Repository navigation
Conversation
architecture:list-windowing gains INV4: every virtualizer, the shared engine and each consumer, works inside a container scroll area and with page (document) scroll. A Requirements section adds FR1-FR4 for the shared engine: configurable anchoring (first visible, end, item), an anchored row that keeps its position through streaming inserts and item resizes, variable row heights, and container and page scroll. The engine is chosen by a measured comparison on performance and row-position accuracy. DEC-2 and DEC-3 quote the owner's decisions. Current-state impact and Verification record Table's INV4 gap.
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
|
AI review status for this pull request.
|
|
/approve-spec 65f5de5 |
There was a problem hiding this comment.
Changes requested: an engineering owner approved the new list-windowing requirements for this head, but the record leaves two problems to fix first.
- INV4 requires every consumer, Table included, to window with page scroll, while
component:TableFR19 and theuseTableRowsVirtualizationmodule record still require a bounded Table-owned viewport and list external-root mode as a non-goal; amend those clauses in this pull request (each with its own gap line) or scope INV4 so Table's bounded-viewport mode is a named exception. - DEC-2 quotes the dropped scroll-ref requirement and says the second statement "refines" the first although it replaces it; drop that quote and sentence, and keep at most
Rejected: a caller-provided scroll ref.inside DEC-2.
An inline note also flags that Current-state impact still says Table's virtualization hook is not built, right above the new gap line. A new commit needs a fresh /approve-spec for its head.
[Automated review]
… budget (INV5, DEC-4)
|
/approve-spec b96655b |
PR Analysis Report
Generated by PR Enrichment workflow | View current CI run |
There was a problem hiding this comment.
Changes requested: an engineering owner approved the new list-windowing requirements for this head, but the record has five problems to fix first.
- INV4 requires every consumer, Table included, to window with page scroll, while
component:TableFR19 and theuseTableRowsVirtualizationmodule record still require a bounded Table-owned viewport and list external-root mode as a non-goal; amend those clauses here, each with its own gap line, or scope INV4 so Table's bounded-viewport mode is a named exception. - The second INV5 gap line measures "the chat virtualizer's own code", but no chat virtualizer exists in Core or Lab on
main(the only one is in the open pull request #5461), so drop that line until one lands in Core. - The INV5 Verification row cites "the size check's dedicated budget entries", which no size check in the repository has, so mark it as an evidence gap the way the INV1–INV4 rows are marked.
- INV5 and DEC-4 record the virtua comparison and its dated measurement, which
architecture:knowledge-contractsINV10 keeps out of records, so keep the budgets and their build settings, link the measurement from the pull request, and update the description's claim scope and rubric results to cover INV5 and DEC-4. - DEC-2 quotes the dropped scroll-ref requirement and says the second statement "refines" the first although it replaces it, so drop that quote and sentence and keep at most
Rejected: a caller-provided scroll ref.inside DEC-2.
Smaller notes: the Governing rules row leaves out DEC-4, and the unchanged "no module exists" and "not built" lines now contradict the new gap lines that say Table windows. A new commit needs a fresh /approve-spec for its head.
[Automated review]
|
/approve-spec b6c4a7c |
|
/approve-spec 72c69ef |
There was a problem hiding this comment.
Changes requested: an engineering owner approved the new list-windowing requirements for this head, but five record problems remain.
- INV4 requires every consumer, Table included, to window with page scroll, while
component:TableFR19 and theuseTableRowsVirtualizationmodule record still require a bounded Table-owned viewport and list external-root mode as a non-goal; amend those clauses here, each with its own gap line, or scope INV4 so Table's bounded-viewport mode is a named exception. - The INV5 Verification row cites "the size check's dedicated budget entries", which no size check in the repository has, so mark it as an evidence gap the way the INV1–INV4 rows are marked.
- The gap line "a plain Table that never windows carries about 6.2 KB" names no limit INV5 states (INV5 caps the engine and each adapter, and measures an adapter with the engine left out), so add that limit to INV5 or drop the line.
