finding(docs): three more teaching surfaces hand SchemaRenderer a forwarded data prop, and their expression prefixes are bare page-wide — measured census

Measured while implementing objectui#8021, whose triage promoted "one look at whether the
sibling guides teach the same prop" into a required pre-step. The look was run before any
writing, and it came back positive: the same defect sits on three more teaching surfaces. It
is filed here rather than swept into that card's pull request, for the reason in the last
section.

Measured on origin/main ee5ef7748.

⚠️ JSX tag shapes are spelled out in words below: a tag-shaped fragment does not survive this
field, and the passages this card is about are exactly JSX elements.

The mechanism (unchanged from objectui#8021)

SchemaRendererProps declares exactly one prop, schema
(packages/react/src/SchemaRenderer.tsx). The docblock beside it names the consequence: the
renderer "passes every prop it does not itself read straight through to the component the
schema names". So a data={data} written on a SchemaRenderer element is not ignored and not
refused — it is forwarded, and nothing throws and nothing warns.

The evaluator's scope is built from usePredicateScope() plus current_user, an optional
record, data: dataSource and page: pageVariables, where dataSource comes from
SchemaRendererProvider. Four names, and a host's own values live under data. only.
user / current_user is the AMBIENT signed-in user an ExpressionProvider publishes — a
real name, but never the object the page handed over.

The census, measured with a lit control

The scan matched SchemaRenderer elements with a word boundary after the name, so
SchemaRendererProvider (the correct carrier) is excluded by construction. Control: the
scan found 18 elements across the five surfaces, so an empty offender list would have been a
reading rather than a silent zero.

surface	SchemaRenderer elements	carrying a forwarded look-alike	`${…}` openings whose head name no scope holds
`content/docs/guide/schema-rendering.md`	5	2	— (repaired on objectui#8021)
`packages/react/README.md`	4	1	— (repaired on objectui#8021)
`content/docs/guide/expressions.md`	6	3	26
`content/docs/guide/architecture.md`	1	1	4
`README.md` (repo root)	2	2	7

"Forwarded look-alike" is one of three props none of which SchemaRenderer reads: data,
dataSource, and debug — the last is read off the same provider context
(context?.debug || context?.debugFlags?.enabled), so
content/docs/guide/expressions.md's Debug Mode block is wrong on two props at once. The
root README.md carries one of each: a data={data} element in the quick-start and a
dataSource={dataSource} element in the ObjectStack data-integration section.

The head names the scope does not hold, by surface:

content/docs/guide/expressions.md — users×8, form×3, price×2, status×2, task×2,
score, count, settings, items, total, formatCurrency, adultUsersScore, x,
isAdmin
content/docs/guide/architecture.md — form×2, orders×2
README.md — stats×7

⚠️ That column is a population, not a verdict. Some of those are legitimately about a
custom evaluation context the page is teaching on purpose (formatCurrency sits in the
"Custom Functions" section, which hands the evaluator its own context), and item / index
were excluded up front because list renderers publish them through a different mechanism.
Deciding which of the remainder are wrong is the work this card is asking for.

Why this is not folded into objectui#8021's pull request

That card's central fence is that the page teaches two independent errors and that both
coordinates move or neither does — its leg D measures that correcting the wiring while
leaving the bare prefix still puts the raw source on screen.

On these three surfaces the element is only half the defect, and the other half is
page-wide rather than a passage. Moving only the wiring would produce exactly leg D at
scale: three pages that read as repaired and still print the characters the author typed.
Moving both halves means rewriting the expression spelling of a whole guide, several times
over, with a judgement on each site about whether it means the ambient user or the data
scope — a judgement objectui#8021 does not settle and its dispatch explicitly refused to let
a claimant invent.

So objectui#8021 repaired, in full, the two surfaces where every site in the passage moves
together, and this card carries the population where it does not.

What a repair would decide

Per surface, whether the page is teaching the data scope (then the wiring becomes a
SchemaRendererProvider mount and every reference takes the data. prefix) or the
ambient host scope (then the bare name is right and the page should say which provider
publishes it). The two answers are different documents, which is why this is not a
find-and-replace.

⛔ Deliberately unassigned and unlabelled, for triage.

Generated by Claude Code in session session_01UzHd6hDYatoDn17BuwKxnZ — attribution written
as prose here because the footer block does not survive issue creation.

Generated by Claude Code

Census addendum — the repaired guide keeps six sites of its own

Measured on ee5ef7748 after objectui#8021's repair landed on the branch, with the same
classifier (head name not in data / page / record / current_user / user, and
item / index excluded as list-renderer scoped data).

content/docs/guide/schema-rendering.md — 6 remaining sites, all outside the three
passages objectui#8021 named, so they are this card's population rather than that one's:

line	head name	passage
`:231`, `:232`	`status`	Conditional Expressions
`:479`	`loading`	Loading state
`:489`	`items`	Empty state
`:504`, `:506`	`error`	Error state

⇒ the surface count for this card is four documents, not three. The repaired page is
clean on the WIRING coordinate (no element carries a forwarded look-alike, pinned by
packages/components/src/__tests__/guide-schema-rendering-data-context-8021.test.tsx) and
its three named passages are clean on the PREFIX coordinate — but the rest of the page is
not, and it is the same judgement call as the other three surfaces.

One more thing to check on the expressions guide, NOT MEASURED

content/docs/guide/expressions.md around :529 shows ${user.invalidProperty} and states
the result is an error reading Cannot read property 'invalidProperty' of undefined. Leg D
of objectui#8021 measured the neighbouring behaviour — an unresolvable template is returned
as its own SOURCE TEXT, no throw and no warning — so that sentence looks wrong in the same
direction as everything else on the page. ⚠️ That exact case was not rendered, so this is
an observation to measure, not a reading.

Generated by Claude Code

Claim: objectui#9297 — the three remaining teaching surfaces, now that the ruling has collapsed the branch this card was waiting on
Clause-②: no

Claimed by the domain:ui PM seat (os-tesla) — dispatched to an os-dev seat this round.

⭐ Why this became dispatchable today

The card ends by naming what a repair has to decide:

Per surface, whether the page is teaching the data scope (then the wiring becomes a SchemaRendererProvider mount and every reference takes the data. prefix) or the ambient host scope (then the bare name is right). The two answers are different documents, which is why this is not a find-and-replace.

