site: non-blog copy sweep — canonical ObjectOS strings, scoped existing-systems claim, approvals path, localized FAQ kicker

[objectstack-fleet]

Fixes #222

Branch. This PR carries 4 commit(s) from the seat's linear stack, replayed onto main @ 78d820e. The replay is patch-equal to git diff 2fcd27a lin/issue-222, with byte-identical diffs. Head: fe39b43. The staged SHAs quoted below are the pre-replay commits; their content is unchanged.

Non-blog copy sweep beyond #209. Every page that says what ObjectStack or ObjectOS is now reads it from src/lib/positioning.ts, every number and capability claim comes from a public source, the "works on your existing systems" claim carries the README's limit everywhere it appears, the agent CTA is the README path, and the capability pages stop answering the same templated FAQ. The three post phrasings carried from the #209 sweep are fixed in all seven authored locales; zh-Hant is regenerated.

Staged on stack/www-wave1 as staged/issue-222 (base 7f69d97d, head dee44d2b), three self-contained commits, replayed with rebase --onto after wave 2 as lin/issue-222 (922605d), plus one follow-up commit on the linear stack, 3a531e1 (head 3a531e1f).

Sources

Facts come from these public sources only; the source codes are used in the tables below.

• S0 src/lib/positioning.ts — T1, T2, T4, the descriptor and E1 (the objectstack README quoted verbatim).
• S1 objectstack README at origin/main 6befe19c: "Try it in five minutes" (npm create objectstack@latest, the scaffolder writes an AGENTS.md, npx os dev --ui), "The runtime runs it" (the 12-row capability table), "Agents are the first users" (MCP on by default at /api/v1/mcp), "Why the mistakes don't ship" (the four gates, "apps small enough for AI to hold whole", the bundled examples/app-crm loads end-to-end), "You own it" (Apache-2.0, the CLI os compile / os serve artifact), "Console renders the real app … Studio authors the same metadata visually".
• S3 objectstack README "What we mean by ontology" + docs.objectos.ai extend-existing-systems: "federating an external datasource is read-only by default and early"; start with a readOnly datasource or a read-only database user and enable writes deliberately, per object; "expect the read path to be the mature one".
• S4 docs.objectos.ai resources/license: the editions table (in-UI AI from Free as a taste, AI governance from Business, clustering / air-gap / SCIM on Enterprise), "Cloud or self-managed" (Business Self-Managed single node, bring your own model or key; Enterprise full private deployment), "The open-source alternative" (Apache-2.0: protocol, kernel, CLI, production runtime, Console UI, MCP server, no feature gate between development and production), contact sales → sales@objectstack.ai.
• S5 docs.objectos.ai build/ai-builder: the AI Builder builds packages, objects, fields, actions, flows from plain-language chat; its built-in tools (create_package, create_object, add_field, modify_field, delete_field, …); every metadata mutation routes through a pending-action queue where approvers see the diff; ai:approve is held by Setup Administrator by default and can be scoped per package in a permission set; every approve/reject lands in the audit log with the conversation id.
• S6 this site's pricing page (src/lib/pricing.ts): Business lists "SSO / SAML".
• S7 docs.objectos.ai configure/ai: the MCP server ships in the open framework, on by default at /api/v1/mcp; the in-product AI runtime (ask, build) ships with ObjectOS; on Self-Managed a licence unlocks it.

Item 1 — retired token figures

zh-Hans carries the same change on every row (the zh-Hans mirror is in the same file under content/pages/zh-Hans/).

