remaining-issues.md

TanStack Table remaining open-issue audit

Generated 2026-07-30 from a fresh scan of the 110 issues still open in TanStack/table, cross-referenced against the two 2026-07-26 audits (issues-gpt.md, issues-claude.md) and re-verified against current v9 alpha source (post module-level-fns perf refactor; 9.0.0-beta.61 era, which includes PR #6458 merged 2026-07-29).

Result

• 61 issues can close now without a new PR — including 2 newly fixed by #6458 and 2 duplicates of a canonical selection issue.
• 49 issues should remain open: 10 verified & urgent, 37 valid, 2 needing investigation.
• 115 of the 225 issues from the July 26 audits were closed between then and today; zero new issues have been filed since. Every currently open issue was covered by both prior audits.
• Where the two audits disagreed (27 issues), each was re-adjudicated against current source: 9 resolved to close, 18 to keep open.
• 17 already-open PRs target keep-open issues (see table below); landing or porting them to alpha would clear roughly a third of the keep-open list.

Audit baseline

• v9 baseline: 9.0.0-beta.61 (source branch includes the module-level-fns perf refactor)
• Method: 9 parallel investigation agents; per-issue checks of issue threads (activity after 2026-07-26), current packages/table-core/src source, and open/merged PRs. No fixes applied, no PRs opened, no repo files modified.
• Verdict rule: disagreements between the prior audits were resolved by re-reading current source; "Verified & Urgent" means the defect was re-confirmed live in current code on a default or documented path.

Summary

Category	Count	Recommended action
Was fixed in V9	14	Close; behavior is correct in current alpha (2 fixed by #6458 this week).
Wrong Library Direction	32	Close with the documented/headless-library boundary.
Junk Issue	11	Close as support-only, malformed, user error, or unreproducible-v8-era.
Couldn't Reproduce	2	Close unless a current minimal reproduction is supplied.
Duplicate	2	Close as duplicates of canonical #6049.
Verified & Urgent	10	Keep open and prioritize.
Valid	37	Keep open; many have fix PRs waiting to land or port.
Maybe & Needs Investigation	2	Keep only for additional triage.
Total	110	61 close candidates; 49 keep open.

Fix-first priority list

Ranked by user impact. Fixing the top four clusters closes or unblocks ~20 issues.

1. #6078 — process is not defined crash (Verified & Urgent, high). Raw process.env.NODE_ENV reads survive into the published dist (packages/table-core/dist/utils.js) because tsdown.config.ts has no define; any bundler-less consumer (importmap/CDN) gets a hard ReferenceError, and v9 dropped the v8 UMD escape hatch. PR #6185 targets main only — needs an alpha port.
2. Global-filter eligibility cluster — canonical #4673 (+ #4783, #4919, #5138). Default getColumnCanGlobalFilter still samples only flatRows[0] and requires string/number, and the AND in column_getCanGlobalFilter means explicit enableGlobalFilter: true can never opt back in — search silently returns nothing. Pending PRs: #6252/#6438 (nullish scan), #6439 (explicit opt-in). One combined fix closes three issues as duplicates.
3. Parent/child selection reconciliation — canonical #6049 (+ #5116, #5398; close #4878/#5416 as duplicates). Deselecting a descendant never removes the ancestor id, so getIsSelected() and getIsSomeSelected() can both be true and the shipped expanding example renders a parent fully checked with an unselected child. One refactor of table_toggleAllRowsSelected/mutateRowIsSelected plausibly fixes five open issues plus the closed-set #4349 family.
4. Auto-reset first-run guard — #5968. v9 dropped v8's registered first-run guard: memo fires onAfterUpdate on the first computation, so the very first getRowModel() schedules table_autoResetExpanded/table_autoResetPageIndex, wiping controlled expanded and any initial pageIndex on mount (and table_resetPageIndex(table, true) hard-codes 0, ignoring initialState).
5. row._valuesCache never invalidated — #4485 / #5363 (high). createCoreRowModel memoizes on [options.data] only, so a new columns/accessorFn silently serves stale values into rendering, sorting, and filtering — the documented "memoize columns" pattern yields wrong data. v8 PR #5582 shows the approach.
6. #6007 — inNumberRange matches null as 0 (Verified & Urgent, high). No resolveDataValue, so null/''/false coerce to 0 and pass any zero-spanning range; this is the auto-selected default filter for number columns. PR #6313 open.
7. #5770 — group-column visibility (Valid, high). Explicit columnVisibility: { groupId: false } is ignored; Kevin already approved fixing it in the alpha ("Only in the alpha. Want to limit breaking changes in main.") and the alpha is the only window for this breaking change.

