-
+
-
-
+
+
-
-
+
+
-
-
+
+
diff --git a/docs/media/landing-story/clips/automate.mp4 b/docs/media/landing-story/clips/automate.mp4
index 5d6d9fb8..18c065e2 100644
Binary files a/docs/media/landing-story/clips/automate.mp4 and b/docs/media/landing-story/clips/automate.mp4 differ
diff --git a/docs/media/landing-story/clips/automate.webm b/docs/media/landing-story/clips/automate.webm
index b337cbbd..dec10c56 100644
Binary files a/docs/media/landing-story/clips/automate.webm and b/docs/media/landing-story/clips/automate.webm differ
diff --git a/docs/media/landing-story/clips/create.mp4 b/docs/media/landing-story/clips/create.mp4
index 2f4182a2..8bafeed8 100644
Binary files a/docs/media/landing-story/clips/create.mp4 and b/docs/media/landing-story/clips/create.mp4 differ
diff --git a/docs/media/landing-story/clips/create.webm b/docs/media/landing-story/clips/create.webm
index 1777a33c..f0d79ae4 100644
Binary files a/docs/media/landing-story/clips/create.webm and b/docs/media/landing-story/clips/create.webm differ
diff --git a/docs/media/landing-story/clips/research.mp4 b/docs/media/landing-story/clips/research.mp4
index 992ca1aa..b67a65fb 100644
Binary files a/docs/media/landing-story/clips/research.mp4 and b/docs/media/landing-story/clips/research.mp4 differ
diff --git a/docs/media/landing-story/clips/research.webm b/docs/media/landing-story/clips/research.webm
index 9da1c2d2..bba7542e 100644
Binary files a/docs/media/landing-story/clips/research.webm and b/docs/media/landing-story/clips/research.webm differ
diff --git a/docs/media/landing-story/clips/ship.mp4 b/docs/media/landing-story/clips/ship.mp4
index 32d1199a..7b955d96 100644
Binary files a/docs/media/landing-story/clips/ship.mp4 and b/docs/media/landing-story/clips/ship.mp4 differ
diff --git a/docs/media/landing-story/clips/ship.webm b/docs/media/landing-story/clips/ship.webm
index 784d5b8a..b7194a7e 100644
Binary files a/docs/media/landing-story/clips/ship.webm and b/docs/media/landing-story/clips/ship.webm differ
diff --git a/docs/media/landing-story/manifest.json b/docs/media/landing-story/manifest.json
index a4398fc7..f264b488 100644
--- a/docs/media/landing-story/manifest.json
+++ b/docs/media/landing-story/manifest.json
@@ -1,78 +1,87 @@
{
"schema": 1,
- "story_id": "sovereignty-workbench-v1",
- "run_id": "20260921T230035Z-f3ed05",
+ "story_id": "react-workbench-v1",
+ "run_id": "20261002T015343Z-6e392a",
"review_status": "approved",
- "reviewed_at": "2026-09-23T10:37:41Z",
+ "reviewed_at": "2026-10-02T01:53:43Z",
"recording_receipt": "recording-receipt.json",
+ "production": {
+ "app": "Row-Bot 5.0.0, the real app process and its React client at /app-v2/",
+ "method": "Scripted Playwright capture (scripts/docs/capture_landing_media.py) in Microsoft Edge at 1440x810 CSS px, dark appearance, on an isolated temporary profile seeded with fictional demo data (a neighbourhood solar co-op); no real user data, accounts, names or paths",
+ "models": "No model ran. The local runtime is the docs capture's display-only stand-in on loopback: it lists qwen3.8:27b, the local model the owner runs, and runs nothing. GPT-5.6 Sol via ChatGPT / Codex is display-only: the harness tells the client it is connected, and nothing is connected or called",
+ "outward_actions": "None: the harness answers the Ship clip's Approve request itself, so no email is sent",
+ "pointer": "Drawn by the harness to show where the scripted clicks land",
+ "stills": "Rendered at 2x and downscaled with Lanczos to 1440x810 WebP",
+ "clips": "Browser screencast frames resampled to 30 fps; VP9 WebM and H.264 MP4 (faststart), no audio"
+ },
"assets": [
{
"scene_id": "hero-app",
"still": "screenshots/research.webp",
- "sha256": "68310dfdebba4089fbadb43f0eae000e966b975a6f1586cb33c4b1c170f0d90c",
- "alt": "Row-Bot using a local model for research and resolving the findings into connected knowledge."
+ "sha256": "21f0ddffcbaa7a70b75de21b468990e11048f8a530fcca47d53a4d4f6fa99e17",
+ "alt": "Row-Bot's Knowledge graph after local-model research, with the co-op's memory and its connections highlighted."
},
{
"scene_id": "research-local",
"still": "screenshots/research.webp",
- "sha256": "68310dfdebba4089fbadb43f0eae000e966b975a6f1586cb33c4b1c170f0d90c",
+ "sha256": "21f0ddffcbaa7a70b75de21b468990e11048f8a530fcca47d53a4d4f6fa99e17",
"clip": "clips/research.webm",
- "clip_sha256": "97e8e0238875d785da1d3ecab244c41705e10257d17b4f37cbd1d249e62483e1",
+ "clip_sha256": "66f945575c014f307b24667aae7da64d4969a4b4224d5f3c06951a6dd4374100",
"clip_fallback": "clips/research.mp4",
- "clip_fallback_sha256": "32aa733d8e7f63be8bd7069bb1ad99c771c2fb9e9a69896eb28b08342e567b1f",
- "alt": "Local-model research with source activity followed by a populated node-and-edge knowledge graph."
+ "clip_fallback_sha256": "fb222cf2667d296c8ca675578432f45e63221f6a51484cecfef1812dae5867dd",
+ "alt": "A local-model conversation with its source trace, followed by the Knowledge graph settling into connected memories."
},
{
"scene_id": "synthesis-sol",
"still": "screenshots/create.webp",
- "sha256": "d8fc8adf27a1b5b73731747fe679e2c8712a999f2366b2dcc5e1a95852cc1cee",
+ "sha256": "c532a3d8e9d326162253497e90a347416ecc559eb54b789133377005a6c74dd8",
"clip": "clips/create.webm",
- "clip_sha256": "b4429bfbd43f6be42bd072af7273d563e989590187365fa4207292e7fad9b466",
+ "clip_sha256": "37c3142237e041bf2a431208d3f97a679718e40b3095913075c7c25a29caeaca",
"clip_fallback": "clips/create.mp4",
- "clip_fallback_sha256": "f7cc3f8b29c4fdff22649a321e34584cdb2bc97ded7038a5cc5bf7410e9a826c",
- "alt": "An explicit frontier-model handoff followed by a real editable Designer artifact."
+ "clip_fallback_sha256": "be7637a203f56c8521dd9d329ac03bec93136f94a2d1ddbd56f01311c419145a",
+ "alt": "A conversation handed to a frontier model, GPT-5.6 Sol, with an editable landing page in its Design panel."
},
{
"scene_id": "knowledge-control",
"still": "screenshots/research.webp",
- "sha256": "68310dfdebba4089fbadb43f0eae000e966b975a6f1586cb33c4b1c170f0d90c",
+ "sha256": "21f0ddffcbaa7a70b75de21b468990e11048f8a530fcca47d53a4d4f6fa99e17",
"clip": "clips/research.webm",
- "clip_sha256": "97e8e0238875d785da1d3ecab244c41705e10257d17b4f37cbd1d249e62483e1",
- "alt": "A genuine Row-Bot knowledge graph with many connected nodes and edges from the research workspace."
+ "clip_sha256": "66f945575c014f307b24667aae7da64d4969a4b4224d5f3c06951a6dd4374100",
+ "alt": "A Row-Bot knowledge graph of 103 connected memories, with one memory's connections labelled."
},
{
"scene_id": "designer-output",
"still": "screenshots/create.webp",
- "sha256": "d8fc8adf27a1b5b73731747fe679e2c8712a999f2366b2dcc5e1a95852cc1cee",
+ "sha256": "c532a3d8e9d326162253497e90a347416ecc559eb54b789133377005a6c74dd8",
"clip": "clips/create.webm",
- "clip_sha256": "b4429bfbd43f6be42bd072af7273d563e989590187365fa4207292e7fad9b466",
- "alt": "A real editable launch visual open in Row-Bot Designer Studio."
+ "clip_sha256": "37c3142237e041bf2a431208d3f97a679718e40b3095913075c7c25a29caeaca",
+ "alt": "An editable landing page open in the conversation's Design panel, its heading then selected for editing."
},
{
"scene_id": "workflow-repeat",
"still": "screenshots/automate.webp",
- "sha256": "c980145f3439741e79b567764ed0423f3ae57af1d4d7069d21ae65c9639f4e8e",
+ "sha256": "1a48703370bd98c111ce0a1fa95f8dc89e15ba02df134fab46c4c1e1c516bf0d",
"clip": "clips/automate.webm",
- "clip_sha256": "5cdadc636b961ad471fecd3ab58413211948871741099eecb5f36235ad3fcccf",
+ "clip_sha256": "460b3b3a72c9bfcf95ec441a45fc417f70a4b58b4df60a4a81dc81103ae5ffef",
"clip_fallback": "clips/automate.mp4",
- "clip_fallback_sha256": "bce62f19eaac16156a18834884679a251d96b671a900ed4044e2cecde31374f3",
- "alt": "A local-model workflow being configured and saved with external delivery disabled."
+ "clip_fallback_sha256": "67ac6fbb2398dbff2459c13f9ed63feb9e62907b25c50ea9af9e589815997bbd",
+ "alt": "Workflow delivery kept in the app with no outside channels, a workflow's local model saved, then its retained run history."
},
{
"scene_id": "approval-boundary",
"still": "screenshots/ship.webp",
- "sha256": "37c1f310ed1201f05ee916d5d7709897aee1288bfa062fbecd904b9e667bdeb6",
+ "sha256": "98214f35cfdc1b9eba3c215e79a8128d28739a091d36140450be0f1ec27c2c2a",
"clip": "clips/ship.webm",
- "clip_sha256": "12bb1ee4e1faf736bc66b4fd3428594afd78cf5a7c5d14f25ef3ab24a4495e60",
+ "clip_sha256": "b4732efdc66ccb0a89caa4e3e7f6bceba3e697ac0ca8547e85db7a429b18bc23",
"clip_fallback": "clips/ship.mp4",
- "clip_fallback_sha256": "7d72c19347d937f52fad94061e7104f5c7ab18d37e01f84f2cdfa8034a89a234",
- "alt": "Row-Bot requesting approval before creating a reviewed local campaign artifact."
+ "clip_fallback_sha256": "1295b6b23d32244918f1d683a3fb86aee3657ade02017122a7d34c8b38cad777",
+ "alt": "Row-Bot waiting for approval before emailing a PDF to riverbend-board@example.com, then the approval being given."
},
{
"scene_id": "model-choice",
"still": "screenshots/create.webp",
- "sha256": "d8fc8adf27a1b5b73731747fe679e2c8712a999f2366b2dcc5e1a95852cc1cee",
- "alt": "The model picker making the local and frontier model boundary explicit."
+ "sha256": "c532a3d8e9d326162253497e90a347416ecc559eb54b789133377005a6c74dd8",
+ "alt": "The composer's model pill showing GPT-5.6 Sol as a cloud model, beside the conversation's Design panel."
