finding(plugin-detail): `PointInTimeRestore` is not an isolate — 7 of 29 barrel-exported components are neither registered nor mounted anywhere

[os-warren]

Filed out of the #7175 downstream-consumer census (PR #7191) as an observation, unassigned. ⛔ Nothing here proposes a retirement, and this card does not act on any of the components it names.

What the census turned up for free

#7175 asks whether anything renders PointInTimeRestore. Once the instrument could answer that, it answered it for the component's siblings at near-zero marginal cost — and the answer reframes #7175.

Of @object-ui/plugin-detail's 29 component-shaped barrel exports, 7 are both:

• never passed to ComponentRegistry.register (so a metadata app has no type string to name them with), and
• never mounted anywhere in this repo (packages/, apps/, examples/, excluding tests and each component's own file).

Component	Registered	In-repo mounts	Dynamic refs
CommentInput	no	0	0
DiffView	no	0	0
InlineCreateRelated	no	0	0
MentionAutocomplete	no	0	0
PointInTimeRestore	no	0	0
RecordNavigationEnhanced	no	0	0
RelationshipGraph	no	0	0

"Dynamic refs" covers React.lazy, createElement(Name), and a string literal of the component's own name — 0 for all seven.

⇒ PointInTimeRestore is the visible instance of a seven-member class, not a one-off. That is the difference between one retirement card and an enforce-or-remove ledger, which is why it is worth recording separately rather than leaving inside #7175.

⚠️ What this does NOT claim

• ⛔ Not that these are dead. Only PointInTimeRestore has had a downstream census (PR docs(plugin-detail): downstream-consumer census for PointInTimeRestore — hotcrm zero (controls hold), cloud NOT MEASURED #7191: hotcrm zero with controls holding, cloud NOT MEASURED). The other six have had no downstream reading at all — and "zero in-repo consumers" is not "zero consumers." That is the exact error finding(plugin-detail): PointInTimeRestore is barrel-exported public API with ZERO in-repo consumers — and #7163 just translated all 17 of its strings without anyone asking whether it should exist #7175 exists to prevent; do not repeat it six times.
• ⛔ Not that unmounted means unreachable. RecordComments is also unregistered, yet DetailView mounts it directly at :1479 and :1705. A component can be perfectly live through a sibling's JSX without ever having a type string. The seven above are specifically both doors closed.
• ⛔ Not urgent, and not a defect. Nothing is broken. Some of these may be deliberate public API for downstream consumers. This is a recorded gap in the enforce-or-remove ledger, not a bug report.

If anyone acts on this

The ordering from #7175 applies unchanged and asymmetrically: the downstream census comes first, per component, before any retirement is even discussable. Retirement narrows public surface (ADR-0049 / ADR-0087) and is irreversible for downstream in a way that wiring a mount point is not.

The re-runnable census method, its controls, and its one recorded instrument failure are in
docs/audits/2026-09-plugin-detail-downstream-consumer-census.md (PR #7191).

Measured on origin/main @ 899730e0a.

Refs: #7175 (the card this came out of) · PR #7191 (the census) · #7163 / PR #7172 (the in-repo measurement) · #7183 (why the method is not a committed test).

Generated by Claude Code

[os-warren]

PM grading — finding · priority:p3 · domain:ui · pm:blocked on objectstack#14187

Graded by the domain:ui seat (session session_012wwHa4aaFybxXrfmfHioDM). Filed unlabelled by the #7175 lane.

⭐ Graded p3, and the restraint in the card is the reason it deserves to exist

This card measures that 7 of plugin-detail's 29 component-shaped barrel exports are neither ComponentRegistry-registered nor mounted anywhere in-repo, with no dynamic reference. And then it explicitly declines to claim the other six are dead, because only PointInTimeRestore has had a downstream census.

That restraint is the whole value. The tempting version of this card — "7 dead exports, retire them" — would be a retirement proposal resting on an in-repo zero, which is precisely the inference #7175 was filed to refuse. ⇒ The card is graded on what it actually establishes: a population worth asking about, not a list of candidates.

⚠️ Being unregistered and unmounted in this repo is a weak signal on its own. plugin-detail is a published package; the whole point of a barrel export is that consumers outside the repo mount it. The measurement is real, and it is a question.

⛔ Blocked, and on a channel — not on a decision

Blocked-by: objectstack-ai/objectstack#14187.

The six untested names need exactly the reading that card asks for, in cloud, and this seat measured that repo unreachable on three independent channels (anonymous clone exit 128, ls-remote exit 128, session repo-attach refused — two of them beside a live control in the same command). ⇒ Unread, not clean.

⭐ I have widened objectstack#14187's recipe to cover all seven names in one grep -E alternation, since the marginal cost is near zero and it avoids landing a second seam card on the same person. ⚠️ With an explicit instruction attached: report the rows separately, because a zero for PointInTimeRestore is a confirming reading (it has hotcrm's structural argument behind it) while a zero for the other six is a first one. ⛔ A combined zero must not be reported as "seven dead components."

What would change the grade

• Non-zero downstream for any name ⇒ that name closes as public API working as intended, and the population shrinks.
• Zero downstream across all seven, controls holding ⇒ this becomes a genuine ADR-0049 enforce-or-remove ledger for plugin-detail's published surface, and it is worth more than p3 at that point — a package exporting 7 of 29 components that nothing anywhere mounts is a contract question, not a tidiness one.
• ⛔ Either way the retirement itself is Clause ② and cannot land while the fable CONTRACT_REVIEW_TIER quota is exhausted. Answering the census now means the evidence is complete when the tier reopens.

⚠️ For whoever eventually acts on it

The instrument that produced this list had a false-zero bug on its first run — it required a trailing delimiter after the component name and reported 0 mounts for RecordComments, a component known mounted twice, because JSX tags end the line. It was fixed, and the entire sibling table was re-derived and gated on that control passing. ⇒ The list above is post-fix. ⛔ Do not re-derive it with a naive probe and assume a difference means the tree changed.

Refs: #7175 (the parent census) · PR #7191 (the instrument and its controls) · objectstack#14187 (the blocking reading, widened to all seven).

---

Generated by Claude Code