fix(docs): [SITE-04] preserve structural chunks and clarify settings

refactor/changes/2026-09-16-SITE-04-setting.md

@@ -0,0 +1,51 @@
+# SITE-04 checkpoint: settings and Markdown chunk boundaries
+
+The bilingual setting guides now document the tree model, actual names/returns,
+callbacks, deferred mounting, lazy nodes, ownership, update recovery and cleanup.
+Root declaration inaccuracies remain compatible and are explained alongside the
+accurate runtime types. The range-to-switch example previously retained range,
+whose rendering precedence kept the slider visible. Both examples now remove
+that old field before updating. Other runnable examples remain unchanged.
+
+The existing documentation regression exposed a separate defect while processing
+the new table: splitTranslation could insert a paragraph break between table rows
+when chunks were joined, causing its own structural validator to reject the result.
+The splitter now keeps tables, nested lists and blockquotes together. A structural
+block above the requested limit fails explicitly; it does not silently exceed the
+limit or weaken validation. Two regressions cover boundary placement and oversized
+blocks. The documentation tooling README explains the behavior and maintenance.
+No player runtime, public declaration, package API or dependency changed.
+
+The [mapping](../core-content-review.json) adds 79 root/runtime setting records.
+Total: 324 reviewed, 639 remaining out of 963. Counts include declaration aliases,
+not independent features. Previous mapped source and guide hashes are unchanged.
+
+## Verification
+
+- Two strict TS examples pass NodeNext with skipLibCheck:false and types:[].
+  All 18 runnable snippets parse; only the two corresponding update examples
+  change. Local links/anchors resolve. Six three-engine page cases and 54 exact
+  code forwards pass; popup targets are isolated rather than executing the editor.
+- Chromium 153.0.8010.12, Firefox 155.0 and Windows WebKit 26.6 each pass 40 DOM
+  assertions against core bytes matched to the recorded archive. Checks include
+  model identity, delayed/canceled mounted callbacks, switch/range semantics,
+  nested selection, stale results, update recovery/listeners, tree rejection,
+  old example reproduction and corrected kind switching, removal and destruction.
+- The exact corrected Chinese example separately reaches ready, runs its own
+  three-second update, and displays a switch with no range input on all engines.
+  No retry or page error. Readiness/DOM evidence is not full playback/device proof.
+- Initial tests: 39 pass, one structure failure. After fixing the splitter and
+  adding regressions: 42 pass, zero fail/skip, 1276.5411ms. The old function at
+  bb2fec8dd reproduces the new-guide failure; the new function preserves structure
+  in two chunks of 3655/2464 characters within the 4000-character limit.
+- Node 24.21.0, Yarn 1.22.22, TypeScript 5.9.3. Final docs build 9.18s
+  (Yarn 10.82s), LLM generation/check, strict docs-tools types (3.52s), root lint
+  (12.32s), site inventory and demo-route checks pass. Screenshots inspected.
+
+See [evidence](../baselines/site04-setting.json). Shared script/test changes make
+all 21 existing library build input fingerprints stale. This is recorded, not
+silently rebound to previous candidates; REL-02 must prepare/verify fresh ones.
+The DOM probe checks unchanged core bytes, not a new release candidate acceptance.
+SITE-04 remains doing, totals 224/22/39. Continue implementation and candidate
+preparation; formal review rounds remain user-directed. No push or publication.
+Revert this checkpoint to restore guides, generated outputs, mapping and tool fix.