objectstack-fleet · Oct 8, 2026 · Oct 8, 2026 · Oct 8, 2026 · Oct 8, 2026
diff --git a/.changeset/11854-vendor-markdown-initial.md b/.changeset/11854-vendor-markdown-initial.md
@@ -0,0 +1,9 @@
+---
+'@object-ui/console': patch
+---
+
+The console's first page load no longer downloads the documentation reader's markdown extras (objectui#11854).
+
+The console's `vendor-markdown` chunk is part of every first page load, because the chat message renderer uses most of the markdown pipeline. The chunk also held the parts only the `markdown` component and the package documentation reader use: code highlighting (`rehype-highlight` with `lowlight` and `highlight.js`), heading anchors (`rehype-slug`, `rehype-autolink-headings`, `github-slugger`), GitHub-style alerts (`remark-github-blockquote-alert`) and their helpers. Every first page load downloaded them, for every user. They now load with `@object-ui/plugin-markdown`, the first time a page renders a `markdown` component or a documentation page. `react-markdown` moves to a small chunk of its own, which the first load still fetches because a lazily rendered markdown field shares a chunk with eagerly loaded components.
+
+What renders is unchanged: chat replies, `markdown` components, markdown fields and documentation pages render as before. Nothing is added to or removed from any package entry, and no route, registry key or translation key changes.
diff --git a/apps/console/vite.config.ts b/apps/console/vite.config.ts
@@ -1368,9 +1368,42 @@ export default defineConfig({
             { name: 'plugin-calendar', test: /[\\/]packages[\\/]plugin-calendar[\\/]/, priority: 70 },
             { name: 'plugin-kanban', test: /[\\/]packages[\\/]plugin-kanban[\\/]/, priority: 70 },
             { name: 'plugin-chatbot', test: /[\\/]packages[\\/]plugin-chatbot[\\/]/, priority: 70 },
-            // react-markdown / remark / micromark family — heavy markdown
-            // pipeline pulled in only by markdown/chatbot plugins.
-            { name: 'vendor-markdown', test: /[\\/]node_modules[\\/](react-markdown|remark-|rehype-|micromark|mdast-|hast-|unified|unist-|vfile|bail|trough|character-entities|decode-named-character-reference|devlop|estree-|comma-separated-tokens|space-separated-tokens|property-information|html-url-attributes|zwitch)/, priority: 85 },
+            //
+            // ## `vendor-markdown` claims only what the first load reaches (objectui#11854)
+            //
+            // The remark / rehype / micromark family. The chatbot's message
+            // renderer (`streamdown`, imported statically by `plugin-chatbot`)
+            // reaches most of it on every page, so this chunk is eager. Without
+            // the tag the group also claimed the family members that only
+            // `plugin-markdown` reaches — `rehype-highlight` with `lowlight` and
+            // `highlight.js`, `rehype-slug`, `rehype-autolink-headings`,
+            // `remark-github-blockquote-alert` and their helpers — and one eager
+            // member made all of them eager: the objectui#11798 capture
+            // mechanism, one group over. `tags: ['$initial']` is the same option,
+            // for the same reason, as on `vendor-objectstack` above
+            // (objectui#11101): the group claims a family member only when a
+            // static import from the entry reaches it, and the rest follows its
+            // importer, `plugin-markdown`'s own group, behind that plugin's lazy
+            // registration and the docs reader's lazy route. The bytes are
+            // recorded once, on `BASELINE` in
+            // `scripts/check-eager-closure-budget.mjs`, ⛔ not here.
+            //
+            // ⛔ Not `includeDependenciesRecursively: false`, the flag
+            // objectui#11798 measured into a chunk cycle on `vendor-charts`; the
+            // tag narrows what the group claims and leaves its capture alone.
+            //
+            // `react-markdown` is the one member with no static path from the
+            // entry that the first load still needs. `MarkdownContent` in
+            // `packages/fields` imports it statically, and that widget sits in
+            // the EAGER `ui-components` chunk although it is only reached through
+            // `React.lazy` — the co-tenancy recorded in
+            // `scripts/vite-ineffective-dynamic-imports.ts` (objectui#5325). Left
+            // unclaimed, `ui-components` (priority 80) would take it by its own
+            // capture, onto a budgeted line; claimed by the tagged group it
+            // cannot be. So it gets a group of its own here, and its chunk loads
+            // with whichever importer loads first.