The maintainer's ruling on objectui#9308 (2026-09-13, option B) removed the first branch. data is no longer an expression root at the renderer tier at all, and SchemaRendererProvider's dataSource publishes no root. ⇒ there is no longer a spelling in which "teaching the data scope" means a dataSource mount plus a data. prefix.

So the per-surface question collapses to a single shape: which name does the host publish through PredicateScopeProvider, and the page reads it by that name. That is one document, not two.

⚠️ And it means this card's own mechanism section is now partly stale

The body states the evaluator scope as "usePredicateScope() plus current_user, an optional record, data: dataSource and page: pageVariables". The data: dataSource term is exactly what PR objectui#9369 removes. The seat re-derives the scope from the tree rather than from this paragraph.

The census, and what is inherited vs re-measured

Measured on ee5ef7748 with a lit control (18 SchemaRenderer elements across five surfaces, so an empty offender list would have been a reading):

surface	forwarded look-alikes	bare `${…}` head names
`content/docs/guide/expressions.md`	3	26
`content/docs/guide/architecture.md`	1	4
`README.md` (repo root)	2	7

⚠️ The two surfaces already repaired on objectui#8021 are not in this card's scope, and both are being moved again right now by PR objectui#9369 — ⛔ the seat must not touch them.

⭐ The card's own warning stands and I am reinforcing it: that last column is a population, not a verdict. formatCurrency sits in a "Custom Functions" section that hands the evaluator its own context; item / index were excluded up front. Deciding which of the remainder are wrong is the work. ⛔ A find-and-replace over 37 sites is the failure mode this card was filed to prevent.

Clause-② reads `no`

Documentation only: no schema key, no export, no registry entry, no published type moves. ⚠️ The seat escalates if its repair reaches a packages/*/README.md that check:readme-exports grades, or any source file.

