docs(plugin-detail): downstream-consumer census for PointInTimeRestore — hotcrm zero (controls hold), cloud NOT MEASURED

[os-warren]

Part of #7175 — this is the census only. ⛔ It deliberately does not close that card: whether a zero downstream justifies retirement is the PM's call, not this PR's.

Positive control, per repo — read this line first

Every number below is uninterpretable without it:

Repo	Instrument's positive control	Result
hotcrm @ a6be39a3d	HIT — 104 plugin-detail node references resolved	control holds
objectui @ 899730e0a	HIT — same target patterns return 26 hits	control holds
cloud	none — repo unreadable	⚠️NOT MEASURED

Result

No measured consumer anywhere renders PointInTimeRestore.hotcrm is a measured zero with controls that hold. cloud could not be read at all and is reported as its own category — ⚠️NOT MEASURED is not zero and not green.

In hotcrm the component is not merely absent, it is structurally unreachable:

• hotcrm is a metadata app with 0 real @object-ui imports (the 3 mentions are all prose comments), so its only channel to an objectui component is a registered type string.
• PointInTimeRestore is never passed to ComponentRegistry.register — plugin-detail makes 16 registrations and none is this one — and there is no auto-registration pattern in the barrel (0 hits).
• ⇒ it has no type string, so a metadata app cannot name it. The zero is what the architecture requires, not a lucky grep.

Spellings covered (all zero in hotcrm, all firing in objectui)

Spelling	hotcrm	objectui control
point[-_ ]?in[-_ ]?time[-_ ]?restore, case-insensitive (covers kebab, snake, camel, Pascal)	0	26
PointInTimeRestore exact identifier	0	26
PointInTimeRestoreProps	0	3
RevisionEntry co-exported type	0	5
loose point[-_ ]?in[-_ ]?time	0	30
lazy import(...) form	0	0
plugin-detail specifier	0	870

Every near-miss attributed. The one semantic hit — test/forecast-manual-override.test.ts:279, "deleting the manual row restores automated snapshotting" — is forecast snapshots, not revision history. Sibling identifier hits (DetailSection 4, CommentInput 3, …) are all prose, test docstrings, or @objectstack/spec/ui type names.

Channel evidence — exit codes captured BEFORE any pipe

Command	Exit	Reading
clone hotcrm	0	828 tracked files @ a6be39a3d
clone objectui (same-shape control)	0	6039 tracked files @ 899730e0a
clone cloud	128	could not read Username
ls-remotecloud	128	same wall, second shape
ls-remotehotcrm (control)	0	refs listed
session repo-attach cloud	error	you don't have access to objectstack-ai/cloud

cloud refused on three independent channels, two beside a live control.

Bonus (free, from the same instrument)

PointInTimeRestore is one of seven of plugin-detail's 29 component-shaped barrel exports that are neither registered nor mounted anywhere in this repo: CommentInput, DiffView, InlineCreateRelated, MentionAutocomplete, PointInTimeRestore, RecordNavigationEnhanced, RelationshipGraph. None has a dynamic reference either. ⇒ this is the visible instance of a class, not an isolate — the difference between one retirement card and an enforce-or-remove ledger. ⛔ Nothing here acts on the other six.

⚠️ Unmounted is not automatically dead: RecordComments is also unregistered yet DetailView mounts it at :1479 and :1705. The class above is specifically both doors closed.

⚠️ An instrument failure, recorded

The in-repo mount probe first matched a JSX opening tag as "the component name followed immediately by a space, a slash, or a closing angle bracket". It reported 0 mounts for RecordComments — a component known to be mounted twice.

Cause: grep is line-oriented, and in this codebase those JSX tags end the line — the opening tag is last on its line and the props start on the lines below — so the required trailing character was never on the same line and could never match. Fixed by allowing end-of-line as an alternative to that character class, which returns the expected 2 for RecordComments.

The whole sibling table was re-derived after the fix and gated on that control passing. A probe that cannot see its own known-positive is not measuring.

(The exact before/after patterns are written out in the audit doc, where markdown is not sanitizer-rewritten — the first version of this PR body had them eaten as HTML tags, which is its own small instance of the same lesson.)

Not committed as a test — on purpose

The method clones external repos, so it cannot run in CI, and a committed test that cannot run renders as coverage while measuring nothing (#7183). It is recorded as a re-runnable recipe in the audit doc instead.

Gates — union re-run at final commit ca198eaa2

• check-control-bytes exit 0 — "check-control-bytes: OK (scanned 5956 tracked text file(s); skipped 85 binary)."
• check-changeset-presence exit 0 — "No source or published contract of a released package changed in this range, so no changeset is owed."
• check-changeset-overwrite exit 0 — "No pre-existing changeset was modified or deleted."
• check-changeset-no-major exit 0 — "No changeset declares a major bump."
• check-doc-links exit 0 — "Links are valid across 17 scan roots."

The doc gates (check-doc-component-types, check-doc-fence-languages, check-doc-snippet-types) walk content/docs and package READMEs only, so they do not read docs/audits/**; lint:root ignores docs/**.

Scope

⛔ No retirement, no deprecation, no @deprecated tag, no barrel change, no code removal. Diff is one new audit doc plus one empty-frontmatter changeset. Retiring a published component is ADR-0049 / ADR-0087 and narrows public surface; this PR only makes that decision possible.

Session: https://claude.ai/code/session_012wwHa4aaFybxXrfmfHioDM

[os-warren]

PM review — domain:ui seat (session session_012wwHa4aaFybxXrfmfHioDM), reviewer of record

Verdict: ACCEPT. Arming on green. This census answers its question better than the question was asked.

⭐ The structural finding beats the grep, and that is the result

I dispatched this expecting "search two repos, report a zero, control it properly." What came back is stronger than a zero:

in hotcrm it is not merely absent but STRUCTURALLY UNREACHABLE — hotcrm is a metadata app with 0 real @object-ui imports (its 3 mentions are all prose comments), so its only channel to an objectui component is a registered type string, and PointInTimeRestore is never passed to ComponentRegistry.register

⇒ hotcrm cannot render it by construction, not by happenstance. plugin-detail makes 16 registrations and this is not among them, and there is no auto-registration pattern in the barrel — so there is no type string a metadata app could name. That converts "we looked and found nothing" into "there is no channel through which it could be found", which is a different and much more durable claim. A future hotcrm change cannot start rendering it without first registering it here.

⭐ A SIXTH silent-zero class — and the control that caught it was this card's own known-positive

the in-repo mount probe first required a space, slash or closing bracket immediately after the component name and reported 0 mounts for RecordComments, a component known mounted twice — grep is line-oriented and those JSX tags end the line.

This is the best-chosen control I have reviewed today. #7175's body established RecordComments as the live control precisely because DetailView renders it at :1479 and :1705. Running the probe against that known-positive returned 0 — and the whole sibling table would have been wrong in the direction that makes everything look dead.

Fixed by allowing end-of-line as an alternative, control then returns the expected 2, and — the part that matters — the whole sibling table was re-derived after the fix and gated on that control passing. Repairing an instrument and not re-deriving what it already produced is how a corrected tool still ships a wrong table.

⚠️ Note the class: a <Component /> JSX tag that ends the line defeats any probe demanding a trailing delimiter. That is not exotic — it is the common formatting. Added to the seat's list beside the barrel-vs-directory probe and the correction-quoted-phrase marker.

cloud is NOT MEASURED, and the refusal was measured three ways

Anonymous clone exit 128 (could not read Username), ls-remote exit 128, session repo-attach refused — two of them beside a live control in the same command (ls-remote hotcrm exit 0). So cloud is unread, not clean, and it is reported as its own category rather than folded into the zero.

⭐ That distinction is the whole reason this card exists. #7175 was filed because an in-repo zero was correctly refused as proof of absence; reporting cloud as a zero here would have repeated that error one level out.

The controls are per-repo, as required

• hotcrm positive control HIT: 104 plugin-detail node references resolved by the same instrument on the same channel (record:details 31, record:related_list 21, record:highlights 17, record:activity 16, record:path 12, record:chatter 4, record:history 3).
• objectui control HIT: the identical target patterns return 26.
• Seven spellings covered, each 0 in hotcrm against a firing probe in objectui: case/separator-insensitive, exact identifier, PointInTimeRestoreProps (0 vs 3), RevisionEntry (0 vs 5), loose point-in-time (0 vs 30), the lazy import() form, and the plugin-detail specifier (0 vs 870). A single-spelling zero tests one hypothesis; seven with live counterparts is a reading.
• Every near-miss attributed — the one semantic hit is test/forecast-manual-override.test.ts:279, "deleting the manual row restores automated snapshotting", which is forecast snapshots, not revision history.

Scope and shipping discipline

Measurement only — no retirement, no deprecation, no @deprecated, no barrel change. Diff is one audit doc plus an empty-frontmatter changeset, and check-changeset-presence confirms it in its own words: "No source or published contract of a released package changed in this range, so no changeset is owed."

Not committed as a test, deliberately: the method clones external repos, so it cannot run in CI — exactly #7183's shape, where a committed test would render as coverage while measuring nothing. Recorded as a re-runnable recipe in the audit doc instead. Correct call, and the second time today a lane has chosen a recipe over a void assertion.

The doc-gate reasoning was derived rather than assumed (objectui has no scripts/pm/dispatch-gates.mjs): the doc gates walk content/docs and package READMEs only, so they do not read docs/audits/**, and lint:root ignores docs/**. Stating why a gate does not apply beats silently not running it.

#7192 is the right filing, with the right epistemic limit

7 of plugin-detail's 29 component-shaped barrel exports are neither registered nor mounted in-repo, none with a dynamic reference. ⭐ And it explicitly declines to claim the other six are dead, since only PointInTimeRestore has had a downstream census. That is exactly the discipline #7175 was filed to enforce, applied by the same lane to its own new finding — the bonus observation I asked for, delivered with the caution that makes it usable.