+            { name: 'vendor-markdown', test: /[\\/]node_modules[\\/](remark-|rehype-|micromark|mdast-|hast-|unified|unist-|vfile|bail|trough|character-entities|decode-named-character-reference|devlop|estree-|comma-separated-tokens|space-separated-tokens|property-information|html-url-attributes|zwitch)/, priority: 85, tags: ['$initial'] },
+            { name: 'vendor-react-markdown', test: /[\\/]node_modules[\\/]react-markdown[\\/]/, priority: 84 },
             // Sentry — only fetched when the RUNTIME serves a DSN on
             // /api/v1/runtime/config (objectstack#12681); a deployment that
             // configured none never requests this chunk at all.

diff --git a/scripts/__tests__/check-eager-closure-budget.test.ts b/scripts/__tests__/check-eager-closure-budget.test.ts
@@ -635,8 +635,11 @@ describe('chunk attribution (objectui#7399)', () => {
       expect(group!.options).toContain(`tags: ['$initial']`);
       // The control: the parse reads the tag off THIS group, not off every
       // group — a tail-blind or table-wide match would satisfy the line above.
+      // objectui#11854 tags `vendor-markdown` the same way, for the same
+      // reason, so the exact set names both, in table order.
       expect(groups.filter((g) => g.options.includes('$initial')).map((g) => g.name)).toEqual([
         'vendor-objectstack',
+        'vendor-markdown',
       ]);
     });
 
@@ -811,10 +814,11 @@ describe('ceiling sensitivity, judged live (objectui#5924)', () => {
     // `@objectstack/*` 17.6.0's spec root shed its migration chain,
     // objectui#11717 UP to 3506.9 on the ruled raise for `@objectstack/*`
     // 17.7.0, ruling 6027998226, objectui#11798 down to 3338.1 when the Studio
-    // builder and the chart engine left the first load) — a rendering derived
+    // builder and the chart engine left the first load, objectui#11854 down to
+    // 3296.2 when the markdown highlighter left it) — a rendering derived
     // in the test would agree with the renderer by construction and pin
     // nothing.
-    expect(result.message).toContain('3338.1');
+    expect(result.message).toContain('3296.2');
   });
 
   it('is exactly one regression wide, from either side of the line', () => {
@@ -1644,7 +1648,7 @@ describe('main', () => {
     // about the FIXTURE while the gate under test behaved correctly. The number
     // this case is actually about is "the report's chunk count, echoed".
     expect(outputs.closure_chunks).toBe(String(fixture.files.length));
-    expect(outputs.closure_gzip_kb).toBe('3338.1');
+    expect(outputs.closure_gzip_kb).toBe('3296.2');
   });
 
   it('exits 1 — a verdict about the BUNDLE — when over budget', () => {
@@ -1737,8 +1741,16 @@ describe('main', () => {
    * past its own line. Before objectui#5490 that run exited 0.
    */
   it('exits 1 when one chunk is over its ceiling while the aggregate is green', () => {
+    // The chunk grows past its line while the rest of the closure shrinks by
+    // the same bytes, so the TOTAL stays on the baseline — the shape the
+    // per-chunk header names: one chunk grows by the whole allowance while the
+    // others shrink. Without the offset the total rode on the aggregate
+    // headroom, and when objectui#11854 lowered the aggregate ceiling the
+    // chunk's overage alone was larger than that headroom, so the aggregate
+    // went red too and the run stopped testing what its name says.