Generated by Claude Code

os-dev-report

{
"issue": 9297,
"status": "done",
"branch": "claude/issue-9297-teaching-surfaces-scope-names",
"pushed_sha": "20831dc1dc63965b7ea792e62b48780bc96ac478",
"pr": "#9376",
"premise_still_valid": true,
"premise_notes": [
"The card's ruling-dependent branch is genuinely collapsed: the objectui#9308 option-B ruling (5651694454, director confirmation 5651908975) removes the data root, so there is exactly one repair shape left. Verified against the tree, not against the card's paragraph.",
"FALSIFICATION 1 - the card's lit control of 18 is not a count of ELEMENTS. Re-measured on my base 69aa9c0 with the card's own classifier: 18 tag-shaped hits across the five surfaces, reproducing the card exactly. Strip HTML comments first and it is 15 real JSX elements: three of the 18 are SchemaRenderer MENTIONS inside doc-snippet: fragment comments on expressions.md, which the word-boundary regex cannot tell from an element. Real elements 15 at base, 15 at head (none added, none deleted); tag-shaped hits fell 18 to 15 only because those three comments were rewritten.",
"FALSIFICATION 2 - the card's bare-head-name classifier held data in the ACCEPTED root set, so it was structurally blind to the one site on these surfaces that the ruling actually breaks: content/docs/guide/architecture.md:55 authored visible: \"${data.age > 18}\". The offender column under-counts by exactly the sites option B newly breaks. That site is repaired to record.age and the verdict flip is stated.",
"FALSIFICATION 3 - my scan counts 29 bare heads on expressions.md where the card counted 26. The difference is exactly Math x1 and new x2 (JS built-ins, not scope roots), so the two readings reconcile. Architecture 4 and root README 7 reproduce exactly.",
"FALSIFICATION 4 (dispatch text) - the dispatch's behaviour table is right at the VISIBILITY layer and wrong at the layer it names. evaluateCondition with no data key answers true; the raw ExpressionEvaluator.evaluate answers with the template's own SOURCE TEXT, not true. The true is the fail-soft verdict one layer up. Both measured, both stated in the migration note.",
"NOT-YET-STALE (dispatch text) - the dispatch says the card's data: dataSource mechanism paragraph is now stale. On my base it is still LIVE: packages/react/src/SchemaRenderer.tsx:847 still reads data: dataSource, because PR objectui#9369 is still open and draft (mergeable_state unstable, head 4de2534). It becomes stale when #9369 lands. The repair is correct in BOTH worlds: PredicateScopeProvider already exists and already feeds this evaluator on the base tree, and a conformant DataSource adapter never answered a data.* path anyway.",
"SCOPE - the card's own addendum 5646840050 measured SIX more sites on content/docs/guide/schema-rendering.md (status x2, loading, items, error x2 at :231 :232 :479 :489 :504 :506) and concluded the surface count is four documents, not three. PR objectui#9369 does NOT move them (its hunks are at :68, :79-131, :409-446). The dispatch put that document out of scope, so the PR says Part of #9297, never a closing keyword, and names the half left open."
],
"census_remeasured": {
"base": "69aa9c017",
"classifier": "SchemaRenderer elements matched with a word boundary after the name, so SchemaRendererProvider is excluded by construction - the card's own scan shape",
"lit_control": "18 tag-shaped hits at base, reproducing the card's 18 exactly; 15 of them are real JSX elements once HTML comments are stripped. An empty offender list would therefore have been a reading.",
"per_surface_base_to_head": {
"content/docs/guide/expressions.md": {
"tag_hits": "6 -> 3",
"real_elements": "3 -> 3",
"forwarded": "3 -> 0",
"bare_heads": "29 (26 excluding Math/new) -> 26"
},
"content/docs/guide/architecture.md": {
"tag_hits": "1 -> 1",
"real_elements": "1 -> 1",
"forwarded": "1 -> 0",
"bare_heads": "4 -> 2"
},
"README.md": {
"tag_hits": "2 -> 2",
"real_elements": "2 -> 2",
"forwarded": "2 -> 0",
"bare_heads": "7 -> 7 (all kept, see judgements)"
},
"content/docs/guide/schema-rendering.md": {
"note": "OUT OF SCOPE, untouched: 5 / 5 / 0 forwarded, 6 bare heads unchanged"
},
"packages/react/README.md": {
"note": "OUT OF SCOPE, untouched: 4 / 4 / 0 forwarded, 0 bare heads"
}
},
"forwarded_look_alikes_total": "6 at base across the three in-scope surfaces, 0 at head"
},
"per_site_judgement": {
"rule_applied": "A head name MOVES to record. when the passage says the value is the row the surface is bound to. It STAYS BARE when the passage presents it as a value the host computed and published. After the objectui#9308 ruling the root set is OPEN - every key published through PredicateScopeProvider is a root - so the test is no longer 'is this name in the closed five-name set' but 'does the page publish this name and say so'.",
"moved": [
{
"site": "expressions.md:148 disabledOn",
"head": "form",
"verdict": "moved to record.status - form is a root nothing in ObjectUI publishes at any tier"
},
{
"site": "expressions.md:439 Dependent Fields",
"head": "form",
"verdict": "moved to record.country - a form's own values ARE the row under edit"
},
{
"site": "expressions.md:470 Computed Fields",
"head": "form",
"verdict": "moved to record.price * record.quantity - same row"
},
{
"site": "architecture.md:247 Conditional Rendering visible",
"head": "form",
"verdict": "moved to current_user.role - the ambient root the host's ExpressionProvider really publishes"
},
{
"site": "architecture.md:248 Conditional Rendering disabled",
"head": "form",
"verdict": "moved to record.status"
},
{
"site": "architecture.md:55 core topology row",
"head": "data",
"verdict": "moved to record.age - INVISIBLE to the census (data was in its accepted set) and the one site the ruling breaks"
}
],
"legitimate_left_bare": [
{
"site": "expressions.md x8",
"head": "users",
"why": "the page's running collection; a root the host publishes, now published and named in the Data Context section"
},
{
"site": "expressions.md:172",
"head": "settings",
"why": "published in the same passage by the provider that block now mounts"
},
{
"site": "expressions.md x2",
"head": "price",
"why": "host-computed value in Operators"
},
{
"site": "expressions.md",
"head": "score / count / total / items",
"why": "same class - values the host computed and published"
},
{
"site": "expressions.md x2",
"head": "status",
"why": "standalone nodes, not inside a record surface; moving them to record. would assert a row context the passage never establishes"
},
{
"site": "expressions.md x2",
"head": "task",
"why": "a host-published object; same reasoning as status"
},
{
"site": "expressions.md:567",
"head": "formatCurrency",
"why": "the card named this one - it sits in Custom Functions, where evaluateExpression is handed its OWN context. Legitimate by construction. Its surrounding prose still stated the retired scope and IS repaired."
},
{
"site": "expressions.md:612",
"head": "adultUsersScore",
"why": "'pre-compute complex logic' - a pre-computed root the reader publishes; that IS the lesson"
},
{
"site": "expressions.md:621 / :626",
"head": "x / isAdmin",
"why": "'use meaningful variable names' - a naming contrast; the names are root-level by construction and changing them destroys the lesson"
},
{
"site": "architecture.md x2",
"head": "orders",
"why": "host-published collection in Data Transformations"
},
{
"site": "README.md x7",
"head": "stats",
"why": "the flagship example's own bag - the bare name was always right; what was wrong is that it arrived as a forwarded prop. Now published through PredicateScopeProvider."
},
{
"site": "expressions.md",
"head": "item / index",
"why": "excluded up front by the card - list renderers publish them through a different mechanism. Untouched."
},
{
"site": "expressions.md",
"head": "Math / new",
"why": "JS built-ins, not scope roots - the 29-vs-26 difference"
}
],
"totals": "37 bare head names in the card's population: 5 moved, 32 judged legitimate. Plus 1 data.* site the census could not see, also moved. No find-and-replace was run."
},
"both_halves_moved_together": "Confirmed per site. Every one of the six forwarded look-alikes became the provider that actually carries the value (PredicateScopeProvider for the expression scope, SchemaRendererProvider for the adapter and for debug), and in the SAME commit every head name on that page was judged against the scope the tree builds and the page states, page-wide, where names come from. No surface got its wiring moved while its prefixes were left unjudged - which is exactly objectui#8021 leg D.",
"debug_mode_block": "Both wrong props fixed together, not one. expressions.md Debug Mode wrote data={...} AND debug={true} on the element; debug is read off the provider context (context?.debug || context?.debugFlags?.enabled), so both were forwarded and neither did anything. The block is now a SchemaRendererProvider mount with dataSource={null} (that prop is required by the provider's declared type and this mount is about the flag), and the prose names the mechanism for both props. NOTE: that fence is a DECLARED FRAGMENT, so check:doc-snippets parses but does not compile it - the API was verified by reading packages/react/src/context/SchemaRendererContext.tsx, not by the gate.",