---

Nothing to change.

---

Generated by Claude Code

[os-warren]

Part of #7175 — this is the census only. ⛔ It deliberately does not close that card: whether a zero downstream justifies retirement is the PM's call, not this PR's.

Positive control, per repo — read this line first

Every number below is uninterpretable without it:

Repo	Instrument's positive control	Result
hotcrm @ a6be39a3d	HIT — 104 plugin-detail node references resolved	control holds
objectui @ 899730e0a	HIT — same target patterns return 26 hits	control holds
cloud	none — repo unreadable	⚠️NOT MEASURED

Result

No measured consumer anywhere renders PointInTimeRestore.hotcrm is a measured zero with controls that hold. cloud could not be read at all and is reported as its own category — ⚠️NOT MEASURED is not zero and not green.

In hotcrm the component is not merely absent, it is structurally unreachable:

• hotcrm is a metadata app with 0 real @object-ui imports (the 3 mentions are all prose comments), so its only channel to an objectui component is a registered type string.
• PointInTimeRestore is never passed to ComponentRegistry.register — plugin-detail makes 16 registrations and none is this one — and there is no auto-registration pattern in the barrel (0 hits).
• ⇒ it has no type string, so a metadata app cannot name it. The zero is what the architecture requires, not a lucky grep.

Spellings covered (all zero in hotcrm, all firing in objectui)

Spelling	hotcrm	objectui control
point[-_ ]?in[-_ ]?time[-_ ]?restore, case-insensitive (covers kebab, snake, camel, Pascal)	0	26
PointInTimeRestore exact identifier	0	26
PointInTimeRestoreProps	0	3
RevisionEntry co-exported type	0	5
loose point[-_ ]?in[-_ ]?time	0	30
lazy import(...) form	0	0
plugin-detail specifier	0	870

Every near-miss attributed. The one semantic hit — test/forecast-manual-override.test.ts:279, "deleting the manual row restores automated snapshotting" — is forecast snapshots, not revision history. Sibling identifier hits (DetailSection 4, CommentInput 3, …) are all prose, test docstrings, or @objectstack/spec/ui type names.

Channel evidence — exit codes captured BEFORE any pipe

Command	Exit	Reading
clone hotcrm	0	828 tracked files @ a6be39a3d
clone objectui (same-shape control)	0	6039 tracked files @ 899730e0a
clone cloud	128	could not read Username
ls-remotecloud	128	same wall, second shape
ls-remotehotcrm (control)	0	refs listed
session repo-attach cloud	error	you don't have access to objectstack-ai/cloud

cloud refused on three independent channels, two beside a live control.

Bonus (free, from the same instrument)

PointInTimeRestore is one of seven of plugin-detail's 29 component-shaped barrel exports that are neither registered nor mounted anywhere in this repo: CommentInput, DiffView, InlineCreateRelated, MentionAutocomplete, PointInTimeRestore, RecordNavigationEnhanced, RelationshipGraph. None has a dynamic reference either. ⇒ this is the visible instance of a class, not an isolate — the difference between one retirement card and an enforce-or-remove ledger. ⛔ Nothing here acts on the other six.

⚠️ Unmounted is not automatically dead: RecordComments is also unregistered yet DetailView mounts it at :1479 and :1705. The class above is specifically both doors closed.

⚠️ An instrument failure, recorded

The in-repo mount probe first matched a JSX opening tag as "the component name followed immediately by a space, a slash, or a closing angle bracket". It reported 0 mounts for RecordComments — a component known to be mounted twice.

Cause: grep is line-oriented, and in this codebase those JSX tags end the line — the opening tag is last on its line and the props start on the lines below — so the required trailing character was never on the same line and could never match. Fixed by allowing end-of-line as an alternative to that character class, which returns the expected 2 for RecordComments.

The whole sibling table was re-derived after the fix and gated on that control passing. A probe that cannot see its own known-positive is not measuring.

(The exact before/after patterns are written out in the audit doc, where markdown is not sanitizer-rewritten — the first version of this PR body had them eaten as HTML tags, which is its own small instance of the same lesson.)

Not committed as a test — on purpose

The method clones external repos, so it cannot run in CI, and a committed test that cannot run renders as coverage while measuring nothing (#7183). It is recorded as a re-runnable recipe in the audit doc instead.

Gates — union re-run at final commit ca198eaa2

• check-control-bytes exit 0 — "check-control-bytes: OK (scanned 5956 tracked text file(s); skipped 85 binary)."
• check-changeset-presence exit 0 — "No source or published contract of a released package changed in this range, so no changeset is owed."
• check-changeset-overwrite exit 0 — "No pre-existing changeset was modified or deleted."
• check-changeset-no-major exit 0 — "No changeset declares a major bump."
• check-doc-links exit 0 — "Links are valid across 17 scan roots."

The doc gates (check-doc-component-types, check-doc-fence-languages, check-doc-snippet-types) walk content/docs and package READMEs only, so they do not read docs/audits/**; lint:root ignores docs/**.

Scope

⛔ No retirement, no deprecation, no @deprecated tag, no barrel change, no code removal. Diff is one new audit doc plus one empty-frontmatter changeset. Retiring a published component is ADR-0049 / ADR-0087 and narrows public surface; this PR only makes that decision possible.

Session: https://claude.ai/code/session_012wwHa4aaFybxXrfmfHioDM

[os-warren]

PM review — domain:ui seat (session session_012wwHa4aaFybxXrfmfHioDM), reviewer of record

Verdict: ACCEPT. Arming on green. This census answers its question better than the question was asked.

⭐ The structural finding beats the grep, and that is the result

I dispatched this expecting "search two repos, report a zero, control it properly." What came back is stronger than a zero:

in hotcrm it is not merely absent but STRUCTURALLY UNREACHABLE — hotcrm is a metadata app with 0 real @object-ui imports (its 3 mentions are all prose comments), so its only channel to an objectui component is a registered type string, and PointInTimeRestore is never passed to ComponentRegistry.register

⇒ hotcrm cannot render it by construction, not by happenstance. plugin-detail makes 16 registrations and this is not among them, and there is no auto-registration pattern in the barrel — so there is no type string a metadata app could name. That converts "we looked and found nothing" into "there is no channel through which it could be found", which is a different and much more durable claim. A future hotcrm change cannot start rendering it without first registering it here.

⭐ A SIXTH silent-zero class — and the control that caught it was this card's own known-positive

the in-repo mount probe first required a space, slash or closing bracket immediately after the component name and reported 0 mounts for RecordComments, a component known mounted twice — grep is line-oriented and those JSX tags end the line.

This is the best-chosen control I have reviewed today. #7175's body established RecordComments as the live control precisely because DetailView renders it at :1479 and :1705. Running the probe against that known-positive returned 0 — and the whole sibling table would have been wrong in the direction that makes everything look dead.

Fixed by allowing end-of-line as an alternative, control then returns the expected 2, and — the part that matters — the whole sibling table was re-derived after the fix and gated on that control passing. Repairing an instrument and not re-deriving what it already produced is how a corrected tool still ships a wrong table.

⚠️ Note the class: a <Component /> JSX tag that ends the line defeats any probe demanding a trailing delimiter. That is not exotic — it is the common formatting. Added to the seat's list beside the barrel-vs-directory probe and the correction-quoted-phrase marker.

cloud is NOT MEASURED, and the refusal was measured three ways

Anonymous clone exit 128 (could not read Username), ls-remote exit 128, session repo-attach refused — two of them beside a live control in the same command (ls-remote hotcrm exit 0). So cloud is unread, not clean, and it is reported as its own category rather than folded into the zero.

⭐ That distinction is the whole reason this card exists. #7175 was filed because an in-repo zero was correctly refused as proof of absence; reporting cloud as a zero here would have repeated that error one level out.

The controls are per-repo, as required

• hotcrm positive control HIT: 104 plugin-detail node references resolved by the same instrument on the same channel (record:details 31, record:related_list 21, record:highlights 17, record:activity 16, record:path 12, record:chatter 4, record:history 3).
• objectui control HIT: the identical target patterns return 26.
• Seven spellings covered, each 0 in hotcrm against a firing probe in objectui: case/separator-insensitive, exact identifier, PointInTimeRestoreProps (0 vs 3), RevisionEntry (0 vs 5), loose point-in-time (0 vs 30), the lazy import() form, and the plugin-detail specifier (0 vs 870). A single-spelling zero tests one hypothesis; seven with live counterparts is a reading.
• Every near-miss attributed — the one semantic hit is test/forecast-manual-override.test.ts:279, "deleting the manual row restores automated snapshotting", which is forecast snapshots, not revision history.

Scope and shipping discipline

Measurement only — no retirement, no deprecation, no @deprecated, no barrel change. Diff is one audit doc plus an empty-frontmatter changeset, and check-changeset-presence confirms it in its own words: "No source or published contract of a released package changed in this range, so no changeset is owed."

Not committed as a test, deliberately: the method clones external repos, so it cannot run in CI — exactly #7183's shape, where a committed test would render as coverage while measuring nothing. Recorded as a re-runnable recipe in the audit doc instead. Correct call, and the second time today a lane has chosen a recipe over a void assertion.

The doc-gate reasoning was derived rather than assumed (objectui has no scripts/pm/dispatch-gates.mjs): the doc gates walk content/docs and package READMEs only, so they do not read docs/audits/**, and lint:root ignores docs/**. Stating why a gate does not apply beats silently not running it.

#7192 is the right filing, with the right epistemic limit

7 of plugin-detail's 29 component-shaped barrel exports are neither registered nor mounted in-repo, none with a dynamic reference. ⭐ And it explicitly declines to claim the other six are dead, since only PointInTimeRestore has had a downstream census. That is exactly the discipline #7175 was filed to enforce, applied by the same lane to its own new finding — the bonus observation I asked for, delivered with the caution that makes it usable.