Where	Before	After	Source
ai.ts metric	under-150k-tokens tile: "A complete CRM, including UI, held in one context window"	Held whole — "Apps small enough for AI to hold whole — the bundled CRM loads end-to-end"	S1
ai.ts build copy	"A complete CRM is under 150k tokens: business logic … under 100k, and UI metadata adds about 50k."	"The definition stays small enough for an agent to hold whole: the objectstack README's bundled CRM — objects, views, a dashboard, a lead-conversion flow, permission sets, actions, translations — loads end-to-end and refactors in one change."	S1
agent-developer.ts metrics	under-150k / under-100k / about-50k token tiles	AGENTS.md "Written by the scaffolder, so the agent starts with the protocol's rules loaded" · 4 gates "Typed, validated, reviewed, governed — between the agent and your users" · Held whole	S1
agent-developer.ts workflow item 1	"Under 150k tokens for a full CRM including UI, the app can be inspected as a whole"	"The app stays small enough for AI to hold whole — the README's bundled CRM loads end-to-end — so it can be inspected as a whole"	S1
forward-deployed-engineers.ts metrics	the same three token tiles	Apache-2.0 "No seat to resell — the client self-hosts the runtime at no license cost" · 1 artifact "os compile builds one deployable JSON artifact; any ObjectStack runtime serves it" · Held whole	S1, S4
forward-deployed-engineers.ts "Security review reads the diff"	"under 150k tokens for a full CRM, with business logic under 100k"	"small enough to hold whole, permissions and audit included"	S1
forward-deployed-engineers.ts checklist	"Keep the whole definition under the 150k-token complete-app budget"	"Keep the whole definition small enough for your agent to hold whole, so it can reason about and refactor it in one change."	S1
platform.ts metric	under-150k-tokens tile	4 gates	S1
product-tour.ts metrics	the same three token tiles	4 gates · 1 definition "Database, REST API, client SDK, UI, and MCP tools all follow from it" · Held whole	S1
glossary application-metadata.ts	"A complete CRM expressed as metadata fits in roughly 150k tokens, so a coding agent can hold the whole system in one context window"	"An application expressed as metadata stays small enough for AI to hold whole — the objectstack README's bundled CRM, objects through translations, loads end-to-end — so a coding agent reasons about the whole system"	S1

The product-tour and platform tiles were listed under #209's sweep, but the sweep branch on the stack is blog-only, so they are taken here; otherwise the grep proof below could not read zero.

Item 2 — old descriptors

Where	Before	After	Source
product-tour.ts connect copy	"ObjectOS gives teams a production platform for operating apps built on it."	"ObjectOS is the commercial runtime environment built on ObjectStack, where teams run and operate the apps built on that model."	S0
platform.ts FAQ 1	"part of the open-source ObjectStack target format and runtime"	"part of open-source ObjectStack, Apache-2.0: the protocol, kernel, CLI, production runtime, Console UI, and MCP server, with no feature gate between development and production"	S4
agent-developer.ts lead	"ObjectStack is the open target format and runtime for AI-written enterprise software, designed for the agent as writer and the human as reviewer. A complete CRM … 150k tokens…"	T2, then "The agent is the writer and you are the reviewer: the scaffolder writes an AGENTS.md so the agent starts with the protocol's rules loaded, Strict TypeScript, Zod schemas, and os validate reject invalid output before it reaches the runtime, and the whole app stays small enough for AI to hold whole."	S0, S1
forward-deployed-engineers.ts description + lead + FAQ 2	"the optional commercial production platform" / "open-source target format and runtime … commercial production platform"	the descriptor from positioning.en, with what clients buy spelled out from the licence page (in-app AI Builder and Ask, AI governance from Business, Cloud or self-managed delivery — Business Self-Managed on a single node; Enterprise for clustering, air-gapped operation, SCIM — and official support)	S0, S4
src/lib/home.ts kicker, all seven authored locales	"For AI-written enterprise software" (and its translations)	positioning[locale].t1 — "The ontology is the software."	S0
src/glossary/ui.ts heading	"The vocabulary of AI-written enterprise software." / "AI 编写的企业软件，用的是这套词汇。"	"The vocabulary of an executable business ontology." / "一份可执行的业务本体，用的是这套词汇。"	S0 (T2)
src/lib/i18n.ts articles lead	"on AI-written enterprise software, runtime governance, and agent-ready applications"	"on executable business ontologies, runtime governance, and agent-operated applications"	S0
src/lib/terms.ts topic-hub description	"with practical notes on AI-written enterprise software"	"with practical notes on executable business ontologies"	S0 — one line outside the claimed file list, same string family as i18n.ts, added so the grep proof reads zero