Open PRs to land or port

PR	Fixes	Status
#6439	#4673/#5138 (explicit enableGlobalFilter opt-in)	Open
#6252 / #6438	#4783/#4919 (scan past nullish first row)	Open
#6431	#4309 (single-sort replace branch)	Open
#6313	#6007 (exclude non-numeric from inNumberRange)	Open
#6361	#5987 (maxLeafRowFilterDepth flatRows)	Open
#6417	#6081 (refresh custom faceted values)	Open
#6443	#5801 (reset expansion on data change)	Open (base beta)
#5790 / #6177	#5173 (isSubRowSelected with unselectable children)	Open
#6075	#6074 (columnFiltersMeta wipe)	Open, targets v8 main — needs alpha port
#5823	#5822 (pinned-row lookup crash)	Open since 2024-12, targets v8 — needs alpha port
#6185	#6078 (process.env in dist)	Open, targets v8 — needs alpha port
#6186	#6077 (Vue empty-string hydration)	Open, targets v8 — needs alpha port
#6116	#6115 (expand-all writes non-expandable ids)	Open, targets v8 — needs alpha port
#6184	#6136 (no-op toggleExpanded rerender)	Open, targets v8 — needs alpha port
#6176	#6175 (example server.allowedHosts)	Open
#5031	#4703-adjacent pageIndex clamp	Open, targets v8
#5582	#4485/#5363 (_valuesCache invalidation approach)	Open, targets v8 — port concept to alpha

Close now

Was fixed in V9 (14)

• #4634 — "Cannot update a component while rendering a different component" when filtering with downshift combobox — The warning's source (v8's useState-backed onStateChange calling setState during render) no longer exists; v9 holds state in Store atoms, and PR #6458 (merged 2026-07-29) defers render-phase state publication to the commit phase, eliminating this warning class.
• #4759 — Cannot toggle selection of grouped row to deselect its children — v9's row_toggleSelected dropped v8's early return and mutateRowIsSelected recurses into subRows while only writing selectable ids, so an unselectable group row now toggles its children on and off as requested.
• #4794 — unnecessary rerenders of every row and cell in examples/react/editable-data — The targeted example no longer exists and the v8 root cause (all state in one useState) is gone: v9 state is atom-backed with useTable selectors and table.Subscribe, demonstrated in the spreadsheet example. The live flexRender-remount subtopic from recent comments deserves a fresh v9-scoped issue when closing.
• #4879 — row selection not working for grouped rows (flat data) — Group rows never enter rowSelection in v9 (toggle-all walks pre-grouped flat rows; mutateRowIsSelected skips unselectable rows), and the example checkbox pattern adopted in #6364 renders both reported flows correctly.
• #5026 — Can't perform a React state update on a component that hasn't mounted yet — Kevin's 2026-07-27 comment attributed this to controlled options.state syncing during render; PR #6458 (merged 2026-07-29) landed exactly that fix, deferring publication of controlled state to the commit phase. v8 will not be patched.
• #5162 — custom Data within expanding example Crashing browser — The v8 crash was the unstable-data auto-reset-expanded render loop; in v9 table_autoResetExpanded fires only from the grouped row model and pagination resets are no-op-guarded, so the loop cannot occur.
• #5605 — cell.getIsAggregated should not check for zero length subrows — cell_getIsAggregated now keys off row.groupingColumnId plus a resolved aggregation fn with no subRows.length term, exactly the requested behavior.
• #5617 — Unexpected behavior of array filterFns (arrIncludes, arrIncludesAll, arrIncludesSome) — filterFn_arrIncludesAll/arrIncludesSome now bail on !Array.isArray(dataValue), and the new filterFn_arrHas covers the scalar-in-filter-array semantic the reporter proposed.
• #5620 — Table v8 expanding - incorrect state of parent checkbox when selecting subrows — The v8 toggleSelected early return that skipped clearing children is gone; deselecting a parent now clears its subrows. The remaining half is the documented indeterminate-state pattern.
• #5836 — Request: Implement Uncontrolled State Behavior for Expanded Rows — v9's atomic state model delivers observe-without-controlling: leave onExpandedChange alone and subscribe to table.atoms.expanded (or use an external atom), so the table keeps owning state while the app gets notifications.
• #5882 — alpha: getting the memo() function when calling column.getFacetedUniqueValues() — Fixed during alpha; column_getFacetedUniqueValues now returns the invoked Map, and the Qwik context it was reported from no longer exists in v9.
• #6158 — Event onPaginationChange is triggered on every history push in RSC — Real in v8 (auto-reset called setPageIndex(0) with no equality check, producing the URL-sync loop); v9's table_resetPageIndex returns early when newPageIndex === currentPageIndex.
• #6224 — alpha: useTable causes "Cannot update a component (SubscribeBound) while rendering" after setOptions moved to render body — Fixed by PR #6458 (merged 2026-07-29, in 9.0.0-beta.61): useTable now calls table_setOptions(..., { syncExternalState: false }) in the render body and publishes captured controlled state in a layout effect via table_publishExternalState, so no store subscriber is notified during render. Ask the reporter to confirm on beta.61.
• #6450 — "Cannot update a component while rendering a different component" while using controlled expanded state — Duplicate root cause of #6224; PR #6458's summary explicitly cites this issue as the report it fixes. Close as fixed, asking for confirmation on beta.61.