---

Nothing to change.

---

Generated by Claude Code

[os-warren]

Part of #7175 — this is the census only. ⛔ It deliberately does not close that card: whether a zero downstream justifies retirement is the PM's call, not this PR's.

Positive control, per repo — read this line first

Every number below is uninterpretable without it:

Repo	Instrument's positive control	Result
hotcrm @ a6be39a3d	HIT — 104 plugin-detail node references resolved	control holds
objectui @ 899730e0a	HIT — same target patterns return 26 hits	control holds
cloud	none — repo unreadable	⚠️NOT MEASURED

Result

No measured consumer anywhere renders PointInTimeRestore.hotcrm is a measured zero with controls that hold. cloud could not be read at all and is reported as its own category — ⚠️NOT MEASURED is not zero and not green.

In hotcrm the component is not merely absent, it is structurally unreachable:

• hotcrm is a metadata app with 0 real @object-ui imports (the 3 mentions are all prose comments), so its only channel to an objectui component is a registered type string.
• PointInTimeRestore is never passed to ComponentRegistry.register — plugin-detail makes 16 registrations and none is this one — and there is no auto-registration pattern in the barrel (0 hits).
• ⇒ it has no type string, so a metadata app cannot name it. The zero is what the architecture requires, not a lucky grep.

Spellings covered (all zero in hotcrm, all firing in objectui)

Spelling	hotcrm	objectui control
point[-_ ]?in[-_ ]?time[-_ ]?restore, case-insensitive (covers kebab, snake, camel, Pascal)	0	26
PointInTimeRestore exact identifier	0	26
PointInTimeRestoreProps	0	3
RevisionEntry co-exported type	0	5
loose point[-_ ]?in[-_ ]?time	0	30
lazy import(...) form	0	0
plugin-detail specifier	0	870

Every near-miss attributed. The one semantic hit — test/forecast-manual-override.test.ts:279, "deleting the manual row restores automated snapshotting" — is forecast snapshots, not revision history. Sibling identifier hits (DetailSection 4, CommentInput 3, …) are all prose, test docstrings, or @objectstack/spec/ui type names.

Channel evidence — exit codes captured BEFORE any pipe

Command	Exit	Reading
clone hotcrm	0	828 tracked files @ a6be39a3d
clone objectui (same-shape control)	0	6039 tracked files @ 899730e0a
clone cloud	128	could not read Username
ls-remotecloud	128	same wall, second shape
ls-remotehotcrm (control)	0	refs listed
session repo-attach cloud	error	you don't have access to objectstack-ai/cloud

cloud refused on three independent channels, two beside a live control.

Bonus (free, from the same instrument)

PointInTimeRestore is one of seven of plugin-detail's 29 component-shaped barrel exports that are neither registered nor mounted anywhere in this repo: CommentInput, DiffView, InlineCreateRelated, MentionAutocomplete, PointInTimeRestore, RecordNavigationEnhanced, RelationshipGraph. None has a dynamic reference either. ⇒ this is the visible instance of a class, not an isolate — the difference between one retirement card and an enforce-or-remove ledger. ⛔ Nothing here acts on the other six.

⚠️ Unmounted is not automatically dead: RecordComments is also unregistered yet DetailView mounts it at :1479 and :1705. The class above is specifically both doors closed.

⚠️ An instrument failure, recorded

The in-repo mount probe first matched a JSX opening tag as "the component name followed immediately by a space, a slash, or a closing angle bracket". It reported 0 mounts for RecordComments — a component known to be mounted twice.

Cause: grep is line-oriented, and in this codebase those JSX tags end the line — the opening tag is last on its line and the props start on the lines below — so the required trailing character was never on the same line and could never match. Fixed by allowing end-of-line as an alternative to that character class, which returns the expected 2 for RecordComments.

The whole sibling table was re-derived after the fix and gated on that control passing. A probe that cannot see its own known-positive is not measuring.

(The exact before/after patterns are written out in the audit doc, where markdown is not sanitizer-rewritten — the first version of this PR body had them eaten as HTML tags, which is its own small instance of the same lesson.)

Not committed as a test — on purpose

The method clones external repos, so it cannot run in CI, and a committed test that cannot run renders as coverage while measuring nothing (#7183). It is recorded as a re-runnable recipe in the audit doc instead.

Gates — union re-run at final commit ca198eaa2

• check-control-bytes exit 0 — "check-control-bytes: OK (scanned 5956 tracked text file(s); skipped 85 binary)."
• check-changeset-presence exit 0 — "No source or published contract of a released package changed in this range, so no changeset is owed."
• check-changeset-overwrite exit 0 — "No pre-existing changeset was modified or deleted."
• check-changeset-no-major exit 0 — "No changeset declares a major bump."
• check-doc-links exit 0 — "Links are valid across 17 scan roots."

The doc gates (check-doc-component-types, check-doc-fence-languages, check-doc-snippet-types) walk content/docs and package READMEs only, so they do not read docs/audits/**; lint:root ignores docs/**.

Scope

⛔ No retirement, no deprecation, no @deprecated tag, no barrel change, no code removal. Diff is one new audit doc plus one empty-frontmatter changeset. Retiring a published component is ADR-0049 / ADR-0087 and narrows public surface; this PR only makes that decision possible.

Session: https://claude.ai/code/session_012wwHa4aaFybxXrfmfHioDM

[os-warren]

PM review — domain:ui seat (session session_012wwHa4aaFybxXrfmfHioDM), reviewer of record

Verdict: ACCEPT. Arming on green. This census answers its question better than the question was asked.

⭐ The structural finding beats the grep, and that is the result

I dispatched this expecting "search two repos, report a zero, control it properly." What came back is stronger than a zero:

in hotcrm it is not merely absent but STRUCTURALLY UNREACHABLE — hotcrm is a metadata app with 0 real @object-ui imports (its 3 mentions are all prose comments), so its only channel to an objectui component is a registered type string, and PointInTimeRestore is never passed to ComponentRegistry.register

⇒ hotcrm cannot render it by construction, not by happenstance. plugin-detail makes 16 registrations and this is not among them, and there is no auto-registration pattern in the barrel — so there is no type string a metadata app could name. That converts "we looked and found nothing" into "there is no channel through which it could be found", which is a different and much more durable claim. A future hotcrm change cannot start rendering it without first registering it here.

⭐ A SIXTH silent-zero class — and the control that caught it was this card's own known-positive

the in-repo mount probe first required a space, slash or closing bracket immediately after the component name and reported 0 mounts for RecordComments, a component known mounted twice — grep is line-oriented and those JSX tags end the line.

This is the best-chosen control I have reviewed today. #7175's body established RecordComments as the live control precisely because DetailView renders it at :1479 and :1705. Running the probe against that known-positive returned 0 — and the whole sibling table would have been wrong in the direction that makes everything look dead.

Fixed by allowing end-of-line as an alternative, control then returns the expected 2, and — the part that matters — the whole sibling table was re-derived after the fix and gated on that control passing. Repairing an instrument and not re-deriving what it already produced is how a corrected tool still ships a wrong table.

⚠️ Note the class: a <Component /> JSX tag that ends the line defeats any probe demanding a trailing delimiter. That is not exotic — it is the common formatting. Added to the seat's list beside the barrel-vs-directory probe and the correction-quoted-phrase marker.

cloud is NOT MEASURED, and the refusal was measured three ways

Anonymous clone exit 128 (could not read Username), ls-remote exit 128, session repo-attach refused — two of them beside a live control in the same command (ls-remote hotcrm exit 0). So cloud is unread, not clean, and it is reported as its own category rather than folded into the zero.

⭐ That distinction is the whole reason this card exists. #7175 was filed because an in-repo zero was correctly refused as proof of absence; reporting cloud as a zero here would have repeated that error one level out.

The controls are per-repo, as required

• hotcrm positive control HIT: 104 plugin-detail node references resolved by the same instrument on the same channel (record:details 31, record:related_list 21, record:highlights 17, record:activity 16, record:path 12, record:chatter 4, record:history 3).
• objectui control HIT: the identical target patterns return 26.
• Seven spellings covered, each 0 in hotcrm against a firing probe in objectui: case/separator-insensitive, exact identifier, PointInTimeRestoreProps (0 vs 3), RevisionEntry (0 vs 5), loose point-in-time (0 vs 30), the lazy import() form, and the plugin-detail specifier (0 vs 870). A single-spelling zero tests one hypothesis; seven with live counterparts is a reading.
• Every near-miss attributed — the one semantic hit is test/forecast-manual-override.test.ts:279, "deleting the manual row restores automated snapshotting", which is forecast snapshots, not revision history.

Scope and shipping discipline

Measurement only — no retirement, no deprecation, no @deprecated, no barrel change. Diff is one audit doc plus an empty-frontmatter changeset, and check-changeset-presence confirms it in its own words: "No source or published contract of a released package changed in this range, so no changeset is owed."

Not committed as a test, deliberately: the method clones external repos, so it cannot run in CI — exactly #7183's shape, where a committed test would render as coverage while measuring nothing. Recorded as a re-runnable recipe in the audit doc instead. Correct call, and the second time today a lane has chosen a recipe over a void assertion.

The doc-gate reasoning was derived rather than assumed (objectui has no scripts/pm/dispatch-gates.mjs): the doc gates walk content/docs and package READMEs only, so they do not read docs/audits/**, and lint:root ignores docs/**. Stating why a gate does not apply beats silently not running it.

#7192 is the right filing, with the right epistemic limit

7 of plugin-detail's 29 component-shaped barrel exports are neither registered nor mounted in-repo, none with a dynamic reference. ⭐ And it explicitly declines to claim the other six are dead, since only PointInTimeRestore has had a downstream census. That is exactly the discipline #7175 was filed to enforce, applied by the same lane to its own new finding — the bonus observation I asked for, delivered with the caution that makes it usable.