Item 3 — "works on your existing systems", with the limit

Where	Before	After	Source
home.ts capability card, seven locales	"Extend systems without replacing them — Add APIs, permissions, workflows, and intelligence on top of databases, ERP, CRM, and custom systems."	"Extend what already runs — read-only first — Federate a database you already run as an external datasource and model its tables as objects. Federation is read-only by default and early: the system of record keeps writing the rows, and writes are enabled deliberately, per object."	S3
home.ts closing copy, seven locales	"Connect one existing system, define its key business objects, …"	"Federate one database you already run — read-only by default — define its key business objects, …"	S3
product-tour.ts connect section	title "Add an object layer without replacing systems of record"; item "Map existing records into business objects … without forcing a rip-and-replace migration."	title "Add an object layer beside systems of record — read-only by default"; copy "ObjectStack can federate a database you already run as an external datasource … read-only by default, and early; the system of record keeps writing the rows"; item "Connect the database you already run — a readOnly datasource or a read-only database user to start … writes are enabled deliberately, per object, once you trust the model."	S3
data-modeling.ts	description "federates existing databases in place"; metric "In place — Existing databases federate without migration"; datasource-sync item; table row "A federation definition"; FAQ "Federation can additionally mount tables … without moving the data."	"federates a database you already run, read-only by default"; metric "Read-only first — A database you already run federates in place — read-only by default, writes opt-in, and early"; "mount its tables as governed objects, read-only by default; writes are enabled deliberately, per object"; "A datasource binding, read-only to start"; FAQ adds "read-only by default, and early: expect the read path to be the mature one."	S3
src/lib/clusters.ts ai-native-app-platform capability	"Integration with existing systems so the platform can extend what already runs instead of forcing a migration."	"Federation of a database you already run — read-only by default, and early — so the platform can read what already runs instead of forcing a migration."	S3
clusters.ts legacy-system-modernization lead, definition, capability 1, FAQ 1	"connect existing systems" / "Connect to existing databases and applications without moving every record first." / "can connect to the existing system of record"	"connect existing systems read-only first" / "Connect an existing database as a read-only datasource first — federation is read-only by default, and early" / "read-only by default, with writes enabled deliberately per object … Federation is early: expect the read path to be the mature one."	S3
clusters.ts manufacturing-ai capability 1, FAQ 2	"Connect production, inventory, equipment, quality, and work order data as business objects." / "connects existing systems and models the relevant records"	"… federated read-only by default from the databases that already run production." / "connects the existing databases as read-only datasources — federation is read-only by default, and early — … so AI can read and analyze"	S3
src/lib/cluster-i18n.ts template capability 4 and FAQ 3, zh-Hans / ja / de / es / fr / ko	"Connect CRM, ERP, databases and custom systems without forcing a migration." / "Usually not. The safer path is to connect existing systems first…"	each locale now says the federation is read-only by default and early, writes enabled deliberately per object	S3

Item 4 — the README agent path

Where	Before	After	Source
agent-developer.ts artifact	"A good rule file makes the target format explicit" with an invented six-line rule file	"The README path — Three commands, and your agent has the rules": the scaffolder installs the AI skills bundle and writes an AGENTS.md; the code block is npm create objectstack@latest my-app && cd my-app, the README's support-desk prompt, npx os dev --ui	S1
agent-developer.ts primary CTA	"Read the docs" → docs.objectos.ai root	"Follow the README path" → https://github.com/objectstack-ai/objectstack#try-it-in-five-minutes	S1
agent-developer.ts description	"Teach a coding agent the open ObjectStack target format, Strict TypeScript definitions, Zod schemas, validation gate, and whole-context proof…"	"The README path for a coding agent: npm create objectstack@latest writes an AGENTS.md, the agent writes typed metadata, os validate rejects what would fail at runtime, and npx os dev --ui renders the app for review." (this line is also the llms.txt entry)	S1
ai.ts "Agent writes source files"	"Claude Code, Cursor, or any coding agent edits ObjectStack definitions in your repo; review happens as ordinary git diffs."	"npm create objectstack@latest scaffolds a project and writes an AGENTS.md, so Claude Code, Cursor, or any coding agent starts with the protocol's rules loaded; it edits ObjectStack definitions in your repo, npx os dev --ui renders them, and review happens as ordinary git diffs."	S1
mcp.ts CTAs	primary "Compare editions and pricing"; secondary "Read the docs" → docs root	primary "Read the MCP docs" → docs.objectos.ai/docs/configure/mcp; secondary "Point your agent at ObjectStack" → /agent-developer/	S7
platform.ts CTAs	primary "Compare editions and pricing"; secondary "Take the product tour"	primary "Point your agent at ObjectStack" → /agent-developer/; secondary "Compare editions and pricing"	AGENTS.md rule 8
home.ts AI link, seven locales	"View the AI security model" (the link goes to /agent-developer/)	"Point your agent at ObjectStack" — the label now says where the existing link goes; the href is in MarketingHome.astro (markup, held out)	—
clusters.ts self-hosted-ai FAQ 4 (new)	—	ends with "To start: npm create objectstack@latest, point your coding agent at the generated project and its AGENTS.md, then npx os dev --ui."	S1

Item 5 — capability pages that repeated each other

Six pages (data-modeling, app-ui, automation, approvals, permissions, analytics) in en and zh-Hans. Each now names the open runtime in its lead, says concretely what ObjectOS adds for that capability, answers a page-specific FAQ instead of the templated one, and opens on a page-specific CTA.