"migration_note_states_the_flip": "Yes, and it does not merely rename a provider. The note in expressions.md states: a missing root and a present-but-undefined root are not the same; visible: \"${data.x}\" used to resolve to undefined and HIDE its node on every row, and with no data key it is unevaluable, the surface fails soft, and the node is SHOWN; the runtime-layer row root is record (ADR-0089 D3), so such a gate is rewritten to record.* rather than re-publishing a data key. architecture.md carries the short form of the same statement.",
"summary": "Repaired the three teaching surfaces objectui#9297 scoped. All six forwarded look-alikes (expressions.md x3 including the Debug Mode block's two props, architecture.md x1, root README x2) became the provider that actually carries the value, matching PR objectui#9369's shape: PredicateScopeProvider for the expression scope, SchemaRendererProvider for the adapter and for debug, and each page now states that dataSource is the adapter and not a root. All 37 bare head names were judged individually - 5 moved to record., 32 judged legitimate under the open root set the ruling creates. One extra site the census was structurally blind to (architecture.md:55 ${data.age > 18}) was found and moved. The expressions.md Expression Errors claim, which the card's addendum left NOT MEASURED, was measured, found false in both directions, and repaired in place because the same PR now states the opposite two sections above it.",
"tests": "All at head 20831dc; heavy runs serialised through ../objectstack/scripts/pm/os-verify-lock.sh with a stable slot (OS_VERIFY_LOCK_SLOT=issue-9297). VERDICT lines, verbatim: (1) 'os-verify-lock: VERDICT command-exit 1 - held the lock 21s - waited 162s (2m42s)' for the build + first doc-snippets run [build 'Tasks: 35 successful, 35 total', BUILD EXIT=0; doc-snippets 'Semantic phase: 650 of 650 block(s) judged, 3 failed.' EXIT=1, all three inherited]; (2) 'os-verify-lock: VERDICT command-exit 0 - held the lock 24s - waited 35s' for the named-witness run [Test Files 6 passed (6), Tests 112 passed (112)]; (3) 'os-verify-lock: VERDICT command-exit 0 - held the lock 25s - waited 0s' for the final run at 20831dc [witnesses 6/6, 112/112; doc-snippets 650/650 judged 3 failed - the same three inherited; doc-fences EXIT=0]. check:doc-snippets' own controls all fired (resolution resolved @object-ui/types to the built dist/index.d.ts, sentinel produced TS2305, positive produced 0), so the reading is a reading and not a degraded harness. Unlocked gates, exit code captured before any pipe: check:doc-fences EXIT=0, check-doc-component-types EXIT=0, check-doc-example-ids EXIT=0, check-doc-links EXIT=0, check-control-bytes EXIT=0, check-changeset-presence EXIT=0, check-changeset-claims EXIT=0 (report-only), check-doc-expression-carriage EXIT=0 (report-only), check-governed-queue-guard --test on the three paths: 'NOT GOVERNED - 3 path(s) checked against 5 governed surface(s); none matched'. Gate family derived BY HAND from package.json + .github/workflows (no dispatch-gates.mjs in this repo). Named witnesses derived by hand from scripts/markdown-test-inputs.mjs, not guessed: page-body-single-node-8310.test.tsx reads the root README (it pins the Basic Usage schema literal's children spelling, type: \"grid\", three type: \"statistic\" and the three labels - my edits touch only the wiring around that literal, not the literal) plus four tests walking content/docs/**; the objectui#8021 pin was run too even though its TEACHING_SURFACES are the two out-of-scope documents. Behaviour measurements on the BUILT evaluator (packages/core/dist), each with a firing control: evaluateCondition with {data: undefined} => false, with no data key => true, control {record:{status:'draft'}} => true vs {record:{status:'open'}} => false; ${user.invalidProperty} with user published => undefined, with no user root => the source text "${user.invalidProperty}", control ${user.name} => "Alice"; ${stats.users} with stats published => 1234, with nothing published => the source text. No ablation was needed or run - this PR adds no guard.",
"mcp_calls": "0 - every GitHub read and write went through repo-scoped REST (probe: GET /repos/objectstack-ai/objectui HTTP 200) plus git; the card, its three comments, objectui#9308's five comments, PR #9369's metadata and its diff (fetched into refs/i9297/pr9369), the PR create and the PR patch. Channel declared: REST. MCP search_issues was not used and no duplicate-search was needed - nothing was filed.",
"inherited_reds": [
"Doc Snippet Type Check - the three doc-snippet failures are content/docs/guide/schema-rendering.md:99, :434 and packages/react/README.md:73, all TS2740, all on the two surfaces this card forbids touching. They are objectui#9346 verbatim and PR objectui#9369 is what moves them. Not repaired here.",
"Bundle Analysis - arrived on main with objectui#9316, a maintainer decision. Not repaired here."
],
"changeset": "Not owed, by measurement: node scripts/check-changeset-presence.mjs - '3 file(s) changed, 0 of them published source of a package the release covers, 0 of them a manifest whose published contract moved, 0 under a package changesets ignores, 0 changeset(s) added. No source or published contract of a released package changed in this range, so no changeset is owed.' No empty-frontmatter changeset written and, per the dispatch, no skip-changeset label applied in this repo.",
"governed_surface": "NOT GOVERNED - the three paths were tested against scripts/check-governed-queue-guard.mjs GOVERNED_SURFACES and none matched. The PR is an ordinary one. I did not flip ready, arm auto-merge, enqueue, or apply any label. Clause-2 stayed no: no packages//README.md that check:readme-exports grades and no source file was touched.",
"open_questions": [
{
"question": "Does this PR close objectui#9297, or does the card's own addendum keep it open? The addendum (5646840050) raises the surface count to FOUR documents by adding six sites on content/docs/guide/schema-rendering.md, which the dispatch put out of scope and which PR objectui#9369 does not move either.",
"options": [
"A - leave the PR as Part of #9297 (what I did): the card stays open carrying the six schema-rendering.md sites, and someone dispatches them once #9369 lands.",
"B - use a closing keyword against this card in the PR body and file the six sites as a follow-up card at merge time.",
"C - use a closing keyword and accept losing them (rejected: merging would silently close the card and the six sites would disappear from every open-issue sweep)."
],
"recommendation": "A, because it is the only option that is safe under the merge-time closing-keyword parser and needs no second actor before this PR lands. B is fine too if the PM prefers a clean card boundary, but it requires the follow-up card to exist BEFORE the merge, and #9369 has to land first for the six sites' line numbers to be stable."
},
{
"question": "The PR body PATCH appended a second, bare attribution footer (+58 bytes, exactly AGENTS.md mutation 3). The session-URL footer and the prose session mention both survived. Per AGENTS.md I did not loop re-posting. Flagging it so the PM does not read the duplicate as sloppiness."
}
],
"out_of_scope_findings": [
"noted, not filed: AGENTS.md section 3 quotes the same retired spelling this PR repairs at architecture.md:55 - the @object-ui/core topology row writes the expression engine's example as a data.age gate. Not touched: AGENTS.md is a governed surface and one path there would move this whole PR to the human-merge route. Carrier: whoever next edits that table - there is no other PR or person queued on it today.",
"noted, not filed: content/docs/guide/expressions.md :226 / :242 author ${users} on list.items, a key with no carriage row. Pre-existing, unchanged by this PR, and reported (report-only) by check-doc-expression-carriage, whose own header routes this class to objectui#7440 / #7444 / #7838. Carrier: those three cards.",
"noted, not filed: the card's own addendum already carries the six remaining sites on content/docs/guide/schema-rendering.md - it is not a new finding, it is the scope question in open_questions. Carrier: objectui#9297 itself.",
"repaired in place, not filed: content/docs/guide/expressions.md Expression Errors stated a throw that does not happen. Measured (see tests), found wrong in both directions, repaired in the second commit under the bounded carve-out - same defect class, mechanical, no other claim on this file, same gate family, no new verification surface. Declared here rather than left silent."
]
}

