afaryy · afaryy · Sep 8, 2026 · Sep 8, 2026 · Sep 8, 2026 · Sep 8, 2026
diff --git a/README.md b/README.md
@@ -41,8 +41,10 @@ Vertex AI supports media analysis and constrained storyboard planning. The API
 validates the plan before FFmpeg renders the selected source media, with optional
 original music from Google Lyria. Google Cloud AI is the product's only AI provider. The official ClickHouse MCP path checks selected-media
 consent evidence before rendering and again before export. The separately deployed
-ADK / Agent Engine planning endpoint is shown below the main flow; the current web
-interface does not call it.
+Google ADK / Vertex AI Agent Engine planning endpoint is shown below the main flow
+and is labelled **Deployed & smoke-test verified**. Solid arrows show the public
+web path; dashed arrows show the separate Agent Engine endpoint. The current web
+interface does not call that endpoint.
 
 [Read the architecture guide](docs/ARCHITECTURE.md) for the workflow, module
 responsibilities, data boundaries, and verified limitations, or
@@ -129,9 +131,7 @@ comma-separated, exact browser origins allowed to call the API.
 - [Application deployment](docs/operations/APP_DEPLOYMENT.md)
 - [Public-domain operations](docs/operations/public-edge.md)
 - [Devpost project page](docs/submission/DEVPOST_PROJECT_PAGE.md)
-- [ST-32 competition compliance checklist](docs/submission/SUBMISSION_CHECKLIST.md)
 - [ST-33 public evidence package](docs/submission/EVIDENCE_PACKAGE.md)
-- [Three-minute demo script](docs/submission/DEMO_SCRIPT.md)
 
 Internal planning artifacts are kept separate from the public product-documentation path.
 

diff --git a/docs/ARCHITECTURE.md b/docs/ARCHITECTURE.md
@@ -27,9 +27,11 @@ return messages and infrastructure connections for readability. In particular,
 Gemini and Lyria return results to the API; Lyria is not downstream of Gemini as a
 separate autonomous agent.
 
-The grey strip shows a **separate deployed planning endpoint**. Its dashed arrows
-do not mean every web request uses Agent Engine. This endpoint has hosted smoke
-evidence, but the current public web journey uses `/storyboards` directly.
+The grey strip is labelled **Google ADK + Vertex AI Agent Engine · Deployed &
+smoke-test verified** and shows the **separate deployed planning endpoint**. Its
+dashed arrows represent that endpoint's ClickHouse preference lookup, verified
+through hosted deployment smoke tests. The footer distinguishes this from the
+solid-arrow public web path: the current web journey uses `/storyboards` directly.
 
 ### 1. You tell the story
 

diff --git a/docs/CAPABILITY_EVIDENCE.md b/docs/CAPABILITY_EVIDENCE.md
@@ -39,8 +39,8 @@ Evidence levels are intentionally separate:
 | Real iPhone journey | Not a code capability | Not proven by emulation | Passed on iPhone 11, iOS 26.6.1, Chrome (version not recorded) | Evidence ready; final recording still pending | [ST-52 iPhone acceptance](qa/ST-52-PHYSICAL-IPHONE-ACCEPTANCE.md) |
 | Real Android journey | Not a code capability | Not proven by emulation | Not tested | Not required for the Web entry | No Android claim is made |
 | Rights-cleared final demo media | Eight fixtures and the generated song are recorded and approved | Final 60-second export reviewed | Owner accepted incidental-background risk and approved the complete song | Evidence ready; keep private artefacts private | [Media rights register](demo/MEDIA_RIGHTS_REGISTER.md) — Yvonne |
-| Final English demo video, at most three minutes | Script and runbook exist | Not applicable | iPhone rehearsal evidence is ready | Pending ST-17 | [Demo script](submission/DEMO_SCRIPT.md), [demo runbook](demo/DEMO_RUNBOOK.md) — Yvonne |
-| Devpost entry and submission confirmation | Draft package exists | Not applicable | iPhone evidence ready | Pending final submission checklist | [Submission checklist](submission/SUBMISSION_CHECKLIST.md), [evidence package](submission/EVIDENCE_PACKAGE.md) — Yvonne; also depends on ST-17, roster, eligibility, track selection, public video, form review, and receipt |
+| Final English demo video, at most three minutes | Script and runbook exist | Not applicable | iPhone rehearsal evidence is ready | Pending ST-17 | Private demo script, [demo runbook](demo/DEMO_RUNBOOK.md) — Yvonne |
+| Devpost entry and submission confirmation | Draft package exists | Not applicable | iPhone evidence ready | Pending final submission checklist | Private submission checklist, [evidence package](submission/EVIDENCE_PACKAGE.md) — Yvonne; also depends on ST-17, roster, eligibility, track selection, public video, form review, and receipt |
 
 The deployed browser evidence does not imply full phone-library access, direct
 posting to social networks, Android or native-screen-reader coverage, a finished

diff --git a/docs/PROJECT_BRIEF.md b/docs/PROJECT_BRIEF.md
@@ -27,8 +27,9 @@ Memory Director is a mobile-first web application that guides a user through:
 
 The core experience combines a simple voice/text request with selected media,
 a validated production plan, and a playable film the user controls. The official
-`mcp-clickhouse` consent/export path can block rendering and export when required
-consent evidence is missing.
+`mcp-clickhouse` path checks consent records before rendering and again before
+export. Either action is blocked when permission is denied or required consent
+evidence is missing.
 