Page	Title before → after	Lead: "In ObjectOS …" →	FAQ "Is X in the open-source ObjectStack? Yes…" → "What is open source here, and what does ObjectOS add?"	Primary CTA	Source
data-modeling	"ObjectOS Data Modeling: …" → "Data Modeling: From Business Objects to Database and APIs"	"On the open-source ObjectStack runtime that is one compact object definition … and on ObjectOS the in-app AI Builder drafts the same objects and fields from a plain-language request."	"All of the data layer … and the object designer, ER diagram, and datasource sync in the Console — Apache-2.0, with no feature gate between development and production. ObjectOS adds the in-app AI Builder, whose built-in tools create packages and objects and add, modify, and delete fields from a chat, queued for human approval before they go live — on Cloud, Business Self-Managed, and Enterprise."	"Read the data modeling docs" → docs/build/data	S4, S5
app-ui	"ObjectOS App Interfaces: …" → "App Interfaces: Permission-Aware Screens Rendered from Metadata"	"On the open-source ObjectStack runtime that is one view definition … On ObjectOS, the in-app AI Builder registers the same views from a plain-language request."	"… all ship in the open-source ObjectStack Console, Apache-2.0. ObjectOS adds the in-app AI Builder, which registers views — a kanban grouped by status, for instance — from a chat request and queues the change for approval; the open runtime leaves that work to your own coding agent and MCP client."	"Read the interface docs" → docs/build/interface	S4, S5
automation	"ObjectOS Automation: …" → "Process Automation: Flows, Triggers, Jobs, and Webhooks for Business Processes"	"exactly the shape ObjectStack flows are built for … and on ObjectOS drafted by the in-app AI Builder from a plain-language request." (also ¥100k → $100k on the English page, four places)	"Flows, all three trigger types, background queues, webhook delivery, and the flow canvas … ship in the open-source ObjectStack Console, Apache-2.0. ObjectOS adds the in-app AI Builder, which drafts flows and actions from a plain-language request and queues every metadata change for human approval, plus AI governance — action audit, guardrails, permission scoping — from Business."	"Read the flows docs" → docs/build/automation/flows	S4, S5
approvals	"ObjectOS Approvals: …" → "Approvals: Multi-Step Sign-Off for Business Changes and AI Actions"	unchanged (no edition claim in the lead)	"… ship in the open-source ObjectStack Console, Apache-2.0. ObjectOS adds a second thing to approve: the in-app AI Builder's metadata changes land in a pending-action queue where approvers see the proposed diff before clicking Approve — only Setup Administrators hold that permission by default — and every approve or reject lands in the audit log with the originating conversation."	"Read the approvals docs" → docs/build/automation/approvals	S4, S5
permissions	"ObjectOS Permissions: …" → "Permissions: Role, Row, and Field-Level Control for AI-Written Apps"	"On the ObjectStack runtime that boundary is one reviewable definition — enforced on every query, API call, and AI tool call, whether the agent is your own over MCP or ObjectOS's in-app AI." Also "ObjectOS never does" → "The ObjectStack runtime never does".	"… are all open-source ObjectStack, Apache-2.0, and apply equally to MCP tool access. ObjectOS adds AI governance for its in-app AI from Business — action audit, guardrails, and permission scoping — and makes approving the AI Builder's changes its own permission, one you can scope per package in a permission set."	"Read the permissions docs" → docs/configure/permissions	S4, S5
analytics	"ObjectOS Analytics: …" → "Analytics and Reporting: Dashboards and Reports over Governed Objects"	"The ObjectStack runtime runs the same questions over the governed objects themselves … and on ObjectOS the "ask your data" assistant answers them in chat, inside the same permissions."	"… are part of the open-source ObjectStack runtime, Apache-2.0 … ObjectOS adds the "ask your data" assistant: a question asked in chat is answered from the same datasets, inside the asking user's permissions, with the query audited and guardrailed from Business. On the open runtime you ask the same questions through any MCP client with your own model."	"Read the dashboards docs" → docs/build/interface/dashboards	S4

Also in item 5:

Where	Before	After	Source
platform.ts ai-native copy	"ObjectOS adds the in-app AI and production experience for teams."	"ObjectOS, the commercial runtime environment built on ObjectStack, adds the in-app AI Builder and Ask on top of that open mechanism, with AI governance from Business and hosted or self-managed delivery."	S0, S4
platform.ts FAQ 2	"Yes. The open-source ObjectStack runtime can be self-hosted. ObjectOS Enterprise also supports private deployment…"	"Yes, three ways. The open-source ObjectStack runtime self-hosts free, with bring-your-own AI over MCP. ObjectOS Business Self-Managed is a licensed single-node deployment that puts the in-app AI on your own infrastructure, bring your own model or key. ObjectOS Enterprise is the full private deployment — a VPC, local servers, or an air-gapped network — with clustering, local models, internal identity, your own secret management, and enterprise support."	S4
mcp.ts lead	"ObjectOS adds the production operations around that same app."	"The MCP server ships in the open framework and is on by default in every edition; ObjectOS is the commercial runtime environment built on ObjectStack, and runs the same server."	S0, S7
clusters.ts self-hosted-ai definition and capability 1	"A self-hosted AI application platform lets enterprises run the governed business runtime in their own infrastructure. The model can be local or external…" / "Run the business object layer, tool registry, approvals, and audit logs in controlled infrastructure."	adds "Two things answer to that description here: the free, Apache-2.0 ObjectStack framework, and licensed self-managed ObjectOS — Business Self-Managed or ObjectOS Enterprise — which adds the in-product AI on top of it." / "… free on the open-source ObjectStack framework, or licensed as self-managed ObjectOS when you want the in-product AI Builder and Ask on your own infrastructure."	S4
clusters.ts self-hosted-ai FAQ 4 (new), and cluster-i18n.ts zh-Hans via a per-cluster EXTRA_FAQ hook	—	"Is self-hosting free, and what does ObjectOS add?" — the licence page's answer (Apache-2.0 framework with MCP-only AI; Business Self-Managed single node with the in-product AI; Enterprise with clustering, air-gap, SCIM, attestations, SLA), E1's export clause, and the README path as the product CTA in text, because the cluster template has no CTA slot (markup, held out)	S4, S0, S1
src/content-pages/ui.ts FAQ heading (all 15 pages)	"Questions this page should answer" / "这一页应该回答的问题"	"Frequently asked questions" / "常见问题"	—

Item 6 — placeholder and copy slips

Where	Before	After	Source
SecurityPage.astro + src/lib/security.ts, seven locales	the five section eyebrows printed the h2 under them ("Data residency / Data residency" …)	new aiKicker, residencyKicker, permissionKicker, deploymentKicker, faqKicker fields — en "How agents act / Where data lives / Who may do what / Where it runs / FAQ", localized in zh-Hans, ja, de, es, fr, ko; the template prints the kicker field in the existing eyebrow element (five expression swaps, no element, attribute or CSS change)	—
SecurityPage.astro + security.ts CTA	en and the other Latin/ja/ko locales linked extend-existing-systems-with-ai as "Read how to connect existing systems", zh linked ai-agent-business-data-security-boundaries	one target in every locale, the agent security-boundaries article (published in all eight); labels follow: "Read the agent security boundaries article" and its translations	—
home.ts AI link	see item 4
automation.ts (en)	"¥100k" in the lead, section title, item and table row	"$100k"	—
customer-stories.ts table	four rows under a "5 patterns" tile	fifth row "Employee service — Policy, Request, Approval, Employee — Employee-data visibility, request routing, policy answers", from the page's own fifth story	—
platform.ts metric	"25+ Runtime services behind every deployed application"	"12 — Typed primitives in the runtime vocabulary, from objects and permissions to AI tools and datasources" (the README's capability table has twelve rows); no public page counts runtime services, so the figure is dropped	S1
platform.ts "Designers & console"	"16+ open-source ObjectStack admin surfaces — …"	"The open-source Console renders the real app and Studio authors the same metadata visually — the object designer, flow canvas, approvals inbox, permission matrix, and audit viewer — …" (no public page counts admin surfaces)	S1
forward-deployed-engineers.ts table "AI access"	"Platform-native assistants only" (an unsourced claim about Foundry)	"Set by the vendor's roadmap" / "Open MCP, on by default — Claude, Cursor, local models"	S1