},
{
"scene_id": "mobile-shell",
@@ -117,7 +126,10 @@
"source_commit": "b614d94f",
"source_sha256": "f1d4b3ade0d97361163bfd7cb5832b94197671ef922da119b8b0b698187718ca",
"source_duration_seconds": 12.042,
- "kept_segment_seconds": [5.333, 9.916],
+ "kept_segment_seconds": [
+ 5.333,
+ 9.916
+ ],
"method": "VP9/WebM packet remux without re-encoding; alpha preserved",
"reason": "Remove the violet opening orbs and retain the cyan Automate motion",
"result_sha256": "00344446d1e9bb4624e7ed4cbbfadfc0972d905372d3fb4fa3d7f2e2a1003369",
diff --git a/docs/media/landing-story/recording-receipt.json b/docs/media/landing-story/recording-receipt.json
index e688257f..cab6780f 100644
--- a/docs/media/landing-story/recording-receipt.json
+++ b/docs/media/landing-story/recording-receipt.json
@@ -1,144 +1,134 @@
{
"schema": 1,
- "story_id": "sovereignty-workbench-v1",
- "editing": "normal-speed editorial cuts with four-frame dissolves; sharpened encodes from matching reviewed source recordings; no global acceleration; Ship approval remains unmodified from the test-machine recording",
+ "story_id": "react-workbench-v1",
+ "run_id": "20261002T015343Z-6e392a",
+ "editing": "normal-speed editorial cuts with four-frame dissolves; scripted takes of the real app, no acceleration",
+ "production": {
+ "app": "Row-Bot 5.0.0, the real app process and its React client at /app-v2/",
+ "method": "Scripted Playwright capture (scripts/docs/capture_landing_media.py) in Microsoft Edge at 1440x810 CSS px, dark appearance, on an isolated temporary profile seeded with fictional demo data (a neighbourhood solar co-op); no real user data, accounts, names or paths",
+ "models": "No model ran. The local runtime is the docs capture's display-only stand-in on loopback: it lists qwen3.8:27b, the local model the owner runs, and runs nothing. GPT-5.6 Sol via ChatGPT / Codex is display-only: the harness tells the client it is connected, and nothing is connected or called",
+ "outward_actions": "None: the harness answers the Ship clip's Approve request itself, so no email is sent",
+ "pointer": "Drawn by the harness to show where the scripted clicks land",
+ "stills": "Rendered at 2x and downscaled with Lanczos to 1440x810 WebP",
+ "clips": "Browser screencast frames resampled to 30 fps; VP9 WebM and H.264 MP4 (faststart), no audio"
+ },
"clips": {
"research": {
- "source_name": "1.mp4",
- "source_sha256": "975367434a6e5f9c1349a9e48d8474385d752fa6214a068fd00f73c6fc40cd84",
- "source_duration_seconds": 47.133,
+ "source": "Screencast of the real Row-Bot 5.0.0 React client",
+ "source_frames": 501,
"segments_seconds": [
[
- 3.0,
- 8.6
+ 0.0,
+ 3.526
],
[
- 44.0,
- 47.05
+ 3.768,
+ 9.775
]
],
- "duration_seconds": 8.533,
+ "duration_seconds": 9.4,
"dimensions": [
1440,
810
],
"fps": 30,
"webm": "media/landing-story/clips/research.webm",
- "webm_sha256": "97e8e0238875d785da1d3ecab244c41705e10257d17b4f37cbd1d249e62483e1",
- "webm_bytes": 602239,
+ "webm_sha256": "66f945575c014f307b24667aae7da64d4969a4b4224d5f3c06951a6dd4374100",
+ "webm_bytes": 791002,
"mp4": "media/landing-story/clips/research.mp4",
- "mp4_sha256": "32aa733d8e7f63be8bd7069bb1ad99c771c2fb9e9a69896eb28b08342e567b1f",
- "mp4_bytes": 339781,
+ "mp4_sha256": "fb222cf2667d296c8ca675578432f45e63221f6a51484cecfef1812dae5867dd",
+ "mp4_bytes": 749915,
"poster": "media/landing-story/screenshots/research.webp",
- "poster_sha256": "68310dfdebba4089fbadb43f0eae000e966b975a6f1586cb33c4b1c170f0d90c",
- "poster_bytes": 89492,
- "editorial_note": "Reviewed source frames retained at natural 30 fps; four-frame dissolves join selected actions."
+ "poster_sha256": "21f0ddffcbaa7a70b75de21b468990e11048f8a530fcca47d53a4d4f6fa99e17",
+ "poster_bytes": 67816,
+ "editorial_note": "Two takes at normal speed joined by a four-frame dissolve: the local-model conversation's source trace, then the Knowledge graph settling and its co-op memory highlighted."
},
"create": {
- "source_name": "2.mp4",
- "source_sha256": "6624ace4e8f218a934899f6a9ff81d0208b2b886e68c75988e55a103a01f8055",
- "source_duration_seconds": 50.0,
+ "source": "Screencast of the real Row-Bot 5.0.0 React client",
+ "source_frames": 325,
"segments_seconds": [
[
- 2.0,
- 4.5
- ],
- [
- 12.0,
- 14.5
- ],
- [
- 22.5,
- 25.5
- ],
- [
- 47.0,
- 48.5
+ 0.0,
+ 9.445
]
],
- "duration_seconds": 9.1,
+ "duration_seconds": 9.433,
"dimensions": [
1440,
810
],
"fps": 30,
"webm": "media/landing-story/clips/create.webm",
- "webm_sha256": "b4429bfbd43f6be42bd072af7273d563e989590187365fa4207292e7fad9b466",
- "webm_bytes": 733183,
+ "webm_sha256": "37c3142237e041bf2a431208d3f97a679718e40b3095913075c7c25a29caeaca",
+ "webm_bytes": 285946,
"mp4": "media/landing-story/clips/create.mp4",
- "mp4_sha256": "f7cc3f8b29c4fdff22649a321e34584cdb2bc97ded7038a5cc5bf7410e9a826c",
- "mp4_bytes": 445845,
+ "mp4_sha256": "be7637a203f56c8521dd9d329ac03bec93136f94a2d1ddbd56f01311c419145a",
+ "mp4_bytes": 276072,
"poster": "media/landing-story/screenshots/create.webp",
- "poster_sha256": "d8fc8adf27a1b5b73731747fe679e2c8712a999f2366b2dcc5e1a95852cc1cee",
- "poster_bytes": 86514,
- "editorial_note": "Reviewed source frames retained at natural 30 fps; four-frame dissolves join selected actions."
+ "poster_sha256": "c532a3d8e9d326162253497e90a347416ecc559eb54b789133377005a6c74dd8",
+ "poster_bytes": 70140,
+ "editorial_note": "One continuous take at normal speed: the hosted model shown in the picker, then the landing page opened for editing in the Design panel."
},
"automate": {
- "source_name": "3.mp4",
- "source_sha256": "b52e3b3ff45824f452c3b20d76ca54a5831f5e4becfce86947154a7d639477cf",
- "source_duration_seconds": 23.133,
+ "source": "Screencast of the real Row-Bot 5.0.0 React client",
+ "source_frames": 532,
"segments_seconds": [
[
- 0.2,
- 2.2
- ],
- [
- 2.4,
- 4.0
+ 0.0,
+ 3.392
],
[
- 7.2,
- 9.8
+ 5.795,
+ 8.919
],
[
- 12.0,
- 14.8
+ 11.296,
+ 13.834
]
],
- "duration_seconds": 8.6,
+ "duration_seconds": 8.8,
"dimensions": [
1440,
810
],
"fps": 30,
"webm": "media/landing-story/clips/automate.webm",
- "webm_sha256": "5cdadc636b961ad471fecd3ab58413211948871741099eecb5f36235ad3fcccf",
- "webm_bytes": 330721,
+ "webm_sha256": "460b3b3a72c9bfcf95ec441a45fc417f70a4b58b4df60a4a81dc81103ae5ffef",
+ "webm_bytes": 317374,
"mp4": "media/landing-story/clips/automate.mp4",
- "mp4_sha256": "bce62f19eaac16156a18834884679a251d96b671a900ed4044e2cecde31374f3",
- "mp4_bytes": 213856,
+ "mp4_sha256": "67ac6fbb2398dbff2459c13f9ed63feb9e62907b25c50ea9af9e589815997bbd",
+ "mp4_bytes": 279166,
"poster": "media/landing-story/screenshots/automate.webp",
- "poster_sha256": "c980145f3439741e79b567764ed0423f3ae57af1d4d7069d21ae65c9639f4e8e",
- "poster_bytes": 39512,
- "editorial_note": "Reviewed source frames retained at natural 30 fps; four-frame dissolves join selected actions."
+ "poster_sha256": "1a48703370bd98c111ce0a1fa95f8dc89e15ba02df134fab46c4c1e1c516bf0d",
+ "poster_bytes": 66846,
+ "editorial_note": "Three takes at normal speed joined by four-frame dissolves: the workflow delivery defaults (no external channels), the workflow's local model saved, then its retained run history."
},
"ship": {
- "source_name": "4.mp4",
- "source_sha256": "e4441c570bb7c4c27137ca61c40a0757597c6fadfb2145e1c6d2cbeb8d786d28",
- "source_duration_seconds": 21.6,
+ "source": "Screencast of the real Row-Bot 5.0.0 React client",
+ "source_frames": 178,
"segments_seconds": [
[
- 9.1,
- 11.7
+ 0.0,
+ 2.887
]
],
- "duration_seconds": 2.6,
+ "duration_seconds": 2.9,
"dimensions": [
1440,
810
],
"fps": 30,
"webm": "media/landing-story/clips/ship.webm",
- "webm_sha256": "12bb1ee4e1faf736bc66b4fd3428594afd78cf5a7c5d14f25ef3ab24a4495e60",
- "webm_bytes": 125411,
+ "webm_sha256": "b4732efdc66ccb0a89caa4e3e7f6bceba3e697ac0ca8547e85db7a429b18bc23",
+ "webm_bytes": 107415,
"mp4": "media/landing-story/clips/ship.mp4",
- "mp4_sha256": "7d72c19347d937f52fad94061e7104f5c7ab18d37e01f84f2cdfa8034a89a234",
- "mp4_bytes": 84508,
+ "mp4_sha256": "1295b6b23d32244918f1d683a3fb86aee3657ade02017122a7d34c8b38cad777",
+ "mp4_bytes": 107139,
"poster": "media/landing-story/screenshots/ship.webp",
- "poster_sha256": "37c1f310ed1201f05ee916d5d7709897aee1288bfa062fbecd904b9e667bdeb6",
- "poster_bytes": 39676,
- "editorial_note": "Reviewed source frames retained at natural 30 fps; four-frame dissolves join selected actions.",
- "approval_moment_seconds": 2.25
+ "poster_sha256": "98214f35cfdc1b9eba3c215e79a8128d28739a091d36140450be0f1ec27c2c2a",
+ "poster_bytes": 62198,
+ "editorial_note": "One continuous take at normal speed: the waiting email approval, then Approve.",
+ "approval_moment_seconds": 2.203
}
}
}
diff --git a/docs/media/landing-story/screenshots/automate.webp b/docs/media/landing-story/screenshots/automate.webp
index e67e4242..86e3de5a 100644
Binary files a/docs/media/landing-story/screenshots/automate.webp and b/docs/media/landing-story/screenshots/automate.webp differ
diff --git a/docs/media/landing-story/screenshots/create.webp b/docs/media/landing-story/screenshots/create.webp
index 8a04dca5..d380e917 100644
Binary files a/docs/media/landing-story/screenshots/create.webp and b/docs/media/landing-story/screenshots/create.webp differ
diff --git a/docs/media/landing-story/screenshots/research.webp b/docs/media/landing-story/screenshots/research.webp
index fa1cc12e..b67085a5 100644
Binary files a/docs/media/landing-story/screenshots/research.webp and b/docs/media/landing-story/screenshots/research.webp differ
diff --git a/docs/media/landing-story/screenshots/ship.webp b/docs/media/landing-story/screenshots/ship.webp
index 7ae70ccf..9262390b 100644
Binary files a/docs/media/landing-story/screenshots/ship.webp and b/docs/media/landing-story/screenshots/ship.webp differ
diff --git a/scripts/docs/capture_landing_media.py b/scripts/docs/capture_landing_media.py
new file mode 100644
index 00000000..a0422179
--- /dev/null
+++ b/scripts/docs/capture_landing_media.py
@@ -0,0 +1,1638 @@
+"""Capture the landing page's story stills and clips from the real React app.
+
+ uv run --with imageio-ffmpeg python scripts/docs/capture_landing_media.py
+ uv run --with imageio-ffmpeg python scripts/docs/capture_landing_media.py --scenes ship
+ uv run python scripts/docs/capture_landing_media.py --serve
+
+Row-Bot runs against an isolated temporary profile seeded with a fictional
+neighbourhood solar co-op: conversations with a tool trace, a knowledge graph,
+local workflows with their runs, a landing page design bound to its
+conversation and an outward email waiting for approval. The local model runtime
+is the docs capture's display-only stand-in on loopback, so nothing runs a
+model. The Create scene's hosted model is display-only too: the harness tells
+the client that ChatGPT / Codex is connected; no account is connected or
+called. In the Ship scene the harness answers the Approve request itself, so
+nothing is sent.
+
+Each scene is driven with Playwright at 1440x810 in the dark appearance, with a
+drawn pointer showing where the scripted clicks land. Stills are rendered at
+twice the density and downscaled; clips are recorded from the browser's
+screencast, resampled to 30 fps and encoded with ffmpeg (``imageio-ffmpeg``,
+added with ``uv run --with``; it is not a project dependency). Output lands in
+``docs/media/landing-story/``, with the manifest and recording receipt rewritten
+to match (``--records`` rewrites only those); scratch stays under
+``.tmp/landing-capture/``. ``--serve`` seeds and starts the app for a manual look
+and stops it when ``.tmp/landing-capture/serve.stop`` appears.
+"""
+
+from __future__ import annotations
+
+import argparse
+import base64
+import hashlib
+import io
+import json
+import os
+import sqlite3
+import subprocess
+import sys
+import tempfile
+import threading
+import time
+from collections import deque
+from contextlib import ExitStack
+from datetime import datetime, timedelta, timezone
+from http.server import ThreadingHTTPServer
+from pathlib import Path
+from typing import Any, Callable
+
+ROOT = Path(__file__).resolve().parents[2]
+if str(ROOT) not in sys.path:
+ sys.path.insert(0, str(ROOT))
+
+from scripts.docs.capture_real_ui_screenshots import ( # noqa: E402
+ _DemoOllamaHandler,
+ _free_port,
+ _launch_app,
+ _wait_ping,
+)
+
+SCRATCH = ROOT / ".tmp" / "landing-capture"
+MEDIA = ROOT / "docs" / "media" / "landing-story"
+WIDTH, HEIGHT = 1440, 810
+FPS = 30
+DISSOLVE_FRAMES = 4
+SCENES = ("research", "create", "automate", "ship")
+
+RESEARCH_THREAD = "landing-research"
+CREATE_THREAD = "landing-create"
+SHIP_THREAD = "landing-ship"
+DIGEST_THREAD = "landing-digest-runs"
+DESIGN_ID = "landing-launch-page"
+SHIP_APPROVAL_ID = "1a4d1a000001"
+# The local model the footage shows, as the owner runs it, used by every local
+# conversation and workflow. Display-only: nothing runs. (The picker sorts its
+# local models by name, so a second one would be listed above it.)
