site: code blocks become generated-snippet placeholders

Fifteen hand-written code blocks across three pages become
`<pre class="example-code" data-snippet="<id>"></pre>`, filled at deploy time from
the payload the MetaObjects repo builds and gates. Eleven on index.html, two on
requirements.html, two on the prompts article — exactly the fifteen ids in that
repo's snippet registry, which is where they were read from rather than typed.

DO NOT PUSH YET. The deploy step that fills them does not exist until the next
task, so these pages would go live with fifteen empty blocks.

Two blocks on index.html stay hand-written, correctly: the assess.md prompt is
something the reader pastes, and the CLI-commands block is shell. That second one
did carry a real error, now fixed — `meta migrate --db … --slug …` under a comment
claiming it diffs the live DB. Without `--from-db` it diffs the committed
snapshot, so the comment described a run the flags did not make; it is now the
CLI's own documented greenfield example.

Prose that the real snippets invalidated:

- The five-language section's per-language labels each promised more than the
  block shows — one file is not the whole generated set. Each now names the file
  and says what else the same run emits.
- "Spring REST + DTO + JPA" for Java, in the Codegen pillar card and the example
  intro. codegen-spring emits DTOs, controllers and repository interfaces; there
  is no `@Entity` and no jakarta.persistence anywhere in the module, and the
  site's own llms.txt already said "the persistence impl is consumer-supplied".
- "The metamodel goes deep" ran on a fictional `blog` package. It is
  `acme::learn` from examples/advanced-modeling now, so Post becomes Program and
  AuthorSummary becomes ProgramSummary. The marked source region is Author and
  its validators, so the label says that rather than promising a projection the
  block does not contain.
- requirements.html's ledger entry was `arena::Bot.status`, introduced as "a real
  entry from a live product, and it is wrong". The block is now the showcase's
  own entry, which resolves and is gated — so it can demonstrate "resolved, not
  trusted" instead of asserting it. The live-product finding is the strongest
  thing on the page and is kept, told as what it now is: a near-identical entry
  that passed every structural check and was still wrong. That is the sharper
  argument anyway — the check is necessary and not sufficient.
- The generated test stub lands beside "what you get, and what you build", where
  the page says a stub is a place to put a proof and not a proof. It carries the
  TypeScript-only caveat next to it.

Palette: `.example-code .ok/.err` for terminal blocks (transcripts would have
shipped unstyled), and `.example-details` for the expander the injector appends —
new markup this site has never carried, so the browser default was the only thing
styling it. The per-page `tok-*` rules are dropped where their only users were the
blocks now injected; `requirements.html` keeps `.tok-err`, which is still used in
prose outside any `<pre>`.

llms.txt and llms-full.txt are re-copied wholesale from the monorepo's docs/llms/,
which is what they mirror. They carried the fifth pillar and were also four minor
lines stale — still describing generators as package imports rather than the
scaffold-and-own copies getting-started.html now documents.

Author: claude

www/articles/prompts-are-code.html

@@ -45,11 +45,6 @@
       padding: 1.25rem 1.4rem; overflow-x: auto; font-family: var(--font-mono);
       font-size: 0.86rem; line-height: 1.6; margin: 0 0 1.6rem; box-shadow: var(--shadow); }
     .article pre code { background: none; color: inherit; padding: 0; font-size: inherit; }
-    .article .tok-key { color: #8FB7E8; }
-    .article .tok-val { color: #A7E0C0; }
-    .article .tok-cmt { color: #7A8AA8; }
-    .article .tok-err { color: #F0A0A0; }
-    .article .tok-ok  { color: #A7E0C0; }
     .article hr { border: none; border-top: 1px solid var(--rule); margin: 2.5rem 0; }
     .article-cta { background: var(--light-gray); border: 1px solid var(--rule);
       border-radius: 10px; padding: 1.5rem 1.75rem; margin: 2.5rem 0 0; }
@@ -136,27 +131,15 @@ <h2>What prompts-as-code looks like in practice</h2>
 
       <p>(It grew out of a mess of my own — a few thousand lines of <code>StringBuilder</code> driving the characters in an LLM game. For the origin story, and the <em>data + text + render</em> decomposition that became the fix, the longer essay is <a href="https://dougmealing.com/writing/the-prompt-is-code/">The prompt is code — and yours is drifting too</a>.)</p>
 
-      <p>A prompt declares its input as a typed payload — a projection of the same entities that drive your database and API:</p>
+      <p>A prompt declares its input as a typed payload — a projection of the same entities that drive your database and API — and names its text by a logical reference, so the text itself stays a versioned file in your repo rather than a row in someone's cloud:</p>
 
-      <pre><code><span class="tok-cmt"># prompt = typed payload + external text</span>
-<span class="tok-key">object.value:</span>
-  <span class="tok-key">name:</span> <span class="tok-val">AuthorBlurbPayload</span>
-  <span class="tok-key">children:</span>
-    - <span class="tok-key">field.string:</span> { name: authorName }
-    - <span class="tok-key">field.string:</span> { name: bio }
-
-<span class="tok-key">template.prompt:</span>
-  <span class="tok-key">name:</span> <span class="tok-val">authorBlurb</span>
-  <span class="tok-key">payloadRef:</span> <span class="tok-val">AuthorBlurbPayload</span>
-  <span class="tok-key">textRef:</span> <span class="tok-val">author/blurb</span>        <span class="tok-cmt"># template text lives in a versioned file, not a cloud</span></code></pre>
+      <pre class="example-code" data-snippet="showcase-prompt"></pre>
 
       <p>From that one declaration you get a generated, typed <code>render</code> function per language — deterministic and byte-identical across all five ports (there's a shared conformance corpus that proves it), so snapshot tests pass and your exact-prefix prompt cache stays warm. Nothing reformats your prompt behind your back.</p>
 
-      <p>And because the payload is <em>declared</em>, the build can check it. Rename <code>authorName</code> in the model — or reference a variable the payload doesn't have — and:</p>
+      <p>And because the payload is <em>declared</em>, the build can check it. Reference a variable the payload doesn't have — rename <code>subscriberName</code> in the model and forget the template, say — and <code>meta verify</code> says so and exits non-zero:</p>
 
-      <pre><code><span class="tok-ok">$</span> meta verify
-<span class="tok-err">[authorBlurb] (prompt) ERR_VAR_NOT_ON_PAYLOAD: displayName</span>
-meta verify — 1 drift error(s) across 1 template(s).   <span class="tok-cmt"># exit 1 — the build fails</span></code></pre>
+      <pre class="example-code" data-snippet="verify-transcript"></pre>
 
       <p>That's the moment a prompt stops being a string and becomes code: <strong>the renamed field is a compile error now</strong>, not a quality regression you discover in three weeks.</p>