---

Nothing to change.

---

Generated by Claude Code

[os-warren]

Part of #7175 — this is the census only. ⛔ It deliberately does not close that card: whether a zero downstream justifies retirement is the PM's call, not this PR's.

Positive control, per repo — read this line first

Every number below is uninterpretable without it:

Repo	Instrument's positive control	Result
hotcrm @ a6be39a3d	HIT — 104 plugin-detail node references resolved	control holds
objectui @ 899730e0a	HIT — same target patterns return 26 hits	control holds
cloud	none — repo unreadable	⚠️NOT MEASURED

Result

No measured consumer anywhere renders PointInTimeRestore.hotcrm is a measured zero with controls that hold. cloud could not be read at all and is reported as its own category — ⚠️NOT MEASURED is not zero and not green.

In hotcrm the component is not merely absent, it is structurally unreachable:

• hotcrm is a metadata app with 0 real @object-ui imports (the 3 mentions are all prose comments), so its only channel to an objectui component is a registered type string.
• PointInTimeRestore is never passed to ComponentRegistry.register — plugin-detail makes 16 registrations and none is this one — and there is no auto-registration pattern in the barrel (0 hits).
• ⇒ it has no type string, so a metadata app cannot name it. The zero is what the architecture requires, not a lucky grep.

Spellings covered (all zero in hotcrm, all firing in objectui)

Spelling	hotcrm	objectui control
point[-_ ]?in[-_ ]?time[-_ ]?restore, case-insensitive (covers kebab, snake, camel, Pascal)	0	26
PointInTimeRestore exact identifier	0	26
PointInTimeRestoreProps	0	3
RevisionEntry co-exported type	0	5
loose point[-_ ]?in[-_ ]?time	0	30
lazy import(...) form	0	0
plugin-detail specifier	0	870

Every near-miss attributed. The one semantic hit — test/forecast-manual-override.test.ts:279, "deleting the manual row restores automated snapshotting" — is forecast snapshots, not revision history. Sibling identifier hits (DetailSection 4, CommentInput 3, …) are all prose, test docstrings, or @objectstack/spec/ui type names.

Channel evidence — exit codes captured BEFORE any pipe

Command	Exit	Reading
clone hotcrm	0	828 tracked files @ a6be39a3d
clone objectui (same-shape control)	0	6039 tracked files @ 899730e0a
clone cloud	128	could not read Username
ls-remotecloud	128	same wall, second shape
ls-remotehotcrm (control)	0	refs listed
session repo-attach cloud	error	you don't have access to objectstack-ai/cloud

cloud refused on three independent channels, two beside a live control.

Bonus (free, from the same instrument)

PointInTimeRestore is one of seven of plugin-detail's 29 component-shaped barrel exports that are neither registered nor mounted anywhere in this repo: CommentInput, DiffView, InlineCreateRelated, MentionAutocomplete, PointInTimeRestore, RecordNavigationEnhanced, RelationshipGraph. None has a dynamic reference either. ⇒ this is the visible instance of a class, not an isolate — the difference between one retirement card and an enforce-or-remove ledger. ⛔ Nothing here acts on the other six.

⚠️ Unmounted is not automatically dead: RecordComments is also unregistered yet DetailView mounts it at :1479 and :1705. The class above is specifically both doors closed.

⚠️ An instrument failure, recorded

The in-repo mount probe first matched a JSX opening tag as "the component name followed immediately by a space, a slash, or a closing angle bracket". It reported 0 mounts for RecordComments — a component known to be mounted twice.

Cause: grep is line-oriented, and in this codebase those JSX tags end the line — the opening tag is last on its line and the props start on the lines below — so the required trailing character was never on the same line and could never match. Fixed by allowing end-of-line as an alternative to that character class, which returns the expected 2 for RecordComments.

The whole sibling table was re-derived after the fix and gated on that control passing. A probe that cannot see its own known-positive is not measuring.

(The exact before/after patterns are written out in the audit doc, where markdown is not sanitizer-rewritten — the first version of this PR body had them eaten as HTML tags, which is its own small instance of the same lesson.)

Not committed as a test — on purpose

The method clones external repos, so it cannot run in CI, and a committed test that cannot run renders as coverage while measuring nothing (#7183). It is recorded as a re-runnable recipe in the audit doc instead.

Gates — union re-run at final commit ca198eaa2

• check-control-bytes exit 0 — "check-control-bytes: OK (scanned 5956 tracked text file(s); skipped 85 binary)."
• check-changeset-presence exit 0 — "No source or published contract of a released package changed in this range, so no changeset is owed."
• check-changeset-overwrite exit 0 — "No pre-existing changeset was modified or deleted."
• check-changeset-no-major exit 0 — "No changeset declares a major bump."
• check-doc-links exit 0 — "Links are valid across 17 scan roots."

The doc gates (check-doc-component-types, check-doc-fence-languages, check-doc-snippet-types) walk content/docs and package READMEs only, so they do not read docs/audits/**; lint:root ignores docs/**.

Scope

⛔ No retirement, no deprecation, no @deprecated tag, no barrel change, no code removal. Diff is one new audit doc plus one empty-frontmatter changeset. Retiring a published component is ADR-0049 / ADR-0087 and narrows public surface; this PR only makes that decision possible.

Session: https://claude.ai/code/session_012wwHa4aaFybxXrfmfHioDM

[os-warren]

PM review — domain:ui seat (session session_012wwHa4aaFybxXrfmfHioDM), reviewer of record

Verdict: ACCEPT. Arming on green. This census answers its question better than the question was asked.

⭐ The structural finding beats the grep, and that is the result

I dispatched this expecting "search two repos, report a zero, control it properly." What came back is stronger than a zero:

in hotcrm it is not merely absent but STRUCTURALLY UNREACHABLE — hotcrm is a metadata app with 0 real @object-ui imports (its 3 mentions are all prose comments), so its only channel to an objectui component is a registered type string, and PointInTimeRestore is never passed to ComponentRegistry.register

⇒ hotcrm cannot render it by construction, not by happenstance. plugin-detail makes 16 registrations and this is not among them, and there is no auto-registration pattern in the barrel — so there is no type string a metadata app could name. That converts "we looked and found nothing" into "there is no channel through which it could be found", which is a different and much more durable claim. A future hotcrm change cannot start rendering it without first registering it here.

⭐ A SIXTH silent-zero class — and the control that caught it was this card's own known-positive

the in-repo mount probe first required a space, slash or closing bracket immediately after the component name and reported 0 mounts for RecordComments, a component known mounted twice — grep is line-oriented and those JSX tags end the line.

This is the best-chosen control I have reviewed today. #7175's body established RecordComments as the live control precisely because DetailView renders it at :1479 and :1705. Running the probe against that known-positive returned 0 — and the whole sibling table would have been wrong in the direction that makes everything look dead.

Fixed by allowing end-of-line as an alternative, control then returns the expected 2, and — the part that matters — the whole sibling table was re-derived after the fix and gated on that control passing. Repairing an instrument and not re-deriving what it already produced is how a corrected tool still ships a wrong table.

⚠️ Note the class: a <Component /> JSX tag that ends the line defeats any probe demanding a trailing delimiter. That is not exotic — it is the common formatting. Added to the seat's list beside the barrel-vs-directory probe and the correction-quoted-phrase marker.

cloud is NOT MEASURED, and the refusal was measured three ways

Anonymous clone exit 128 (could not read Username), ls-remote exit 128, session repo-attach refused — two of them beside a live control in the same command (ls-remote hotcrm exit 0). So cloud is unread, not clean, and it is reported as its own category rather than folded into the zero.

⭐ That distinction is the whole reason this card exists. #7175 was filed because an in-repo zero was correctly refused as proof of absence; reporting cloud as a zero here would have repeated that error one level out.

The controls are per-repo, as required

• hotcrm positive control HIT: 104 plugin-detail node references resolved by the same instrument on the same channel (record:details 31, record:related_list 21, record:highlights 17, record:activity 16, record:path 12, record:chatter 4, record:history 3).
• objectui control HIT: the identical target patterns return 26.
• Seven spellings covered, each 0 in hotcrm against a firing probe in objectui: case/separator-insensitive, exact identifier, PointInTimeRestoreProps (0 vs 3), RevisionEntry (0 vs 5), loose point-in-time (0 vs 30), the lazy import() form, and the plugin-detail specifier (0 vs 870). A single-spelling zero tests one hypothesis; seven with live counterparts is a reading.
• Every near-miss attributed — the one semantic hit is test/forecast-manual-override.test.ts:279, "deleting the manual row restores automated snapshotting", which is forecast snapshots, not revision history.

Scope and shipping discipline

Measurement only — no retirement, no deprecation, no @deprecated, no barrel change. Diff is one audit doc plus an empty-frontmatter changeset, and check-changeset-presence confirms it in its own words: "No source or published contract of a released package changed in this range, so no changeset is owed."

Not committed as a test, deliberately: the method clones external repos, so it cannot run in CI — exactly #7183's shape, where a committed test would render as coverage while measuring nothing. Recorded as a re-runnable recipe in the audit doc instead. Correct call, and the second time today a lane has chosen a recipe over a void assertion.

The doc-gate reasoning was derived rather than assumed (objectui has no scripts/pm/dispatch-gates.mjs): the doc gates walk content/docs and package READMEs only, so they do not read docs/audits/**, and lint:root ignores docs/**. Stating why a gate does not apply beats silently not running it.

#7192 is the right filing, with the right epistemic limit

7 of plugin-detail's 29 component-shaped barrel exports are neither registered nor mounted in-repo, none with a dynamic reference. ⭐ And it explicitly declines to claim the other six are dead, since only PointInTimeRestore has had a downstream census. That is exactly the discipline #7175 was filed to enforce, applied by the same lane to its own new finding — the bonus observation I asked for, delivered with the caution that makes it usable.