+LOCAL_MODELS = ("qwen3.8:27b",)
+LOCAL_MODEL = f"model:ollama:{LOCAL_MODELS[0]}"
+HOSTED_MODEL = "model:codex:gpt-5.6-sol"
+CO_OP = "Riverbend Community Solar"
+# The capture's clock: conversations, runs and the approval are dated around a
+# late-morning "now", and the browser's clock runs from the same moment.
+CLOCK_HOUR, CLOCK_MINUTE = 10, 40
+
+
+def _anchor() -> datetime:
+ return datetime.now().replace(hour=CLOCK_HOUR, minute=CLOCK_MINUTE, second=0, microsecond=0)
+
+
+# Seeding (runs in its own process, bound to the capture profile) -------------
+
+def _call(name: str, args: dict, call_id: str) -> dict:
+ return {"name": name, "args": args, "id": call_id, "type": "tool_call"}
+
+
+def _conversations() -> list[dict]:
+ from langchain_core.messages import AIMessage, HumanMessage, ToolMessage
+
+ def step(name: str, args: dict, call_id: str, result: str) -> list:
+ return [AIMessage(content="", tool_calls=[_call(name, args, call_id)]),
+ ToolMessage(content=result, name=name, tool_call_id=call_id)]
+
+ return [
+ {
+ "id": RESEARCH_THREAD,
+ "title": "Community solar research",
+ "minutes": 28,
+ "model": LOCAL_MODEL,
+ "messages": [
+ HumanMessage(content=(
+ "Research how community solar co-ops share savings with their members. "
+ "Use public sources, keep the evidence visible, and save what we learn to Knowledge."
+ )),
+ *step("search_memory", {"query": "Riverbend community solar"}, "r1",
+ "4 memories about Riverbend Community Solar"),
+ *step("duckduckgo_search", {"query": "community solar co-op member savings"}, "r2",
+ "8 results from public sources"),
+ *step("wikipedia", {"query": "community solar"}, "r3",
+ "3 articles: community solar, net metering, cooperatives"),
+ *step("read_url", {"url": "https://en.wikipedia.org/wiki/Net_metering"}, "r4",
+ "Net metering ยท encyclopedia article"),
+ *step("read_url", {"url": "https://en.wikipedia.org/wiki/Cooperative"}, "r5",
+ "Cooperative ยท encyclopedia article"),
+ *step("save_memory", {"subject": "Bill credits", "category": "concept"}, "r6",
+ "Saved Bill credits"),
+ *step("save_memory", {"subject": "Subscription model", "category": "concept"}, "r7",
+ "Saved Subscription model"),
+ *step("link_memories", {"subject": CO_OP}, "r8",
+ f"Linked 12 memories to {CO_OP}"),
+ AIMessage(content=(
+ "Here is what the public sources agree on:\n\n"
+ "- **Shared ownership**: members subscribe to, or buy, a share of one larger array "
+ "instead of fitting panels at home.\n"
+ "- **Bill credits**: each share's output shows up as a credit on the member's bill.\n"
+ "- **Subscribe or buy**: subscriptions avoid an upfront cost; bought shares pay back over time.\n"
+ "- **Renters can join**: no roof of their own is needed.\n"
+ "- **Local rules decide the credit**: confirm the calculation with the grid operator.\n\n"
+ "I saved 12 memories and linked them to *Riverbend Community Solar* in Knowledge."
+ )),
+ ],
+ },
+ {
+ "id": CREATE_THREAD,
+ "title": "Launch page direction",
+ "minutes": 9,
+ "model": HOSTED_MODEL,
+ "messages": [
+ HumanMessage(content=(
+ "Turn the research into a launch page for Riverbend Community Solar: a warm hero, "
+ "how shares work, and a clear invitation to join."
+ )),
+ *step("search_memory", {"query": "Riverbend community solar research"}, "c1",
+ "12 memories from Community solar research"),
+ *step("create_design", {"name": "Riverbend launch page", "mode": "landing"}, "c2",
+ "Created Riverbend launch page"),
+ AIMessage(content=(
+ "The launch page is open in the Design panel: a sunrise hero with the invitation to join, "
+ "three steps for how shares work, and what members get back. Edit any section directly, "
+ "or tell me what to change."
+ )),
+ ],
+ },
+ {
+ "id": SHIP_THREAD,
+ "title": "Share the launch page",
+ "minutes": 2,
+ "model": LOCAL_MODEL,
+ "messages": [
+ HumanMessage(content="Who should review the launch page before it goes public?"),
+ *step("search_memory", {"query": "Riverbend board review"}, "s9",
+ "3 memories: Board chair, Board meets on the first Tuesday, Board pack"),
+ AIMessage(content=(
+ "The board reviews anything members will read. Their shared address, "
+ "riverbend-board@example.com, reaches all five board members."
+ )),
+ HumanMessage(content="Check the launch page against the research before we share it."),
+ *step("search_memory", {"query": "Riverbend launch page facts"}, "s0",
+ "9 memories behind the launch page"),
+ AIMessage(content=(
+ "It matches the research: shares, bill credits and renter access are all explained, "
+ "and every figure on the page comes from Knowledge. The open evening date is on it too."
+ )),
+ HumanMessage(content="Export it as a PDF and email it to the co-op board for review."),
+ *step("export_to_pdf", {"name": "riverbend-launch-page.pdf"}, "s1",
+ "Saved riverbend-launch-page.pdf (3 pages)"),
+ AIMessage(
+ content="The PDF is ready. Emailing it leaves this computer, so it waits for your approval.",
+ tool_calls=[_call("send_gmail_message", {
+ "to": ["riverbend-board@example.com"],
+ "subject": "Launch page for review",
+ "attachments": ["riverbend-launch-page.pdf"],
+ }, "s2")],
+ ),
+ ],
+ },
+ {
+ "id": DIGEST_THREAD,
+ "title": "Weekly policy digest",
+ "minutes": 60 * 26,
+ "model": LOCAL_MODEL,
+ "thread_type": "workflow",
+ "messages": [
+ HumanMessage(content="Check public sources for community solar rule changes and summarise them."),
+ AIMessage(content=(
+ "**Policy digest**\n\n- No change to how bill credits are calculated this week.\n"
+ "- The small-grants round is open until the end of the month.\n"
+ "- Two new questions for the grid connection FAQ."
+ )),
+ ],
+ },
+ {
+ "id": "landing-agenda",
+ "title": "Board meeting agenda",
+ "minutes": 95,
+ "model": LOCAL_MODEL,
+ "messages": [
+ HumanMessage(content="Draft the agenda for Tuesday's board meeting."),
+ AIMessage(content=(
+ "1. Survey results: 14 suitable roofs\n2. Launch page review\n"
+ "3. Grant application timeline\n4. Open evening at Market Hall"
+ )),
+ ],
+ },
+ {
+ "id": "landing-survey",
+ "title": "Rooftop survey checklist",
+ "minutes": 60 * 23,
+ "model": LOCAL_MODEL,
+ "messages": [
+ HumanMessage(content="Make a checklist for the rooftop survey day."),
+ AIMessage(content=(
+ "- Roof direction and shading\n- Access and safety\n"
+ "- Structural notes for the surveyor\n- Photos of each roof"
+ )),
+ ],
+ },
+ {
+ "id": "landing-volunteers",
+ "title": "Volunteer onboarding notes",
+ "minutes": 60 * 24 * 3,
+ "model": LOCAL_MODEL,
+ "messages": [
+ HumanMessage(content="Summarise what new volunteers need to know."),
+ AIMessage(content=(
+ "Volunteers help at the open evening, on survey days and with member questions. "
+ "Training night covers all three."
+ )),
+ ],
+ },
+ {
+ "id": "landing-grant",
+ "title": "Grant application outline",
+ "minutes": 60 * 24 * 8,
+ "model": LOCAL_MODEL,
+ "messages": [
+ HumanMessage(content="Outline the small-grants application."),
+ AIMessage(content=(
+ "1. The co-op and who it serves\n2. The first array\n3. Costs the grant covers\n"
+ "4. Community benefit"
+ )),
+ ],
+ },
+ ]
+
+
+def _seed_conversations(anchor: datetime) -> None:
+ from row_bot import threads
+
+ for item in _conversations():
+ threads.create_thread(
+ item["title"],
+ thread_id=item["id"],
+ thread_type=item.get("thread_type", ""),
+ approval_mode="approve",
+ model_override=item["model"],
+ name_source="manual",
+ )
+ threads.append_checkpoint_messages(item["id"], item["messages"])
+ moment = (anchor - timedelta(minutes=item["minutes"])).isoformat()
+ with sqlite3.connect(threads.DB_PATH) as conn:
+ conn.execute(
+ "UPDATE thread_meta SET created_at = ?, updated_at = ? WHERE thread_id = ?",
+ (moment, moment, item["id"]),
+ )
+
+
+def _tasks_db() -> sqlite3.Connection:
+ return sqlite3.connect(Path(os.environ["ROW_BOT_DATA_DIR"]) / "tasks.db")
+
+
+def _seed_workflows(anchor: datetime) -> None:
+ from row_bot import tasks
+
+ digest = tasks.create_task(
+ "Weekly policy digest",
+ prompts=[
+ "Check public sources for changes to community solar rules this week.",
+ "Summarise anything that affects Riverbend in five plain-language bullets.",
+ "Save new facts to Knowledge and link them to Riverbend Community Solar.",
+ ],
+ description="Tracks community solar rule changes every Monday and keeps each run for review.",
+ icon="๐ฐ",
+ schedule="weekly:monday:08:00",
+ model_override=LOCAL_MODEL,
+ safety_mode="block",
+ channels=[],
+ persistent_thread_id=DIGEST_THREAD,
+ )
+ questions = tasks.create_task(
+ "Member questions round-up",
+ prompts=["Collect this week's member questions.", "Draft short answers for the FAQ."],
+ description="Gathers member questions into draft FAQ answers.",
+ icon="๐ฌ",
+ schedule="weekly:friday:16:00",
+ model_override=LOCAL_MODEL,
+ safety_mode="block",
+ channels=[],
+ )
+ survey = tasks.create_task(
+ "Survey photo sorter",
+ prompts=["Sort the new roof survey photos by site.", "List any roof that needs a second visit."],
+ description="Files survey photos by roof and flags second visits.",
+ icon="๐ ",
+ schedule="daily:18:00",
+ model_override=LOCAL_MODEL,
+ safety_mode="block",
+ channels=[],
+ )
+ tasks.create_task(
+ "Open evening reminder",
+ prompts=["Remind me to confirm the Market Hall booking."],
+ description="A reminder before the open evening.",
+ icon="๐",
+ schedule="weekly:wednesday:09:00",
+ notify_only=True,
+ notify_label="Confirm the Market Hall booking",
+ channels=[],
+ )
+
+ def run(task_id: str, thread: str, name: str, steps: int, days: float, message: str) -> None:
+ run_id = tasks._record_run_start(task_id, thread, steps, name, "bolt")
+ tasks._update_run_progress(run_id, steps)
+ tasks._finish_run(run_id, "completed", message)
+ started = anchor - timedelta(days=days)
+ with _tasks_db() as conn:
+ conn.execute(
+ "UPDATE task_runs SET started_at = ?, finished_at = ? WHERE id = ?",
+ (started.isoformat(), (started + timedelta(minutes=3)).isoformat(), run_id),
+ )
+
+ for week, note in enumerate((
+ "Two rule updates summarised; 3 facts saved.",
+ "No changes this week.",
+ "Grant round opened; 2 facts saved.",
+ "One consultation found; 1 fact saved.",
+ "No changes this week.",
+ "Connection FAQ updated; 2 facts saved.",
+ "Credit rules unchanged; 1 fact saved.",
+ "Three updates summarised; 4 facts saved.",
+ )):
+ run(digest, DIGEST_THREAD, "Weekly policy digest", 3, 1.1 + 7 * week, note)
+ for week in range(5):
+ run(questions, f"landing-questions-{week}", "Member questions round-up", 2, 4.7 + 7 * week,
+ "Five questions answered in draft.")
+ for day in range(9):
+ run(survey, f"landing-survey-{day}", "Survey photo sorter", 2, 0.7 + day, "Photos filed by roof.")