Wrong Library Direction (32)

• #4498 — [React] Row selection does not reset after upgrading from v7 to v8 — Row selection is id-keyed and intentionally persists across data changes; resetRowSelection(), toggleAllRowsSelected(false), and stable getRowId are the supported answers, already in the thread.
• #4628 — When sorting on date grouped column of year is not taken into account — Group rows sort by the column's aggregated value, and default aggregationFn: 'auto' resolves to sum for numbers, so summed timestamps look random; aggregationFn: 'min'/'max' on the date column yields the expected order (test-verified). Intentional pipeline semantics with a config fix.
• #4771 — getIsAllRowsSelected returns unexpected result when using manual pagination — getIsAllRowsSelected is defined over the loaded rows; the page-aware APIs (getIsAllPageRowsSelected/getToggleAllPageRowsSelectedHandler) are the documented answer.
• #4781 — Pagination and Row Selection — Server-side "select all" cannot materialize ids the client never received; v9 also added toggleAllRowsSelected(false, { deselectAll: true }) for clearing unloaded pages.
• #4785 — rowSpan doesn't work with useFlexLayout and useResizeColumns — Targets v7's removed plugin hooks; flex/absolute layouts fundamentally cannot preserve native rowSpan semantics, and v7 is frozen.
• #4797 — Manual Pagination doesnt reset PageIndex when Manual Column Filtering. (server-side, v8) — With manual pagination the table cannot distinguish a filter change from a page fetch; the docs explicitly transfer reset policy to the app's filter handler.
• #4812 — Row selection acting strange when using fully controlled table — Under manual pagination row.id defaults to row.index, so ids repeat per page; a stable getRowId is the documented requirement and resolved the report.
• #4848 — incorrect Row ID in "Pagination Controlled" react example — Same root story as #4812; index-default row ids are by design and the targeted v8 example no longer exists in v9.
• #4872 — Use columnOrder and initialize it, then the group column is always at the end — columnOrder orders leaf column ids by design; seed with getAllLeafColumns().map(c => c.id). Optional docs clarification only.
• #4880 — columnSizing state gets out of sync with actual column sizes when used with a full width table — columnSizing stores logical widths and table-core never reads DOM geometry; a width: 100% table's browser-distributed widths will always diverge. The actionable ask underneath is #4825 (flex-redistribution resize) — close into it.
• #4905 — Filtered row disappears if its cell value was changed — Filtered row models are pure derivations of data plus filter state; a row edited out of an active filter correctly disappears (reporter concedes as much).
• #4906 — Column pinning inconsistence — Group-column getIsPinned is a documented column-level summary; the per-region header-group APIs are the correct rendering signal, which the reporter found.
• #4970 — Expanded rows become hidden within the table — Default paginateExpandedRows: true paginates sub-rows alongside parents; paginateExpandedRows: false is the documented option and resolved it for two commenters.
• #4994 — Incorrect pagination behavior with autoResetPageIndex set to true when deleting rows — Despite the title, the reporter runs autoResetPageIndex: false and expects the table to still adjust pageIndex when page count shrinks; the pagination guide documents that opting out means handling resets yourself. The adjacent real clamp gap is tracked by #4703 and PR #5031.
• #5120 — Fixed width is not applying for fix width column along with Full width table, Column Resizing and Column visibility. — column.getSize() is correct; a w-full table with table-layout: auto treats widths as hints. Use table-layout: fixed with getSize() widths.
• #5305 — Vue Query's onPaginationChange has incorrect union type/unexpected behavior — Updater<T> = T | ((old: T) => T) is the library-wide state contract; the correct narrowing pattern is shown in the v9 Vue table-state guide and in-thread.