Item 7 — plan details

Where	Before	After	Why this value
ai.ts table	—	already lists Cloud Free's taste of AI and "Business (Cloud or Self-Managed)" on the stack (#216); unchanged here	—
platform.ts FAQ 2	omits Business Self-Managed	names all three self-hosting paths (see item 5)	S4
trust-center.ts identity row	"Project identity and permission metadata" / "SSO, SCIM, admin roles, session policy mapping" — SSO under the Enterprise packet	"Project identity and permission metadata; SSO / SAML from ObjectOS Business" / "SCIM and identity lifecycle, admin roles, session policy mapping" — the pricing page (S6) puts SSO / SAML on Business; the licence page (S4) gates SCIM and identity lifecycle on Enterprise and does not gate SSO; the trust center follows both	S4, S6
src/lib/pricing.ts Enterprise contact	mailto:support@objectstack.ai	mailto:sales@objectstack.ai — the licence page (S4) names sales@ for "contact sales" on Enterprise, the Runtime Subscription and OEM terms three times; support@ is the billing and refunds address the refund policy names	S4
src/pages/llms.txt.ts pricing line	"Open-source ObjectStack and ObjectOS Cloud and Enterprise plans."	"Open-source ObjectStack, the ObjectOS Cloud plans (Free, Team, Business), Business Self-Managed, and ObjectOS Enterprise." The FDE and agent-developer lines pick up the new page descriptions.	S4

Prices, the annual-billing wording and the plan feature lists in pricing.ts are untouched (#214); the signup and "Start" CTA targets are untouched (#215).

Carried post phrasings (first comment on #222)

Post	Before	After	Locales
ai-ontology-open-protocol line 145	"ObjectOS adds the commercial production experience around the same application — browser-based Build and Ask, …"	"ObjectOS, the commercial runtime environment built on ObjectStack, adds browser-based Build and Ask, team review and approval, managed cloud or private deployment, SSO, operations, and support around the same application — …"	en, zh-Hans, ja, de, es, fr, ko; zh-Hant regenerated. Line 180 of the card's reading already carried the canonical form on the stack after #209's sweep, so only this sentence moved.
self-hosted-ai-app-platform section heading	"ObjectStack Runtime and the Optional ObjectOS Layer"	"The ObjectStack Runtime, and ObjectOS Built on It" (and the six translations)	all seven
extend-existing-systems-with-ai description and the opening claim	"Connect ObjectOS to the database you already run, …"	"On self-managed ObjectOS (Enterprise), connect the database you already run — read-only by default — …"; the TL;DR / one-sentence paragraph states the assumption: a self-managed ObjectOS deployment, ObjectOS Enterprise on your own infrastructure, where the database is reachable	all seven; de/es/fr descriptions shortened to stay under the 230-char lint ceiling

All three posts already carried updated: 2026-10-06T10:00:00+08:00 from the stack, so no frontmatter date moved.

Approvals: AI metadata changes are their own reviewed path (seat review, commit dee44d2)

The approvals page, the approval-workflow glossary entry and llms.txt said AI-proposed structural changes ride "the same queue" / "the same approvals inbox" as record approvals, which the new approvals FAQ already contradicted. Source: docs.objectos.ai build/ai-builder (objectos origin/main 5d2f837, lines 67–91, "Human-in-the-Loop approval"): tool calls that mutate metadata route through a pending-action queue (/api/v1/ai/pending-actions), a separate queue from flow approval requests; approvers see the proposed diff; ai:approve is held only by Setup Administrator by default and can be scoped per package in a permission set; every approve/reject lands in sys_audit_log with the originating conversation id; it is an ObjectOS (in-app AI Builder) feature, not the open-source console's. The record-approvals inbox is the open-source console's (Apache-2.0). en and zh-Hans; zh-Hant derived.