+    const over = PER_CHUNK_GZIP_CEILINGS['vendor-objectstack'] + 1;
     const { code, outputs } = run(
-      budgeted({ 'vendor-objectstack': PER_CHUNK_GZIP_CEILINGS['vendor-objectstack'] + 1 }),
+      budgeted({ 'vendor-objectstack': over }, -(over - PER_CHUNK_BASELINE['vendor-objectstack'])),
     );
     expect(code).toBe(1);
     expect(outputs.closure_status).toBe('pass');

diff --git a/scripts/check-eager-closure-budget.mjs b/scripts/check-eager-closure-budget.mjs
@@ -417,12 +417,14 @@ import { isEntrypoint } from './invoked-as.mjs';
  *     header for both builds and the three control rows that show the bytes
  *     LEFT rather than moved.
  *
- * Headroom above {@link BASELINE} is 26,363 bytes — 0.29x
- * {@link REGRESSION_THIS_GATE_MUST_CATCH_BYTES}, on the pair objectui#11798
- * re-pinned when the Studio builder and the chart engine left the first load,
- * lowering the ceiling by exactly the bytes that left (the header's
- * objectui#11798 entry says why that band sits under the 0.50x design point);
- * before it, the pair
+ * Headroom above {@link BASELINE} is 11,020 bytes — 0.12x
+ * {@link REGRESSION_THIS_GATE_MUST_CATCH_BYTES}, on the pair objectui#11854
+ * re-pinned when the markdown highlighter and the docs-only markdown plugins
+ * left the first load, lowering the ceiling by exactly the bytes that left
+ * (the header's objectui#11854 entry says why that band sits under the 0.50x
+ * design point); before it, the pair objectui#11798 re-pinned when the Studio
+ * builder and the chart engine left the first load carried 26,363 bytes
+ * (0.29x), the pair
  * objectui#11717 re-pinned under the ruling recorded in the header's fourth
  * raise carried 45,972 bytes (0.50x), and the pair objectui#11438 re-pinned
  * when the `@objectstack/*` 17.6.0 bump paid back the rest of the spec-root
@@ -930,8 +932,81 @@ import { isEntrypoint } from './invoked-as.mjs';
  * the budgeted chunks moved by at most 2 bytes across the two builds, so
  * {@link PER_CHUNK_GZIP_CEILINGS} and {@link PER_CHUNK_BASELINE} stand. Not
  * {@link REGRESSION_THIS_GATE_MUST_CATCH_BYTES}. No exemption was added.
+ *
+ * ## ⭐ LOWERED WHEN THE MARKDOWN HIGHLIGHTER LEFT THE FIRST LOAD (objectui#11854)
+ *
+ * From 3,444,622 to 3,386,364: down by 58,258 bytes, exactly what
+ * objectui#11854 took out of the closure. Lowered by the amount recovered, as
+ * the objectui#11798 entry above was: ⛔ not re-derived as the new reading plus
+ * half a regression, which would also have absorbed `main`'s drift since the
+ * `c1d378ec` reading (a raise under another name), and ⛔ not rounded.
+ *
+ * WHAT LEFT — two console builds, one container, one instrument, each under
+ * `scripts/pm/os-verify-lock.sh`: the console's own
+ * `apps/console/vite.config.ts` run through Vite's `build()` with one read-only
+ * module-graph dump appended, reading the `eager-closure.json` the build
+ * writes. `9cb4e29` is `main`; `406f760` is that tree plus the chunk-group
+ * change to `vendor-markdown` in `apps/console/vite.config.ts`, and nothing
+ * else. Gzipped bytes:
+ *
+ *   | chunk                         |  `9cb4e29` |  `406f760` |   delta |
+ *   |-------------------------------|-----------:|-----------:|--------:|
+ *   | `vendor-markdown`             |    163,717 |    104,286 | -59,431 |
+ *   | `vendor-react-markdown`, new  |          — |      1,144 |  +1,144 |
+ *   | `ui-components`               |    284,424 |    284,433 |      +9 |
+ *   | everything else               |  2,985,461 |  2,985,481 |     +20 |
+ *   | ⇒ aggregate                   |  3,433,602 |  3,375,344 | -58,258 |
+ *
+ * The builds weigh 333 of 2456 chunks (`9cb4e29`) and 334 of 2457
+ * (`406f760`); the one new chunk is `vendor-react-markdown`, and it is eager.
