site(video): captions for the re-cuts, and the drift blurb gains the fourth surface

The drift-detection description and its aria-label still named three surfaces —
code, prompts, schema — and said any mismatch fails the build. The video's own
output shows the fourth: a requirement nothing implements warns at exit 0, and
fails once the implementation it names is gone.

Captions follow the re-cut drift-detection, adopt-existing and runtime-ui
videos. getting-started's caption text is unchanged (only a slide was fixed).

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01JGmCLHbQKuLBPjjhWC43jk

Author: claude

www/video/adopt-existing.vtt

@@ -121,38 +121,38 @@ Customers matches the model.
 Orders isn't modeled yet — so that's your next table to adopt, whenever you're ready.
 
 31
-00:02:18.476 --> 00:02:20.004
+00:02:18.476 --> 00:02:20.055
 This is low-risk by design.
 
 32
-00:02:20.004 --> 00:02:29.223
-The generated entity module imports only Drizzle and Zod; the routes mount through one ordinary open-source package — so adopting it is reversible, table by table.
+00:02:20.055 --> 00:02:30.110
+At runtime the generated entity module needs only Drizzle and Zod; the routes mount through one ordinary open-source package — so adopting it is reversible, table by table.
 
 33
-00:02:29.223 --> 00:02:35.106
+00:02:30.110 --> 00:02:36.191
 The rest of your app is untouched, and you keep your escape hatches for anything metadata can't express.
 
 34
-00:02:36.506 --> 00:02:44.305
+00:02:37.591 --> 00:02:45.389
 That's adoption: assess the fit, describe one table, choose your generators, generate, adopt — no big rewrite, no magic.
 
 35
-00:02:44.305 --> 00:02:50.999
+00:02:45.389 --> 00:02:52.084
 Then, table by table, the model becomes the spine, and meta verify keeps it honest as you grow into it.
 
 36
-00:02:52.399 --> 00:02:55.003
+00:02:53.484 --> 00:02:56.087
 You don't have to start over to start.
 
 37
-00:02:55.003 --> 00:03:00.347
+00:02:56.087 --> 00:03:01.432
 Have your agent assess the fit, run meta init, describe a table, and meta gen.
 
 38
-00:03:00.347 --> 00:03:04.185
+00:03:01.432 --> 00:03:05.269
 That's the MetaObjects series — more at metaobjects.dev.
 
 39
-00:03:04.185 --> 00:03:05.555
+00:03:05.269 --> 00:03:06.640
 Thanks for watching.
 

www/video/drift-detection.vtt

@@ -25,150 +25,150 @@ The model gained a field, nobody re-ran the generator, and one command caught it
 Let me show you how this works across four surfaces — your code, your prompts, your schema, and your claims.
 
 7
-00:00:29.560 --> 00:00:33.697
+00:00:29.560 --> 00:00:33.543
 Meta verify checks one thing: does everything still agree with the model?
 
 8
-00:00:33.697 --> 00:00:43.332
+00:00:33.543 --> 00:00:42.820
 It does it across four surfaces — your generated code, your prompt templates, your database schema, and the requirements you declare: what the software is supposed to do.
 
 9
-00:00:43.332 --> 00:00:45.429
-Drift on any of them fails the build.
+00:00:42.820 --> 00:00:50.842
+Drift in the code, the prompts or the schema fails the build; a requirement warns when nothing implements it, and fails when what it named is gone.
 
 10
-00:00:46.829 --> 00:00:49.188
+00:00:52.242 --> 00:00:54.601
 Before we run it, here's that one model.
 
 11
-00:00:49.188 --> 00:00:56.031
+00:00:54.601 --> 00:01:01.444
 The Task entity carries its validators right on the fields — that drives the code, and its source drives the schema.
 
 12
-00:00:56.031 --> 00:01:00.396
+00:01:01.444 --> 00:01:05.809
 The prompt has a typed payload and a template — that's the second surface.
 
 13
-00:01:00.396 --> 00:01:05.292
+00:01:05.809 --> 00:01:10.705
 And the requirements at the bottom claim what's built — each one names a real node.
 
 14
-00:01:05.292 --> 00:01:12.253
+00:01:10.705 --> 00:01:17.666
 And look at the references between them: the payload extends a real Task field, the template points at a real payload.
 
 15
-00:01:12.253 --> 00:01:17.739
+00:01:17.666 --> 00:01:23.152
 Those are resolved, not trusted — rename or delete the target, and the model won't even load.
 