+ with _tasks_db() as conn:
+ for task_id, days in ((digest, 1.1), (questions, 4.7), (survey, 0.7)):
+ conn.execute("UPDATE tasks SET last_run = ? WHERE id = ?",
+ ((anchor - timedelta(days=days)).isoformat(), task_id))
+
+
+# Knowledge: (key, type, subject, description)
+_ENTITIES = [
+ ("coop", "organisation", CO_OP, "A neighbourhood energy co-op planning its first shared solar array."),
+ ("tenants", "organisation", "Mill Street Tenants Association", "Helps renters join without a roof of their own."),
+ ("library", "organisation", "Riverbend Library", "Offers its south-facing roof and a meeting room."),
+ ("school", "organisation", "Northside Primary School", "Its gym roof is shortlisted for a second array."),
+ ("credit", "organisation", "Riverbend Credit Union", "Discussing low-interest loans for member shares."),
+ ("grid", "organisation", "Regional grid operator", "Approves the connection and sets the credit rules."),
+ ("city", "organisation", "City sustainability office", "Runs the small-grants round."),
+ ("riverbend", "place", "Riverbend", "The neighbourhood the co-op serves."),
+ ("libroof", "place", "Library roof", "South-facing, with room for the first array."),
+ ("gymroof", "place", "Northside gym roof", "Large and flat; needs a structural check."),
+ ("millst", "place", "Mill Street", "Terraced housing, mostly rented."),
+ ("market", "place", "Market Hall", "Venue for the open evening."),
+ ("depot", "place", "Old tram depot", "The largest roof in the neighbourhood."),
+ ("chair", "person", "Board chair", "Chairs the monthly board meeting."),
+ ("treasurer", "person", "Treasurer", "Keeps the share register and the savings model."),
+ ("volco", "person", "Volunteer coordinator", "Runs training night and the survey rota."),
+ ("surveyor", "person", "Site surveyor", "Checks each roof's direction, shade and structure."),
+ ("liaison", "person", "Member liaison", "Answers member questions."),
+ ("openeve", "event", "Open evening", "The launch event for founding members."),
+ ("vote", "event", "Board vote on the first array", "Decides between the library and the depot."),
+ ("surveyday", "event", "Roof survey day", "Volunteers and the surveyor visit each shortlisted roof."),
+ ("deadline", "event", "Grant deadline", "The small-grants round closes at the end of the month."),
+ ("training", "event", "Volunteer training night", "Covers the open evening, surveys and member questions."),
+ ("launchday", "event", "Launch day", "The launch page goes live and the member drive starts."),
+ ("launchpage", "project", "Launch page", "The public page inviting neighbours to join."),
+ ("drive", "project", "Member drive", "Aims for 120 founding member households."),
+ ("survey", "project", "Rooftop survey", "Shortlists roofs for the first two arrays."),
+ ("grant", "project", "Grant application", "Asks the city to cover survey and legal costs."),
+ ("battery", "project", "Battery pilot", "A small storage trial at the library."),
+ ("digest", "project", "Weekly policy digest", "A local workflow that tracks rule changes."),
+ ("boardpack", "project", "Board pack", "Papers for the board vote."),
+ ("cs", "concept", "Community solar", "Many households share the output of one array."),
+ ("shared", "concept", "Shared ownership", "Members own the array together."),
+ ("credits", "concept", "Bill credits", "A share's output appears as a credit on the member's bill."),
+ ("subscription", "concept", "Subscription model", "Members pay monthly instead of buying a share."),
+ ("shares", "concept", "Member shares", "A one-off purchase that pays back over time."),
+ ("netmeter", "concept", "Net metering", "Exported power offsets power bought from the grid."),
+ ("vnm", "concept", "Virtual net metering", "Credits from one array spread across many bills."),
+ ("fit", "concept", "Feed-in tariff", "A set price paid for exported power."),
+ ("ppa", "concept", "Power purchase agreement", "A long-term contract to buy the array's output."),
+ ("interconnect", "concept", "Grid connection", "Permission and equipment to connect the array."),
+ ("capacity", "concept", "Capacity factor", "How much of its rated output an array delivers."),
+ ("payback", "concept", "Payback period", "How long a share takes to repay its cost."),
+ ("energycoop", "concept", "Energy cooperative", "A member-owned energy organisation."),
+ ("control", "concept", "Democratic member control", "Members decide together."),
+ ("onevote", "concept", "One member, one vote", "Every member has an equal say."),
+ ("renters", "concept", "Renter access", "People without a roof can still take part."),
+ ("equity", "concept", "Energy equity", "Fair access to clean, affordable power."),
+ ("carveout", "concept", "Low-income places", "Shares set aside for households on low incomes."),
+ ("storage", "concept", "Battery storage", "Keeps daytime power for the evening."),
+ ("peak", "concept", "Peak shaving", "Using storage to cut demand at the busiest times."),
+ ("inverter", "concept", "Inverter", "Turns the panels' direct current into mains power."),
+ ("suitability", "concept", "Roof suitability", "Direction, shade, strength and access."),
+ ("gridcap", "concept", "Grid capacity", "How much new generation the local network can take."),
+ ("benefit", "concept", "Community benefit fund", "Surplus income set aside for local projects."),
+ ("tariff", "concept", "Green tariff", "A supplier's renewable electricity offer."),
+ ("localfirst", "concept", "Local-first research", "Research runs on this computer and keeps its sources."),
+ ("evidence", "concept", "Visible evidence", "Every claim keeps a link to its source."),
+ ("f_south", "fact", "Library roof faces south", "Good for a first array."),
+ ("f_roofs", "fact", "Survey found 14 suitable roofs", "Out of 31 roofs checked."),
+ ("f_target", "fact", "Founding target: 120 households", "Set by the board in spring."),
+ ("f_instal", "fact", "Shares can be paid in instalments", "Agreed with the credit union."),
+ ("f_monthly", "fact", "Credits appear on the monthly bill", "Confirmed in the grid connection FAQ."),
+ ("f_noupfront", "fact", "Subscriptions need no upfront cost", "From the public sources."),
+ ("f_renters", "fact", "Renters can join without a roof", "From the public sources."),
+ ("f_months", "fact", "Connection approval takes months", "Start the application early."),
+ ("f_tuesday", "fact", "Board meets on the first Tuesday", "Monthly, at the library."),
+ ("f_thursday", "fact", "Open evening is on a Thursday", "At Market Hall, from 18:30."),
+ ("f_vote", "fact", "Each member gets one vote", "Whatever the size of their share."),
+ ("f_surplus", "fact", "Surplus goes to the community fund", "Agreed in the draft rules."),
+ ("f_gym", "fact", "Gym roof needs a structural check", "Booked for survey day."),
+ ("f_depot", "fact", "Depot roof is the largest site", "About three times the library roof."),
+ ("f_grant", "fact", "Grant covers survey costs", "And part of the legal fees."),
+ ("f_loans", "fact", "Credit union offers share loans", "Repaid over two years."),
+ ("f_rules", "fact", "Credit rules unchanged this month", "From the latest policy digest."),
+ ("f_faq", "fact", "Two new connection FAQ questions", "From the latest policy digest."),
+ ("m_cs", "media", "Community solar article", "Public source read during research."),
+ ("m_nm", "media", "Net metering article", "Public source read during research."),
+ ("m_coop", "media", "Cooperative article", "Public source read during research."),
+ ("m_guide", "media", "Co-op starter guide", "A public guide to setting up an energy co-op."),
+ ("m_faq", "media", "Grid connection FAQ", "The grid operator's public questions and answers."),
+ ("m_member", "media", "Member survey results", "What 86 neighbours said about joining."),
+ ("m_photos", "media", "Roof survey photos", "Photos from survey day, filed by roof."),
+ ("m_design", "media", "Launch page design", "The editable design in the Design panel."),
+ ("m_deck", "media", "Board deck: first array", "Slides for the board vote."),
+ ("m_flyer", "media", "Open evening flyer", "Printed for the library and Market Hall."),
+ ("m_d38", "media", "Policy digest: week 38", "Weekly digest from the local workflow."),
+ ("m_d39", "media", "Policy digest: week 39", "Weekly digest from the local workflow."),
+ ("m_d40", "media", "Policy digest: week 40", "Weekly digest from the local workflow."),
+ ("p_plain", "preference", "Prefers plain-language summaries", "Short sentences, no jargon."),
+ ("p_delivery", "preference", "Keep delivery off for drafts", "Results stay in the app until reviewed."),
+ ("p_metric", "preference", "Uses metric units", "Square metres and kilowatt-hours."),
+ ("p_cite", "preference", "Cite sources inline", "Every claim names where it came from."),
+ ("p_monday", "preference", "Digest on Monday mornings", "Ready before the week starts."),
+ ("p_tone", "preference", "Warm, neighbourly tone", "For anything members read."),
+ ("s_roof", "skill", "Roof suitability checks", "Direction, shade and access."),
+ ("s_writing", "skill", "Plain-language writing", "Turns rules into everyday words."),
+ ("s_model", "skill", "Savings modelling", "Share price, credits and payback."),
+ ("s_grant", "skill", "Grant writing", "Structure, budget and benefit."),
+ ("s_events", "skill", "Event planning", "Venues, rotas and flyers."),
+ ("k_local", "self_knowledge", f"Research runs on {LOCAL_MODELS[0]}", "On this computer, with sources kept."),
+ ("k_hosted", "self_knowledge", "Hosted models only when chosen", "A frontier model joins only when picked."),
+ ("k_drafts", "self_knowledge", "Drafts stay on this computer", "Nothing is sent without approval."),
+]
+
+_RELATIONS = [
+ ("coop", "riverbend", "based_in"), ("coop", "cs", "uses"), ("coop", "energycoop", "builds_on"),
+ ("tenants", "coop", "member_of"), ("library", "coop", "member_of"), ("school", "coop", "participates_in"),
+ ("credit", "coop", "participates_in"), ("coop", "grid", "uses"), ("city", "grant", "participates_in"),
+ ("libroof", "riverbend", "located_in"), ("gymroof", "riverbend", "located_in"),
+ ("millst", "riverbend", "located_in"), ("market", "riverbend", "located_in"),
+ ("depot", "riverbend", "located_in"), ("library", "libroof", "owns"), ("school", "gymroof", "owns"),
+ ("tenants", "millst", "based_in"), ("chair", "coop", "leads"), ("treasurer", "coop", "member_of"),
+ ("volco", "coop", "member_of"), ("surveyor", "survey", "works_on"), ("liaison", "drive", "works_on"),
+ ("treasurer", "s_model", "has_skill"), ("surveyor", "s_roof", "has_skill"), ("volco", "s_events", "has_skill"),
+ ("liaison", "s_writing", "has_skill"), ("chair", "vote", "leads"), ("treasurer", "grant", "works_on"),
+ ("volco", "training", "leads"), ("openeve", "market", "located_in"), ("openeve", "drive", "part_of"),
+ ("vote", "boardpack", "uses"), ("surveyday", "survey", "part_of"), ("deadline", "grant", "deadline_for"),
+ ("training", "volco", "created_by"), ("launchday", "launchpage", "scheduled_for"),
+ ("launchday", "drive", "part_of"), ("launchpage", "coop", "part_of"), ("launchpage", "m_design", "uses"),
+ ("launchpage", "drive", "part_of"), ("launchpage", "p_tone", "uses"), ("launchpage", "credits", "cites"),
+ ("launchpage", "shares", "cites"), ("drive", "coop", "part_of"), ("drive", "f_target", "uses"),
+ ("survey", "coop", "part_of"), ("survey", "suitability", "uses"), ("survey", "m_photos", "uses"),
+ ("grant", "coop", "part_of"), ("grant", "f_grant", "cites"), ("grant", "s_grant", "uses"),
+ ("battery", "libroof", "located_in"), ("battery", "storage", "uses"), ("battery", "peak", "uses"),
+ ("digest", "coop", "part_of"), ("digest", "k_local", "uses"), ("digest", "p_delivery", "uses"),
+ ("digest", "p_monday", "uses"), ("boardpack", "m_deck", "uses"), ("boardpack", "vote", "part_of"),
+ ("shared", "cs", "part_of"), ("credits", "cs", "part_of"), ("subscription", "cs", "part_of"),
+ ("shares", "cs", "part_of"), ("vnm", "netmeter", "extends"), ("credits", "vnm", "uses"),
+ ("netmeter", "credits", "builds_on"), ("fit", "netmeter", "contradicts"), ("ppa", "cs", "part_of"),
+ ("interconnect", "grid", "uses"), ("interconnect", "gridcap", "uses"), ("capacity", "payback", "builds_on"),
+ ("payback", "shares", "part_of"), ("energycoop", "control", "uses"), ("control", "onevote", "builds_on"),
+ ("renters", "cs", "part_of"), ("equity", "renters", "builds_on"), ("carveout", "equity", "part_of"),
+ ("storage", "peak", "uses"), ("inverter", "cs", "part_of"), ("suitability", "libroof", "cites"),
+ ("benefit", "energycoop", "part_of"), ("tariff", "cs", "contradicts"), ("localfirst", "evidence", "uses"),
+ ("localfirst", "k_local", "uses"), ("evidence", "p_cite", "builds_on"), ("shared", "energycoop", "builds_on"),
+ ("subscription", "shares", "contradicts"), ("credits", "f_monthly", "cites"),
+ ("f_south", "libroof", "part_of"), ("f_roofs", "survey", "part_of"), ("f_target", "drive", "part_of"),
+ ("f_instal", "shares", "part_of"), ("f_instal", "credit", "cites"), ("f_monthly", "m_faq", "cites"),
+ ("f_noupfront", "subscription", "part_of"), ("f_noupfront", "m_cs", "cites"),
+ ("f_renters", "renters", "part_of"), ("f_renters", "m_cs", "cites"), ("f_months", "interconnect", "part_of"),
+ ("f_months", "m_faq", "cites"), ("f_tuesday", "chair", "part_of"), ("f_thursday", "openeve", "part_of"),
+ ("f_vote", "onevote", "part_of"), ("f_vote", "m_coop", "cites"), ("f_surplus", "benefit", "part_of"),
+ ("f_gym", "gymroof", "part_of"), ("f_gym", "surveyday", "scheduled_for"), ("f_depot", "depot", "part_of"),
+ ("f_grant", "city", "cites"), ("f_loans", "credit", "part_of"), ("f_loans", "shares", "part_of"),
+ ("f_rules", "m_d40", "extracted_from"), ("f_rules", "credits", "part_of"),
+ ("f_faq", "m_d40", "extracted_from"), ("f_faq", "m_faq", "extends"), ("m_cs", "cs", "cites"),
+ ("m_nm", "netmeter", "cites"), ("m_coop", "energycoop", "cites"), ("m_guide", "energycoop", "cites"),
+ ("m_guide", "control", "cites"), ("m_faq", "grid", "created_by"), ("m_member", "drive", "part_of"),
+ ("m_member", "renters", "cites"), ("m_photos", "surveyday", "part_of"), ("m_design", "launchpage", "part_of"),
+ ("m_deck", "vote", "part_of"), ("m_flyer", "openeve", "part_of"), ("m_d38", "digest", "created_by"),
+ ("m_d39", "digest", "created_by"), ("m_d40", "digest", "created_by"), ("m_d39", "m_d38", "extends"),
+ ("m_d40", "m_d39", "extends"), ("m_d38", "f_months", "cites"), ("p_plain", "s_writing", "uses"),
+ ("p_cite", "evidence", "uses"), ("p_metric", "s_model", "uses"), ("p_tone", "m_flyer", "uses"),
+ ("p_plain", "launchpage", "uses"), ("s_model", "payback", "uses"), ("s_model", "credits", "uses"),
+ ("s_roof", "suitability", "uses"), ("s_grant", "grant", "uses"), ("s_events", "openeve", "uses"),
+ ("s_writing", "m_flyer", "uses"), ("k_local", "localfirst", "part_of"), ("k_hosted", "launchpage", "uses"),
+ ("k_drafts", "p_delivery", "uses"), ("depot", "vote", "participates_in"), ("libroof", "vote", "participates_in"),
+ ("gymroof", "survey", "part_of"), ("depot", "survey", "part_of"), ("libroof", "survey", "part_of"),
+ ("millst", "renters", "uses"), ("tenants", "renters", "interested_in"), ("tenants", "carveout", "interested_in"),
+ ("school", "battery", "interested_in"), ("library", "battery", "participates_in"),
+ ("grid", "interconnect", "owns"), ("grid", "netmeter", "uses"), ("city", "equity", "interested_in"),
+ ("coop", "benefit", "owns"), ("coop", "control", "uses"), ("coop", "shares", "uses"),
+ ("coop", "subscription", "uses"), ("coop", "localfirst", "uses"), ("treasurer", "payback", "studies"),
+ ("chair", "m_deck", "authored"), ("liaison", "m_member", "authored"), ("volco", "m_flyer", "authored"),
+]
+
+
+def _entity_id(key: str) -> str:
+ # Fixed ids give the graph the same layout on every capture; this seed
+ # spreads the co-op's labelled neighbours without overlapping labels.