• #5405 — Unable to access Map entries with array keys by value in getFacetedUniqueValues function — Identity-keyed Map entries for array values are the documented default; the getUniqueValues column option exists for exactly this case and is documented in the v9 faceting guide.
• #5455 — useQwikTable(options) causes vite to crash in dev mode — The Qwik adapter could not survive Qwik's serialization model and was removed from the repo (commit 620590e58); there is no Qwik surface in v9 to fix.
• #5500 — sortingFn: 'alphanumeric' returns numbers last — Numbers-after-letters is an intentional branch in the chunked comparator, preserved to avoid silently reordering existing tables; digits-first is a two-line custom sort fn shown in the issue.
• #5536 — useQwikTable undefined sorting handlers [QwikCity] — Qwik's QRL serialization boundary dropped the method-bearing instance; the Qwik adapter was removed in v9 (same territory as #5455).
• #5594 — Grouping by not using subRows — Grouping deliberately operates on the current level and leaves a getSubRows hierarchy intact; v9's pluggable row-model factories make a recursive groupBy a supported userland customization.
• #5710 — when i useing thead Grouping. only set subHeader enableColumnFilter work, parent Column didn't work — Filtering applies to leaf columns with an accessor; a group column has no row value to filter, confirmed by the maintainer in-thread (2024-08-15).
• #5792 — [Feature] Keep sorting order after changing row information — Row models are pure derivations with no mutation tracking; manualSorting or row pinning give full ordering control in userland.
• #5800 — Function getIsSomeSelected() return false when has one child — Documented as "some but not all": with one child selected, all sub-rows are selected, so false is correct and getIsAllSubRowsSelected() returns true.
• #5870 — Column resizing doesn't respect clientWidth, bad experience with limited width tables — Core resize math is absolute-pixel correct; the runaway values came from the v8 percentage-based example (not in v9), and clamping to rendered clientWidth requires DOM measurement a headless core deliberately does not do.
• #5935 — New Rows bring from infinite scroll are coming with Expand = False if I collapse any row — ExpandedState is either true or an explicit id map by design; the supported escape hatch for defaulting newly-loaded rows to expanded is the documented getIsRowExpanded option.
• #6010 — From Docs Full width Resizable example - on resizing a particular column its effecting columns previous to it — The handler writes only the dragged column's size; the neighbor shift is browser auto table-layout on a width: 100% table. A spreadsheet-style algorithm is a rendering strategy the consumer owns (or lands via #4825).
• #6061 — sortUndefined and Excel like sorting — sortUndefined matches literal undefined only, by documentation and implementation; normalizing null/'' in an accessorFn is the intended pattern, already posted in-thread.
• #6091 — [Docs] Document how to do pivoting — Pivoting was a v6/v7 plugin removed in v8; there is no inconsistent doc to fix and no built-in pivot API planned. Grouping + aggregation + userland column generation is the answer.
• #6102 — [v7+] Collapse header group — Feature request against EOL v7 for UI-affordance state the library deliberately does not own; buildable from columnVisibility plus a placeholder column.
• #6109 — Incorrect TanStack Table behavior when using virtualized rows inside hidden tabs — The reporter's own stack trace and proposed guard live in @tanstack/react-virtual's measurement of a display: none container; refile at TanStack/virtual.
• #6243 — Chrome renderer freeze when TanStack row model gates a Suspense child — Table's role is supplying a synchronous rows.length; the freeze mechanism is a thrown timer-backed promise under Suspense in a Chrome production preview, and no row-model path can wedge a renderer process. Point to stable data refs and redirect to facebook/react or Chromium.

