diff --git a/hugo.toml b/hugo.toml
index 12098fa..87ea77c 100644
--- a/hugo.toml
+++ b/hugo.toml
@@ -9,4 +9,4 @@ disableKinds = ["taxonomy", "term", "RSS", "sitemap", "404"]
   name = "parser"
   url = "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/go-ruby-parser/parser"
   status = "ready"
-  blurb = "The front-end: `parser.Parse(src) (*ast.Program, error)`, with sub-packages `token`, `ast` and `lexer`. A stateful lexer (MRI `spaceSeen` semantics), a recursive-descent + Pratt expression parser with a scope stack, and a typed AST for a broad, practical subset of **Ruby 4.0** — literals, operators, control flow, `case`/`in` **pattern matching**, methods/blocks, classes/modules/metaprogramming. 100% covered (the gate runs on all three OS lanes; the suite also runs on all six 64-bit targets and on js/wasm and wasip1/wasm), differential-tested against MRI Ruby 4.0.5. Current release **v0.2.0** (2026-09-21): fourteen grammar gaps closed, and every one of the 68 files in the pinned ruby/spec `language/` corpus now parses."
+  blurb = "The front-end: `parser.Parse(src) (*ast.Program, error)`, with sub-packages `token`, `ast` and `lexer`. A stateful lexer (MRI `spaceSeen` semantics), a recursive-descent + Pratt expression parser with a scope stack, and a typed AST for a broad, practical subset of **Ruby 4.0** — literals, operators, control flow, `case`/`in` **pattern matching**, methods/blocks, classes/modules/metaprogramming. 100% covered (the gate runs on all three OS lanes; the suite also runs on all six 64-bit targets and on js/wasm and wasip1/wasm), differential-tested against MRI Ruby 4.0.5. Current release **v0.4.0** (2026-09-26): all 80 `_spec.rb` files in the pinned ruby/spec `language/` corpus parse, and the three limitations the docs used to list — paren-less command calls with keyword/splat/block args, default block parameters, and the positional `Class(a)` find-pattern — all parse now. `BEGIN { }` / `END { }` remain the one refusal."
diff --git a/layouts/index.html b/layouts/index.html
index 8ccc59e..f867939 100644
--- a/layouts/index.html
+++ b/layouts/index.html
@@ -88,7 +88,7 @@ <h1>go-ruby-parser</h1>
       <span class="pill">stateful lexer</span>
       <span class="pill">recursive-descent + Pratt</span>
       <span class="pill">typed AST</span>
-      <span class="pill">v0.2.0 · 2026-09-21</span>
+      <span class="pill">v0.4.0 · 2026-09-26</span>
       <span class="pill">Ruby 4.0 subset</span>
       <span class="pill">pattern matching</span>
       <span class="pill">heredocs</span>
@@ -119,10 +119,13 @@ <h1>go-ruby-parser</h1>
       <strong>100% coverage</strong> is enforced in CI on all three OS lanes, and the suite also runs
       on all <strong>six 64-bit targets</strong> — amd64 and arm64 natively, riscv64 / loong64 /
       ppc64le / s390x under QEMU — plus <code>js/wasm</code> and <code>wasip1/wasm</code>.
-      The current release is <a href="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/go-ruby-parser/parser/tree/v0.2.0"><strong>v0.2.0</strong></a>
-      (2026-09-21), which closed fourteen grammar gaps: every one of the 68 files in the pinned
-      ruby/spec <code>language/</code> corpus now parses, thirteen of which had not parsed at all
-      before.</p>
+      The current release is <a href="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/go-ruby-parser/parser/tree/v0.4.0"><strong>v0.4.0</strong></a>
+      (2026-09-26). Measured on 2026-09-26: <strong>all 80</strong> <code>_spec.rb</code> files in the
+      pinned ruby/spec <code>language/</code> corpus parse, and of 47 top-level CRuby 4.0.5 stdlib
+      files MRI accepts, this parser accepts 46. On parse speed it is roughly
+      <strong>1.9× quicker than <code>Ripper.sexp</code></strong> and about
+      <strong>1.6× slower than the C parser built into CRuby</strong>
+      (<code>RubyVM::AbstractSyntaxTree.parse</code>).</p>
 
     <section class="grid">
       {{ range .Site.Params.repos }}