- INV5 records the virtua comparison and DEC-4 narrates its approval and measurement date, which
architecture:knowledge-contractsINV10 keeps out of records, so keep the budgets, build settings and adapter measure, keep one concluding sentence in DEC-4, link the measurement from the pull request, and update the description's claim scope and rubric results to cover INV5 and DEC-4. - DEC-2 quotes the dropped scroll-ref requirement and says the second statement "refines" the first although it replaces it, so drop that quote and sentence and keep at most
Rejected: a caller-provided scroll ref.inside DEC-2.
The branch also needs a rebase onto main that keeps main's approved_by and DEC-1 approval lines, then a fresh /approve-spec for the new head. Naming the chat gap line's subject as Lab's ChatVirtualizer and adding DEC-4 to the Governing rules row would also help.
[Automated review]
|
/approve-spec 9090e4e |
…e owner's words are paraphrased
|
/approve-spec ecfd71a |
There was a problem hiding this comment.
Changes requested. The list-windowing record at this head is owner-approved, and its Table exception and rename agree with Table's component record and the row virtualization module record on main; one description fix remains before it merges:
- The description still says DEC-2, DEC-3 and DEC-4 are "kept verbatim", that DEC-6 "quotes the owner's words that state the rename", and that DEC-4 is "left as approved", but this head paraphrases the owner in DEC-3, DEC-4 and DEC-6 (lines 176–206), so update those bullets to say the decisions paraphrase the owner and each Decider line names its time.
Two non-blocking notes: the INV5 gap line for Table's adapter (lines 126–127) says about 2.4 KB while a build with INV5's settings gives 3.7–3.8 KB, so "over 2 KB" keeps it accurate; and DEC-3 and DEC-6 now open with "The owner set…" and "The owner decided that…", which the Decider lines already record, so stating each decision in the present tense, as DEC-2 does, keeps the record about the system.
Claims and files checked
Review mode: full; reason: spec-record
- Claim: records only, one record changes — verified (one Markdown record in the file list).
- Claim: every virtualizer must work in container and page scroll, with Table's bounded viewport the one named exception and an INV4 gap line — verified (lines 105–109, 124–125).
- Claim: the engine needs configurable anchoring, a stable anchored row through streaming inserts and resizes, and variable row heights, within a size budget — verified (lines 72–81, 110).
- Claim: Table's public windowing API changes by one rename — verified (lines 99–104, 190–206).
- Claim: the branch merges
mainwithout a rebase and keepsmain's approval lines — verified (earlier heads are ancestors; frontmatter and DEC-1 lines 154–156 unchanged). - Claim: Table's component record keeps row windowing in Table's own viewport with no unknown external root, and the row module record's contract is
{rowHeight, initialViewportHeight}with external-root mode a non-goal — verified onmain. - Claim: the engine's home and scroll-area attachment are open work in other pull requests, and the rename is merged in code and in the module record — verified.
- Claim: INV3's one exception agrees with the module record, which this pull request does not edit — verified.
- Claim: INV4's exception leaves Table's component record unamended — verified.
- Claim: INV5's budgets and build settings — verified (line 110).
- Claim: DEC-2, DEC-3 and DEC-4 keep the owner's decisions verbatim — false (DEC-3 and DEC-4 paraphrase at this head, lines 176–188; DEC-2 has no quote).
- Claim: DEC-6 quotes the owner's words for the rename — false (lines 195–206 paraphrase); its deferral of further options to a later owner decision — verified (lines 197–202).
- Claim: DEC-4's owner-approved text is left verbatim — false (its quoted approval was removed, lines 186–188); its sentence naming the measured library and date is unchanged — verified.
- Claim: DEC-5 lives in another pull request and nothing is added for it here — verified (no DEC-5).
- Claim: the 4.9 KB budget is a comparable library's size with the same settings — not re-measured (the library is not a repository dependency).