Junk Issue (11)

• #4901 — Table row count change causes scrolling on Chrome when table is height limited — Pure CSS/Radix layout problem in the reporter's shadcn markup; both in-thread fixes confirmed working by follow-up commenters.
• #4917 — Manual Pagination doesn't update table.getRowModel().rows - it shows all results — User error answered three times: manualPagination: true declares that data is already the current page.
• #4929 — Infinite loop caused by combining virtual scroll and grouping — Root-caused in-thread to unstable references (data: data || [] per render) triggering the documented auto-reset loop; not a grouping/virtualization interaction.
• #4942 — Cell's filterFn never gets called — Column filterFn runs for columnFilters; global filtering deliberately resolves the table-level globalFilterFn. The v9 guide now states this explicitly.
• #5033 — VSCode Intellisense is very slow with @tanstack/react-table 8.9.2 — Repro field is literally *; the reporter's own update attributes the slowdown to the surrounding file's other imports. v9 fully rewrote the type graph; invite a refile with a reproducible v9 project.
• #5045 — Order + filter with formatted data — Support question with a dead sandbox; raw accessor + cell formatter + custom per-column filterFn/sortingFn already covers it.
• #5047 — react-beautiful-dnd does not work with react-table v8 — Works locally per the reporter, fails only in the sandbox, filed upstream at hello-pangea/dnd#661 and abandoned; Table owns no drag/drop layout.
• #5093 — Dropdown in header fires the sorting fn. — Consumer-side event bubbling from a sort toggle attached to the <th>; e.stopPropagation() stands and the follow-up supplied no repro.
• #5352 — When using virtual tables, high memory usage has always been unable to be reduced. — Feb 2024 v8 solid-table heap screenshots from a third-party Storybook demo; no minimal repro, zero comments in 2.5 years, and v9's prototype-based rows rewrote the allocation profile being measured. Ask for a v9 repro with measurements.
• #6018 — Can react-table v7 be safely used with React 19? — Support question about an EOL major (last release 2022); the supported path is v8/v9.
• #6024 — Performance Issue on table! — Vague 3,000×3,000-cell v8 report with no profile; pagination/virtualization is the answer and nothing is actionable without a fresh v9 repro.