 16
-00:01:17.739 --> 00:01:20.099
+00:01:23.152 --> 00:01:25.512
 So the model can't drift against itself.
 
 17
-00:01:21.499 --> 00:01:23.073
+00:01:26.912 --> 00:01:28.486
 Surface one: your code.
 
 18
-00:01:23.073 --> 00:01:27.863
+00:01:28.486 --> 00:01:33.276
 Verify regenerates from the model and compares it to what's committed.
 
 19
-00:01:27.863 --> 00:01:31.559
+00:01:33.276 --> 00:01:36.972
 The model moved and the code didn't — drift, exit one.
 
 20
-00:01:32.959 --> 00:01:34.462
+00:01:38.372 --> 00:01:39.875
 Surface two: your prompts.
 
 21
-00:01:34.462 --> 00:01:38.506
+00:01:39.875 --> 00:01:43.919
 Every variable in a prompt template has to exist on its typed payload.
 
 22
-00:01:38.506 --> 00:01:44.747
+00:01:43.919 --> 00:01:50.160
 Rename a field and leave the template behind, and the build fails — instead of the prompt quietly degrading.
 
 23
-00:01:46.147 --> 00:01:47.702
+00:01:51.560 --> 00:01:53.115
 Surface three: your schema.
 
 24
-00:01:47.702 --> 00:01:50.581
+00:01:53.115 --> 00:01:55.994
 Verify diffs the model against your live database.
 
 25
-00:01:50.581 --> 00:01:57.146
+00:01:55.994 --> 00:02:02.559
 A column that drifted out of band — dropped, renamed, or added by hand — gets caught before it reaches production.
 
 26
-00:01:58.546 --> 00:02:01.847
+00:02:03.959 --> 00:02:07.260
 And surface four is the one no test can replace.
 
 27
-00:02:01.847 --> 00:02:06.043
+00:02:07.260 --> 00:02:11.456
 A test exercises code that exists; an absence can't fail one.
 
 28
-00:02:06.043 --> 00:02:10.308
+00:02:11.456 --> 00:02:15.721
 So you declare requirements in the model — and mark them live.
 
 29
-00:02:10.308 --> 00:02:16.774
+00:02:15.721 --> 00:02:22.187
 Mark one live while nothing implements it, verify gives a warning — a capability nobody built.
 
 30
-00:02:16.774 --> 00:02:19.885
+00:02:22.187 --> 00:02:25.298
 And it resolves the link rather than trusting it.
 
 31
-00:02:19.885 --> 00:02:25.219
+00:02:25.298 --> 00:02:30.632
 Rename the node a live claim points at, and the warning becomes an error — exit one.
 
 32
-00:02:25.219 --> 00:02:30.616
+00:02:30.632 --> 00:02:36.029
 It checks that the claim points at something real, not that the something is correct.
 
 33
-00:02:32.016 --> 00:02:38.870
+00:02:37.429 --> 00:02:44.283
 When everything's in sync, one command says so: code, prompts, schema, and claims all agree with the model.
 
 34
-00:02:40.270 --> 00:02:43.951
+00:02:45.683 --> 00:02:49.364
 Put that one line in your CI, and drift can't merge.
 
 35
-00:02:43.951 --> 00:02:52.516
+00:02:49.364 --> 00:02:57.929
 A pull request that diverges from the model fails the check — so an agent has to change the model first, then regenerate.
 
 36
-00:02:52.516 --> 00:02:55.064
+00:02:57.929 --> 00:03:00.477
 The model stays the source of truth.
 
 37
-00:02:56.464 --> 00:02:57.754
+00:03:01.877 --> 00:03:03.118
 That's the whole idea.
 
 38
-00:02:57.754 --> 00:03:03.206
+00:03:03.118 --> 00:03:08.363
 AI code is locally coherent and globally divergent — drift you can't see by reading one file.
 
 39
-00:03:03.206 --> 00:03:08.307
-Meta verify checks four surfaces against one model, and any mismatch is a failed build.
+00:03:08.363 --> 00:03:16.427
+Meta verify checks four surfaces against one model: drift fails the build, and a requirement nothing has built yet is warned about, not buried.
 
 40
-00:03:08.307 --> 00:03:11.239
+00:03:16.427 --> 00:03:19.247
 The AI writes the code; the model keeps it honest.
 