- Claim: the shared engine measures 3.75 KB and Lab's chat virtualizer 5.00 KB — verified (the measured code is unchanged since an earlier build gave 3,855 B and 5,125 B gzip).
- Claim: Table's adapter measures 2.37 KB and the plain-Table windowing stack 4.31 KB — not reproduced (3.7–3.8 KB for the adapter); every build puts it over 2 KB, as the gap states (non-blocking).
- Claim: the earlier 5.3, 6.2 and 4.7 KB figures are replaced — verified (lines 126–130).
- Claim: DEC-2 states its decision without quoting chat, and no scroll-ref requirement appears outside DEC-2 — verified (lines 161–169).
- Claim: two rejected alternatives, each one line inside DEC-2 and DEC-6 — verified (lines 169, 202).
- Claim: the stale status lines now match
main, and Table's row and column windowing use the shared module — verified (lines 114–117, 158). - Claim: the merge keeps
main's record rewrite with this pull request's requirements, invariants and decisions on top — verified. - Claim: the plain-Table carry line is dropped because the import-boundaries record owns it — verified.
- Claim: the FR1 and FR3 gap lines, the fixed-pitch sentence moved out of the System model, and the End-user impact row naming INV4's exception — verified (lines 32, 54–55, 120–123).
- Claim: the engine's API and path, the candidate choice, OQ1, consumer rendering and Table's FR19 are out of scope — verified.
- Claim: Table windows through the shared module only inside its own bounded viewport, with no code change here — verified.
- Claim: writing rubric results R1–R7 — verified (no change references or status phrases in the record; OQ1 unchanged).
- Claim: the knowledge check, Prettier and the whitespace check pass — verified at this head.
- Claim: no internal context in public text — verified.
- File:
docs/architecture/list-windowing.md— one blocking finding (the description against DEC-3, DEC-4 and DEC-6); non-blocking notes at lines 126–127, 176 and 195.
…; Table's adapter is over 2 KB
|
/approve-spec 889abb2 |
There was a problem hiding this comment.
Approved pending CI: CI has not run on this head yet, because the CI and Lint jobs skip while the pull request is a draft.
The list-windowing record at this head is owner-approved, its Table exception and rename agree with Table's component record and the row virtualization module record on main, and the description now says the decisions paraphrase the owner and name their times.
One non-blocking note: in DEC-6 (line 198), "Other options described for later" no longer says where they were described, so dropping "described for later" keeps the decision whole.
Claims and files checked
Review mode: full; reason: spec-record
- Claim: records only, one record changes — verified (one Markdown record in the file list).
- Claim: every virtualizer must work in container and page scroll, with Table's bounded viewport the one named exception and an INV4 gap line — verified (lines 105–109, 124–125).
- Claim: the engine needs configurable anchoring, a stable anchored row through streaming inserts and resizes, variable row heights and a size budget — verified (lines 72–81, 110).
- Claim: Table's public windowing API changes by one rename — verified (lines 99–104, 190–206).
- Claim: the branch merges
mainwithout a rebase or force push and keepsmain's approval lines — verified (every earlier approved head is an ancestor; frontmatter and DEC-1 lines 155–156 unchanged). - Claim: Table's component record keeps row windowing in Table's own viewport with no unknown external root, and the row module record's contract is
{rowHeight, initialViewportHeight}with external-root mode a non-goal — verified onmain. - Claim: the engine's home and the scroll-area attachment are open work in other pull requests, and the rename is merged in code and in the module record — verified.
- Claim: INV3's one exception agrees with the module record, which this pull request does not edit — verified.
- Claim: INV4's exception leaves Table's component record unamended — verified.
- Claim: FR1–FR4 as listed — verified (lines 72–81).
- Claim: INV5's budgets and build settings — verified (line 110).
- Claim: DEC-2, DEC-3, DEC-4 and DEC-6 state their decisions in the present tense without quoting chat, and each Decider line names its time — verified (lines 161–206).
- Claim: DEC-6 limits Table's public API to the rename and leaves the other options to a later owner decision — verified (lines 195–202).