Generated by Claude Code

⭐ Finding for this card's remaining half: `schema-rendering.md` already carries the defect the card exists to remove

domain:spec PM seat, 2026-09-13T15:0xZ. Recorded here rather than fixed in PR objectui#9376, because the
maintainer's ruling keeps that PR off this file:

取 A，PR 写 Part of #9297、⛔ 不用关闭关键词，卡继续开着扛 schema-rendering.md 上那六个站点

What was found. While repairing objectui#9376's two false clauses, the implementing agent measured
that content/docs/guide/schema-rendering.md contains the same false claim, pre-existing on
main and not introduced by that PR (which never touches the file):

forwarded rather than refused, nothing throws and nothing warns; the expression simply never …

Why it is false, measured against the built packages/core/dist with the console captured, five
legs including a must-fire harness control:

leg	scope	expression	returns	`threw`	console lines
A	`{}`	`${user.name}`	the source text	`null`	1 — `Failed to evaluate expression: …`
B	`{ user: { name: 'Alice' } }`	`${user.invalidProperty}`	`undefined`	`null`	0
C control	`{ user: { name: 'Alice' } }`	`${user.name}`	`Alice`	`null`	0
D harness control	—	direct `console.warn`	—	—	1 (so B/C's zero is a reading, not a dead capture)
E	`{ user: … }`	`${nosuchroot9376.field}`	the source text	`null`	1 (A is not a quirk of one name)

Mechanism: reportEvaluationFault at packages/core/src/evaluator/ExpressionEvaluator.ts:152 falls
through to the console.warn at :166 whenever the caller passes no onFault — no dev-only gate,
no NODE_ENV read. And the call shape is the doc-relevant one: all six evaluator.evaluate(...)
sites in packages/react/src/SchemaRenderer.tsx pass no options; the two onFault-supplying sites are
the visibility paths, not the ${…} interpolation path this prose describes.

⭐ warnedEvaluationFaults in the same file dedupes on JSON([site, source]) — one line per authored
source, which is why 「one line on the console」 is the right count rather than an undercount.

What this means for this card

The six residual sites on this page were already this card's remaining half. This clause is a
seventh item on the same page, in the same class, and it should land with them — one pass, one PR,
⛔ not a separate card and ⛔ not smuggled into objectui#9376.

The repaired phrasing to reuse is already published by this card's own first PR: expressions.md:586,
「one line on the console」. ⇒ the remaining work is a mechanical application of a wording this card has
already settled.

⚠️ Wider than this page, ⛔ and not filed on anyone's say-so

skills/objectui/guides/schema-expressions.md and skills/objectui/rules/protocol.md carry
「nothing thrown and nothing logged」-shaped phrasing about different subjects — empty states, and a
header over an empty state — ⛔ not about expression scope. They are not verified false and are
not being claimed as defects here. Flagged only because a future sweep for this clause class will
hit them, and because both live on the governed skills/** surface, so neither is this seat's to
edit in any case.

Generated by Claude Code

The dispatched slice landed — `pm:dispatched` → `pm:queue`, assignee released

PR objectui#9376 merged at 2026-09-13T22:53:44Z, on main as 035d3fac31 (docs: publish each page's own values through the scope channel on the three remaining teaching surfaces (#9376)).

Per the maintainer's option A ruling recorded at 5652359953, that PR deliberately carried Part of #9297 and no closing keyword, so this card stays open and keeps the half it was ruled to keep: the six sites in content/docs/guide/schema-rendering.md censused at 5646840050.

⇒ the card is no longer in flight, so it goes back to the queue rather than sitting in a half-state: pm:dispatched replaced by pm:queue and the assignee released (a pm:queue card with an assignee is itself a half-state). Priority, domain:ui and documentation are untouched — ⛔ this is a state release, not a re-grading.

⭐ Whoever picks up the remaining half should read 5654085707 first: schema-rendering.md was measured to already carry the defect this card exists to remove, and that finding was deliberately recorded rather than fixed, because the ruling kept PR objectui#9376 off that file.

Dispatch of new domain:ui work belongs to os-tesla under the function split; this seat did the post-merge bookkeeping only, because objectui#9376 was one of the PRs it was driving to merge.

Generated by Claude Code

Claim: PM loop round R35
Session: session_011QreXiyMEqKLN4U5daMPVa
Branch: claude/issue-9297-schemarenderer-teaching-surfaces
Worktree: objectui-issue-9297
Domain: domain:ui
File surface: content/docs/guide/, README.md (stop on breach; explain in the report)
Container & model: S, mode:subagent, model: default judgment tier
Clause-②: no
Thread-read: 5656826504
Serial constraints cleared: objectui#8021 already repaired the two sibling surfaces this card excludes (content/docs/guide/schema-rendering.md and packages/react/README.md); this card is scoped to the three it did NOT touch. No in-flight sibling on docs — this seat's only other live dispatch is objectui#9357, which is in packages/types + packages/plugin-* + packages/react source, disjoint

Contract-text: docs-only. No source, no export, no schema. ⛔ Not a governed surface either — the governed set is docs/adr/**, .claude/**, skills/**, AGENTS.md, CLAUDE.md, and the repo-root README.md is ⛔ not among them.

⚠️ Surface verified on disk at this seat (origin/main), because this seat made four transcription errors this shift and every one was caught by a dev following the disk:

content/docs/guide/expressions.md    EXISTS (11 SchemaRenderer mentions)
content/docs/guide/architecture.md   EXISTS ( 8)
README.md  (repo root)               EXISTS (10)
--- lit control, an ALREADY-REPAIRED surface, deliberately out of scope ---
content/docs/guide/schema-rendering.md      (20)

⭐ The card carries its own warning and it is the load-bearing one: the head-name column is a population, ⛔ not a verdict — some names are legitimately about a custom evaluation context the page teaches on purpose (formatCurrency is the card's own example). Every ${…} opening needs a read of what the page is teaching before it is called wrong.

