ci(vnext): align worker fixture budget

Author: Light2Dark

scripts/worker-placement.mjs

@@ -38,8 +38,8 @@ const BIGQUERY_GZIP_LIMIT = 50 * 1024;
 const CORE_TOTAL_GZIP_LIMIT = 48 * 1024;
 const CORE_TOTAL_RAW_LIMIT = 180 * 1024;
 const POSTGRESQL_GZIP_LIMIT = 68 * 1024;
-const WORKER_TOTAL_GZIP_LIMIT = 128 * 1024;
-const WORKER_TOTAL_RAW_LIMIT = 570 * 1024;
+const WORKER_TOTAL_GZIP_LIMIT = 160 * 1024;
+const WORKER_TOTAL_RAW_LIMIT = 700 * 1024;
 const MIME_TYPES = new Map([
   [".css", "text/css; charset=utf-8"],
   [".html", "text/html; charset=utf-8"],

state/artifacts/reports/pr-197-browser-ci.md

@@ -10,25 +10,34 @@
 ## Root cause
 
 The framework-independent vNext bundle grew from the session walking skeleton
-to the completed relation service and catalog scheduler. Its exact packed
-consumer output is 40,214 gzip bytes and still contains no parser modules.
+to the completed relation service and catalog scheduler. The exact local
+packed-consumer measurement is 40,121 gzip/143,265 raw bytes; hosted Linux CI
+measured 40,214 gzip bytes. It still contains no parser modules.
 
 The worker-placement script retained the obsolete 16 KiB skeleton limit, while
 the capability charter already budgets 75 KiB for core plus the future
-CodeMirror adapter.
+CodeMirror adapter. The complete worker fixture also embeds the page-side core,
+so after the first correction its obsolete aggregate ceiling failed at 150,847
+gzip bytes. The exact local aggregate is 150,985 gzip/669,106 raw bytes; its
+PostgreSQL and BigQuery lazy grammar closures are unchanged.
 
 ## Fix
 
 Allocate 48 KiB gzip and 180 KiB raw to the framework-independent core. This
 passes the current measured bundle with bounded headroom and reserves 27 KiB of
 the declared compressed budget for the separate CodeMirror adapter.
 
+Allocate 160 KiB gzip and 700 KiB raw to the complete worker evidence fixture.
+This is an aggregate test-fixture ceiling, not a product bundle promise: it
+includes the page-side core plus both lazy parser grammars. The dedicated
+dialect-closure ceilings remain unchanged and continue to detect parser growth.
+
 ## Friction and future guidance
 
-The local isolated worktree shares dependencies by symlink, so the
-worker-placement script cannot perform its nested package-manager install
-without replacing that link. Hosted CI remains authoritative for the exact
-tarball fixture.
+The local isolated worktree shares dependencies by symlink, so its root install
+must run with CI semantics to avoid an interactive module-directory prompt.
+Hosted CI remains authoritative for platform-sensitive compressed sizes; the
+local exact-tarball report records the raw sizes and dependency graph.
 
 Future vertical slices that change public entry-point reachability should
 compare emitted bundle composition with both the per-entry allocation and the

test/worker-placement/README.md

@@ -9,23 +9,23 @@ package's dependency. The fixture workspace pins Vite's floating transitive
 versions to the exact versions in the root lock, so the nested frozen install
 can run offline after a clean root CI install.
 
-The minified Vite 8 packed-consumer baseline is 15,028 gzip/47,141 raw
+The minified Vite 8 packed-consumer baseline is 40,121 gzip/143,265 raw
 bytes for the complete parser-free core, 67,573 gzip bytes for the PostgreSQL
 transitive graph, 50,470 gzip bytes for the BigQuery transitive graph, and
-125,937 gzip/572,982 raw bytes for the complete worker build output. The core
-measurement includes the four authenticated relation-dialect runtimes and
-their reserved-word tables. The PostgreSQL and BigQuery figures each include
-their transitive shared chunks; the report also identifies those shared chunks
-explicitly.
+150,985 gzip/669,106 raw bytes for the complete worker build output. The core
+measurement includes the four authenticated relation-dialect runtimes, their
+reserved-word tables, and relation-completion/session orchestration. The
+PostgreSQL and BigQuery figures each include their transitive shared chunks;
+the report also identifies those shared chunks explicitly.
 
-The fail-closed ceilings retain approximately 9% gzip and 13% raw headroom for
-the core, 4% gzip and 2% raw headroom for the complete worker output, and the
-existing tight dialect-graph headroom:
+The fail-closed ceilings retain approximately 18% gzip and 22% raw headroom
+for the core, 8% gzip and 7% raw headroom for the complete worker output, and
+the existing tight dialect-graph headroom:
 
-- Complete parser-free core: 16 KiB gzip and 52 KiB raw
+- Complete parser-free core: 48 KiB gzip and 180 KiB raw
 - PostgreSQL transitive graph: 68 KiB gzip
 - BigQuery transitive graph: 50 KiB gzip
-- Complete worker build output: 128 KiB gzip and 570 KiB raw
+- Complete worker build output: 160 KiB gzip and 700 KiB raw
 
 These are provisional placement limits, not product bundle promises. The
 orchestration script fails closed when they are exceeded, when the dialects no