+ * The +20 is import bookkeeping spread over 19 chunks, none by more than 5,
+ * ⛔ not attributed further; every `vendor-icon-*` chunk is unmoved.
+ *
+ *   - THE CAPTURE. On `9cb4e29` `vendor-markdown` held 296 modules, and 53 of
+ *     them no static import from the entry reaches: `highlight.js` (39 of the
+ *     53, most of the bytes), `lowlight`, `rehype-highlight`, `rehype-slug`,
+ *     `github-slugger`, `rehype-autolink-headings`,
+ *     `remark-github-blockquote-alert`, the five hast and unist helpers they
+ *     import, and `react-markdown`. The chunk was eager for the other 243,
+ *     which the chat message renderer reaches, so the group's capture made all
+ *     296 eager. With `tags: ['$initial']` the group holds the 243 alone (the
+ *     comment on the group in `apps/console/vite.config.ts` says why), and 52
+ *     of the 53 follow their importer into `plugin-markdown`'s lazy chunk,
+ *     which went from 4,342 to 62,888 gzipped bytes across the two builds.
+ *   - `react-markdown`, the 53rd, stays eager in the new chunk: `MarkdownContent`
+ *     in `packages/fields` imports it statically, and that widget sits in the
+ *     eager `ui-components` chunk although only `React.lazy` reaches it (the
+ *     objectui#5325 co-tenancy, recorded in
+ *     `scripts/vite-ineffective-dynamic-imports.ts`). A chunk of its own keeps
+ *     its bytes off the budgeted `ui-components` line, which would otherwise
+ *     take it by its own capture.
+ *
+ * ⛔ What it did NOT recover. The 243 modules left in `vendor-markdown` are the
+ * pipeline the chat message renderer runs on every page: `streamdown` imports
+ * `remark-parse`, `remark-gfm`, `remark-rehype`, `rehype-raw`,
+ * `rehype-sanitize`, `rehype-harden`, `unified` and `hast-util-to-jsx-runtime`
+ * statically, and `plugin-chatbot`, which imports `streamdown` statically, is
+ * eager. Moving them would lazy-load a module the first load reaches, which is
+ * objectui#11798's question under the objectui#6795 ruling, ⛔ not a chunk
+ * group's.
+ *
+ * Headroom 11,020 bytes = 0.12x {@link REGRESSION_THIS_GATE_MUST_CATCH_BYTES}
+ * over {@link BASELINE}: exactly the headroom `main` had on `9cb4e29` before
+ * this change (3,444,622 − 3,433,602), because the ceiling came down by the
+ * aggregate's own delta. It sits under the 0.29x the objectui#11798 entry left
+ * because `9cb4e29` weighed 15,343 bytes more than the `c1d378ec` reading that
+ * entry was sized on — `main`'s drift since, ⛔ not attributed here. No build
+ * that passed before this edit and measures under 3,386,364 fails after it,
+ * and the `9cb4e29` build, which is this change reverted, lands over it.
+ *
+ * ⛔ What moved with it, and what did not. None of the four per-chunk rows:
+ * `ui-components` moved by 9 bytes across the two builds and the other three
+ * budgeted chunks by none, so {@link PER_CHUNK_GZIP_CEILINGS} and
+ * {@link PER_CHUNK_BASELINE} stand; `vendor-markdown` has no row and gets
+ * none. Not {@link REGRESSION_THIS_GATE_MUST_CATCH_BYTES}. No exemption was
+ * added.
  */
-export const MAX_EAGER_CLOSURE_GZIP_BYTES = 3_444_622;
+export const MAX_EAGER_CLOSURE_GZIP_BYTES = 3_386_364;
 
 /**
  * The measurement the ceiling above was derived from. Exported so the two
@@ -944,6 +1019,28 @@ export const BASELINE = Object.freeze({
   /**
    * `emitEagerClosureReport`'s `eagerGzipBytes` on this commit.