+ return hashlib.sha256(f"v28-{key}".encode()).hexdigest()[:12]
+
+
+def _seed_knowledge(anchor: datetime) -> None:
+ from row_bot import knowledge_graph as kg
+
+ previous = kg._skip_reindex
+ kg._skip_reindex = True
+ try:
+ for key, entity_type, subject, description in _ENTITIES:
+ kg.save_entity(entity_type, subject, description, tags="riverbend,demo",
+ properties={"provenance": "landing demo fixture"}, source="landing-demo",
+ entity_id=_entity_id(key))
+ for source, target, relation in _RELATIONS:
+ # Fixed relation ids too: the graph reads its edges in id order, and
+ # that order steers the layout.
+ kg.add_relation(_entity_id(source), _entity_id(target), relation, confidence=0.95,
+ properties={"provenance": "landing demo fixture"}, source="landing-demo",
+ relation_id=_entity_id(f"{source}>{target}"))
+ finally:
+ kg._skip_reindex = previous
+ moment = (anchor - timedelta(minutes=26)).isoformat()
+ with sqlite3.connect(Path(os.environ["ROW_BOT_DATA_DIR"]) / "memory.db") as conn:
+ conn.execute("UPDATE entities SET created_at = ?, updated_at = ?", (moment, moment))
+ conn.execute("UPDATE relations SET created_at = ?, updated_at = ?", (moment, moment))
+
+
+def _launch_page_html() -> str:
+ """The fictional co-op's launch page, as the Design panel stores it."""
+ ink, cream, sun, muted = "#1F2A1C", "#FFF8EC", "#F59E0B", "#5B6656"
+ eyebrow = "margin:0 0 14px;font-size:20px;font-weight:700;letter-spacing:4px;color:#B45309"
+ steps = "".join(
+ "
"
+ f"
{number}
"
+ f"
{title}
"
+ f"
{detail}
"
+ for number, title, detail, color in (
+ ("01", "Join for a share", "Subscribe monthly or buy a share outright. Renters welcome.", "#D97706"),
+ ("02", "We build together", "One array on the library roof, owned by its members.", "#16A34A"),
+ ("03", "Credits on your bill", "Your share's sunshine appears on every monthly bill.", "#0E7490"),
+ )
+ )
+ stats = "".join(
+ f"
{value}
"
+ f"
{label}
"
+ for value, label in (("120", "founding households"), ("14", "suitable roofs"), ("1", "vote per member"))
+ )
+ benefits = "".join(
+ f"
"
+ f"
{title}
"
+ f"
{detail}
"
+ for title, detail, bg in (
+ ("Lower bills", "Credits from your share arrive on every bill, all year.", "#FFEFC7"),
+ ("A local fund", "Surplus goes to projects the members choose together.", "#E3F1DE"),
+ ("A real say", "Every member votes on what the co-op builds next.", "#DCEEF3"),
+ ("No roof needed", "Renters and flat owners join on the same terms.", "#F7E3D6"),
+ )
+ )
+ questions = "".join(
+ f"
"
+ f"
{q}
"
+ f"
{a}
"
+ for q, a in (
+ ("Do I need my own roof?", "No. Your share is part of the array on the library roof."),
+ ("Can I pay over time?", "Yes. Shares can be paid in instalments with the credit union."),
+ ("What happens to the surplus?", "It goes to the community fund, and members decide how it is spent."),
+ )
+ )
+ return (
+ f""
+ ""
+ ""
+ f"
COMMUNITY-OWNED ENERGY
"
+ "
"
+ "Power your street, together.
"
+ "
"
+ "Riverbend Community Solar puts one shared array on the library roof. Every member owns a piece, "
+ "and every share earns credits on the bill.
"
+ f"
Become a founding member"
+ "Come to the open evening
"
+ "
"
+ "
"
+ ""
+ f"
HOW SHARES WORK
"
+ "
"
+ "Three steps to clean, local power
"
+ f"
{steps}
"
+ f""
+ "
Owned by neighbours, "
+ "run by members
One member, one "
+ "vote. Surplus goes to the community fund for local projects.
"
+ f"
{stats}
"
+ ""
+ f"
WHAT MEMBERS GET BACK
"
+ "
More than cheaper power
"
+ f"
{benefits}
"
+ ""
+ f"
QUESTIONS
Asked at every street meeting
"
+ f"
{questions}
"
+ f""
+ "
Open evening, Thursday 18:30
"
+ "
Market Hall, Riverbend. Bring your questions and your "
+ f"neighbours.
Save my place"
+ f""
+ ""
+ )
+
+
+def _seed_ship_goal(anchor: datetime) -> None:
+ """The Ship conversation works toward a goal, shown in its details beside the approval."""
+ from row_bot.goals import start_goal
+
+ goal = start_goal(SHIP_THREAD, "Share the reviewed launch page with the co-op board", max_turns=6)
+ progress = ["Checked the page against the research", "Exported riverbend-launch-page.pdf"]
+ started = (anchor - timedelta(minutes=4)).isoformat()
+ with _tasks_db() as conn:
+ conn.execute(
+ "UPDATE thread_goals SET turns_used = ?, last_progress = ?, evidence_json = ?, blockers_json = ?, "
+ "last_reason = ?, created_at = ?, updated_at = ?, window_started_at = ? WHERE id = ?",
+ (2, progress[-1], json.dumps(progress), json.dumps(["Waiting for approval to send the email"]),
+ "The email waits for your approval.", started, started, started, goal["id"]),
+ )
+
+
+def _seed_design() -> None:
+ from row_bot.conversation_resources import bind, list_bindings
+ from row_bot.designer.state import BrandConfig, DesignerPage, DesignerProject, ProjectBrief
+ from row_bot.designer.storage import save_project
+
+ save_project(DesignerProject(
+ id=DESIGN_ID,
+ name="Riverbend launch page",
+ mode="landing",
+ aspect_ratio="landing",
+ pages=[DesignerPage(title="Home", route_id="home", kind="screen", html=_launch_page_html(),
+ notes="Hero, how shares work, ownership, member benefits and the open evening.")],
+ brand=BrandConfig(primary_color="#1F2A1C", accent_color="#F59E0B"),
+ brief=ProjectBrief(output_type="Landing page", audience="Riverbend neighbours",
+ tone="Warm and neighbourly", length="One page"),
+ thread_id=CREATE_THREAD,
+ ))
+ snapshot = list_bindings(CREATE_THREAD)
+ bind(CREATE_THREAD, "artifact", DESIGN_ID, expected_revision=snapshot.revision, role="primary")
+
+
+def _seed_model_catalog() -> None:
+ """The picker's local models, read from the display-only runtime, as Quick Choices."""
+ from row_bot.providers.model_catalog_cache import refresh_model_catalog_cache
+ from row_bot.providers.selection import add_quick_choice_for_model
+
+ refresh_model_catalog_cache(reason="landing_demo", force=True, provider_id="ollama")
+ for model_id in LOCAL_MODELS:
+ add_quick_choice_for_model(model_id, provider_id="ollama")
+
+
+def _bind_profile(data_dir: Path, ollama_host: str = "") -> None:
+ os.environ["ROW_BOT_DATA_DIR"] = str(data_dir)
+ if ollama_host:
+ os.environ["OLLAMA_HOST"] = ollama_host
+ os.environ.setdefault("ROW_BOT_DOCS_CAPTURE", "1")
+ src = str(ROOT / "src")
+ if src not in sys.path:
+ sys.path.insert(0, src)
+
+
+def seed_profile(data_dir: Path, ollama_host: str, anchor: datetime) -> None:
+ """Seed the fictional co-op into the capture profile (in this process)."""
+ _bind_profile(data_dir, ollama_host)
+ from scripts.docs import seed_real_app_demo_data as demo
+ from row_bot.docs_capture import scan_demo_data_safety
+
+ demo._seed_app_config(data_dir, first_run=False)
+ demo._write_json(data_dir / "model_settings.json", {"model": LOCAL_MODEL})
+ _seed_conversations(anchor)
+ _seed_workflows(anchor)
+ _seed_knowledge(anchor)
+ _seed_design()
+ _seed_ship_goal(anchor)
+ _seed_model_catalog()
+ errors = scan_demo_data_safety(data_dir)
+ if errors:
+ raise SystemExit("\n".join(errors))
+
+
+def add_ship_approval(data_dir: Path, anchor: datetime) -> None:
+ """The email waits for Approve or Deny in its conversation."""
+ _bind_profile(data_dir)
+ from row_bot.tasks import create_approval_request
+
+ interrupt = {
+ "tool": "send_gmail_message",
+ "label": "Send email",
+ "description": "Email riverbend-launch-page.pdf to riverbend-board@example.com",
+ "args": {
+ "to": ["riverbend-board@example.com"],
+ "subject": "Launch page for review",
+ "attachments": ["riverbend-launch-page.pdf"],
+ },
+ "tool_call_id": "s2",
+ "risk_class": "medium",
+ "scope": "One email to riverbend-board@example.com, with the PDF attached.",
+ }
+ _token, approval_id = create_approval_request(
+ "landing-ship-pass", "", "conversation", interrupt["description"],
+ resume_kind="conversation", source_thread_id=SHIP_THREAD, parent_thread_id=SHIP_THREAD,
+ approval_payload_json={
+ "interrupt": interrupt,
+ "interrupt_ids": [],
+ "pass_id": "landing-ship-pass",
+ "model_selection": {"provider_id": "ollama", "model_ref": LOCAL_MODEL},
+ },
+ )
+ with _tasks_db() as conn:
+ conn.execute("UPDATE approval_requests SET id = ?, requested_at = ? WHERE id = ?",
+ (SHIP_APPROVAL_ID, (anchor - timedelta(minutes=1)).isoformat(), approval_id))
+
+
+# The app and the browser --------------------------------------------------------
+
+class _LandingOllamaHandler(_DemoOllamaHandler):
+ """The docs capture's display-only local runtime, listing the landing footage's models."""
