and painting in pre-wrap/break-spaces