 A separate deployed Agent Engine planner demonstrates an approved, read-only
 ClickHouse preference lookup. It is not connected to the public web interface.

diff --git a/docs/assets/architecture/memorydirector-architecture.drawio b/docs/assets/architecture/memorydirector-architecture.drawio
diff --git a/docs/assets/architecture/memorydirector-architecture.png b/docs/assets/architecture/memorydirector-architecture.png
diff --git a/docs/research/agentic-cinema-link-archive.md b/docs/research/agentic-cinema-link-archive.md
@@ -7,7 +7,7 @@ Retrieved 2026-08-16. This is a practical archive of every unique destination ex
 - The challenge asks for a functional, production-ready Gemini + Google Cloud Agent Builder agent or multi-agent network for a media-and-entertainment workflow.
 - A project must actively use the selected partner at runtime. For ClickHouse, that means the official `mcp-clickhouse` server connected to ClickHouse Cloud or a self-hosted cluster.
 - The resources page explicitly includes multimodal video analysis, audio/music generation (Lyria 3), Gemini TTS, native ADK/Agent Engine, function calling, Cloud Run, and Secret Manager.
-- The live official rules, overview, and dates page were rechecked on 7 September 2026 and now agree on the entry deadline: **2026-09-09 14:00 PDT**. The earlier September 7 conflict has been corrected on Devpost. See the current [ST-32 compliance checklist](../submission/SUBMISSION_CHECKLIST.md) and recheck the live rules immediately before submission.
+- The live official rules, overview, and dates page were rechecked on 7 September 2026 and now agree on the entry deadline: **2026-09-09 14:00 PDT**. The earlier September 7 conflict has been corrected on Devpost. The ST-32 compliance checklist is maintained privately; recheck the live rules immediately before submission.
 - The rules prohibit non-Google AI models, AI APIs, and agent frameworks in the submitted project. Use only Google Cloud AI plus the chosen partner’s built-in AI capabilities.
 
 ## Challenge pages

diff --git a/docs/submission/DEMO_SCRIPT.md b/docs/submission/DEMO_SCRIPT.md
diff --git a/docs/submission/DEVPOST_PROJECT_PAGE.md b/docs/submission/DEVPOST_PROJECT_PAGE.md
@@ -13,7 +13,7 @@ The browser only sees files the user selects. Media processing requires consent,
 ## How we built it
 We built a mobile-first Next.js and React interface backed by a FastAPI service on Google Cloud Run. Gemini on Vertex AI supports media analysis and turns a plain-language request into a constrained storyboard. The API validates the plan before FFmpeg renders a 60-second portrait MP4 from the selected source media. Google Lyria supports the original music option, with instrumental and no-music fallbacks.
 
-ClickHouse Cloud is part of the runtime through the official mcp-clickhouse server: consent/export checks can block rendering and export, and scoped events provide evidence of those decisions. We also deployed a separate Google ADK / Vertex AI Agent Engine production-planning endpoint that invokes an approved, read-only ClickHouse preference tool. That endpoint has passed deployment smoke checks; the current public web interface uses the direct Gemini storyboard path rather than the separate Agent Engine endpoint. Private Cloud Storage holds consented media.
+ClickHouse Cloud is part of the runtime through the official mcp-clickhouse server: consent-record checks before rendering and export can block either action if permission is denied, and scoped events provide evidence of those decisions. We also deployed a separate Google ADK / Vertex AI Agent Engine production-planning endpoint that invokes an approved, read-only ClickHouse preference tool. That endpoint has passed deployment smoke checks; the current public web interface uses the direct Gemini storyboard path rather than the separate Agent Engine endpoint. Private Cloud Storage holds consented media.
 
 We built automated deployment workflows with GitHub Actions and Terraform for Google Cloud infrastructure, containerized services, public-domain configuration, ClickHouse schema and access setup, and ADK Agent Engine deployment with runtime verification. Workload Identity Federation supports deployment authentication, and Secret Manager holds runtime credentials.
 

diff --git a/docs/submission/EVIDENCE_PACKAGE.md b/docs/submission/EVIDENCE_PACKAGE.md
@@ -24,8 +24,8 @@ documentation and approved evidence, not the deployed application image.
 | Demo language | English or English-subtitled | Blocked on final video |
 
 Use [`DEVPOST_PROJECT_PAGE.md`](DEVPOST_PROJECT_PAGE.md) as the source for the
-long description, [`DEMO_SCRIPT.md`](DEMO_SCRIPT.md) for the recording, and
-[`SUBMISSION_CHECKLIST.md`](SUBMISSION_CHECKLIST.md) for rules compliance.
+long description. The recording script and internal submission checklist are
+maintained privately by the project owner.
 The [capability evidence matrix](../CAPABILITY_EVIDENCE.md) is the cross-document
 source of truth for implemented, deployed, physical-device, and final-submission
 claims.