Where	Before	After	Source
approvals.ts description (:8)	"… a three-tab approvals inbox, escalation, and record locking — the same queue that gates AI-written structural changes before they ship."	"… a three-tab approvals inbox, escalation, and record locking in the open-source console — and, on ObjectOS, a separate reviewed path for the AI Builder's metadata changes: a pending-action queue with the diff, its own approve permission, and an audit entry."	ai-builder.mdx 67–91; license (console is Apache-2.0)
approvals.ts lead (:12)	"The schema change an AI just proposed rides the exact same queue."	"A schema change the ObjectOS AI Builder just proposed waits the same way — in its own pending-action queue, where an approver reads the diff before it goes live."	ai-builder.mdx 67–75
approvals.ts section title (:133)	"The same queue that reviews people reviews AI"	"People and AI are both reviewed before anything ships"	—
approvals.ts section copy (:135)	"… structural changes from an agent do not ship — they queue."	"… a structural change from an agent does not ship — it waits for a person. On the open runtime that is your git diff; on ObjectOS it is the AI Builder's pending-action queue."	ai-builder.mdx 67–69
approvals.ts "Structural changes queue" (:139)	"When an agent proposes new objects, fields, flows, or permission changes, the change lands in the same approvals inbox as a compact diff."	"When the ObjectOS AI Builder proposes new packages, objects, fields, actions, or flows, the change lands in a pending-action queue — separate from record approvals — where approvers see the proposed diff before clicking Approve; every approve or reject lands in the audit log with the originating conversation."	ai-builder.mdx 7–9, 67–75, 89–91
approvals.ts "The diff is the request" (:143)	"Reviewers see exactly what would change — schema, authority, automation — not a description of it."	+ "Approving is its own permission, ai:approve: Setup Administrators hold it by default, and a permission set can scope it per package."	ai-builder.mdx 77–87
approvals.ts table row (:158)	"The approval queue in front of deployment"	"The AI Builder's pending-action queue, in front of deployment"	ai-builder.mdx 67
approvals.ts checklist (:167)	"AI-proposed structural changes cannot bypass the queue."	"AI-proposed structural changes cannot bypass review — the pending-action queue on ObjectOS, the git diff on the open runtime."	ai-builder.mdx 67
zh-Hans approvals.ts :8 :12 :133 :135 :139 :143 :158 :167	"…也要过同一个队列" / "走的也是这同一个队列" / "审人的队列，同样审 AI" / "落入同一个审批收件箱" / "部署之前的那道审批队列" / "无法绕过队列"	"…都在开源控制台里；在 ObjectOS 上，AI Builder 的元数据变更另走一条受审路径：带 diff 的待审动作队列、独立的批准权限与审计记录" / "在它自己的待审动作队列里，审批人读完 diff 才会生效" / "人和 AI 都要过审，才有东西上线" / "落入一个独立于记录审批的待审动作队列 … 每一次批准或驳回都连同发起它的对话写入审计日志" / "批准是一项独立权限 ai:approve：默认只有 Setup Administrator 持有，也可以在权限集里按包限定范围" / "AI Builder 的待审动作队列，横在部署之前" / "无法绕过审阅——ObjectOS 上是待审动作队列，开源运行时上是 git diff"	same
glossary approval-workflow.ts en/zh-Hans (:13)	"This is also the queue that AI-proposed structural changes land in — the diff itself becomes the request —" / "这也是 AI 提出的结构性变更所进入的同一个队列——那份 diff 本身就是请求——"	"AI-proposed structural changes get a reviewed path of their own: on ObjectOS the AI Builder's metadata changes wait in a pending-action queue where the diff itself is the request and approving is its own permission, ai:approve —" / "AI 提出的结构性变更另有一条受审路径：在 ObjectOS 上，AI Builder 的元数据变更在待审动作队列里等人，那份 diff 本身就是请求，批准是一项独立权限 ai:approve——"	ai-builder.mdx 67–87
src/pages/llms.txt.ts (:84)	"and a structural change routes to the same approval queue a human change would."	"and a structural change it proposes waits for a person too: on open-source ObjectStack as a git diff, on ObjectOS in the AI Builder's pending-action queue, where an approver reads the diff and approving is its own permission."	ai-builder.mdx 67–87

Sweep: the same claim grepped over every non-blog locale source (same approvals inbox|same queue|same approval queue|同一个审批收件箱|同一个队列|…) reads 0 after the change; the built output reads 0 across all eight locales; no blog post carries it (0 files), so there is no post-side finding to report. The approvals page was screenshotted again at 1440 and 390 in en and zh-Hans (looked at: the longer "Structural changes queue" card and the new section title fit; 12 overflow probes, 0 overflow).

Security kickers: localized FAQ kicker in ja, de, es, fr, ko (linear-stack follow-up, commit 3a531e1)

A side effect that only shows when #219 and #222 are combined: #219's securityEnglishFields marks any field whose locale value equals the English value as lang="en", and #222 had left faqKicker: 'FAQ' in the five translated locales, so each of /{de,es,fr,ja,ko}/security/ rendered one English eyebrow. The kicker now follows the site's own per-locale FAQ wording (es keeps the cluster label; the others get the plain phrase). The other four kickers were already translated in every locale — a field-by-field comparison of the ja/de/es/fr/ko blocks against en finds faqKicker as the only equal string.