Couldn't Reproduce (2)

• #4852 — Resizing a column makes other columns resized when changing the size property — A core test shows resizing a leaf writes exactly one columnSizing entry; residual visual shifting is native full-width CSS layout (conceptually tracked under #4825).
• #6048 — enableSortingRemoval doesnt seem to be working with manualSorting — The toggle cycle never consults manualSorting and a 4-case matrix test cycles removal correctly; neither the reporter nor the Feb 2026 confirmer supplied a repro. The plausible real mechanism (auto-sort-dir flip on server data swap) is tracked by #5147/#5832.

Duplicates (2)

• #4878 — row selection not working for nested rows (getSubRows) — Half fixed (the #6364 example checkbox pattern covers select-all-children); the remaining half (deselecting a child leaves the parent id in rowSelection) is exactly the root cause tracked by canonical #6049. Close with a pointer.
• #5416 — Selection of subrows not working correctly — Both symptoms reduce to the missing upward reconciliation in mutateRowIsSelected; identical root cause to #6049, which states the scenarios more precisely.

Keep open

Verified & Urgent (10)

• #4673 — Global filter doesn't run when accessor key points to an object — high; canonical for the global-filter cluster. Default getColumnCanGlobalFilter samples only flatRows[0] and requires string/number (globalFilteringFeature.ts:30-37), and the AND in column_getCanGlobalFilter means explicit enableGlobalFilter: true can never override — a custom globalFilterFn is silently skipped. PR #6439 open. This issue exercises both facets (nullish scan + opt-in), so keep it as canonical.
• #4783 — Global Filter Fails When First Row Value is Undefined — high. Same first-row-sampling defect: a column whose row-0 value is undefined/null is silently excluded from global filtering, returning empty search results on the default path. PRs #6252/#6438 open. Closeable as duplicate of #4673 once one fix merges.
• #4919 — Filter not working if accessorFn returns null for any row — high. The column-filter half is fixed in v9 (column_getAutoFilterFn scans past nullish), but the global-filter half persists. Same cluster and pending PRs as #4783.
• #5173 — row.getIsSomeSelected returning true for row with disabled subrows — medium. isSubRowSelected (rowSelectionFeature.utils.ts:909-941) still initializes allChildrenSelected = true and never clears it when all children are non-selectable, returning 'all' with zero selected rows and propagating bogus someSelected upward. Two open fix PRs (#5790, #6177) — land one on alpha.
• #5363 — getValue cache not invalidating when accessorFn is updated — high. row_getValue serves row._valuesCache forever and createCoreRowModel memoizes on [options.data] only, so swapping columns/accessorFn silently feeds stale values into rendering, sorting, and filtering. v8 PR #5582 shows the fix approach; same root cause as #4485/#4702.
• #5822 — Pinning aggregation rows then ungrouping causes a hard error — medium. table_getPinnedRows still calls the throwing table.getRow(rowId, true) unguarded (rowPinningFeature.utils.ts:126), so a pinned synthetic group id crashes getTopRows()/getBottomRows() after ungrouping. Reporter's fix PR #5823 open and unreviewed since 2024-12.
• #5968 — Invoking table.getRowModel() causes rows to collapse — high. v9 dropped v8's first-run auto-reset guard: memo fires onAfterUpdate on the first computation and createGroupedRowModel forces rowInputsChanged = true on that first run, so the first getRowModel() schedules table_autoResetExpanded/table_autoResetPageIndex — wiping controlled expanded and any initial pageIndex on mount (table_resetPageIndex(table, true) hard-codes 0, ignoring initialState). Canonical survivor of the #5970/#6025/#6030 cluster (all since closed).
• #6007 — When using filterVariant: 'range' and filtering with a min value of 0, null values are also included in the results — high. filterFn_inNumberRange has no resolveDataValue, so null/''/false coerce to 0 and match any zero-spanning range — and it is the auto-selected default filter for number columns. Sibling inDateRange already does this correctly. PR #6313 open.
• #6049 — Row Selection: Incorrect Parent Row Selection Behavior with Nested Data — high; canonical for the selection-reconciliation cluster. Deselecting descendants never removes the ancestor id and row_getIsSelected reads only its own id, so a parent can report getIsSelected() and getIsSomeSelected() true simultaneously; the shipped expanding example renders it fully checked with an unselected child, and getSelectedRowModel() keeps the stale parent. Close #4878/#5416 into this.
• #6078 — process is not defined when used in Vanilla JS (without Node.js) — high. Raw process.env.NODE_ENV reads persist in source (~14 sites) and survive into the published build (packages/table-core/dist/utils.js) because tsdown.config.ts has no define; any bundler-less consumer gets a hard ReferenceError, and v9 dropped the v8 UMD escape hatch. PR #6185 targets v8 main only — needs an alpha fix.