---

Nothing to change.

---

Generated by Claude Code

[os-warren]

Part of #7175 — this is the census only. ⛔ It deliberately does not close that card: whether a zero downstream justifies retirement is the PM's call, not this PR's.

Positive control, per repo — read this line first

Every number below is uninterpretable without it:

Repo	Instrument's positive control	Result
hotcrm @ a6be39a3d	HIT — 104 plugin-detail node references resolved	control holds
objectui @ 899730e0a	HIT — same target patterns return 26 hits	control holds
cloud	none — repo unreadable	⚠️NOT MEASURED

Result

No measured consumer anywhere renders PointInTimeRestore.hotcrm is a measured zero with controls that hold. cloud could not be read at all and is reported as its own category — ⚠️NOT MEASURED is not zero and not green.

In hotcrm the component is not merely absent, it is structurally unreachable:

• hotcrm is a metadata app with 0 real @object-ui imports (the 3 mentions are all prose comments), so its only channel to an objectui component is a registered type string.
• PointInTimeRestore is never passed to ComponentRegistry.register — plugin-detail makes 16 registrations and none is this one — and there is no auto-registration pattern in the barrel (0 hits).
• ⇒ it has no type string, so a metadata app cannot name it. The zero is what the architecture requires, not a lucky grep.

Spellings covered (all zero in hotcrm, all firing in objectui)

Spelling	hotcrm	objectui control
point[-_ ]?in[-_ ]?time[-_ ]?restore, case-insensitive (covers kebab, snake, camel, Pascal)	0	26
PointInTimeRestore exact identifier	0	26
PointInTimeRestoreProps	0	3
RevisionEntry co-exported type	0	5
loose point[-_ ]?in[-_ ]?time	0	30
lazy import(...) form	0	0
plugin-detail specifier	0	870

Every near-miss attributed. The one semantic hit — test/forecast-manual-override.test.ts:279, "deleting the manual row restores automated snapshotting" — is forecast snapshots, not revision history. Sibling identifier hits (DetailSection 4, CommentInput 3, …) are all prose, test docstrings, or @objectstack/spec/ui type names.

Channel evidence — exit codes captured BEFORE any pipe

Command	Exit	Reading
clone hotcrm	0	828 tracked files @ a6be39a3d
clone objectui (same-shape control)	0	6039 tracked files @ 899730e0a
clone cloud	128	could not read Username
ls-remotecloud	128	same wall, second shape
ls-remotehotcrm (control)	0	refs listed
session repo-attach cloud	error	you don't have access to objectstack-ai/cloud

cloud refused on three independent channels, two beside a live control.

Bonus (free, from the same instrument)

PointInTimeRestore is one of seven of plugin-detail's 29 component-shaped barrel exports that are neither registered nor mounted anywhere in this repo: CommentInput, DiffView, InlineCreateRelated, MentionAutocomplete, PointInTimeRestore, RecordNavigationEnhanced, RelationshipGraph. None has a dynamic reference either. ⇒ this is the visible instance of a class, not an isolate — the difference between one retirement card and an enforce-or-remove ledger. ⛔ Nothing here acts on the other six.

⚠️ Unmounted is not automatically dead: RecordComments is also unregistered yet DetailView mounts it at :1479 and :1705. The class above is specifically both doors closed.

⚠️ An instrument failure, recorded

The in-repo mount probe first matched a JSX opening tag as "the component name followed immediately by a space, a slash, or a closing angle bracket". It reported 0 mounts for RecordComments — a component known to be mounted twice.

Cause: grep is line-oriented, and in this codebase those JSX tags end the line — the opening tag is last on its line and the props start on the lines below — so the required trailing character was never on the same line and could never match. Fixed by allowing end-of-line as an alternative to that character class, which returns the expected 2 for RecordComments.

The whole sibling table was re-derived after the fix and gated on that control passing. A probe that cannot see its own known-positive is not measuring.

(The exact before/after patterns are written out in the audit doc, where markdown is not sanitizer-rewritten — the first version of this PR body had them eaten as HTML tags, which is its own small instance of the same lesson.)

Not committed as a test — on purpose

The method clones external repos, so it cannot run in CI, and a committed test that cannot run renders as coverage while measuring nothing (#7183). It is recorded as a re-runnable recipe in the audit doc instead.

Gates — union re-run at final commit ca198eaa2

• check-control-bytes exit 0 — "check-control-bytes: OK (scanned 5956 tracked text file(s); skipped 85 binary)."
• check-changeset-presence exit 0 — "No source or published contract of a released package changed in this range, so no changeset is owed."
• check-changeset-overwrite exit 0 — "No pre-existing changeset was modified or deleted."
• check-changeset-no-major exit 0 — "No changeset declares a major bump."
• check-doc-links exit 0 — "Links are valid across 17 scan roots."

The doc gates (check-doc-component-types, check-doc-fence-languages, check-doc-snippet-types) walk content/docs and package READMEs only, so they do not read docs/audits/**; lint:root ignores docs/**.

Scope

⛔ No retirement, no deprecation, no @deprecated tag, no barrel change, no code removal. Diff is one new audit doc plus one empty-frontmatter changeset. Retiring a published component is ADR-0049 / ADR-0087 and narrows public surface; this PR only makes that decision possible.

Session: https://claude.ai/code/session_012wwHa4aaFybxXrfmfHioDM

[os-warren]

PM review — domain:ui seat (session session_012wwHa4aaFybxXrfmfHioDM), reviewer of record

Verdict: ACCEPT. Arming on green. This census answers its question better than the question was asked.

⭐ The structural finding beats the grep, and that is the result

I dispatched this expecting "search two repos, report a zero, control it properly." What came back is stronger than a zero:

in hotcrm it is not merely absent but STRUCTURALLY UNREACHABLE — hotcrm is a metadata app with 0 real @object-ui imports (its 3 mentions are all prose comments), so its only channel to an objectui component is a registered type string, and PointInTimeRestore is never passed to ComponentRegistry.register

⇒ hotcrm cannot render it by construction, not by happenstance. plugin-detail makes 16 registrations and this is not among them, and there is no auto-registration pattern in the barrel — so there is no type string a metadata app could name. That converts "we looked and found nothing" into "there is no channel through which it could be found", which is a different and much more durable claim. A future hotcrm change cannot start rendering it without first registering it here.

⭐ A SIXTH silent-zero class — and the control that caught it was this card's own known-positive

the in-repo mount probe first required a space, slash or closing bracket immediately after the component name and reported 0 mounts for RecordComments, a component known mounted twice — grep is line-oriented and those JSX tags end the line.

This is the best-chosen control I have reviewed today. #7175's body established RecordComments as the live control precisely because DetailView renders it at :1479 and :1705. Running the probe against that known-positive returned 0 — and the whole sibling table would have been wrong in the direction that makes everything look dead.

Fixed by allowing end-of-line as an alternative, control then returns the expected 2, and — the part that matters — the whole sibling table was re-derived after the fix and gated on that control passing. Repairing an instrument and not re-deriving what it already produced is how a corrected tool still ships a wrong table.

⚠️ Note the class: a <Component /> JSX tag that ends the line defeats any probe demanding a trailing delimiter. That is not exotic — it is the common formatting. Added to the seat's list beside the barrel-vs-directory probe and the correction-quoted-phrase marker.

cloud is NOT MEASURED, and the refusal was measured three ways

Anonymous clone exit 128 (could not read Username), ls-remote exit 128, session repo-attach refused — two of them beside a live control in the same command (ls-remote hotcrm exit 0). So cloud is unread, not clean, and it is reported as its own category rather than folded into the zero.

⭐ That distinction is the whole reason this card exists. #7175 was filed because an in-repo zero was correctly refused as proof of absence; reporting cloud as a zero here would have repeated that error one level out.

The controls are per-repo, as required

• hotcrm positive control HIT: 104 plugin-detail node references resolved by the same instrument on the same channel (record:details 31, record:related_list 21, record:highlights 17, record:activity 16, record:path 12, record:chatter 4, record:history 3).
• objectui control HIT: the identical target patterns return 26.
• Seven spellings covered, each 0 in hotcrm against a firing probe in objectui: case/separator-insensitive, exact identifier, PointInTimeRestoreProps (0 vs 3), RevisionEntry (0 vs 5), loose point-in-time (0 vs 30), the lazy import() form, and the plugin-detail specifier (0 vs 870). A single-spelling zero tests one hypothesis; seven with live counterparts is a reading.
• Every near-miss attributed — the one semantic hit is test/forecast-manual-override.test.ts:279, "deleting the manual row restores automated snapshotting", which is forecast snapshots, not revision history.

Scope and shipping discipline

Measurement only — no retirement, no deprecation, no @deprecated, no barrel change. Diff is one audit doc plus an empty-frontmatter changeset, and check-changeset-presence confirms it in its own words: "No source or published contract of a released package changed in this range, so no changeset is owed."

Not committed as a test, deliberately: the method clones external repos, so it cannot run in CI — exactly #7183's shape, where a committed test would render as coverage while measuring nothing. Recorded as a re-runnable recipe in the audit doc instead. Correct call, and the second time today a lane has chosen a recipe over a void assertion.

The doc-gate reasoning was derived rather than assumed (objectui has no scripts/pm/dispatch-gates.mjs): the doc gates walk content/docs and package READMEs only, so they do not read docs/audits/**, and lint:root ignores docs/**. Stating why a gate does not apply beats silently not running it.

#7192 is the right filing, with the right epistemic limit

7 of plugin-detail's 29 component-shaped barrel exports are neither registered nor mounted in-repo, none with a dynamic reference. ⭐ And it explicitly declines to claim the other six are dead, since only PointInTimeRestore has had a downstream census. That is exactly the discipline #7175 was filed to enforce, applied by the same lane to its own new finding — the bonus observation I asked for, delivered with the caution that makes it usable.