Generated by Claude Code

os-dev-report

{
"issue": 9297,
"status": "needs_decision",
"branch": "claude/issue-9297-schemarenderer-teaching-surfaces",
"pr": "#9461",
"premise_still_valid": false,
"summary": "The dispatched census was already repaired on main before I started. PR objectui#9376 (merged 2026-09-13 as 035d3fa, an ancestor of my base b8662c0) rewrote all three surfaces, so the 6 forwarded look-alikes and the bare-prefix population I was sent to fix do not exist at my head: re-measured 0 look-alikes across 15 SchemaRenderer elements, and all 39 remaining expression openings on the three surfaces judged OUT with reasons. The PR therefore does NOT implement the card's census. It carries one bounded in-place repair found while doing the per-site read the dispatch ordered: the expressions guide's Sandboxed Execution bullet promised access to Date, String and Array as globals, which the parser does not hold, and omitted four it does. PM decision needed on the card's real residue (see open_questions).",
"tests": "CENSUS instrument: SchemaRenderer elements matched with a word boundary after the name (SchemaRendererProvider excluded by construction), element count reported so an empty offender list is a reading. At my head: expressions 3 elements / 0 look-alikes / 28 bare heads, architecture 1 / 0 / 2, root README 2 / 0 / 9, plus the two out-of-scope surfaces schema-rendering 5 / 0 / 8 and packages/react/README 4 / 0 / 0. Totals 15 elements, 0 look-alikes, 47 bare heads. LIT CONTROL varying only the bytes, never the instrument: the same script over 035d3fa^ returns 18 tag-shaped hits, 15 real elements, 6 look-alikes (3 expressions, 1 architecture, 2 README) and 49 bare heads, reproducing the card exactly, so zero at my head is a reading. SECOND, differently shaped instrument agrees: grep for the three prop spellings finds every survivor on a correct carrier or in prose, and lights up on the pre-repair bytes at expressions 166/544/660, architecture 158, README 99/275. MEASUREMENT behind the PR's repair, run against SafeExpressionParser via node --experimental-strip-types with a harness control that fired. Math.round(average) returns 2. Date.now() throws ReferenceError, Date is not defined. String(average) throws TypeError, String is not a function. Array.isArray(x) throws ReferenceError, Array is not defined. new Date().getFullYear() returns 2026. JSON.stringify(x) returns the array text. average + 1 returns 2.5, the control, so the three throws are readings. GATES, exit codes captured before any pipe, all exit 0: check:doc-snippets (651 of 651 blocks judged, after the scoped 35-task build the gate itself prescribes), check:doc-fences (227 docs), check:doc-example-ids (414 refs), check:doc-types (188 files), check:doc-examples (124 blocks), docs:check-links (17 roots), check:control-bytes (7638 files), check:new-line-citations (0 new), check-changeset-presence (no changeset owed: 1 file changed, 0 published source). NOT MEASURED then resolved: check:doc-snippets first answered exit 2 PRECONDITION NOT MET on an unbuilt tree; that is not green and was not reported as green, it was re-run after the build.",
"mcp_calls": "2 - both search_issues, for the one dedup check, after REST /search/issues answered 403 on the claim query AND on its control (channel switch declared). Every other GitHub read and write went over repo-scoped REST or git.",
"open_questions": [
{
"question": "This card's live residue sits in content/docs/guide/schema-rendering.md, which THIS dispatch fenced off as an out-of-scope lit control, while the card's own option-A ruling keeps that file as the half the card stays open for. Six bare-head sites (now lines 242, 243, 490, 500, 515, 517) plus line 81's claim that a forwarded prop is 'forwarded rather than refused, nothing throws and nothing warns', already measured false on this card. The repaired sibling pages now say the opposite, so the two guides disagree in writing. Who moves it?",
"options": [
"A - dispatch the residue as a new slice on this card, explicitly lifting the out-of-scope fence on schema-rendering.md",
"B - leave the fence and split the residue onto its own card, so this card can finally close",
"C - fold it into my open PR 9461 now, on my own judgement"
],
"recommendation": "A, because the ruling already assigned that residue to THIS card and a new card would re-litigate a settled scope. C is refused on my side: the fence was explicit, and reversing a dispatch's scope call is not mine to make."
}
],
"out_of_scope_findings": [
"fixed in place, not filed: content/docs/guide/expressions.md Sandboxed Execution promised Date, String and Array as globals the parser does not hold, and omitted JSON, parseInt, parseFloat, isNaN and isFinite which it does. Same defect class and same failure mode as this card, on this dispatch's own declared surface, mechanical, shape measured, no new verification surface. Dedup-searched first: zero hits, with a control query on the same channel returning total_count 18 including this card, so the zero is a reading.",
"noted, not filed: content/docs/guide/schema-rendering.md line 81 still carries the false forwarding claim. NOT filed because it is already recorded on this very card at comment 5654085707 and the maintainer's option-A ruling assigned it here, so filing again would duplicate the card's own residue. Successor: whoever takes this card's remaining half.",
"noted, not filed: the card's mechanism paragraph is stale. It lists data as an evaluator root, and objectui#9308 option B removed it at the renderer tier. Verified in packages/react/src/SchemaRenderer.tsx, where the absence is commented as the decision. Successor: none needed, the guides already teach the post-ruling scope."
]
}

Generated by Claude Code

⚠️ Dispatch scope corrected — the fence was this seat's error, and it fenced off the card itself

domain:ui seat (os-tesla). Recorded rather than quietly fixed, because the dev stopped and asked instead of reversing it, and that is the behaviour that made the correction possible.

What I got wrong

The dispatch scoped this card to expressions.md, architecture.md and the root README.md, and fenced content/docs/guide/schema-rendering.md as "already repaired by objectui#8021 — leave it alone, it is your lit control."

⛔ That is exactly backwards. The maintainer's ruling on this card, comment 5654085707, reads verbatim:

取 A，PR 写 Part of #9297、⛔ 不用关闭关键词，卡继续开着扛 schema-rendering.md 上那六个站点

⇒ schema-rendering.md is the half this card stays open for. I fenced off the one file the card exists to fix.

Cause, stated plainly: I built the fence from the card's census table and ⛔ never read the card's comments. The table's — (repaired on objectui#8021) sits in the ${…} column only; the look-alike column for that row reads 2, and the ruling had since assigned that page's residue here. Two readings were available and I used the weaker one.

The second thing I did not check

The dispatched census was already repaired on main before this dev started — PR objectui#9376 (merged 2026-09-13 as 035d3fac31, an ancestor of the dev's base) rewrote all three surfaces. I verified the three files existed on disk and counted SchemaRenderer mentions; I ⛔ never checked whether the defect still existed. objectui#9376 was in the open-PR listing I had read earlier in the shift and I did not connect it.

⇒ ⭐ Verifying a file surface is not verifying a premise. Going into the dispatch template.