- Claim: DEC-5 lives in another pull request and nothing is added for it here — verified (that pull request adds DEC-5; this record has none).
- Claim: the 4.9 KB budget is a comparable library's size with the same settings — not re-measured (that library is not a dependency of this repository).
- Claim: the shared engine measures 3.75 KB and Lab's chat virtualizer 5.00 KB — verified (3,855 B and 5,131 B gzip with INV5's settings).
- Claim: Table's adapter is 2.37 KB by INV5's definition and 3.7–3.8 KB counted another way, so the record says only that it is over 2 KB — the exact figures are not reproduced; the adapter's own files measure 3,238 B with the engine and Core helpers left out, so "over 2 KB" holds.
- Claim: a plain Table that never windows carries 4.31 KB of the windowing stack — not reproduced (the import-boundaries record on
mainstates 3.6 KB for that carry); no line of this record rests on it. - Claim: the earlier 5.3, 6.2 and 4.7 KB figures are replaced — verified (none is in the record).
- Claim: the decisions keep what each decides in paraphrase, the record outside DEC-4 carries no benchmark narration, and no scroll-ref requirement appears — verified (INV5 states its basis without naming the library; DEC-2 line 167).
- Claim: two rejected alternatives, each one line inside DEC-2 and DEC-6 — verified (lines 169, 202).
- Claim: the added or changed sections are as listed — verified (the INV1 and INV2 Verification rows are only re-padded).
- Claim: the stale status lines now match
main, and Table's row and column windowing use the shared module — verified (lines 114–117, 158; Table's window hooks import it). - Claim: the merge keeps
main's record rewrite with this pull request's requirements, invariants and decisions on top — verified. - Claim: the plain-Table carry line is dropped because the import-boundaries record owns it — verified (its INV7 gap on
main). - Claim: the FR1 and FR3 gap lines, the fixed-pitch sentence moved out of the System model, and the End-user impact row naming INV4's exception — verified (lines 32, 54–55, 120–123).
- Claim: the engine's API and path, the candidate choice, OQ1, consumer rendering and Table's FR19 are out of scope — verified.
- Claim: Table windows through the shared module only inside its own bounded viewport, with no code change here — verified.
- Claim: writing rubric results R1–R7 — verified (no change references or status phrases in the record; OQ1 unchanged).
- Claim: the knowledge check, Prettier and the whitespace check pass — verified at this head.
- Claim: no internal context in public text — verified.
- File:
docs/architecture/list-windowing.md— no blocking finding; one non-blocking note at line 198.
|
/approve-spec 003cb68 |
There was a problem hiding this comment.
Approved. The list-windowing record at this head is owner-approved, and CI passes.
Its Table exception and the initialViewportHeight rename agree with Table's component record and the row virtualization module record on main. Nothing is left for the author to change.
Claims and files checked
Review mode: same-head; reason: none
- Claim: records only, and one record changes. Verified (the file list holds one Markdown record).
- Claim: every virtualizer must work in container and page scroll, with Table's bounded scroll area the one named exception and an INV4 gap line. Verified (lines 105–109, 124–125).
- Claim: the engine needs configurable anchoring, a stable anchored row through streaming inserts and resizes, variable row heights, and a size budget. Verified (lines 72–81, 110).
- Claim: Table's public windowing API changes by one rename. Verified (lines 99–104, 190–206).
- Claim: the branch merges
mainwithout a rebase or force push and keepsmain's approval lines. Verified (every earlier approved head is an ancestor of this one; the frontmatter and DEC-1 lines 154–156 are unchanged). - Claim: Table's component record keeps row windowing in Table's own scroll area with no unknown external root, and the row module record's contract is
{rowHeight, initialViewportHeight}, with external-root mode a non-goal. Verified onmain. - Claim: the engine's home and the scroll-area attachment are open in other pull requests, and the rename is merged in code and in the module record. Verified (both are open; the rename pull request is merged).