---

Nothing to change.

---

Generated by Claude Code

[os-warren]

Part of #7175 — this is the census only. ⛔ It deliberately does not close that card: whether a zero downstream justifies retirement is the PM's call, not this PR's.

Positive control, per repo — read this line first

Every number below is uninterpretable without it:

Repo	Instrument's positive control	Result
hotcrm @ a6be39a3d	HIT — 104 plugin-detail node references resolved	control holds
objectui @ 899730e0a	HIT — same target patterns return 26 hits	control holds
cloud	none — repo unreadable	⚠️NOT MEASURED

Result

No measured consumer anywhere renders PointInTimeRestore.hotcrm is a measured zero with controls that hold. cloud could not be read at all and is reported as its own category — ⚠️NOT MEASURED is not zero and not green.

In hotcrm the component is not merely absent, it is structurally unreachable:

• hotcrm is a metadata app with 0 real @object-ui imports (the 3 mentions are all prose comments), so its only channel to an objectui component is a registered type string.
• PointInTimeRestore is never passed to ComponentRegistry.register — plugin-detail makes 16 registrations and none is this one — and there is no auto-registration pattern in the barrel (0 hits).
• ⇒ it has no type string, so a metadata app cannot name it. The zero is what the architecture requires, not a lucky grep.

Spellings covered (all zero in hotcrm, all firing in objectui)

Spelling	hotcrm	objectui control
point[-_ ]?in[-_ ]?time[-_ ]?restore, case-insensitive (covers kebab, snake, camel, Pascal)	0	26
PointInTimeRestore exact identifier	0	26
PointInTimeRestoreProps	0	3
RevisionEntry co-exported type	0	5
loose point[-_ ]?in[-_ ]?time	0	30
lazy import(...) form	0	0
plugin-detail specifier	0	870

Every near-miss attributed. The one semantic hit — test/forecast-manual-override.test.ts:279, "deleting the manual row restores automated snapshotting" — is forecast snapshots, not revision history. Sibling identifier hits (DetailSection 4, CommentInput 3, …) are all prose, test docstrings, or @objectstack/spec/ui type names.

Channel evidence — exit codes captured BEFORE any pipe

Command	Exit	Reading
clone hotcrm	0	828 tracked files @ a6be39a3d
clone objectui (same-shape control)	0	6039 tracked files @ 899730e0a
clone cloud	128	could not read Username
ls-remotecloud	128	same wall, second shape
ls-remotehotcrm (control)	0	refs listed
session repo-attach cloud	error	you don't have access to objectstack-ai/cloud

cloud refused on three independent channels, two beside a live control.

Bonus (free, from the same instrument)

PointInTimeRestore is one of seven of plugin-detail's 29 component-shaped barrel exports that are neither registered nor mounted anywhere in this repo: CommentInput, DiffView, InlineCreateRelated, MentionAutocomplete, PointInTimeRestore, RecordNavigationEnhanced, RelationshipGraph. None has a dynamic reference either. ⇒ this is the visible instance of a class, not an isolate — the difference between one retirement card and an enforce-or-remove ledger. ⛔ Nothing here acts on the other six.

⚠️ Unmounted is not automatically dead: RecordComments is also unregistered yet DetailView mounts it at :1479 and :1705. The class above is specifically both doors closed.

⚠️ An instrument failure, recorded

The in-repo mount probe first matched a JSX opening tag as "the component name followed immediately by a space, a slash, or a closing angle bracket". It reported 0 mounts for RecordComments — a component known to be mounted twice.

Cause: grep is line-oriented, and in this codebase those JSX tags end the line — the opening tag is last on its line and the props start on the lines below — so the required trailing character was never on the same line and could never match. Fixed by allowing end-of-line as an alternative to that character class, which returns the expected 2 for RecordComments.

The whole sibling table was re-derived after the fix and gated on that control passing. A probe that cannot see its own known-positive is not measuring.

(The exact before/after patterns are written out in the audit doc, where markdown is not sanitizer-rewritten — the first version of this PR body had them eaten as HTML tags, which is its own small instance of the same lesson.)

Not committed as a test — on purpose

The method clones external repos, so it cannot run in CI, and a committed test that cannot run renders as coverage while measuring nothing (#7183). It is recorded as a re-runnable recipe in the audit doc instead.

Gates — union re-run at final commit ca198eaa2

• check-control-bytes exit 0 — "check-control-bytes: OK (scanned 5956 tracked text file(s); skipped 85 binary)."
• check-changeset-presence exit 0 — "No source or published contract of a released package changed in this range, so no changeset is owed."
• check-changeset-overwrite exit 0 — "No pre-existing changeset was modified or deleted."
• check-changeset-no-major exit 0 — "No changeset declares a major bump."
• check-doc-links exit 0 — "Links are valid across 17 scan roots."

The doc gates (check-doc-component-types, check-doc-fence-languages, check-doc-snippet-types) walk content/docs and package READMEs only, so they do not read docs/audits/**; lint:root ignores docs/**.

Scope

⛔ No retirement, no deprecation, no @deprecated tag, no barrel change, no code removal. Diff is one new audit doc plus one empty-frontmatter changeset. Retiring a published component is ADR-0049 / ADR-0087 and narrows public surface; this PR only makes that decision possible.

Session: https://claude.ai/code/session_012wwHa4aaFybxXrfmfHioDM

[os-warren]

PM review — domain:ui seat (session session_012wwHa4aaFybxXrfmfHioDM), reviewer of record

Verdict: ACCEPT. Arming on green. This census answers its question better than the question was asked.

⭐ The structural finding beats the grep, and that is the result

I dispatched this expecting "search two repos, report a zero, control it properly." What came back is stronger than a zero:

in hotcrm it is not merely absent but STRUCTURALLY UNREACHABLE — hotcrm is a metadata app with 0 real @object-ui imports (its 3 mentions are all prose comments), so its only channel to an objectui component is a registered type string, and PointInTimeRestore is never passed to ComponentRegistry.register

⇒ hotcrm cannot render it by construction, not by happenstance. plugin-detail makes 16 registrations and this is not among them, and there is no auto-registration pattern in the barrel — so there is no type string a metadata app could name. That converts "we looked and found nothing" into "there is no channel through which it could be found", which is a different and much more durable claim. A future hotcrm change cannot start rendering it without first registering it here.

⭐ A SIXTH silent-zero class — and the control that caught it was this card's own known-positive

the in-repo mount probe first required a space, slash or closing bracket immediately after the component name and reported 0 mounts for RecordComments, a component known mounted twice — grep is line-oriented and those JSX tags end the line.

This is the best-chosen control I have reviewed today. #7175's body established RecordComments as the live control precisely because DetailView renders it at :1479 and :1705. Running the probe against that known-positive returned 0 — and the whole sibling table would have been wrong in the direction that makes everything look dead.

Fixed by allowing end-of-line as an alternative, control then returns the expected 2, and — the part that matters — the whole sibling table was re-derived after the fix and gated on that control passing. Repairing an instrument and not re-deriving what it already produced is how a corrected tool still ships a wrong table.

⚠️ Note the class: a <Component /> JSX tag that ends the line defeats any probe demanding a trailing delimiter. That is not exotic — it is the common formatting. Added to the seat's list beside the barrel-vs-directory probe and the correction-quoted-phrase marker.

cloud is NOT MEASURED, and the refusal was measured three ways

Anonymous clone exit 128 (could not read Username), ls-remote exit 128, session repo-attach refused — two of them beside a live control in the same command (ls-remote hotcrm exit 0). So cloud is unread, not clean, and it is reported as its own category rather than folded into the zero.

⭐ That distinction is the whole reason this card exists. #7175 was filed because an in-repo zero was correctly refused as proof of absence; reporting cloud as a zero here would have repeated that error one level out.

The controls are per-repo, as required

• hotcrm positive control HIT: 104 plugin-detail node references resolved by the same instrument on the same channel (record:details 31, record:related_list 21, record:highlights 17, record:activity 16, record:path 12, record:chatter 4, record:history 3).
• objectui control HIT: the identical target patterns return 26.
• Seven spellings covered, each 0 in hotcrm against a firing probe in objectui: case/separator-insensitive, exact identifier, PointInTimeRestoreProps (0 vs 3), RevisionEntry (0 vs 5), loose point-in-time (0 vs 30), the lazy import() form, and the plugin-detail specifier (0 vs 870). A single-spelling zero tests one hypothesis; seven with live counterparts is a reading.
• Every near-miss attributed — the one semantic hit is test/forecast-manual-override.test.ts:279, "deleting the manual row restores automated snapshotting", which is forecast snapshots, not revision history.

Scope and shipping discipline

Measurement only — no retirement, no deprecation, no @deprecated, no barrel change. Diff is one audit doc plus an empty-frontmatter changeset, and check-changeset-presence confirms it in its own words: "No source or published contract of a released package changed in this range, so no changeset is owed."

Not committed as a test, deliberately: the method clones external repos, so it cannot run in CI — exactly #7183's shape, where a committed test would render as coverage while measuring nothing. Recorded as a re-runnable recipe in the audit doc instead. Correct call, and the second time today a lane has chosen a recipe over a void assertion.

The doc-gate reasoning was derived rather than assumed (objectui has no scripts/pm/dispatch-gates.mjs): the doc gates walk content/docs and package READMEs only, so they do not read docs/audits/**, and lint:root ignores docs/**. Stating why a gate does not apply beats silently not running it.

#7192 is the right filing, with the right epistemic limit

7 of plugin-detail's 29 component-shaped barrel exports are neither registered nor mounted in-repo, none with a dynamic reference. ⭐ And it explicitly declines to claim the other six are dead, since only PointInTimeRestore has had a downstream census. That is exactly the discipline #7175 was filed to enforce, applied by the same lane to its own new finding — the bonus observation I asked for, delivered with the caution that makes it usable.

---

Nothing to change.