+
+ def do_GET(self) -> None: # noqa: N802 - http.server naming
+ if self.path.startswith("/api/tags"):
+ self._reply(200, {"models": [self._model(name) for name in LOCAL_MODELS]})
+ else:
+ super().do_GET()
+
+ def do_POST(self) -> None: # noqa: N802 - http.server naming
+ length = int(self.headers.get("Content-Length") or 0)
+ try:
+ request = json.loads(self.rfile.read(length) or b"{}")
+ except ValueError:
+ request = {}
+ name = str(request.get("model") or request.get("name") or "")
+ if self.path.startswith("/api/show") and name in LOCAL_MODELS:
+ details = self._model(name)["details"]
+ family = details["family"]
+ self._reply(200, {
+ "details": details,
+ "model_info": {"general.architecture": family, f"{family}.context_length": 32768},
+ "capabilities": ["completion", "tools"],
+ })
+ else:
+ self._reply(503, {"error": "the landing capture never runs a model"})
+
+
+def _start_demo_ollama(stack: ExitStack) -> str:
+ """Serve the display-only local runtime on loopback for the app's lifetime."""
+ server = ThreadingHTTPServer(("127.0.0.1", 0), _LandingOllamaHandler)
+ threading.Thread(target=server.serve_forever, name="landing-demo-ollama", daemon=True).start()
+ stack.callback(server.server_close)
+ stack.callback(server.shutdown)
+ return f"http://127.0.0.1:{server.server_address[1]}"
+
+
+class App:
+ """The real app on the isolated profile; stopped when the stack closes."""
+
+ def __init__(self, stack: ExitStack, data_dir: Path, ollama_host: str) -> None:
+ self.port = _free_port()
+ self.proc, secret = _launch_app(self.port, data_dir, stack, ollama_host=ollama_host)
+ stack.callback(self.stop)
+ _wait_ping(self.port, self.proc, 120, launcher_secret=secret)
+ self.base = f"http://127.0.0.1:{self.port}/app-v2"
+
+ def stop(self) -> None:
+ if self.proc.poll() is None:
+ self.proc.terminate()
+ try:
+ self.proc.wait(timeout=15)
+ except subprocess.TimeoutExpired:
+ self.proc.kill()
+ self.proc.wait(timeout=15)
+
+
+def _helper(*args: str) -> None:
+ subprocess.run([sys.executable, str(Path(__file__).resolve()), *args], cwd=str(ROOT), check=True)
+
+
+_HOSTED_CHOICE = {
+ "provider_id": "codex",
+ "model_ref": HOSTED_MODEL,
+ "label": "GPT-5.6 Sol - ChatGPT / Codex",
+ "available": True,
+ "unavailable_reason": None,
+ "billing": "subscription",
+}
+
+
+def _hosted_model_is_display_only(page) -> None:
+ """Show GPT-5.6 Sol as a connected choice; nothing is connected or called."""
+
+ def ready(workspace: dict) -> None:
+ selection = (workspace.get("controls") or {}).get("model_selection") or {}
+ if selection.get("provider_id") == "codex":
+ workspace["model_status"] = {"state": "ready", "reason": "", "fix": None,
+ "local": False, "sees_images": True}
+ for action in workspace.get("actions") or []:
+ action.update(ready=True, code=None)
+
+ def handshake(route) -> None:
+ response = route.fetch()
+ body = response.json()
+ body["models"] = [*body.get("models", []), _HOSTED_CHOICE]
+ route.fulfill(response=response, json=body)
+
+ def opened(route) -> None:
+ response = route.fetch()
+ body = response.json()
+ ready(body.get("workspace") or {})
+ route.fulfill(response=response, json=body)
+
+ def workspace(route) -> None:
+ response = route.fetch()
+ body = response.json()
+ ready(body)
+ route.fulfill(response=response, json=body)
+
+ page.route("**/api/v1/handshake", handshake)
+ page.route("**/api/v1/conversations/*/open", opened)
+ page.route("**/api/v1/conversations/*/workspace", workspace)
+
+
+def _approval_is_answered_here(page) -> None:
+ """Answer Approve in the harness: the app shows its submitted state and nothing is sent."""
+
+ def answer(route) -> None:
+ command = json.loads(route.request.post_data or "{}")
+ if command.get("type") != "approval.resolve":
+ route.abort()
+ return
+ route.fulfill(status=200, json={"command_id": command.get("command_id", ""), "status": "accepted",
+ "approval_id": SHIP_APPROVAL_ID})
+
+ page.route("**/api/v1/approvals/*/commands", answer)
+
+
+_POINTER = """
+(() => {
+ if (window.top !== window) return;
+ const install = () => {
+ if (document.getElementById('landing-pointer')) return;
+ const pointer = document.createElement('div');
+ pointer.id = 'landing-pointer';
+ pointer.innerHTML = '';
+ pointer.style.cssText = 'position:fixed;left:0;top:0;z-index:2147483647;pointer-events:none;opacity:0;'
+ + 'transition:opacity .25s ease;filter:drop-shadow(0 2px 3px rgba(0,0,0,.45));will-change:transform';
+ const ring = pointer.querySelector('span');
+ ring.style.cssText = 'position:absolute;left:-12px;top:-12px;width:28px;height:28px;border-radius:50%;'
+ + 'border:2px solid rgba(255,255,255,.85);opacity:0;transform:scale(.4)';
+ document.documentElement.appendChild(pointer);
+ window.__pointer = (x, y, show) => {
+ pointer.style.transform = `translate(${x - 2}px, ${y - 2}px)`;
+ if (show != null) pointer.style.opacity = show ? '1' : '0';
+ };
+ window.__press = () => ring.animate(
+ [{opacity: .9, transform: 'scale(.4)'}, {opacity: 0, transform: 'scale(1.25)'}],
+ {duration: 420, easing: 'ease-out'});
+ };
+ if (document.documentElement) install();
+ document.addEventListener('DOMContentLoaded', install);
+})();
+"""
+
+
+def _ease(t: float) -> float:
+ return 4 * t ** 3 if t < 0.5 else 1 - (-2 * t + 2) ** 3 / 2
+
+
+class Director:
+ """Human-paced pointer moves, clicks and scrolls, with the drawn pointer."""
+
+ def __init__(self, page, start: tuple[float, float] = (WIDTH * 0.72, HEIGHT * 0.62)) -> None:
+ self.page = page
+ self.x, self.y = start
+
+ def place(self, x: float, y: float, show: bool | None = None) -> None:
+ self.x, self.y = x, y
+ self.page.mouse.move(x, y)
+ self.page.evaluate("([x, y, show]) => window.__pointer && window.__pointer(x, y, show)",
+ [x, y, show])
+
+ def show(self) -> None:
+ self.place(self.x, self.y, True)
+
+ def hide(self) -> None:
+ self.place(self.x, self.y, False)
+
+ def move(self, x: float, y: float, ms: int = 800) -> None:
+ start = time.monotonic()
+ x0, y0 = self.x, self.y
+ while True:
+ t = min(1.0, (time.monotonic() - start) * 1000 / ms)
+ k = _ease(t)
+ self.place(x0 + (x - x0) * k, y0 + (y - y0) * k)
+ if t >= 1:
+ return
+ self.page.wait_for_timeout(12)
+
+ def to(self, locator, ms: int = 800, dx: float = 0, dy: float = 0) -> None:
+ box = locator.bounding_box()
+ if not box:
+ raise RuntimeError(f"not on screen: {locator}")
+ self.move(box["x"] + box["width"] / 2 + dx, box["y"] + box["height"] / 2 + dy, ms)
+
+ def click(self, locator=None, ms: int = 800, settle: int = 120) -> None:
+ if locator is not None:
+ self.to(locator, ms)
+ self.page.wait_for_timeout(settle)
+ self.page.evaluate("() => window.__press && window.__press()")
+ self.page.mouse.down()
+ self.page.wait_for_timeout(70)
+ self.page.mouse.up()
+
+ def scroll(self, dy: float, ms: int = 600) -> None:
+ steps = max(1, ms // 16)
+ for _ in range(steps):
+ self.page.mouse.wheel(0, dy / steps)
+ self.page.wait_for_timeout(16)
+
+ def wait(self, ms: int) -> None:
+ self.page.wait_for_timeout(ms)
+
+
+class Recorder:
+ """The page's screencast, saved frame by frame with its timestamps."""
+
+ def __init__(self, context, page, folder: Path) -> None:
+ self.folder = folder
+ self.frames: list[tuple[float, Path]] = []
+ self.marks: dict[str, float] = {}
+ self.cdp = context.new_cdp_session(page)
+ self.cdp.on("Page.screencastFrame", self._frame)
+ self.recording = False
+
+ def _frame(self, params: dict) -> None:
+ if self.recording:
+ path = self.folder / f"{len(self.frames):05d}.png"
+ path.write_bytes(base64.b64decode(params["data"]))
+ self.frames.append((float(params["metadata"]["timestamp"]), path))
+ try:
+ self.cdp.send("Page.screencastFrameAck", {"sessionId": params["sessionId"]})
+ except Exception: # noqa: BLE001 - the page may already be closing
+ pass
+
+ def start(self, page) -> None:
+ self.recording = True
+ self.cdp.send("Page.startScreencast", {"format": "png", "everyNthFrame": 1})
+ deadline = time.monotonic() + 5
+ while not self.frames and time.monotonic() < deadline:
+ page.wait_for_timeout(20)
+ # Repaint once so the first frame is current.
+ page.evaluate("() => new Promise(r => requestAnimationFrame(() => requestAnimationFrame(r)))")
+
+ def mark(self, name: str) -> None:
+ self.marks[name] = time.time()
+
+ def stop(self, page) -> None:
+ page.wait_for_timeout(200)
+ self.cdp.send("Page.stopScreencast")
+ self.recording = False
+
+ def sample(self, start: float, end: float) -> list[Path]:
+ """Frames at 30 fps between two marks, each the newest shown at its moment."""
+ frames = sorted(self.frames)
+ picked: list[Path] = []
+ index = 0
+ count = round((end - start) * FPS)
+ for n in range(count):
+ moment = start + n / FPS
+ while index + 1 < len(frames) and frames[index + 1][0] <= moment:
+ index += 1
+ picked.append(frames[index][1])
+ return picked
+
+
+def _new_context(browser, *, scale: int, clock: Callable[[], datetime], pointer: bool):
+ context = browser.new_context(
+ viewport={"width": WIDTH, "height": HEIGHT},
+ device_scale_factor=scale,
+ color_scheme="dark",
+ reduced_motion="no-preference",
+ locale="en-GB",
+ timezone_id="Europe/London",
+ )
+ # The approval counts as announced (no floating notice over the scene), and
+ # the sidebar's agents row is folded so the conversation list ends tidily.
+ context.add_init_script(
+ "try { localStorage.setItem('row-bot.approvals-announced.v1', '[\"%s\"]');"
+ " localStorage.setItem('row-bot.sidebar-agents.v1', 'collapsed') } catch (e) {}"
+ % SHIP_APPROVAL_ID
+ )
+ if pointer:
+ context.add_init_script(_POINTER)
+ page = context.new_page()
+ page.clock.set_system_time(clock())
+ _hosted_model_is_display_only(page)
+ _approval_is_answered_here(page)
+ return context, page
+
+
+def _open(page, base: str, route: str, text: str) -> None:
+ page.goto(base + route, wait_until="load", timeout=60_000)
+ page.get_by_text(text, exact=False).first.wait_for(timeout=30_000)
+ page.wait_for_timeout(800)
+
+
+def _close_details(page) -> None:
+ """Close the conversation's details rail (its goal card) if it is open."""
+ toggle = page.locator('button[aria-label="Conversation details"]').first
+ if toggle.get_attribute("aria-pressed") == "true":
+ toggle.click()
+ page.wait_for_timeout(400)
+ _blur(page)
+
+
+def _blur(page) -> None:
+ page.evaluate("() => document.activeElement && document.activeElement.blur && document.activeElement.blur()")
+
+
+def _park(page) -> None:
+ page.mouse.move(WIDTH - 2, HEIGHT - 2)
+
+
+# Scenes ---------------------------------------------------------------------------
+
+def _graph_hub(page, scale: int) -> tuple[float, float]:
+ """Where the co-op's memory sits: the largest organisation-coloured node."""