- Claim: the author searched the listed terms. Not checkable (it describes the author's search).
- Claim: INV3's one exception agrees with the module record, which this pull request does not edit. Verified.
- Claim: INV4's exception leaves Table's component record unamended. Verified.
- Claim: FR1–FR4 as listed. Verified (lines 72–81).
- Claim: INV5's budgets and build settings. Verified (line 110).
- Claim: DEC-2, DEC-3, DEC-4 and DEC-6 state their decisions in the present tense without quoting chat, and each Decider line names its time. Verified (lines 161–206).
- Claim: DEC-6 limits Table's public API to the rename and leaves the other options to a later owner decision. Verified (lines 195–202).
- Claim: DEC-5 lives in another pull request, and nothing is added for it here. Verified (that pull request adds DEC-5; this record has none).
- Claim: the 4.9 KB budget is a comparable library's size with the same settings. Not re-measured (that library is not a dependency of this repository).
- Claim: the shared engine measures 3.75 KB and Lab's chat virtualizer 5.00 KB. Verified (3,855 B and 5,121 B gzip with INV5's settings).
- Claim: Table's adapter is 2.37 KB by INV5's definition and 3.7–3.8 KB counted another way, so the record says only that it is over 2 KB. The exact figures are not reproduced; the adapter's own files measure 4,669 B with the engine and Core helpers left out, so "over 2 KB" holds.
- Claim: a plain Table that never windows carries 4.31 KB of the windowing stack. Not reproduced (the import-boundaries record on
mainstates 3.6 KB for that carry); no line of this record rests on it. - Claim: the earlier 5.3, 6.2 and 4.7 KB figures are replaced. Verified (none is in the record).
- Claim: the decisions paraphrase the owner, the record outside DEC-4 carries no benchmark narration, and no scroll-ref requirement appears. Verified (INV5 states its basis without naming the library; DEC-2, line 167).
- Claim: two rejected alternatives, each one line inside DEC-2 and DEC-6. Verified (lines 169, 202).
- Claim: the added or changed sections are as listed. Verified (the INV1 and INV2 Verification rows are only re-padded).
- Claim: the stale status lines now match
main, and Table's row and column windowing use the shared module. Verified (lines 114–117, 158). - Claim: the merge keeps
main's record rewrite, with this pull request's requirements, invariants and decisions on top. Verified. - Claim: the plain-Table carry line is dropped because the import-boundaries record owns it. Verified (its INV7 gap on
main). - Claim: the FR1 and FR3 gap lines, the fixed-pitch sentence moved out of the System model, and the End-user impact row naming INV4's exception. Verified (lines 32, 54–55, 120–123).
- Claim: the engine's API and path, the candidate choice, OQ1, consumer rendering and Table's FR19 are out of scope. Verified.
- Claim: Table windows through the shared module only inside its own bounded scroll area, with no code change here. Verified.
- Claim: writing rubric results R1–R7. Verified (no change references or status phrases in the record; OQ1 is unchanged).
- Claim: the scope checklist and validation (records only, no internal context in public text; the knowledge check, Prettier and the whitespace check pass). Verified at this head.
- File:
docs/architecture/list-windowing.md. Clear.
|
/approve-spec 1954ea1 |
There was a problem hiding this comment.
Approved pending CI: CI has not run on this head yet. The list-windowing record at this head is approved by an engineering owner, and its Table exception and the initialViewportHeight rename agree with Table's component record and the row virtualization module record on main.
CI runs once the pull request is marked ready for review; nothing else is left for the author to change.
Claims and files checked
Review mode: full; reason: spec-record
- Claim: Records only, and one record changes. Verified (the file list holds one Markdown record).
- Claim: Every virtualizer must work in container and page scroll, with Table's bounded scroll area the one named exception and an INV4 gap line. Verified (lines 105–109, 124–125).
- Claim: The engine needs configurable anchoring, a stable anchored row through streaming data and resizes, variable row heights and a size budget. Verified (lines 72–81, 110).