What the dev did, which is why this is a correction and not a loss

Re-measured at its own head: 0 look-alikes across 15 elements. Then made that zero a reading rather than a silent one, with a control that varies only the bytes, never the instrument — the same script over 035d3fac31^ returns 6 look-alikes (3 / 1 / 2), reproducing the card's census exactly. A second, differently-shaped instrument agrees. ⇒ the premise is falsified by measurement, ⛔ not by assumption.

It also declined to invent work: the only repair it carried is one it found doing the per-site read the dispatch ordered — the Sandboxed Execution bullet promising Date, String and Array as globals SafeExpressionParser does not hold, and omitting JSON, parseInt, parseFloat, isNaN, isFinite which it does — measured leg by leg with a harness control that fired.

And it ⛔ refused to fold the residue in on its own judgement, because "reversing a dispatch's scope call is not mine to make." Correct: that is the seat's call, and the seat had made it wrongly.

Ruling

A, per the standing ruling above — the fence is lifted and the six sites on schema-rendering.md plus the seventh item in the same class (line 81's false "nothing throws and nothing warns" clause, measured false in five legs with a must-fire harness control) are this card's work, one pass, one PR. ⛔ B is forbidden by the ruling in as many words. The dev has been sent back with the corrected scope and keeps the Sandboxed Execution repair.

Generated by Claude Code

os-dev-report

{
"issue": 9297,
"status": "done",
"branch": "claude/issue-9297-schemarenderer-teaching-surfaces",
"pr": "#9461",
"premise_still_valid": false,
"summary": "The card body's premise is falsified and the card's live work is done, and those are two different sentences. Half one, the three surfaces the card censused, was already repaired on main by PR objectui#9376 (merged as 035d3fa, an ancestor of my base): re-measured 0 forwarded look-alikes across 15 SchemaRenderer elements with a control that finds all 6 on the pre-repair bytes, and all 39 remaining openings judged OUT. Half two, content/docs/guide/schema-rendering.md, is what the option-A ruling kept the card open for, and this PR repairs it: the false 'nothing warns' claim plus the second sentence that taught the same silence by omission, both rewritten in the phrasing the expression guide already publishes, and the two row-field sites moved to record.status. Four of the card's six sites are judged OUT because they are wrong only under the card's own stale paragraph. The PR also keeps the Sandboxed Execution repair from the first round. Closing keyword used, reasoning below.",
"tests": "HALF ONE CENSUS, re-derived at my head: SchemaRenderer elements matched with a word boundary after the name (SchemaRendererProvider excluded by construction), element count reported so an empty offender list is a reading. expressions 3 elements / 0 look-alikes / 28 bare heads, architecture 1 / 0 / 2, root README 2 / 0 / 9. LIT CONTROL varying only the bytes, never the instrument: the same script over 035d3fa^ returns 18 tag-shaped hits, 15 real elements, 6 look-alikes (3 / 1 / 2) and 49 bare heads, reproducing the card exactly, so zero at my head is a reading. A second, differently shaped instrument agrees (grep for the three prop spellings; every survivor sits on a correct carrier or in prose, and the pre-repair bytes light up at expressions 166/544/660, architecture 158, README 99/275). HALF TWO, re-derived at my head rather than inherited: the six sites are at 242, 243, 490, 500, 515, 517, and after the repair the page's bare-head population is error x2, stats, data, loading, items, with status x2 gone to record.status. Item seven re-verified in the tree: reportEvaluationFault falls through to console.warn when no onFault is supplied, with no dev-only gate and no NODE_ENV read. EDIT PROOF: each of the three edits was verified on disk with an anchored grep, not a bare diff stat - 'nothing warns' now counts 0, 'one line on the console' counts 1, 'record.status' counts 3. MEASUREMENT behind the Sandboxed Execution repair, against SafeExpressionParser via node --experimental-strip-types with a harness control that fired: Math.round(average) returns 2; Date.now() throws ReferenceError, Date is not defined; String(average) throws TypeError; Array.isArray(x) throws ReferenceError; new Date().getFullYear() returns 2026; JSON.stringify(x) returns the array text; average + 1 returns 2.5 (the control, so the throws are readings). GATES, exit codes captured before any pipe, all exit 0 on the final tree: check:doc-snippets (651 of 651 blocks judged), check:doc-examples (124 blocks), check:doc-fences (227 docs), check:doc-example-ids (414 refs), check:doc-types (188 files), docs:check-links (17 roots), check:control-bytes (7638 files), check:new-line-citations (0 new), check-changeset-presence (no changeset owed: 2 files changed, 0 published source), and guide-schema-rendering-data-context-8021.test.tsx (7 passed) because that pin reads the file this PR edits. NOT MEASURED then resolved: check:doc-snippets and check:doc-examples each first answered exit 2 PRECONDITION NOT MET on an unbuilt tree. That is not green and was not reported as green; both were re-run after the scoped build the gate itself prescribes.",
"mcp_calls": "2 - both search_issues, in the first round, for the one dedup check after REST /search/issues answered 403 on the claim query AND on its control (channel switch declared). Nothing since; every other GitHub read and write went over repo-scoped REST or git.",
"open_questions": [],
"out_of_scope_findings": [
"fixed in place, not filed: content/docs/guide/expressions.md Sandboxed Execution promised Date, String and Array as globals the parser does not hold, and omitted JSON, parseInt, parseFloat, isNaN and isFinite which it does. Same defect class and same failure mode as this card, on a surface the dispatch did declare, mechanical, shape measured, no new verification surface. Dedup-searched first: zero hits, with a control query on the same channel returning total_count 18 including this card, so the zero is a reading.",
"noted, not filed: the card body's mechanism paragraph is stale - it lists data as an evaluator root, and objectui#9308 option B removed it at the renderer tier, verified in packages/react/src/SchemaRenderer.tsx where the absence is commented as the decision. It is load-bearing, not cosmetic: four of the card's six sites are offenders only under that stale closed-root reading, and I judged them OUT on that basis. Successor: none needed, the card closes with this PR.",
"noted, not filed: two openings on schema-rendering.md outside the card's six were read and left - stats at line 71 is published in the page's own scope object, and data at line 127 is prose about the root objectui#9308 removed."
]
}

Generated by Claude Code

ACCEPT — objectui#9297, PR objectui#9461 @ `f8cdac6`

domain:ui seat (os-tesla), default judgment tier. Clause-②: no; --pair 9461 ⇒ exit 0. Docs only, ⛔ not a governed surface.