+ from PIL import Image
+
+ image = Image.open(io.BytesIO(page.screenshot())).convert("RGB")
+ width, height = image.size
+ pixels = image.load()
+ target = (213, 81, 129)
+ left = int(310 * scale)
+ mask = {
+ (x, y)
+ for y in range(int(60 * scale), height, 1)
+ for x in range(left, int(1250 * scale))
+ if sum(abs(a - b) for a, b in zip(pixels[x, y], target)) < 24
+ }
+ best: list[tuple[int, int]] = []
+ while mask:
+ seed = mask.pop()
+ blob, queue = [seed], deque([seed])
+ while queue:
+ x, y = queue.popleft()
+ for neighbour in ((x + 1, y), (x - 1, y), (x, y + 1), (x, y - 1)):
+ if neighbour in mask:
+ mask.remove(neighbour)
+ blob.append(neighbour)
+ queue.append(neighbour)
+ if len(blob) > len(best):
+ best = blob
+ if not best:
+ raise RuntimeError("the co-op's memory is not on the graph")
+ return (sum(x for x, _ in best) / len(best) / scale, sum(y for _, y in best) / len(best) / scale)
+
+
+def _setup_research(page, base: str) -> None:
+ _open(page, base, f"/conversations/{RESEARCH_THREAD}", "Used 8 tools")
+ _close_details(page)
+
+
+def _settled_hub(page, base: str, scale: int) -> tuple[float, float]:
+ page.goto(base + "/?tab=knowledge", wait_until="load", timeout=60_000)
+ page.locator('[aria-label^="Interactive knowledge graph"]').wait_for(timeout=30_000)
+ page.wait_for_timeout(7000)
+ return _graph_hub(page, scale)
+
+
+def _setup_create(page, base: str) -> None:
+ _open(page, base, f"/conversations/{CREATE_THREAD}", "Launch page direction")
+ preview = page.locator('iframe[title="Page preview: Home"]')
+ page.wait_for_timeout(1000)
+ if not preview.is_visible():
+ # The conversation's details list its design; opening it docks the panel.
+ details = page.locator('button[aria-label="Conversation details"]').first
+ if details.get_attribute("aria-pressed") != "true":
+ details.click()
+ page.locator('button[title="Open Riverbend launch page"]').first.click()
+ preview.wait_for(timeout=30_000)
+ page.locator('button[aria-label="Toggle navigation"]').first.click()
+ page.wait_for_timeout(1500)
+ # Widen the panel so the page reads at a glance beside the conversation.
+ handle = page.locator('[role="separator"][aria-label="Resize side panel"]').first
+ box = handle.bounding_box()
+ page.mouse.move(box["x"] + 1, box["y"] + 400)
+ page.mouse.down()
+ page.mouse.move(box["x"] - 150, box["y"] + 400, steps=8)
+ page.mouse.move(620, box["y"] + 400, steps=8)
+ page.mouse.up()
+ page.wait_for_timeout(1500)
+ _blur(page)
+
+
+def _setup_automate(page, base: str) -> None:
+ _open(page, base, "/?tab=workflows", "Weekly policy digest")
+ page.wait_for_timeout(600)
+
+
+def _workflow_menu(page, d: Director | None, item: str) -> None:
+ row = page.locator('[data-home-tab="workflows"]').get_by_text("Weekly policy digest").first
+ more = page.locator('button[aria-label="More actions for Weekly policy digest"]')
+ if d is None:
+ row.hover()
+ more.click()
+ page.get_by_role("menuitem", name=item).click()
+ return
+ d.to(row, ms=600, dx=420)
+ d.click(more, ms=450)
+ d.wait(300)
+ d.click(page.get_by_role("menuitem", name=item), ms=450)
+
+
+def _run_history(page, base: str) -> None:
+ """The workflow's run drawer, as the Automate still shows it."""
+ _setup_automate(page, base)
+ _workflow_menu(page, None, "Run history")
+ page.get_by_text("History").first.wait_for(timeout=15_000)
+ page.wait_for_timeout(900)
+ _blur(page)
+
+
+def _setup_ship(page, base: str) -> None:
+ # The details beside it show the goal; they also keep the approval clear of
+ # Buddy, who stands over the right of the landing page's Ship scene.
+ _open(page, base, f"/conversations/{SHIP_THREAD}", "Send an email?")
+ details = page.locator('button[aria-label="Conversation details"]').first
+ if details.get_attribute("aria-pressed") != "true":
+ details.click()
+ page.get_by_text("Share the reviewed launch page").first.wait_for(timeout=15_000)
+ page.wait_for_timeout(600)
+ _blur(page)
+
+
+SETUP = {"research": _setup_research, "create": _setup_create, "automate": _setup_automate, "ship": _setup_ship}
+
+
+def _still(browser, base: str, scene: str, clock: Callable[[], datetime], out: Path) -> None:
+ from PIL import Image
+
+ context, page = _new_context(browser, scale=2, clock=clock, pointer=False)
+ try:
+ if scene == "research":
+ x, y = _settled_hub(page, base, 2)
+ page.mouse.move(x - 40, y + 30)
+ page.mouse.move(x, y, steps=6)
+ page.wait_for_timeout(1200)
+ elif scene == "automate":
+ _run_history(page, base)
+ _park(page)
+ else:
+ SETUP[scene](page, base)
+ _park(page)
+ page.wait_for_timeout(1500)
+ raw = page.screenshot(animations="disabled")
+ finally:
+ context.close()
+ image = Image.open(io.BytesIO(raw)).convert("RGB").resize((WIDTH, HEIGHT), Image.Resampling.LANCZOS)
+ image.save(out, "WEBP", quality=86, method=6)
+
+
+def _record_research(page, base: str, d: Director, rec: Recorder) -> list[tuple[str, str]]:
+ # Find the settled co-op memory first; reloading forgets the layout, so
+ # the recorded graph settles again from the start.
+ x, y = _settled_hub(page, base, 1)
+ _setup_research(page, base)
+ _park(page)
+ page.wait_for_timeout(1000)
+ trace = page.get_by_text("Used 8 tools").first
+ d.place(980, 560)
+ rec.start(page)
+ rec.mark("a0")
+ d.wait(400)
+ d.show()
+ d.click(trace, ms=900)
+ d.wait(400)
+ d.move(d.x + 140, d.y + 60, 800)
+ d.wait(700)
+ rec.mark("a1")
+ page.locator('a[aria-label="Home"]').first.click()
+ page.get_by_role("tab", name="Knowledge").click()
+ rec.mark("b0")
+ d.place(1190, 700)
+ d.wait(3500)
+ d.move(x + 60, y + 70, 700)
+ d.move(x, y, 450)
+ d.wait(1250)
+ rec.mark("b1")
+ rec.stop(page)
+ return [("a0", "a1"), ("b0", "b1")]
+
+
+def _record_create(page, base: str, d: Director, rec: Recorder) -> list[tuple[str, str]]:
+ pill = page.locator('button[aria-label="Model"]').first
+ d.place(470, 560)
+ rec.start(page)
+ rec.mark("s")
+ d.wait(700)
+ d.show()
+ d.click(pill, ms=900)
+ d.wait(500)
+ d.to(page.locator(".model-picker").get_by_text("GPT-5.6 Sol").first, ms=600)
+ d.wait(1000)
+ page.keyboard.press("Escape")
+ d.wait(400)
+ d.click(page.get_by_role("radio", name="Edit"), ms=900)
+ d.wait(900)
+ # The preview is the page scaled into the panel: map the headline into it.
+ box = page.locator('iframe[title="Page preview: Home"]').bounding_box()
+ left, top, width, height, page_width = page.frame_locator('iframe[title="Page preview: Home"]').locator(
+ "h1").first.evaluate("e => { const r = e.getBoundingClientRect(); "
+ "return [r.x, r.y, r.width, r.height, document.documentElement.clientWidth]; }")
+ scale = box["width"] / page_width
+ d.move(box["x"] + (left + width * 0.42) * scale, box["y"] + (top + height * 0.28) * scale, 800)
+ d.wait(300)
+ d.click()
+ d.wait(1500)
+ rec.mark("e")
+ rec.stop(page)
+ return [("s", "e")]
+
+
+def _record_automate(page, base: str, d: Director, rec: Recorder) -> list[tuple[str, str]]:
+ # Everything that matters stays above and left of Buddy, who stands over
+ # the lower right of the landing page's Automate scene. Three takes: the
+ # delivery defaults, the workflow's local model saved, and its run history.
+ d.place(760, 470)
+ rec.start(page)
+ rec.mark("a0")
+ d.wait(400)
+ d.show()
+ d.click(page.locator('button[aria-label="Delivery defaults"]'), ms=700)
+ page.get_by_text("No external channels are set up.").first.wait_for(timeout=15_000)
+ d.move(1085, 232, 500)
+ d.wait(1250)
+ page.keyboard.press("Escape")
+ d.wait(200)
+ rec.mark("a1")
+ _workflow_menu(page, d, "Workflow settings")
+ page.get_by_text("Readiness is checked when it runs").first.wait_for(timeout=15_000)
+ rec.mark("b0")
+ d.wait(150)
+ d.to(page.get_by_text(f"{LOCAL_MODELS[0]} - Ollama Local").first, ms=650)
+ d.wait(1000)
+ d.click(page.get_by_role("button", name="Save settings"), ms=600)
+ page.get_by_text("4 workflows").first.wait_for(timeout=15_000)
+ d.wait(250)
+ rec.mark("b1")
+ _workflow_menu(page, d, "Run history")
+ page.get_by_text("History").first.wait_for(timeout=15_000)
+ _blur(page) # The drawer focuses its first button, whose tooltip would cover the history.
+ rec.mark("c0")
+ d.wait(200)
+ d.move(1075, 400, 650)
+ d.wait(1650)
+ rec.mark("c1")
+ rec.stop(page)
+ return [("a0", "a1"), ("b0", "b1"), ("c0", "c1")]
+
+
+def _record_ship(page, base: str, d: Director, rec: Recorder) -> list[tuple[str, str]]:
+ approve = page.locator('aside.approval-card button[aria-label="Approve"]').first
+ # The landing page turns Buddy to "success" 2.25 s into this clip
+ # (SHIP_APPROVAL_AT in docs/landing-story.js), so Approve lands there.
+ d.place(830, 690)
+ rec.start(page)
+ rec.mark("s")
+ d.wait(1100)
+ d.show()
+ d.to(approve, ms=950)
+ d.wait(120)
+ rec.mark("click")
+ d.click(settle=0)
+ d.wait(560)
+ rec.mark("e")
+ rec.stop(page)
+ return [("s", "e")]
+
+
+RECORD = {"research": _record_research, "create": _record_create, "automate": _record_automate,
+ "ship": _record_ship}
+
+
+def _clip_frames(rec: Recorder, cuts: list[tuple[str, str]], folder: Path) -> tuple[Path, int]:
+ """The edited clip as 30 fps PNG frames: segments joined by short dissolves."""
+ from PIL import Image
+
+ segments = [rec.sample(rec.marks[a], rec.marks[b]) for a, b in cuts]
+ frames: list[Any] = list(segments[0])
+ for segment in segments[1:]:
+ tail, head = frames[-DISSOLVE_FRAMES:], segment[:DISSOLVE_FRAMES]
+ blends = [
+ Image.blend(Image.open(a).convert("RGB"), Image.open(b).convert("RGB"), (i + 1) / (DISSOLVE_FRAMES + 1))
+ for i, (a, b) in enumerate(zip(tail, head))
+ ]
+ frames = frames[:-DISSOLVE_FRAMES] + blends + list(segment[DISSOLVE_FRAMES:])
+ for index, frame in enumerate(frames):
+ target = folder / f"{index:05d}.png"
+ if isinstance(frame, Path):
+ image = Image.open(frame).convert("RGB")
+ else:
+ image = frame
+ if image.size != (WIDTH, HEIGHT):
+ image = image.resize((WIDTH, HEIGHT), Image.Resampling.LANCZOS)
+ image.save(target)
+ return folder, len(frames)
+
+
+def _ffmpeg() -> str:
+ import imageio_ffmpeg
+
+ return imageio_ffmpeg.get_ffmpeg_exe()
+
+
+def _encode(frames: Path, name: str, out_dir: Path) -> None:
+ ffmpeg = _ffmpeg()
+ source = ["-framerate", str(FPS), "-i", str(frames / "%05d.png")]
+ webm = out_dir / f"{name}.webm"
+ mp4 = out_dir / f"{name}.mp4"
+ log = SCRATCH / f"{name}-vp9"
+ common = [ffmpeg, "-v", "error", "-y", *source, "-an", "-pix_fmt", "yuv420p"]
+ vp9 = ["-c:v", "libvpx-vp9", "-b:v", "0", "-crf", "36", "-row-mt", "1", "-tile-columns", "2",
+ "-deadline", "good", "-cpu-used", "1", "-g", "150", "-passlogfile", str(log)]
+ subprocess.run([*common, *vp9, "-pass", "1", "-f", "webm", os.devnull], check=True)
+ subprocess.run([*common, *vp9, "-pass", "2", str(webm)], check=True)
+ subprocess.run([*common, "-c:v", "libx264", "-preset", "veryslow", "-crf", "25", "-profile:v", "high",
+ "-tune", "stillimage", "-g", "150", "-movflags", "+faststart", str(mp4)], check=True)
+ for leftover in SCRATCH.glob(f"{name}-vp9*.log"):
+ leftover.unlink()
+
+
+def capture(scenes: list[str]) -> dict[str, Any]:
+ from playwright.sync_api import sync_playwright
+
+ stills_dir = MEDIA / "screenshots"
+ clips_dir = MEDIA / "clips"
+ anchor = _anchor()
+ started = time.time()
+
+ def clock() -> datetime:
+ return anchor + timedelta(seconds=time.time() - started)
+
+ timings: dict[str, Any] = {}
+ SCRATCH.mkdir(parents=True, exist_ok=True)
+ with ExitStack() as stack:
+ profile = stack.enter_context(tempfile.TemporaryDirectory(prefix="profile-", dir=SCRATCH,
+ ignore_cleanup_errors=True))
+ data_dir = Path(profile).resolve()
+ ollama = _start_demo_ollama(stack)
+ _helper("--seed", str(data_dir), "--ollama-host", ollama, "--anchor", anchor.isoformat())
+ pw = stack.enter_context(sync_playwright())
+ browser = pw.chromium.launch(channel="msedge", headless=True)
+ stack.callback(browser.close)
+ groups = [[s for s in scenes if s != "ship"], [s for s in scenes if s == "ship"]]
+ for group in groups:
+ if not group:
+ continue
+ with ExitStack() as run:
+ if group == ["ship"]:
+ _helper("--add-approval", str(data_dir), "--anchor", anchor.isoformat())
+ app = App(run, data_dir, ollama)
+ for scene in group:
+ print(f"{scene}: still", flush=True)
+ _still(browser, app.base, scene, clock, stills_dir / f"{scene}.webp")
+ print(f"{scene}: clip", flush=True)
+ # The screencast and the edited frames are scratch; only the encodes are kept.