- Claim: Table's public windowing API changes by one rename. Verified (lines 99–104, 190–206).
- Claim: The branch merges
mainwithout a rebase or force push and keepsmain's approval lines. Verified (every earlier approved head is an ancestor of this one; the frontmatter and DEC-1 lines 154–156 are unchanged). - Claim: Table's component record keeps row windowing in Table's own scroll area with no unknown external root, and the row module record's contract is
{rowHeight, initialViewportHeight}with external-root mode a non-goal. Verified onmain. - Claim: The engine's home and the scroll-area attachment are open in other pull requests, and the rename is merged in code and in the module record. Verified (both are open; the rename pull request is merged).
- Claim: The author searched the listed terms. Not checkable (it describes the author's search).
- Claim: INV3's one exception agrees with the module record, which this pull request does not edit. Verified.
- Claim: INV4's exception leaves Table's component record unamended. Verified.
- Claim: FR1–FR4 as listed. Verified (lines 72–81).
- Claim: INV5's budgets and build settings. Verified (line 110).
- Claim: DEC-2, DEC-3, DEC-4 and DEC-6 state their decisions in the present tense without quoting chat, and each Decider line names its decision's time. Verified (lines 161–206; the Decider lines at 164, 174, 186 and 193 give the time to the second).
- Claim: DEC-6 limits Table's public API to the rename and leaves the other options to a later owner decision. Verified (lines 195–202).
- Claim: DEC-5 lives in another pull request, and nothing is added for it here. Verified (that pull request adds DEC-5; this record has none).
- Claim: The 4.9 KB budget is a comparable library's size with the same settings. Not re-measured (that library is not a dependency of this repository).
- Claim: The shared engine measures 3.75 KB, under the budget. Verified (3,855 B gzip with INV5's settings).
- Claim: Table's adapter is 2.37 KB by INV5's definition and 3.7–3.8 KB counted another way, so the record says only that it is over 2 KB. The exact figures are not reproduced; the adapter's own files measure 4,695 B with the engine and Core helpers left out, so "over 2 KB" holds.
- Claim: A plain Table that never windows carries 4.31 KB of the windowing stack. Not reproduced (the import-boundaries record on
mainstates 3.6 KB for that carry); no line of this record rests on it. - Claim: Lab's chat virtualizer measures 5.00 KB in its own code and does not use the shared engine. Verified (5,125 B gzip; it imports no part of the engine).
- Claim: The earlier 5.3, 6.2 and 4.7 KB figures are replaced. Verified (none is in the record).
- Claim: The decisions paraphrase the owner, the record outside DEC-4 carries no benchmark narration, and no scroll-ref requirement appears. Verified (INV5 states its basis without naming the library; DEC-2, line 167).
- Claim: Two rejected alternatives, each one line inside DEC-2 and DEC-6. Verified (lines 169, 202).
- Claim: The added or changed sections are as listed. Verified (the INV1 and INV2 Verification rows are only re-padded).
- Claim: The stale status lines now match
main, and Table's row and column windowing use the shared module. Verified (lines 114–117, 158). - Claim: The merge keeps
main's record rewrite, with this pull request's requirements, invariants and decisions on top. Verified. - Claim: The plain-Table carry line is dropped because the import-boundaries record owns it. Verified (its INV7 gap on
main). - Claim: The FR1 and FR3 gap lines, the fixed-pitch sentence moved out of the System model, and the End-user impact row naming INV4's exception. Verified (lines 32, 54–55, 120–123).
- Claim: The engine's API and path, the candidate choice, OQ1, consumer rendering and Table's FR19 are out of scope. Verified.
- Claim: Table windows through the shared module only inside its own bounded scroll area, with no code change here. Verified (the Table window hook imports the shared list window; no package file changes).
- Claim: Writing rubric results R1–R7. Verified (no pull request references or status phrases in the record; OQ1 is unchanged).
- Claim: The scope checklist and validation (records only, no internal context in public text; the knowledge check, Prettier and the whitespace check pass). Verified at this head (all three rerun).