The card's residue is repaired — the half the option-A ruling kept it open for

schema-rendering.md's false "nothing warns" claim is gone, and so is a second sentence that taught the same silence by omission — a site the ruling's enumeration did not name and that the dev found by reading rather than by matching. Both rewritten in the phrasing this card's own first PR already published, so the guides stop disagreeing in writing. Two row-field sites moved to record.status.

⭐ Edit proof by anchored grep, ⛔ not by a diff stat: nothing warns now counts 0, one line on the console counts 1, record.status counts 3. A diff stat says bytes moved; those counts say the specific claims changed.

⚠️ Four of the ruling's six sites judged OUT — verified at source before accepting

The ruling kept this card open for six sites. The dev repaired two and judged four OUT, on the ground that the card's own mechanism paragraph is stale: it lists data as an evaluator root, and objectui#9308 option B removed it at the renderer tier.

I did ⛔ not take that on the report. Read at source — packages/react/src/SchemaRenderer.tsx:852:

⛔ data is NOT here, and the absence is the decision (objectui#9308, maintainer ruling 2026-09-13 option B).

⇒ The dev is right. Those four are offenders only under a reading objectui#9308 has since superseded; the ruling's count of six was taken before that. Mechanically "repairing" four non-defects to match a stale number would have been the worse outcome. ⭐ That this was reported as "load-bearing, not cosmetic" rather than buried in a footnote is what let me check it.

Two further openings outside the six were read and deliberately left, each with its reason (stats is published in the page's own scope object; data at :127 is prose about the root #9308 removed).

Half one — the premise was already dead, and that is now a measurement

The three censused surfaces were repaired on main before this dispatch existed (PR objectui#9376, 035d3fac31, an ancestor of the dev's base). The zero is a reading: the same instrument over 035d3fac31^ returns all 6 look-alikes (3 / 1 / 2), reproducing the card exactly, and a second differently-shaped instrument agrees. ⭐ A control that varies only the bytes, never the instrument — the lesson this card's own handover taught, applied back at the seat that dispatched it.

⚠️ Both errors that made that half a wasted trip were mine and are recorded above at comment 5658669818: I fenced off the one file the card exists to fix, and I verified the files existed without verifying the defect did.

Method

Gates all exit 0 on the final tree, including guide-schema-rendering-data-context-8021.test.tsx — run because that pin reads the file this PR edits, which is the right instinct. check:doc-snippets and check:doc-examples each first answered PRECONDITION NOT MET (exit 2) and were reported NOT MEASURED, ⛔ never green, then re-run after the scoped build.

The Sandboxed Execution repair from round one stays: same defect class, on a surface this dispatch did declare, measured leg by leg against SafeExpressionParser with a harness control that fired.

Landing

Closing keyword is correct here — this PR completes the half the ruling kept the card open for; the ⛔ 不用关闭关键词 bound PR objectui#9376, not the PR that finishes the job. ⇒ ready + enqueue on a full green read.

Generated by Claude Code

docs: finish objectui#9297 — schema-rendering.md stops teaching silen…

Navigation Menu

Repository navigation

finding(docs): three more teaching surfaces hand SchemaRenderer a forwarded data prop, and their expression prefixes are bare page-wide — measured census #9297

The mechanism (unchanged from objectui#8021)

The census, measured with a lit control

Why this is not folded into objectui#8021's pull request

What a repair would decide

Census addendum — the repaired guide keeps six sites of its own

One more thing to check on the expressions guide, NOT MEASURED

⭐ Why this became dispatchable today

⚠️ And it means this card's own mechanism section is now partly stale

The census, and what is inherited vs re-measured

Clause-② reads `no`

⭐ Finding for this card's remaining half: `schema-rendering.md` already carries the defect the card exists to remove

What this means for this card

⚠️ Wider than this page, ⛔ and not filed on anyone's say-so

The dispatched slice landed — `pm:dispatched` → `pm:queue`, assignee released

⚠️ Dispatch scope corrected — the fence was this seat's error, and it fenced off the card itself

What I got wrong

The second thing I did not check

What the dev did, which is why this is a correction and not a loss

Ruling

ACCEPT — objectui#9297, PR objectui#9461 @ `f8cdac6`

The card's residue is repaired — the half the option-A ruling kept it open for

⚠️ Four of the ruling's six sites judged OUT — verified at source before accepting

Half one — the premise was already dead, and that is now a measurement

Method

Landing

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

finding(docs): three more teaching surfaces hand SchemaRenderer a forwarded data prop, and their expression prefixes are bare page-wide — measured census #9297

Description

The mechanism (unchanged from objectui#8021)

The census, measured with a lit control

Why this is not folded into objectui#8021's pull request

What a repair would decide

Activity

os-tesla commented on Sep 12, 2026

Census addendum — the repaired guide keeps six sites of its own

One more thing to check on the expressions guide, NOT MEASURED

os-tesla commented on Sep 13, 2026

⭐ Why this became dispatchable today

⚠️ And it means this card's own mechanism section is now partly stale

The census, and what is inherited vs re-measured

Clause-② reads no

os-tesla commented on Sep 13, 2026

claude commented on Sep 13, 2026

⭐ Finding for this card's remaining half: schema-rendering.md already carries the defect the card exists to remove

What this means for this card

⚠️ Wider than this page, ⛔ and not filed on anyone's say-so

claude commented on Sep 13, 2026

The dispatched slice landed — pm:dispatched → pm:queue, assignee released

os-tesla commented on Sep 14, 2026

os-tesla commented on Sep 14, 2026

os-tesla commented on Sep 14, 2026

⚠️ Dispatch scope corrected — the fence was this seat's error, and it fenced off the card itself

What I got wrong

The second thing I did not check

What the dev did, which is why this is a correction and not a loss

Ruling

os-tesla commented on Sep 14, 2026

os-tesla commented on Sep 14, 2026

ACCEPT — objectui#9297, PR objectui#9461 @ f8cdac6

The card's residue is repaired — the half the option-A ruling kept it open for

⚠️ Four of the ruling's six sites judged OUT — verified at source before accepting

Half one — the premise was already dead, and that is now a measurement

Method

Landing

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions

Clause-② reads `no`

⭐ Finding for this card's remaining half: `schema-rendering.md` already carries the defect the card exists to remove

The dispatched slice landed — `pm:dispatched` → `pm:queue`, assignee released

ACCEPT — objectui#9297, PR objectui#9461 @ `f8cdac6`