[finding] objectql's error-code constants are all on the batteries barrel only, while /core publishes the SchemaRegistry that raises three of them #16260

Activity

1. claude

   added

   findingRecorded observation — held, not dispatchable until the findings triage round grades itRecorded observation — held, not dispatchable until the findings triage round grades it

   domain:engineRetired 2026-10-22, alias of domain:server (engine surface); in-flight cards keep it until they landRetired 2026-10-22, alias of domain:server (engine surface); in-flight cards keep it until they land

   on Sep 6, 2026
2. zhuangjianguo

   added 2 commits that reference this issue on Sep 6, 2026

   feat(objectql): publish DriverConnectError and DatasourceUnavailableE…

   ...

   3fd854f

   feat(objectql): the two transaction-seam refusals publish their code …

   ...

   8c871e6
3. os-zhuang commented on Sep 7, 2026

   

   os-zhuang

   on Sep 7, 2026 – with Claude

   Contributor

   More actions

   Triage: closing not planned — outside the three admission classes (#16351 / SKILL.md:354-355), on the boundary this seat has applied throughout today's round: incomplete is not admissible; wrong is.

   Against each class:

   • (a) reproducible defect — ⛔ none. A /core embedder that catches NAMESPACE_CONFLICT by re-spelling the literal works. Nothing refuses, nothing misbehaves; the constant is merely not offered on that entry.
   • (b) contract violation — ⛔ no. /core is the lean ADR-0076 entry and nowhere declares that it publishes the code constants for every verb it exposes. objectql: 11 error classes still spell their code as an inline literal, so a consumer cannot follow the by code, not instanceof convention the docs already teach #16159's direction (make codes importable) is a programme, not a contract this omission breaches.
   • (c) metadata trap — ⛔ not applicable.

   ⭐ The measurement is exemplary and should be reused whenever this is revisited: the table is read from the built declarations (dist/index.d.ts / dist/core.d.ts), ⛔ not from the source barrels — explicitly because at least one constant reaches index.d.ts without being spelled in index.ts. And it carries a firing control: NAV_CONTRIBUTION_GROUP_MISSING is on the lean entry, with a docblock in core.ts saying why. ⇒ The nine "no"s are readings, not an artefact of looking in the wrong file. The surface ablation (core.d.ts 79 → 79 while index.d.ts 244 → 247 across PR #16259) is a second independent confirmation.

   承接者: a plausible one exists, and I am naming it rather than claiming it. #16159's remaining slices touch packages/objectql/src/index.ts and could carry option (b) — one sentence in core.ts's header making the omission read as a decision rather than an oversight — at essentially no cost. ⚠️ But the card itself says the question "would still be open afterwards for all nine", so I am ⛔ not recording that as a guarantee. If a #16159 slice author reads this: (b) is cheap, it is in your file, and the card's own view is that it "may be the whole answer."

   What I am uncertain about: whether a /core embedder re-spelling 'NAMESPACE_CONFLICT' is a real population or a hypothetical one. ⛔ Not measured — the card names the exposure structurally (a /core holder has SchemaRegistry, hence installPackage and registerObject, hence those three refusals) without demonstrating one exists. That is the gap between this and an admissible card.

   ⛔ What this closure does NOT authorize. ⛔ Nobody exports the nine constants from /core on this card's evidence. That is a published-surface widening per entry point and a policy decision about all nine, ⛔ not a mechanical repeat of #16159. ⚠️ And the ADR-0076 D2 boundary-ratchet reading — that it governs protocol/plugin imports and so would not object to a bare string constant — is explicitly unconfirmed with the ratchet's owner; core-boundary.ratchet.test.ts is where that gets settled, not in a PR description.

   Reopen conditions (any one, no approval needed):

   1. A /core embedder — in this repo, cloud, or a reported integration — is found re-spelling one of these literals. check:error-code-provenance records exactly that as a stamp site in the consumer's own package, so this is a query, not a guess. That instance makes it class (a).
   2. One of the nine code strings is renamed while /core consumers can only hardcode it. At that point the omission has cost someone a silent break.
   3. The ratchet's owner rules that code constants may cross to /core, which turns (a) from a policy question into a mechanical follow-up.

   Reversibility: fully reversible, ⛔ nothing decays. All three options stay equally open; no surface changes; the NAV_CONTRIBUTION_GROUP_MISSING precedent and its docblock remain in the tree as the standing argument that this can be done.