 41
-00:03:12.639 --> 00:03:16.402
+00:03:20.647 --> 00:03:24.410
 Add meta verify to your build, and make drift fail the build.
 
 42
-00:03:16.402 --> 00:03:20.411
+00:03:24.410 --> 00:03:28.420
 This is part of the MetaObjects series — more at metaobjects.dev.
 
 43
-00:03:20.411 --> 00:03:21.645
+00:03:28.420 --> 00:03:29.653
 Thanks for watching.
 

www/video/runtime-ui.vtt

@@ -109,54 +109,54 @@ Load the model, and it drives a whole lot: dynamic CRUD, validation, relationshi
 That tier ships today in TypeScript, Java, Kotlin, and Python; C# is on the roadmap.
 
 28
-00:01:53.379 --> 00:01:56.805
+00:01:53.379 --> 00:01:56.876
 So when do you generate, and when do you drive it live?
 
 29
-00:01:56.805 --> 00:02:03.719
-Generate for build-time artifacts — typed, committed — and the entity tier imports only your ORM and validator.
+00:01:56.876 --> 00:02:04.568
+Generate when you want build-time artifacts: typed, committed code whose entity tier runs on just your ORM and validator.
 
 30
-00:02:03.719 --> 00:02:12.439
+00:02:04.568 --> 00:02:13.468
 Drive it live when codegen can't reach: a shipped app you can't redeploy, an agent that connects tomorrow, behavior that changes on the fly.
 
 31
-00:02:12.439 --> 00:02:15.304
+00:02:13.468 --> 00:02:16.392
 A snapshot can't reach them; a live model can.
 
 32
-00:02:16.704 --> 00:02:17.718
+00:02:17.792 --> 00:02:18.806
 That's the idea.
 
 33
-00:02:17.718 --> 00:02:22.216
+00:02:18.806 --> 00:02:23.304
 Codegen is a snapshot of your model; runtime metadata is the live feed.
 
 34
-00:02:22.216 --> 00:02:26.082
+00:02:23.304 --> 00:02:27.170
 Same model — generate the code, or drive it live, often both.
 
 35
-00:02:26.082 --> 00:02:28.553
+00:02:27.170 --> 00:02:29.641
 Change the model, and behavior follows.
 
 36
-00:02:28.553 --> 00:02:31.341
+00:02:29.641 --> 00:02:32.429
 Nothing to regenerate, nothing to recompile.
 
 37
-00:02:32.741 --> 00:02:34.897
+00:02:33.829 --> 00:02:35.985
 Drive your app from the model.
 
 38
-00:02:34.897 --> 00:02:37.629
+00:02:35.985 --> 00:02:38.717
 Build grid ships in runtime-web today.
 
 39
-00:02:37.629 --> 00:02:42.302
+00:02:38.717 --> 00:02:43.390
 This is part of the MetaObjects series — more at metaobjects.dev.
 
 40
-00:02:42.302 --> 00:02:43.739
+00:02:43.390 --> 00:02:44.827
 Thanks for watching.
 

www/videos.html

@@ -130,13 +130,13 @@ <h2 class="section-label">Adopt an existing app — one table at a time</h2>
     <section class="video-section" id="drift-detection" style="max-width:900px;margin:0 auto;padding:2.5rem 1.5rem 0;">
       <h2 class="section-label">Drift detection — make drift fail the build</h2>
       <p style="color:#8b9ab5;max-width:720px;">
-        <code>meta verify</code> checks that your code, your prompts, and your database schema still agree with the model — any mismatch fails the build.
+        <code>meta verify</code> checks your code, your prompts, and your database schema against the model, and drift in any of them fails the build. It also checks your declared requirements: a claim nothing implements warns, and one whose implementation is gone fails.
       </p>
       <figure style="margin:0;">
         <video controls preload="metadata" playsinline
                poster="/video/drift-detection-poster.png"
                width="1280" height="720"
-               aria-label="Drift detection: meta verify checks code, prompts, and schema against one model and fails the build on any mismatch."
+               aria-label="Drift detection: meta verify checks code, prompts, schema, and declared requirements against one model. Drift fails the build; an unimplemented claim warns."
                style="width:100%;max-width:860px;height:auto;border-radius:14px;border:1px solid rgba(255,255,255,0.08);margin:1rem auto 0;display:block;">
           <source src="https://metaobjects.com/video/drift-detection.mp4" type="video/mp4">
           <track kind="captions" src="/video/drift-detection.vtt" srclang="en" label="English">