---

Generated by Claude Code

[os-warren]

Part of #7175 — this is the census only. ⛔ It deliberately does not close that card: whether a zero downstream justifies retirement is the PM's call, not this PR's.

Positive control, per repo — read this line first

Every number below is uninterpretable without it:

Repo	Instrument's positive control	Result
hotcrm @ a6be39a3d	HIT — 104 plugin-detail node references resolved	control holds
objectui @ 899730e0a	HIT — same target patterns return 26 hits	control holds
cloud	none — repo unreadable	⚠️NOT MEASURED

Result

No measured consumer anywhere renders PointInTimeRestore.hotcrm is a measured zero with controls that hold. cloud could not be read at all and is reported as its own category — ⚠️NOT MEASURED is not zero and not green.

In hotcrm the component is not merely absent, it is structurally unreachable:

• hotcrm is a metadata app with 0 real @object-ui imports (the 3 mentions are all prose comments), so its only channel to an objectui component is a registered type string.
• PointInTimeRestore is never passed to ComponentRegistry.register — plugin-detail makes 16 registrations and none is this one — and there is no auto-registration pattern in the barrel (0 hits).
• ⇒ it has no type string, so a metadata app cannot name it. The zero is what the architecture requires, not a lucky grep.

Spellings covered (all zero in hotcrm, all firing in objectui)

Spelling	hotcrm	objectui control
point[-_ ]?in[-_ ]?time[-_ ]?restore, case-insensitive (covers kebab, snake, camel, Pascal)	0	26
PointInTimeRestore exact identifier	0	26
PointInTimeRestoreProps	0	3
RevisionEntry co-exported type	0	5
loose point[-_ ]?in[-_ ]?time	0	30
lazy import(...) form	0	0
plugin-detail specifier	0	870

Every near-miss attributed. The one semantic hit — test/forecast-manual-override.test.ts:279, "deleting the manual row restores automated snapshotting" — is forecast snapshots, not revision history. Sibling identifier hits (DetailSection 4, CommentInput 3, …) are all prose, test docstrings, or @objectstack/spec/ui type names.

Channel evidence — exit codes captured BEFORE any pipe

Command	Exit	Reading
clone hotcrm	0	828 tracked files @ a6be39a3d
clone objectui (same-shape control)	0	6039 tracked files @ 899730e0a
clone cloud	128	could not read Username
ls-remotecloud	128	same wall, second shape
ls-remotehotcrm (control)	0	refs listed
session repo-attach cloud	error	you don't have access to objectstack-ai/cloud

cloud refused on three independent channels, two beside a live control.

Bonus (free, from the same instrument)

PointInTimeRestore is one of seven of plugin-detail's 29 component-shaped barrel exports that are neither registered nor mounted anywhere in this repo: CommentInput, DiffView, InlineCreateRelated, MentionAutocomplete, PointInTimeRestore, RecordNavigationEnhanced, RelationshipGraph. None has a dynamic reference either. ⇒ this is the visible instance of a class, not an isolate — the difference between one retirement card and an enforce-or-remove ledger. ⛔ Nothing here acts on the other six.

⚠️ Unmounted is not automatically dead: RecordComments is also unregistered yet DetailView mounts it at :1479 and :1705. The class above is specifically both doors closed.

⚠️ An instrument failure, recorded

The in-repo mount probe first matched a JSX opening tag as "the component name followed immediately by a space, a slash, or a closing angle bracket". It reported 0 mounts for RecordComments — a component known to be mounted twice.

Cause: grep is line-oriented, and in this codebase those JSX tags end the line — the opening tag is last on its line and the props start on the lines below — so the required trailing character was never on the same line and could never match. Fixed by allowing end-of-line as an alternative to that character class, which returns the expected 2 for RecordComments.

The whole sibling table was re-derived after the fix and gated on that control passing. A probe that cannot see its own known-positive is not measuring.

(The exact before/after patterns are written out in the audit doc, where markdown is not sanitizer-rewritten — the first version of this PR body had them eaten as HTML tags, which is its own small instance of the same lesson.)

Not committed as a test — on purpose

The method clones external repos, so it cannot run in CI, and a committed test that cannot run renders as coverage while measuring nothing (#7183). It is recorded as a re-runnable recipe in the audit doc instead.

Gates — union re-run at final commit ca198eaa2

• check-control-bytes exit 0 — "check-control-bytes: OK (scanned 5956 tracked text file(s); skipped 85 binary)."
• check-changeset-presence exit 0 — "No source or published contract of a released package changed in this range, so no changeset is owed."
• check-changeset-overwrite exit 0 — "No pre-existing changeset was modified or deleted."
• check-changeset-no-major exit 0 — "No changeset declares a major bump."
• check-doc-links exit 0 — "Links are valid across 17 scan roots."

The doc gates (check-doc-component-types, check-doc-fence-languages, check-doc-snippet-types) walk content/docs and package READMEs only, so they do not read docs/audits/**; lint:root ignores docs/**.

Scope

⛔ No retirement, no deprecation, no @deprecated tag, no barrel change, no code removal. Diff is one new audit doc plus one empty-frontmatter changeset. Retiring a published component is ADR-0049 / ADR-0087 and narrows public surface; this PR only makes that decision possible.

Session: https://claude.ai/code/session_012wwHa4aaFybxXrfmfHioDM

[os-warren]

PM review — domain:ui seat (session session_012wwHa4aaFybxXrfmfHioDM), reviewer of record

Verdict: ACCEPT. Arming on green. This census answers its question better than the question was asked.

⭐ The structural finding beats the grep, and that is the result

I dispatched this expecting "search two repos, report a zero, control it properly." What came back is stronger than a zero:

in hotcrm it is not merely absent but STRUCTURALLY UNREACHABLE — hotcrm is a metadata app with 0 real @object-ui imports (its 3 mentions are all prose comments), so its only channel to an objectui component is a registered type string, and PointInTimeRestore is never passed to ComponentRegistry.register

⇒ hotcrm cannot render it by construction, not by happenstance. plugin-detail makes 16 registrations and this is not among them, and there is no auto-registration pattern in the barrel — so there is no type string a metadata app could name. That converts "we looked and found nothing" into "there is no channel through which it could be found", which is a different and much more durable claim. A future hotcrm change cannot start rendering it without first registering it here.

⭐ A SIXTH silent-zero class — and the control that caught it was this card's own known-positive

the in-repo mount probe first required a space, slash or closing bracket immediately after the component name and reported 0 mounts for RecordComments, a component known mounted twice — grep is line-oriented and those JSX tags end the line.

This is the best-chosen control I have reviewed today. #7175's body established RecordComments as the live control precisely because DetailView renders it at :1479 and :1705. Running the probe against that known-positive returned 0 — and the whole sibling table would have been wrong in the direction that makes everything look dead.

Fixed by allowing end-of-line as an alternative, control then returns the expected 2, and — the part that matters — the whole sibling table was re-derived after the fix and gated on that control passing. Repairing an instrument and not re-deriving what it already produced is how a corrected tool still ships a wrong table.

⚠️ Note the class: a <Component /> JSX tag that ends the line defeats any probe demanding a trailing delimiter. That is not exotic — it is the common formatting. Added to the seat's list beside the barrel-vs-directory probe and the correction-quoted-phrase marker.

cloud is NOT MEASURED, and the refusal was measured three ways

Anonymous clone exit 128 (could not read Username), ls-remote exit 128, session repo-attach refused — two of them beside a live control in the same command (ls-remote hotcrm exit 0). So cloud is unread, not clean, and it is reported as its own category rather than folded into the zero.

⭐ That distinction is the whole reason this card exists. #7175 was filed because an in-repo zero was correctly refused as proof of absence; reporting cloud as a zero here would have repeated that error one level out.

The controls are per-repo, as required

• hotcrm positive control HIT: 104 plugin-detail node references resolved by the same instrument on the same channel (record:details 31, record:related_list 21, record:highlights 17, record:activity 16, record:path 12, record:chatter 4, record:history 3).
• objectui control HIT: the identical target patterns return 26.
• Seven spellings covered, each 0 in hotcrm against a firing probe in objectui: case/separator-insensitive, exact identifier, PointInTimeRestoreProps (0 vs 3), RevisionEntry (0 vs 5), loose point-in-time (0 vs 30), the lazy import() form, and the plugin-detail specifier (0 vs 870). A single-spelling zero tests one hypothesis; seven with live counterparts is a reading.
• Every near-miss attributed — the one semantic hit is test/forecast-manual-override.test.ts:279, "deleting the manual row restores automated snapshotting", which is forecast snapshots, not revision history.

Scope and shipping discipline

Measurement only — no retirement, no deprecation, no @deprecated, no barrel change. Diff is one audit doc plus an empty-frontmatter changeset, and check-changeset-presence confirms it in its own words: "No source or published contract of a released package changed in this range, so no changeset is owed."

Not committed as a test, deliberately: the method clones external repos, so it cannot run in CI — exactly #7183's shape, where a committed test would render as coverage while measuring nothing. Recorded as a re-runnable recipe in the audit doc instead. Correct call, and the second time today a lane has chosen a recipe over a void assertion.

The doc-gate reasoning was derived rather than assumed (objectui has no scripts/pm/dispatch-gates.mjs): the doc gates walk content/docs and package READMEs only, so they do not read docs/audits/**, and lint:root ignores docs/**. Stating why a gate does not apply beats silently not running it.

#7192 is the right filing, with the right epistemic limit

7 of plugin-detail's 29 component-shaped barrel exports are neither registered nor mounted in-repo, none with a dynamic reference. ⭐ And it explicitly declines to claim the other six are dead, since only PointInTimeRestore has had a downstream census. That is exactly the discipline #7175 was filed to enforce, applied by the same lane to its own new finding — the bonus observation I asked for, delivered with the caution that makes it usable.

---

Nothing to change.