    *
+   * `406f760` is objectui#11854's branch commit that narrows `vendor-markdown`
+   * to what the first load reaches: `main` at `9cb4e29` plus that one
+   * chunk-group change in `apps/console/vite.config.ts`, re-pinned with the
+   * ceiling it was lowered with (the objectui#11854 entry on
+   * {@link MAX_EAGER_CLOSURE_GZIP_BYTES} carries its table and its `9cb4e29`
+   * control, ⛔ not restated here). It is a branch commit for the reason the
+   * paragraphs below give for `c1d378ec`: the pull request's own diff is what
+   * moved the figure. Measured by the console's own
+   * `apps/console/vite.config.ts` run through Vite's `build()` with one
+   * read-only module-graph dump appended, reading the `eager-closure.json` the
+   * build writes, under `scripts/pm/os-verify-lock.sh`, in the same container
+   * as the `9cb4e29` control build. The console's ordinary build of `57a1066`
+   * (`CI=true pnpm exec turbo run build --filter=@object-ui/console...
+   * --concurrency=2`, exit 0) wrote the same figure to the byte, which is this
+   * instrument's calibration. The commits after `406f760` on the branch (this
+   * re-pin, its unit test and the changeset) reach no bundler input, so they
+   * cannot move the figure it pins.
+   *
+   * ⚠️ The paragraph below describes the reading this one superseded, the
+   * previous baseline `c1d378ec` (3,418,259 across 333 of 2456 chunks), kept as
+   * its provenance.
+   *
    * `c1d378ec` is objectui#11798's branch tip: `main` at `455c6466` merged into
    * the branch that takes the Studio builder and the chart engine out of the
    * first load, re-pinned with the ceiling it was lowered with (the
@@ -1026,10 +1123,10 @@ export const BASELINE = Object.freeze({
    * and not from `main` once it is squash-merged — the dead end objectui#9355
    * added `squashMerge` below to route around.
    */
-  gzipBytes: 3_418_259,
-  chunks: 333,
-  totalChunks: 2456,
-  commit: 'c1d378ec',
+  gzipBytes: 3_375_344,
+  chunks: 334,
+  totalChunks: 2457,
+  commit: '406f760',
 
   /**
    * The squash merge that carried the reading above onto `main`, recorded when
@@ -1038,7 +1135,7 @@ export const BASELINE = Object.freeze({
    * measurement gets a handle rather than a dead end.
    *
    * `null` since objectui#10996, and `null` again after objectui#11101,
-   * objectui#11438, objectui#11717 and objectui#11798, deliberately. This field can only ever be BACK-FILLED: a squash sha does
+   * objectui#11438, objectui#11717, objectui#11798 and objectui#11854, deliberately. This field can only ever be BACK-FILLED: a squash sha does
    * not exist until the pull request merges, so the change that re-pins the
    * field above ⛔ cannot write its own here. ⛔ Do not guess one: a wrong sha in
    * this position is worse than an absent one, because it RESOLVES, and a
@@ -1714,8 +1811,12 @@ export const PER_CHUNK_GZIP_CEILINGS = Object.freeze({
  * objectui#11717's `c1e32e96`, the 17.7.0 bump's build, re-pinned both, as
  * objectui#11438's `2ba091c`, objectui#11101's `4acbea07` and objectui#11073's
  * `048e7f6` had before it. objectui#11798 re-pinned the aggregate onto
- * BASELINE's `c1d378ec` and left this key at `c1e32e96`: that chunk moved by 0
- * bytes across objectui#11798's two builds, and its ceiling did not move. The
+ * `c1d378ec` and left this key at `c1e32e96`: that chunk moved by 0
+ * bytes across objectui#11798's two builds, and its ceiling did not move.
+ * objectui#11854 re-pinned the aggregate onto BASELINE's `406f760` and left
+ * every key where it was: `vendor-objectstack`, `framework` and
+ * `i18n-locale-en` moved by 0 bytes across objectui#11854's two builds and
+ * `ui-components` by 9, and no ceiling moved. The
  * aggregate is the later reading for every key: `c1e32e96` for
  * `vendor-objectstack`, `bbf6b02d9` (2026-09-13, objectui#9251) for `ui-components`,
  * and `3f775eeb8` for `framework`. ⚠️ `i18n-locale-en`'s commit was `755d34a5f` when it