   ⛔ Not a blocker for #16159 and ⛔ not inside its completion scope — the card says so, and this close ratifies it: #16159 is discharged when the remaining inline literals become constants, regardless of this question.

   ---

   Generated by Claude Code
4. os-zhuang

   closed this as not plannedon Sep 7, 2026
5. os-zhuang commented on Sep 7, 2026

   

   os-zhuang

   on Sep 7, 2026 – with Claude

   Contributor

   More actions

   ⚠️ Closure reasoning, posted after the close — my error, recorded rather than backdated. I closed this without the mandatory reasoning comment. That is now an explicit rule (PR #16557, landed after I closed this: 「仍然关,但关闭理由必须写明『承接者:无』」), and it was already this seat's standing practice. The verdict below is the one the close was made on; ⛔ nothing here is a new judgement invented afterwards.

   Closed not planned — outside the three admission classes (#16351 / SKILL.md:349-351).

   Against each class:

   • (a) reproducible defect — ⛔ none, and the card says so in its own second line: "Observation, not a runtime defect." A /core embedder that re-spells 'NAMESPACE_CONFLICT' as a literal works correctly today.
   • (b) violation of a declared contract — ⛔ none. objectql: 11 error classes still spell their code as an inline literal, so a consumer cannot follow the by code, not instanceof convention the docs already teach #16159's direction (make codes importable so consumers stop re-spelling) is a direction, not a declared contract that ./core breaches. ADR-0076 D2's boundary ratchet governs protocol/plugin imports and ⚠️ on its face does not object to a bare string constant — but the card is careful to say that reading has not been confirmed with the ratchet's owner, so ⛔ it cannot be cited as a violated contract either.
   • (c) metadata trap — ⛔ not applicable.

   Under the boundary now written into the rule — 「(a) 的分界是不完整 vs 错误」 — this is incomplete, not wrong: nine constants that could be exported from a second entry point and are not. ⛔ Nothing published is false.

   ⭐ The measurement is good and survives the close: the table was read from the built declarations (dist/index.d.ts / dist/core.d.ts) rather than from the source barrels — deliberately, because at least one constant reaches index.d.ts without being spelled in index.ts. And NAV_CONTRIBUTION_GROUP_MISSING is a genuine firing control: one code constant IS on the lean entry, with a docblock in core.ts saying why. ⇒ The nine "no"s are readings, not a barrel that exports nothing. Corroborated by the surface ablation (core.d.ts 79 → 79 while index.d.ts 244 → 247 across PR #16259).

   承接者: ⛔ none for the question as asked. ⚠️ I checked rather than assumed, and the honest answer has two halves: #16159's remaining slices touch index.ts and could physically carry option (b)'s one-line header sentence in core.ts — but the card itself states that this question survives #16159's completion for all nine constants, so ⛔ that card is not a vehicle for the decision, only for a sentence someone might choose to write. ⇒ No carrier for the ruling.

   What I am uncertain about: whether any real @objectstack/objectql/core embedder exists and catches these refusals. ⛔ Not measured, by me or by the card. If several do, option (a) stops being a widening-for-tidiness and becomes a consumer requirement — which is why the reopen condition below is written in those terms rather than in terms of the export table.

   Reopen conditions (either, no approval needed):

   1. A /core embedder — in this repo, in objectui, in cloud, or reported — is measured re-spelling one of the nine literals. That is a named consumer paying the cost, which is class (a) territory.
   2. One of the nine wire codes is renamed. At that moment every re-spelled literal breaks silently, and the missing export becomes the reason it broke.

   Reversibility: fully reversible; nothing decays while closed. ⭐ Option (b) — a sentence in core.ts's header saying the omission is deliberate, so it reads as a decision rather than an oversight — remains available to anyone editing that file for any reason, costs nothing, and the card is right that it "may be the whole answer." ⛔ It does not need this card to be open.

   ⛔ Correctly out of #16159's scope: that card is discharged when the remaining inline literals become constants, and this question would still be open afterwards for all nine.

   ---

   Generated by Claude Code
6. akarma-synetal

   added 2 commits that reference this issue on Sep 9, 2026

   feat(objectql): publish DriverConnect, DatasourceUnavailable and Summ…

   ...

   ddfbf04

   feat(objectql): the two transaction-seam refusals publish their error…

   ...

   eddd612