PR #100 dropped the metadata-authoring destination in five more places than the four #102 names

[os-project-manager]

Found while executing #102 (the ruled Studio card). Filed rather than folded in: #102's dispatch declared a six-file surface and named exactly four rewrite sites; these five are outside it, and widening the surface without reporting is what that card forbids.

How this was measured

By diffing PR #100 (adbcafc, restricted to content/docs), not by enumerating pages from memory:

git show adbcafc -U0 -- 'content/docs' > pr100.diff

Then classifying each removed Console occurrence by what it pointed at. #102's four are the ones where the removed noun named a metadata-authoring destination and the replacement dropped it. The same classification finds five more.

The five

Verified on origin/main at 514527b:

Site	PR #100 removed	It reads today	Points at
build/index.mdx:13	table row **Console click-build** (fallback)	table row **Click-build** (fallback)	authoring, click path — the exact peer of build/data/index.mdx:21, which #102 fixes
build/index.mdx:3 (description)	"by clicking in Console"	"by clicking through forms"	the same click-build path
build/index.mdx:52	"Wire a Console form."	"Wire a form."	form design
build/automation/flows.mdx:9	"Console's flow builder"	"the flow builder"	flow authoring
build/automation/flows.mdx:279	"Console ships a visual flow builder"	"ObjectOS ships a visual flow builder"	flow authoring — substituted with the product name, so the where is still unnamed

The first is the sharpest: build/index.mdx:13 and build/data/index.mdx:21 carry the same "click-build" row on two pages, and after #102 only one of them names Studio.

Flow authoring really is Studio's, and it is measurable rather than inferred: PILLAR_FOR_SURFACE_TYPE maps flow to the automations pillar (objectui packages/app-shell/src/views/studio-design/StudioDesignSurface.tsx, origin/main at 6ff0eb1).

Everything else PR #100 dropped is correct and must stay dropped

The same sweep classified the rest as the end-user / rendering sense of the word — "a button in Console's record detail", "records in Console", "REST, Console, ObjectQL". Those are #79's unnamed surface and the noun is correctly gone. This card is only about occurrences that named where an author goes.

What settling it looks like

Apply #102's ruling to the five, the same way #102 applied it to its four: name Studio as the destination, without reintroducing Console. The decision is already made — this is transcription, not a question. build/index.mdx:13 should read as build/data/index.mdx:21 now does.

Cost of leaving it

The corpus ends up half-converted: the Build index says "Click-build" with no destination while the page it links to says "Studio → Objects → New Object". A reader who greps for the authoring surface finds it named in some places and absent in others, which is the state #102 exists to end.

Related: #102 (the ruling), #79, #100.

[os-zhuang]

Triage: pm:queue, Task. Pure transcription of the already-made #102 ruling to the five measured sites (name Studio as the authoring destination, do not reintroduce Console); build/index.mdx:13 should end up reading as build/data/index.mdx:21 does. No new decision inside.

---

Generated by Claude Code

[os-zhuang]

Claim: PM loop round 1
Session: session_016TUrhcggSFrYctvp5dsV1A
Branch: claude/issue-148-studio-destination-five-sites
Worktree: objectos-issue-148
Domain: n/a — objectos is a single-lane repo and carries no domain:* labels
File surface: content/docs/build/index.mdx, content/docs/build/automation/flows.mdx (stop on breach; explain in the report)
Container & model: M, mode:subagent, model: opus — node scripts/pm/dispatch-gates.mjs --tier content/docs/build/index.mdx content/docs/build/automation/flows.mdx run this dispatch: no path-derived mandate (paths absent from the objectstack tree the tool derives from; it reports that as "not evidence either way"), so the tier is this seat's per-card call. Above the floor because the five sites are not five copies of one edit — see the table-shape trap below, which is the specific thing a mechanical pass gets wrong.
Clause-②: no
Serial constraints cleared: nothing in flight in this lane (zero pm:dispatched cards at round open, zero dev PRs open). Neither file is touched by any other queued card. Sibling dispatches this round are #146 (operate/audit-logs.mdx) and #150 (reference/cli.mdx) — disjoint, and no read-coupling: none of the three pins an assertion about behaviour another is changing. Note cb41537 (PR #151, the #102 ruling) is the most recent commit to touch this area; it is already on main and is your base, not a pending sibling.