Valid (37)

High priority

• #4485 — React: accessor function don't trigger when data is outside of exported component — Same _valuesCache defect as #5363, kept for the columns-swap case: a brand-new columns array with a new accessorFn still serves stale values, so the documented "memoize columns on a dependency" pattern silently yields wrong data. One invalidation change fixes both.
• #5770 — Parent columns in groups are not respecting the isVisible value we're passing in — column_getIsVisible short-circuits on column.columns.length and returns childColumns.some(getIsVisible), so an explicit columnVisibility: { groupId: false } is ignored whenever any child is visible. Kevin approved the fix direction in-thread ("Only in the alpha. Want to limit breaking changes in main."); the alpha is the only window and nothing has landed.

Medium priority

• #4309 — toggleSorting preserve the multi selection — In non-multi mode, replace is chosen only when the column is not the last sorted entry (rowSortingFeature.utils.ts:253-262), so clicking the last sorted column preserves all other sorts while clicking an earlier one replaces — same gesture, two behaviors. PR #6431 open.
• #4825 — implement full width resize handler — The redistribute-delta-into-next-column mode still doesn't exist and is pure ColumnSizingState math (no DOM measurement), so the headless-boundary objection fails. Canonical tracking issue for #4880/#4852/#6010-family demand.
• #4946 — enableMultiRemove attribute is invalid — column_toggleSorting calls column_getNextSortingOrder(column) with no multi argument, so the documented option is dead on every path including the recommended handler.
• #5008 — Median aggregation fails if null values — aggregationFn_median bails to undefined on the first non-number while mean/sum/min/max skip nullish values; now documented as-is, making it intentional-by-inertia rather than by design.
• #5116 — getToggleAllRowsSelectedHandler doesn't take into account enableSubRowSelection property — Four select-all paths disagree: table_toggleAllRowsSelected ignores row_getCanSelectSubRows, the page variant honors it, and table_getIsAllPageRowsSelected still requires sub-rows selected, so the page header checkbox can never render checked with enableSubRowSelection: false. Contradicts docs; distinct from the #6049 cluster.
• #5138 — column.enableGlobalFilter is not considered in the default implemenation of table.getColumnCanGlobalFilter — The opt-in half of the #4673 cluster; PR #6439 targets exactly this. Workaround exists (custom getColumnCanGlobalFilter), hence medium; closeable as duplicate once the fix lands.
• #5147 — Sorting direction not updating properly with manual sorting and same values on current page — column_getAutoSortDir derives direction from flatRows[0] only, so a manual-sorting server data swap flips the toggle cycle and silently drops a state. sortDescFirst is the documented workaround; shares root cause with #5832.
• #5202 — The header object has the 'rowSpan' property, but there is an issue where it is not applied — Header.rowSpan is 0 for every header in every table shape (buildHeaderGroups.ts updateHeaderSpans); dead public API that should be implemented or removed before v9 stable.
• #5397 — there is a bug when you try sticky pin column group — table_getColumnOffsets builds offsets from visible leaf columns only, so column_getStart/getAfter fall through to ?? 0 for group columns and pinned group headers stack at offset 0. header.getStart() is correct, giving a workaround.
• #5398 — Toggle select all rows doesn't respect canSelect() condition — Deselect branches unconditionally delete while select branches guard on row_getCanSelect; a commenter documents the guard existed in PR #5061 commit b76bb8d and was dropped in 42f7595 — a regression, not a design decision.
• #5497 — Column visibility APIs do not work with column groups — column_toggleVisibility writes a dead state key for group ids while column_getIsVisible derives from children, and column_getCanHide advertises hideability it cannot deliver.
• #5653 — Custom sortingFn not being run when row key is undefined — The default sortUndefined: 1 short-circuits before invoking the user's sortFn for undefined pairs, and docs/framework/react/guide/sorting.md:430 describes false incorrectly ("considered tied" vs actually passing undefined through). At minimum a docs defect plus a legitimate pass-through-mode request.
• #5778 — ColumnGrouping.getDefaultColumnDef should not return aggregatedCell — The aggregation feature unconditionally injects a default aggregatedCell and FlexRender resolves aggregatedCell ?? cell, so a user's cell formatter can never win on aggregated cells without the aggregatedCell: null workaround. Small breaking change that belongs in the alpha.
• #5801 — autoResetExpanded not reset the expanded state on data change in V8 — In v9, table_autoResetExpanded is wired only from the grouped row model, so expanding-without-grouping never auto-resets on data change. Fix PR #6443 (base beta) implements the audit's fix note; close via that PR when it lands.
• #5832 — Unexpected sorting behavior arises due to the implementation in the @tanstack/table-core library — Same column_getAutoSortDir first-row sampling as #5147: a leading null flips the auto direction and the cycle silently drops a state.
• #5833 — flatRows contains duplicate sub-rows when pagination is enabled — _createPaginatedRowModel's walk pushes each row then recurses into subRows, but rows is already expandRows()-flattened, so expanded children land in flatRows twice.
• #5908 — Wrong type for GlobalFilterTableState — globalFilter: any lets a nested { globalFilter } object type-check via satisfies and silently misbehave, and the any poisons TableState. Widen to unknown or a threaded generic during the alpha.
• #5971 — Table<RowType> is not assignable to Table<unknown> — Re-verified with a scratch tsc 6.0.3 compile: assignment still fails through optionsStore atom invariance into contravariant option callbacks. The variance programme already fixed Table→Table_Internal; finishing the job fits current maintainer direction.
• #5987 — getFacetedRowModel inconsistently respects maxLeafRowFilterDepth: 0 — filterRowModelFromRoot keeps parents' original subRows but never pushes those descendants into flatRows/rowsById, so facet counts under-count. PR #6361 open.
• #6074 — filterFromLeafRows causes columnFiltersMeta to get wiped — filterRowModelFromLeafs copies only columnFilters when rebuilding rows, dropping columnFiltersMeta and breaking the fuzzy-filter + rank-sort recipe. One-line fix; PR #6075 (v8) needs an alpha port.
• #6081 — Custom getFacetedUniqueValues does not update on rerenders — Two cache layers freeze server-side faceting results (per-column factory cache + memoDeps: [facetedRowModel.flatRows]), making the documented "Custom Server-Side Faceting" recipe non-functional. PR #6417 open.
• #6101 — filterFn.autoRemove ignored for empty strings — shouldAutoRemoveFilter ORs the custom autoRemove result with a hardcoded empty-string check, so a custom autoRemove returning false for '' can never win — filtering for empty strings is impossible.
• #6115 — Expanding API including rows that dont include a subRow — row_toggleExpanded materializes expanded === true by writing every id from rowsById with no expandability filter, polluting state and skewing getExpandedDepth. PR #6116 (v8) needs an alpha port.
• #6230 — HeaderContext and CellContext types should use ReactTable — CellContext.table is typed as core Table while the runtime object in React carries Subscribe/FlexRender, and docs push table.Subscribe inside cell renderers, so users hit type errors on real APIs. Kevin acknowledged the gap in-thread (2026-04-29); fix lives in the adapter via module augmentation, not core.