Locale	Before	After	Source
ja	faqKicker: 'FAQ'	よくある質問	the page's own FAQ wording ("セキュリティレビュー FAQ" heading stays)
de	faqKicker: 'FAQ'	Häufige Fragen	—
es	faqKicker: 'FAQ'	Preguntas frecuentes	src/lib/cluster-i18n.ts es label faq: 'Preguntas frecuentes'
fr	faqKicker: 'FAQ'	Questions fréquentes	—
ko	faqKicker: 'FAQ'	자주 묻는 질문	—

Built-HTML check over /{de,es,fr,ja,ko}/security/ at 3a531e1 against the 922605d build: eyebrows with lang="en" 0 on each page (was 1 each); meta robots (noindex, nofollow) and link canonical (/en/security/) byte-identical to 922605d — the pages still inherit other English parts, so their fallback status is unchanged by design. /de/security/ and /ja/security/ screenshotted at 1440 and 390 and looked at; 12 overflow probes, 0 overflow.

Proofs

Source side, non-blog files (content/pages, content/glossary, src/lib, src/content-pages, src/glossary, src/components, src/pages), every locale:

retired token figures  (150k, under-100k, ~50k, 约 50k, 100k tokens, 50k tokens, 15 万 token, 150k-token, roughly 150k)  hits=0
old descriptors        (… target format and runtime|production platform|commercial production|AI-written enterprise software|面向 AI 所写企业软件|商业生产平台|…)  hits=0  (positioning.ts's own comment excluded)
carried post phrasings (commercial production experience|optional ObjectOS|…|Connect ObjectOS to the database you already run)  hits=0 across the three posts, all eight files each
item 6/7 strings       (Questions this page should answer|¥100k|'25+'|16+ …|Platform-native|仅平台自带|support@objectstack)  hits=0

Built output (dist/, 49 non-blog pages per locale × 8 locales): old descriptors 0, item-6 strings 0. The retired-figure grep finds 31 non-blog HTML files, every one of them the business-app-in-16k-tokens post card (its title "A Complete CRM Under 150k" and its description) rendered on three glossary term pages and the locale homes through article cards. That post is blog content and is not touched here: the #209 ruling retired the figures from positioning copy, and a post's own measurement stays, as the #209 sweep recorded ("these are the posts' own measurements"). The authored non-blog copy is clean.

Gates

• pnpm content:lint (default and --published): ✓ content lint passed (339 files, 44 glossary terms checked).
• pnpm check: Result (143 files): 0 errors, 0 warnings, 0 hints — VERDICT command-exit 0.
• pnpm build (gen:zh-hant → content:lint --published → svg:lint → astro build → content:lint --dist): [build] Complete!, ✓ content lint --dist passed (478 built blog pages checked), ✓ content lint --dist links passed (89519 internal links in 880 built pages resolve to built files) — VERDICT command-exit 0.
• pnpm seo:smoke: SEO smoke test passed (870 HTML pages checked).
• pnpm svg:lint: svg-text-fit: OK (112 SVGs, 2244 text boxes).

Browser check

Playwright 1.56.1 with the preinstalled Chromium against astro preview of the build: home, product-tour, agent-developer, data-modeling, security × en, zh-Hans. Full-page screenshots at 1440×750 and 390×844 (20 files, looked at); horizontal-overflow probe at all six AGENTS.md viewports (1366, 1440, 1512, 1920, 390, 360): 60 probes, 0 with document wider than the viewport, all 200. The artifact code blocks scroll inside their own box at phone width, as before. No copy was shortened for layout and no CSS was touched.

Held out, on purpose

• Price figures, annual-billing wording, plan feature lists ([Decision] Pricing headline: www shows $20 / $45 per AI seat "billed annually (about two months free)", docs License & Pricing shows $24 / $54 per month with no annual price — which is the source, and is the annual discount real? #214); signup / "Start" CTA targets ([Decision] Where does "Get started" go? Neither site has an ObjectOS Cloud signup URL, so every plan CTA loops www → docs → www #215).
• MarketingHome.astro's AI-link href (markup) — the label was changed instead.
• A real CTA slot on cluster pages (markup) — the product path is in the self-hosted FAQ text.
• astro.config.mjs, scripts/lib/, CSS.

---

Generated by Claude Code

🤖 Generated with Claude Code

https://claude.ai/code/session_01FeA1nwBz1ohH65dvffUGKr