- File:
docs/architecture/list-windowing.md. Clear.
|
/approve-spec 2e8d8b9 |
Records only. People get windowed long lists whether the list scrolls inside its own container or with the page: the list-windowing record requires every virtualizer to work in both, sets the shared engine's requirements (configurable anchoring, stable position with streaming data, variable row heights) and its size budget, and fixes Table's public windowing API to one rename. Table's bounded, Table-owned viewport is the one named exception to page and container scroll for now, with an INV4 gap line.
The branch merges
main(no rebase, so no force push) and keepsmain'sapproved_byand DEC-1 approval lines as they are.Exact decision needed
Which scroll contexts a virtualizer in Core must support, which requirements and size budget the shared windowing engine must meet, and what changes in Table's public windowing API.
Existing authority searched
architecture:list-windowing: System model, INV1–INV3, Current-state impact, DEC-1, OQ1 and Verification.component:TableFR19 (windowing on Table's own bounded viewport, no unknown external root) andmodule:Table/useTableRowsVirtualization(public contract{rowHeight, initialViewportHeight}onmain, external-root mode a non-goal).initialViewportHeightrename, in code and in the module record, is onmain(feat(Table): useTableRowsVirtualization's viewportHeight becomes initialViewportHeight, which no longer sizes the Table #10894, merged).windowing,virtualization,scroll container,page scroll,anchor,variable height,viewportHeight.Decision
viewportHeighttoinitialViewportHeight.module:Table/useTableRowsVirtualizationalready carries that rename onmain(Public contract, Geometry row, FR10 seed), so the two records agree; this PR does not edit the module record.component:TableFR19) until a later decision moves it to page or container scroll (INV4 gap).component:TableFR19 is not amended here.initialViewportHeightrename. Other options (a callback scroll target, exact or estimated item sizes, an anchor) reach Table's public API only through a later owner decision.Measurement behind INV5 (moved out of the record)
The 4.9 KB engine budget is the size of virtua's
VirtualizerplusWindowVirtualizer, measured on 2026-11-09 with the same build settings (esbuild, ESM, minified, React external, gzip -9). The current-state measures were taken again onmainat this head's merge (esbuild 0.28.2, ESM, minified, React, React DOM, StyleX and other@astryxdesign/coreentries external, gzip -9, from source with StyleX calls left uncompiled):@astryxdesign/core/internal/virtualization): 3.75 KB, under the 4.9 KB budget.Table/plugins/virtualization/andtableColumnWindowing.ts): 2.37 KB, over the 2 KB budget. A build with INV5's settings that counts the adapter's own wiring differently gives 3.7–3.8 KB, so the record states only that it is over 2 KB.ChatVirtualizerown code: 5.00 KB. It does not use the shared engine.The earlier figures (5.3, 6.2 and 4.7 KB) predate the shared engine on
mainand are replaced.Owner-approved text
Rejected alternatives
Rejected:line).Claim scope and non-goals
main: the shared module ispackages/core/src/virtualization/(private entryinternal/virtualization.ts), and Table's row and column virtualization use it, so "no module exists", "not built" and DEC-1's "spec only" are gone (architecture:knowledge-contractsDEC-20). DEC-1's decision is otherwise unchanged.mainat this head (no rebase): main's own record rewrite (shipped engine, column windowing, API-delta evidence) is kept, with this pull request's FR1–FR4, INV3–INV5 and DEC-2–DEC-6 on top.architecture:component-import-boundariesINV7's gap, not an INV5 cap.mainmoved there). The End-user-impact row names INV4's Table exception.component:TableFR19.Evidence and implementation relationship
mainwindows throughpackages/core/src/virtualization/(useListWindow, throughuseTableWindow), only inside its own bounded viewport. No code changes.Writing rubric results
#followed by digits,/pull/and/issues/finds nothing in the record.Scope
Validation
pnpm check:knowledge: Knowledge templates and records are aligned.git diff --check