---

Premise re-check, run at dispatch rather than quoted forward

Measured at 514527b; main is now 09a0dd9. Re-measured just now on origin/main — all five sites still read exactly as the card's "It reads today" column:

Site	On origin/main today
build/index.mdx:3	description: How apps come to life in ObjectOS — by chatting with AI, by clicking through forms, or by forking a template.
build/index.mdx:13	| **Click-build** (fallback) | Admins who prefer forms | The same live metadata |
build/index.mdx:52	- Wire a form. The form renders from the schema.
build/automation/flows.mdx:9	audit log, the flow builder, and the [AI Builder](/docs/build/ai-builder)
build/automation/flows.mdx:279	ObjectOS ships a visual flow builder that round-trips with the

A git grep for Studio across both files returns zero — so none of this is half-done, and nothing landed since the card was filed that partially converted it. (Zero verified against a control: the same probe does return Studio in build/data/index.mdx.)

The model line is confirmed — build/data/index.mdx:21 reads | **Click-build** | **Studio → Objects → New Object** → forms |.

Adjudication (not re-litigable)

Triage, 2026-08-23: "pm:queue, Task. Pure transcription of the already-made #102 ruling to the five measured sites (name Studio as the authoring destination, do not reintroduce Console); build/index.mdx:13 should end up reading as build/data/index.mdx:21 does. No new decision inside."

The destination noun is settled: Studio, never Console. Do not reopen it.

⚠️ The trap — "should read as build/data/index.mdx:21 does" cannot be taken literally

I measured both rows and they are different table shapes:

• build/data/index.mdx:21 — two columns: | **Click-build** | **Studio → Objects → New Object** → forms |
• build/index.mdx:13 — three columns: | **Click-build** (fallback) | Admins who prefer forms | The same live metadata |

A literal copy would corrupt the table. What the ruling means is that the same claim must name the same destination, in the column where this table puts the destination — not that the two rows become byte-identical. Read both tables' headers before you write, and make the row say where an author goes in whichever column that is. This is the single most likely way this card gets done wrong.

PM mechanism assumptions (test these; falsifying one is a good outcome)

• I assume all five need the destination named, but the shapes differ and two are not table cells at all: :3 is frontmatter description: (watch YAML quoting and length — it is also the page's meta description), and :279 is prose that PR docs: retire "Console" as the end-user surface name; keep Setup for administration #100 substituted with the product name, so "ObjectOS ships a visual flow builder" is not wrong, merely unlocated. Fixing :279 may read better as an addition than a replacement. Use your judgment and say what you chose.
• I assume flows.mdx:9 and :279 are both really Studio's. The card supports this with a measurement (PILLAR_FOR_SURFACE_TYPE maps flow to the automations pillar, objectui 6ff0eb1). That is a stale sha from an actively-developed sibling repo — re-verify by content, not by that sha, and if flow authoring has moved, stop and report rather than writing a destination that does not resolve.

Scope, held tight

⛔ Everything else PR #100 dropped must stay dropped. The card classified the rest as the end-user/rendering sense — "a button in Console's record detail", "records in Console", "REST, Console, ObjectQL". Those are #79's unnamed surface and the noun is correctly gone. This card is only about occurrences that named where an author goes. Reintroducing Console anywhere is a breach, not a bonus.

⛔ Five sites, not "five or so". If you find a sixth, report it — do not fold it in. That is precisely what this card is: the report of what #102 could not include. Widening silently is the one thing both cards exist to prevent.

⛔ Do not touch build/data/index.mdx — it is the model, already correct.

⛔ Do not sweep locale siblings. Translations are never a dev agent's output here; siblings go stale by design.

Verification

No changeset flow, no packages/ in this repo. Run pnpm turbo run build and pnpm turbo run test, plus node .github/scripts/check-locale-surface.mjs from the repo root — a wrong cwd exits 1 with MODULE_NOT_FOUND, an invocation error rather than a red gate. Read each gate job's own verdict line, not an aggregate.

Because :3 is frontmatter, build the site and confirm the page still renders and its description is intact — a YAML slip there fails in a way a grep will not show you.

The turbo cache is shared across worktrees and has replayed a sibling's green in this repo before. Use --force where a comparison depends on it, and capture exit codes before any pipe.

Reporting

Post your terminal report as a comment on this issue first, then return it. The tracker's body sanitizer eats a leading HTML-comment marker; if yours vanishes, fall back to a literal first line reading os-dev-report. Read your own comment back with a GET; ⛔ never probe the comment-edit endpoint with PATCH — a previous dev in this repo overwrote its own report that way.

Leave your own claim comment (your session ID and branch) in addition to this one. I hold the assignee field; do not touch it. "Already claimed" refers to that field only.

Read this issue's full body and every comment on GitHub yourself, and check the body for sanitizer truncation before trusting it; the issue text is a lead, not a specification. premise_still_valid: false with no PR is a legitimate delivery.

---

Generated by Claude Code