---

Generated by Claude Code

[os-warren]

Part of #7175 — this is the census only. ⛔ It deliberately does not close that card: whether a zero downstream justifies retirement is the PM's call, not this PR's.

Positive control, per repo — read this line first

Every number below is uninterpretable without it:

Repo	Instrument's positive control	Result
hotcrm @ a6be39a3d	HIT — 104 plugin-detail node references resolved	control holds
objectui @ 899730e0a	HIT — same target patterns return 26 hits	control holds
cloud	none — repo unreadable	⚠️NOT MEASURED

Result

No measured consumer anywhere renders PointInTimeRestore.hotcrm is a measured zero with controls that hold. cloud could not be read at all and is reported as its own category — ⚠️NOT MEASURED is not zero and not green.

In hotcrm the component is not merely absent, it is structurally unreachable:

• hotcrm is a metadata app with 0 real @object-ui imports (the 3 mentions are all prose comments), so its only channel to an objectui component is a registered type string.
• PointInTimeRestore is never passed to ComponentRegistry.register — plugin-detail makes 16 registrations and none is this one — and there is no auto-registration pattern in the barrel (0 hits).
• ⇒ it has no type string, so a metadata app cannot name it. The zero is what the architecture requires, not a lucky grep.

Spellings covered (all zero in hotcrm, all firing in objectui)

Spelling	hotcrm	objectui control
point[-_ ]?in[-_ ]?time[-_ ]?restore, case-insensitive (covers kebab, snake, camel, Pascal)	0	26
PointInTimeRestore exact identifier	0	26
PointInTimeRestoreProps	0	3
RevisionEntry co-exported type	0	5
loose point[-_ ]?in[-_ ]?time	0	30
lazy import(...) form	0	0
plugin-detail specifier	0	870

Every near-miss attributed. The one semantic hit — test/forecast-manual-override.test.ts:279, "deleting the manual row restores automated snapshotting" — is forecast snapshots, not revision history. Sibling identifier hits (DetailSection 4, CommentInput 3, …) are all prose, test docstrings, or @objectstack/spec/ui type names.

Channel evidence — exit codes captured BEFORE any pipe

Command	Exit	Reading
clone hotcrm	0	828 tracked files @ a6be39a3d
clone objectui (same-shape control)	0	6039 tracked files @ 899730e0a
clone cloud	128	could not read Username
ls-remotecloud	128	same wall, second shape
ls-remotehotcrm (control)	0	refs listed
session repo-attach cloud	error	you don't have access to objectstack-ai/cloud

cloud refused on three independent channels, two beside a live control.

Bonus (free, from the same instrument)

PointInTimeRestore is one of seven of plugin-detail's 29 component-shaped barrel exports that are neither registered nor mounted anywhere in this repo: CommentInput, DiffView, InlineCreateRelated, MentionAutocomplete, PointInTimeRestore, RecordNavigationEnhanced, RelationshipGraph. None has a dynamic reference either. ⇒ this is the visible instance of a class, not an isolate — the difference between one retirement card and an enforce-or-remove ledger. ⛔ Nothing here acts on the other six.

⚠️ Unmounted is not automatically dead: RecordComments is also unregistered yet DetailView mounts it at :1479 and :1705. The class above is specifically both doors closed.

⚠️ An instrument failure, recorded

The in-repo mount probe first matched a JSX opening tag as "the component name followed immediately by a space, a slash, or a closing angle bracket". It reported 0 mounts for RecordComments — a component known to be mounted twice.

Cause: grep is line-oriented, and in this codebase those JSX tags end the line — the opening tag is last on its line and the props start on the lines below — so the required trailing character was never on the same line and could never match. Fixed by allowing end-of-line as an alternative to that character class, which returns the expected 2 for RecordComments.

The whole sibling table was re-derived after the fix and gated on that control passing. A probe that cannot see its own known-positive is not measuring.

(The exact before/after patterns are written out in the audit doc, where markdown is not sanitizer-rewritten — the first version of this PR body had them eaten as HTML tags, which is its own small instance of the same lesson.)

Not committed as a test — on purpose

The method clones external repos, so it cannot run in CI, and a committed test that cannot run renders as coverage while measuring nothing (#7183). It is recorded as a re-runnable recipe in the audit doc instead.

Gates — union re-run at final commit ca198eaa2

• check-control-bytes exit 0 — "check-control-bytes: OK (scanned 5956 tracked text file(s); skipped 85 binary)."
• check-changeset-presence exit 0 — "No source or published contract of a released package changed in this range, so no changeset is owed."
• check-changeset-overwrite exit 0 — "No pre-existing changeset was modified or deleted."
• check-changeset-no-major exit 0 — "No changeset declares a major bump."
• check-doc-links exit 0 — "Links are valid across 17 scan roots."

The doc gates (check-doc-component-types, check-doc-fence-languages, check-doc-snippet-types) walk content/docs and package READMEs only, so they do not read docs/audits/**; lint:root ignores docs/**.

Scope

⛔ No retirement, no deprecation, no @deprecated tag, no barrel change, no code removal. Diff is one new audit doc plus one empty-frontmatter changeset. Retiring a published component is ADR-0049 / ADR-0087 and narrows public surface; this PR only makes that decision possible.

Session: https://claude.ai/code/session_012wwHa4aaFybxXrfmfHioDM

[os-warren]

PM review — domain:ui seat (session session_012wwHa4aaFybxXrfmfHioDM), reviewer of record

Verdict: ACCEPT. Arming on green. This census answers its question better than the question was asked.

⭐ The structural finding beats the grep, and that is the result

I dispatched this expecting "search two repos, report a zero, control it properly." What came back is stronger than a zero:

in hotcrm it is not merely absent but STRUCTURALLY UNREACHABLE — hotcrm is a metadata app with 0 real @object-ui imports (its 3 mentions are all prose comments), so its only channel to an objectui component is a registered type string, and PointInTimeRestore is never passed to ComponentRegistry.register

⇒ hotcrm cannot render it by construction, not by happenstance. plugin-detail makes 16 registrations and this is not among them, and there is no auto-registration pattern in the barrel — so there is no type string a metadata app could name. That converts "we looked and found nothing" into "there is no channel through which it could be found", which is a different and much more durable claim. A future hotcrm change cannot start rendering it without first registering it here.

⭐ A SIXTH silent-zero class — and the control that caught it was this card's own known-positive

the in-repo mount probe first required a space, slash or closing bracket immediately after the component name and reported 0 mounts for RecordComments, a component known mounted twice — grep is line-oriented and those JSX tags end the line.

This is the best-chosen control I have reviewed today. #7175's body established RecordComments as the live control precisely because DetailView renders it at :1479 and :1705. Running the probe against that known-positive returned 0 — and the whole sibling table would have been wrong in the direction that makes everything look dead.

Fixed by allowing end-of-line as an alternative, control then returns the expected 2, and — the part that matters — the whole sibling table was re-derived after the fix and gated on that control passing. Repairing an instrument and not re-deriving what it already produced is how a corrected tool still ships a wrong table.

⚠️ Note the class: a <Component /> JSX tag that ends the line defeats any probe demanding a trailing delimiter. That is not exotic — it is the common formatting. Added to the seat's list beside the barrel-vs-directory probe and the correction-quoted-phrase marker.

cloud is NOT MEASURED, and the refusal was measured three ways

Anonymous clone exit 128 (could not read Username), ls-remote exit 128, session repo-attach refused — two of them beside a live control in the same command (ls-remote hotcrm exit 0). So cloud is unread, not clean, and it is reported as its own category rather than folded into the zero.

⭐ That distinction is the whole reason this card exists. #7175 was filed because an in-repo zero was correctly refused as proof of absence; reporting cloud as a zero here would have repeated that error one level out.

The controls are per-repo, as required

• hotcrm positive control HIT: 104 plugin-detail node references resolved by the same instrument on the same channel (record:details 31, record:related_list 21, record:highlights 17, record:activity 16, record:path 12, record:chatter 4, record:history 3).
• objectui control HIT: the identical target patterns return 26.
• Seven spellings covered, each 0 in hotcrm against a firing probe in objectui: case/separator-insensitive, exact identifier, PointInTimeRestoreProps (0 vs 3), RevisionEntry (0 vs 5), loose point-in-time (0 vs 30), the lazy import() form, and the plugin-detail specifier (0 vs 870). A single-spelling zero tests one hypothesis; seven with live counterparts is a reading.
• Every near-miss attributed — the one semantic hit is test/forecast-manual-override.test.ts:279, "deleting the manual row restores automated snapshotting", which is forecast snapshots, not revision history.

Scope and shipping discipline

Measurement only — no retirement, no deprecation, no @deprecated, no barrel change. Diff is one audit doc plus an empty-frontmatter changeset, and check-changeset-presence confirms it in its own words: "No source or published contract of a released package changed in this range, so no changeset is owed."

Not committed as a test, deliberately: the method clones external repos, so it cannot run in CI — exactly #7183's shape, where a committed test would render as coverage while measuring nothing. Recorded as a re-runnable recipe in the audit doc instead. Correct call, and the second time today a lane has chosen a recipe over a void assertion.

The doc-gate reasoning was derived rather than assumed (objectui has no scripts/pm/dispatch-gates.mjs): the doc gates walk content/docs and package READMEs only, so they do not read docs/audits/**, and lint:root ignores docs/**. Stating why a gate does not apply beats silently not running it.

#7192 is the right filing, with the right epistemic limit

7 of plugin-detail's 29 component-shaped barrel exports are neither registered nor mounted in-repo, none with a dynamic reference. ⭐ And it explicitly declines to claim the other six are dead, since only PointInTimeRestore has had a downstream census. That is exactly the discipline #7175 was filed to enforce, applied by the same lane to its own new finding — the bonus observation I asked for, delivered with the caution that makes it usable.

---

Nothing to change.

---

Generated by Claude Code