+ raw = Path(run.enter_context(tempfile.TemporaryDirectory(prefix=f"{scene}-raw-", dir=SCRATCH)))
+ edited = Path(run.enter_context(tempfile.TemporaryDirectory(prefix=f"{scene}-", dir=SCRATCH)))
+ context, page = _new_context(browser, scale=1, clock=clock, pointer=True)
+ try:
+ SETUP[scene](page, app.base)
+ _park(page)
+ page.wait_for_timeout(1200)
+ rec = Recorder(context, page, raw)
+ cuts = RECORD[scene](page, app.base, Director(page), rec)
+ finally:
+ context.close()
+ frames, count = _clip_frames(rec, cuts, edited)
+ _encode(frames, scene, clips_dir)
+ first = rec.marks[cuts[0][0]]
+ timings[scene] = {
+ "segments_seconds": [[round(rec.marks[a] - first, 3), round(rec.marks[b] - first, 3)]
+ for a, b in cuts],
+ "frames": count,
+ "raw_frames": len(rec.frames),
+ }
+ if "click" in rec.marks:
+ timings[scene]["click_seconds"] = round(rec.marks["click"] - first, 3)
+ saved = SCRATCH / "timings.json"
+ merged = json.loads(saved.read_text(encoding="utf-8")) if saved.exists() else {}
+ saved.write_text(json.dumps({**merged, scene: timings[scene]}, indent=2), encoding="utf-8")
+ return timings
+
+
+STORY_ID = "react-workbench-v1"
+PRODUCTION = {
+ "app": "Row-Bot 5.0.0, the real app process and its React client at /app-v2/",
+ "method": (
+ "Scripted Playwright capture (scripts/docs/capture_landing_media.py) in Microsoft Edge at 1440x810 CSS px, "
+ "dark appearance, on an isolated temporary profile seeded with fictional demo data (a neighbourhood solar "
+ "co-op); no real user data, accounts, names or paths"
+ ),
+ "models": (
+ "No model ran. The local runtime is the docs capture's display-only stand-in on loopback: it lists "
+ f"{LOCAL_MODELS[0]}, the local model the owner runs, and runs nothing. GPT-5.6 Sol via ChatGPT / Codex "
+ "is display-only: the harness tells the client it is connected, and nothing is connected or called"
+ ),
+ "outward_actions": "None: the harness answers the Ship clip's Approve request itself, so no email is sent",
+ "pointer": "Drawn by the harness to show where the scripted clicks land",
+ "stills": "Rendered at 2x and downscaled with Lanczos to 1440x810 WebP",
+ "clips": "Browser screencast frames resampled to 30 fps; VP9 WebM and H.264 MP4 (faststart), no audio",
+}
+CLIP_NOTES = {
+ "research": "Two takes at normal speed joined by a four-frame dissolve: the local-model conversation's "
+ "source trace, then the Knowledge graph settling and its co-op memory highlighted.",
+ "create": "One continuous take at normal speed: the hosted model shown in the picker, then the landing page "
+ "opened for editing in the Design panel.",
+ "automate": "Three takes at normal speed joined by four-frame dissolves: the workflow delivery defaults (no "
+ "external channels), the workflow's local model saved, then its retained run history.",
+ "ship": "One continuous take at normal speed: the waiting email approval, then Approve.",
+}
+# The landing page's story assets: scene id -> (scene, has clip, has MP4 fallback, alt text).
+STORY_ASSETS = {
+ "hero-app": ("research", False, False,
+ "Row-Bot's Knowledge graph after local-model research, with the co-op's memory and its "
+ "connections highlighted."),
+ "research-local": ("research", True, True,
+ "A local-model conversation with its source trace, followed by the Knowledge graph "
+ "settling into connected memories."),
+ "synthesis-sol": ("create", True, True,
+ "A conversation handed to a frontier model, GPT-5.6 Sol, with an editable landing page "
+ "in its Design panel."),
+ "knowledge-control": ("research", True, False,
+ "A Row-Bot knowledge graph of 103 connected memories, with one memory's connections "
+ "labelled."),
+ "designer-output": ("create", True, False,
+ "An editable landing page open in the conversation's Design panel, its heading then "
+ "selected for editing."),
+ "workflow-repeat": ("automate", True, True,
+ "Workflow delivery kept in the app with no outside channels, a workflow's local model "
+ "saved, then its retained run history."),
+ "approval-boundary": ("ship", True, True,
+ "Row-Bot waiting for approval before emailing a PDF to riverbend-board@example.com, "
+ "then the approval being given."),
+ "model-choice": ("create", False, False,
+ "The composer's model pill showing GPT-5.6 Sol as a cloud model, beside the conversation's "
+ "Design panel."),
+}
+
+
+def _file_record(path: Path) -> tuple[str, int]:
+ return hashlib.sha256(path.read_bytes()).hexdigest(), path.stat().st_size
+
+
+def write_records(*, approve: bool = False) -> None:
+ """Hash the published media into the manifest and the recording receipt.
+
+ New media wait for the owner's review (``pending_review``); ``approve``
+ records the owner's approval of the media as published now.
+ """
+ import imageio_ffmpeg
+
+ timings = json.loads((SCRATCH / "timings.json").read_text(encoding="utf-8"))
+ digest = hashlib.sha256()
+ clips: dict[str, Any] = {}
+ for scene in SCENES:
+ still = MEDIA / "screenshots" / f"{scene}.webp"
+ webm, mp4 = MEDIA / "clips" / f"{scene}.webm", MEDIA / "clips" / f"{scene}.mp4"
+ frames, _seconds = imageio_ffmpeg.count_frames_and_secs(str(webm))
+ record: dict[str, Any] = {
+ "source": "Screencast of the real Row-Bot 5.0.0 React client",
+ "source_frames": timings[scene]["raw_frames"],
+ "segments_seconds": timings[scene]["segments_seconds"],
+ "duration_seconds": round(frames / FPS, 3),
+ "dimensions": [WIDTH, HEIGHT],
+ "fps": FPS,
+ }
+ for key, path in (("webm", webm), ("mp4", mp4), ("poster", still)):
+ sha, size = _file_record(path)
+ digest.update(sha.encode())
+ record.update({key: f"media/landing-story/{path.parent.name}/{path.name}",
+ f"{key}_sha256": sha, f"{key}_bytes": size})
+ record["editorial_note"] = CLIP_NOTES[scene]
+ if "click_seconds" in timings[scene]:
+ record["approval_moment_seconds"] = timings[scene]["click_seconds"]
+ clips[scene] = record
+ run_id = datetime.now(timezone.utc).strftime("%Y%m%dT%H%M%SZ") + "-" + digest.hexdigest()[:6]
+ receipt = {
+ "schema": 1,
+ "story_id": STORY_ID,
+ "run_id": run_id,
+ "editing": ("normal-speed editorial cuts with four-frame dissolves; scripted takes of the real app, "
+ "no acceleration"),
+ "production": PRODUCTION,
+ "clips": clips,
+ }
+ (MEDIA / "recording-receipt.json").write_text(json.dumps(receipt, indent=2) + "\n", encoding="utf-8", newline="\n")
+
+ manifest_path = MEDIA / "manifest.json"
+ manifest = json.loads(manifest_path.read_text(encoding="utf-8"))
+ assets = []
+ for asset in manifest["assets"]:
+ if asset["scene_id"] not in STORY_ASSETS:
+ assets.append(asset)
+ continue
+ scene, has_clip, has_fallback, alt = STORY_ASSETS[asset["scene_id"]]
+ entry = {"scene_id": asset["scene_id"], "still": f"screenshots/{scene}.webp",
+ "sha256": clips[scene]["poster_sha256"]}
+ if has_clip:
+ entry.update(clip=f"clips/{scene}.webm", clip_sha256=clips[scene]["webm_sha256"])
+ if has_fallback:
+ entry.update(clip_fallback=f"clips/{scene}.mp4", clip_fallback_sha256=clips[scene]["mp4_sha256"])
+ entry["alt"] = alt
+ assets.append(entry)
+ reviewed_at = datetime.now(timezone.utc).strftime("%Y-%m-%dT%H:%M:%SZ") if approve else None
+ manifest.update(story_id=STORY_ID, run_id=run_id, review_status="approved" if approve else "pending_review",
+ reviewed_at=reviewed_at, production=PRODUCTION, assets=assets)
+ first = ("schema", "story_id", "run_id", "review_status", "reviewed_at", "recording_receipt", "production")
+ ordered = {key: manifest[key] for key in first}
+ ordered.update({key: value for key, value in manifest.items() if key not in ordered})
+ manifest_path.write_text(json.dumps(ordered, indent=2, ensure_ascii=False) + "\n", encoding="utf-8", newline="\n")
+
+
+def serve() -> None:
+ anchor = _anchor()
+ with ExitStack() as stack:
+ data_dir = Path(stack.enter_context(tempfile.TemporaryDirectory(
+ prefix="profile-", dir=SCRATCH, ignore_cleanup_errors=True))).resolve()
+ ollama = _start_demo_ollama(stack)
+ _helper("--seed", str(data_dir), "--ollama-host", ollama, "--anchor", anchor.isoformat())
+ _helper("--add-approval", str(data_dir), "--anchor", anchor.isoformat())
+ app = App(stack, data_dir, ollama)
+ stop = SCRATCH / "serve.stop"
+ stop.unlink(missing_ok=True)
+ (SCRATCH / "serve.json").write_text(json.dumps({"base": app.base, "anchor": anchor.isoformat()}),
+ encoding="utf-8")
+ print(f"Row-Bot is running at {app.base}/; create {stop.relative_to(ROOT)} to stop it.", flush=True)
+ while app.proc.poll() is None and not stop.exists():
+ time.sleep(0.5)
+ stop.unlink(missing_ok=True)
+
+
+def main() -> int:
+ parser = argparse.ArgumentParser(description=__doc__.splitlines()[0])
+ parser.add_argument("--scenes", nargs="*", default=list(SCENES), choices=SCENES)
+ parser.add_argument("--serve", action="store_true", help="seed and start the app for a manual look")
+ parser.add_argument("--records", action="store_true",
+ help="only rewrite the manifest and recording receipt from the published media")
+ parser.add_argument("--approve", action="store_true",
+ help="with --records: record the owner's approval of the published media")
+ parser.add_argument("--seed", metavar="DATA_DIR", help=argparse.SUPPRESS)
+ parser.add_argument("--add-approval", metavar="DATA_DIR", help=argparse.SUPPRESS)
+ parser.add_argument("--ollama-host", default="", help=argparse.SUPPRESS)
+ parser.add_argument("--anchor", default="", help=argparse.SUPPRESS)
+ args = parser.parse_args()
+ anchor = datetime.fromisoformat(args.anchor) if args.anchor else _anchor()
+ if args.seed:
+ seed_profile(Path(args.seed), args.ollama_host, anchor)
+ elif args.add_approval:
+ add_ship_approval(Path(args.add_approval), anchor)
+ elif args.serve:
+ serve()
+ elif args.records:
+ write_records(approve=args.approve)
+ else:
+ print(json.dumps(capture(args.scenes), indent=2))
+ write_records()
+ return 0
+
+
+if __name__ == "__main__":
+ raise SystemExit(main())
+
+ 
+
+ 
+
+ 