Low priority

• #4512 — Column resizing on scaled transformation — header_getResizeHandler captures raw viewport clientX with no scale input, so resizing drifts by the factor k under transform: scale(k). Needs an API decision (scale option or derived scale).
• #4939 — setting enableExpanding has no effect. — row_toggleExpanded and table_toggleAllRowsExpanded never consult row_getCanExpand, unlike the selection twin which guards its imperative path. Cross-feature inconsistency, trivial fix; the documented handler path works.
• #5019 — Solid Table: Examples from docs cause unnecessary rerenders of components when data updates — Every v9 Solid example still uses <For>, which recreates all row DOM on data change, and the Solid guides contain no <Index>/<Key>/createStore guidance. Docs plus one fine-grained example; the in-thread solution works.
• #5275 — A table doesn't get updated when a defaultColumn changes — constructColumn bakes the default def in at construction while getAllColumns memoizes on [options.columns] only, yet getDefaultColumnDef itself re-memoizes on [options.defaultColumn] — the table reports the new default while columns keep the old one. Defect or docs gap; fix has a tradeoff around inline defaultColumn objects.
• #5850 — Table: Row selection is not cleaned up when table data is removed. — By-design as filed, but the thread converged on table.removeRowSelection(ids) which Kevin explicitly agreed to add to the v9 alpha; no such helper exists yet. Keep as an accepted enhancement.
• #5864 — Add mode() aggregation function to find most common unique value. — Tree-shakeable individual aggregation fns make this near-zero-cost; the reporter supplied a tested implementation.
• #5909 — TableState does not account for options.state properties being set as undefined — table_syncExternalStateToBaseAtoms still writes undefined slices into base atoms with no guard (survives the #6458 refactor), violating the non-optional TableState type. Downstream code is defensive, so a type-contract violation rather than a visible failure.
• #6077 — [Vue] Hydration error in FlexRender when column is an empty string — Vue flexRender returns string results verbatim, so an empty-string cell produces an SSR text-node mismatch. PR #6186 (v8) needs an alpha port.
• #6136 — Calling row.toggleExpanded(bool) triggers re-render even without state change — table_setExpanded invokes onExpandedChange unconditionally even for no-op updates, so controlled-state consumers get spurious callbacks; the uncontrolled atom path already bails on identity. PR #6184 (v8) needs an alpha port.
• #6175 — Unable to view examples in Codesandbox without forking — No example Vite config on alpha sets server.allowedHosts, so CodeSandbox devbox embeds are blocked by Vite's host check. PR #6176 open.
• #6302 — Accessor key does not have the same typing as Form's name key — accessorKey is (string & {}) | keyof TData while columnHelper.accessor already constrains to DeepKeys<TData>; the library ships the requested type and simply doesn't use it on the raw ColumnDef form. Gated on a TS-perf check against wide TData.

Maybe & Needs Investigation (2)

• #5696 — Table using insane amount of memory even with virtualisation when scrolling — row_getAllCells still materializes a Cell per leaf column per row into row._cellsCache (now a WeakMap) retained for the lifetime of a stable columns reference, so column virtualization does not avoid rows×cols cell construction. Prior scratch estimate ~122 B/cell (~470 MB extrapolated at 2000×2000); needs a fresh browser heap profile separating table vs virtual vs renderer cost.
• #6011 — virtual columns experimental can not scroll smoothly when nearing the end — The repro is this repo's own shipped virtualized-columns-experimental example, so even if the root cause is TanStack Virtual measurement, the fix is plausibly example-side (row virtualizer measureElement changing total height mid-drag). One cheap experiment (fixed row estimateSize or overflow-y: scroll) decides between fixing the example and transferring to the Virtual repo.

Root-cause clusters (updated from the July 26 audits)

#	Root cause	Open issues	Status
1	v9 dropped v8's first-run auto-reset guard; first getRowModel() wipes controlled/initial expanded/pageIndex	#5968 (canonical; #5970/#6025/#6030 already closed)	Unfixed; table_resetPageIndex(table, true) also ignores initialState
2	Global-filter eligibility decided from flatRows[0]; explicit enableGlobalFilter: true cannot override	#4673 (canonical), #4783, #4919, #5138	PRs #6252/#6438/#6439 open
3	row._valuesCache never invalidated when columns/accessorFn change	#4485, #5363 (#4702 closed)	v8 PR #5582 open; port concept to alpha
4	Parent/child selection never reconciled upward	#6049 (canonical), #5116, #5398; close #4878/#5416 as dups	Unfixed; one refactor fixes the family
5	column_getAutoSortDir samples flatRows[0]	#5147, #5832	Unfixed; sortDescFirst workaround
6	Non-multi sort toggle preserves other sorts on the last-sorted column	#4309 (#6070 closed)	PR #6431 open
7	Group (non-leaf) columns second-class in visibility/offsets	#5497, #5770, #5397 (#4872/#5577 closed as docs-direction)	Unfixed; #5770 has maintainer-approved direction
8	v8-targeted fix PRs never ported to alpha	#6074, #6077, #6078, #6115, #6136, #5822	Six open PRs against main need alpha ports

Changes vs the July 26 audits

• #6224 / #6450: were Verified & Urgent in both audits; fixed by PR #6458 (merged 2026-07-29) — useTable now defers controlled-state publication to a layout effect. Close both.
• #5026 / #4634: same fix family; the 2026-07-27 maintainer comment plus #6458 flips them from open questions to closeable.
• #6078: was Maybe/Valid; escalated to Verified & Urgent — the process.env reads demonstrably survive into the published v9 dist and there is no workaround.
• #4994: was Verified & Urgent in the GPT audit; flips to close — the reporter explicitly opted out with autoResetPageIndex: false, and the docs assign reset responsibility to the app; the real clamp gap is tracked by #4703 (closed) and PR #5031.
• #4878 / #5416: keep-verdicts consolidated into close-as-duplicate of canonical #6049.
• #5850: GPT's close verdict overturned — Kevin committed in-thread to adding removeRowSelection(ids) to the v9 alpha; the helper does not exist yet.