[Decision] what 「is empty」 means on a text column and on a multi-value column: null only (the spec's lowering today), or null OR '' / [] — three objectui builders disagree, and a stored sharing rule's rows depend on the answer

Unblocked: 5865751001 · #20399 ruled A (batch #234 item 1, 5865693155) · 2026-09-28T07:53Z
Ruled: 5861435168 · letter B · 2026-09-28T00:55Z

Unblocked by ruling 5865693155 on #20399 (A, batch #234 item 1; note 5865751001) — was blocked by #20399 from 2026-09-28T06:54Z. Execution: declare $empty per the ruling.

Path: permissions that actually hold | 缺项 (no item authors 「is empty」 on a sharing-rule criterion) | P2

Filing gate: ② a decision only the maintainer can make: it is the meaning of a declared operator in @objectstack/spec, and changing it re-scopes stored rules. Filed and graded by the triage seat (objectstack-wide, seat post #6015, session_01W89enF2dYV7K4N2Fbfj33f) while answering pm:retriage on objectstack-ai/objectui#10813, whose stop valve tripped. ⛔ Not a claim.

Grade: needs-user-decision · priority:p2 · domain:spec · area:records.

维护者速读

事情:「为空」这个筛选条件,在平台里有三种意思。

协议(spec)把它翻成「值是 null 或缺失」。
objectui 的数据集桥写成「字段不存在」。
最常用的那个筛选控件(共享规则、汇总字段的过滤都用它)写成「null 或空字符串」。

同一个条件,在不同地方挑出的记录不一样。

为什么要你拍板:开发实测过,清空的文本框存的是空字符串 '',不是 null。平台底层有一条测试专门钉住「空字符串不能变成 null」。所以如果照协议现在的写法(只认 null),用户眼里空着的文本记录,「为空」反而筛不出来。而改控件的写法,会改变已存共享规则能看到的记录。这是协议语义加存量数据的决定,席位不能自己选。

⭐ 数值、日期、布尔字段里存了 '',是写入环节的缺陷,不需要你裁,已单独立卡 #20308,在写入口统一转成 null。本卡只问文本字段和多值字段。

选项:

A:「为空」只认 null。写入口把文本的 '' 也转成 null,并迁移存量数据。代价:推翻那条「空字符串不能变成 null」的底层约定,还要做数据迁移。
B(荐):协议声明「为空」按字段类型展开:文本类 = null 或 '';多值类(多选、标签、多值查找)= null 或 [];其它类型 = null。三个构建器都改成读协议的这一个展开。代价:协议里多一张按类型展开的表,但只写这一处。
C:不动,把三种意思写进文档。

席位意见:荐 B。

用户眼里「空着」就是空。主流平台也这么处理:Salesforce 把空白文本存成 null,ServiceNow、Airtable 的「为空」都匹配空字符串(⚠️ 本席未重新实测这三家)。
B 不动存储,不需要迁移,也保住了「空字符串不能变成 null」的底层约定。
已测到零条存量规则会因为 B 改变结果(objectui、objectstack 都是 0 条;生产环境读不到)。

⚠️ 没测的:生产环境里有没有共享规则用了「为空」,读不到。

你要做的:回一个字母 —— A / B / C。

os-decision-facets

① 项目长远合理性:一个操作符在协议里只该有一个意思,并且只写在一处。B 把「为空」按类型的展开写进协议,三个构建器改成读它,以后不会再出现第四种意思。A 也只剩一个意思,但要推翻已钉住的存储约定。C 让分歧永久存在。
② 实际业务拉动:共享规则和汇总字段过滤用的正是那个控件。今天「为空」在多值字段上还会直接报错:控件写出 $in: [''],而 SQL 驱动拒绝对 JSON 列做 $in。这是真实的、会报错的路径,不是理论问题。
③ 防 AI 写代码/写元数据犯错:三种意思意味着 AI 在不同构建器里写同一个条件,得到不同的记录集。B 让协议自己回答「为空」是什么,构建器只照读。
④ 创业阶段不扩散:B 不是新能力,是给已声明的操作符补上它缺的类型展开。A 需要一次数据迁移,是三者里最重的。

Governing text:

packages/spec/src/data/filter.zod.ts:2307-2308 ('is_empty': '$null', 'is_not_empty': '$null');
the spec's value round-trip fixture VALUE_ROUNDTRIP_CASES row str_empty (「empty string must not become null」);
objectui#10813's measurements (dev report 5858799523, review 5859057830, release 5859079825).

After the ruling

B: domain:spec declares the per-type lowering once (filter.zod.ts). objectui#10813 then converges FilterConditionField and datasetFilterCondition.ts on it, including the multi-value arm that today writes $in: [''] onto JSON columns.
A: the write-door card (record write door: '' skips every type check, so a number, boolean, date, datetime or time column stores an empty string — normalise it to null at the door (seam from objectui#10813) #20308) extends its normalisation to string columns, with a data migration, and the fixture row changes by ruling.
C: the three meanings are documented where each builder offers the operator.

Refs: objectstack-ai/objectui#10813 · objectstack-ai/objectui#10790 · PR objectstack-ai/objectui#10834 (closed).

Unblocked: #20399 ruled A (batch #234 item 1, 5865693155) · pm:blocked → pm:queue · 2026-09-28T07:51Z

Director seat (objectstack#12708, summon #30 续 2, session_01AsCNgFBs8HCjwhyHQsFbx3). The maintainer's word on #20399: 「同意」 to A. This card's Blocked-by: #20399 is resolved by that ruling; the body line is replaced in the same stroke.

Re-claim on the dev's kept branch as ruling B here plus ruling A there: declare $empty: boolean in FILTER_OPERATORS and FieldOperatorsSchema with the ruled per-type table as its describe; one expansion function keyed on the field definition (type plus multiple); the is_empty / is_not_empty lowering keeps emitting $null until the flip card lands; ruling 乙 on #19757 untouched ({ tags: [] } still refused). Clause-②: yes (widening), minor. Then file one sibling card per compile-surface lane, each Blocked-by: this card's spec PR, and the flip card last. Full parameters and pins are on the ruling.

Claim: PM loop round 1
Session: session_01B3TqpoQbTAfG7G74GMDWNW
Account: hotlong (the seat's linked user as GET /user answers it; the card's assignee; seated on the maintainer's ruling 「放行，以 hotlong 坐席」, seat post #6017 marker 5864561291)
Branch: claude/issue-20311-is-empty-per-type-lowering (kept from the first dispatch; remote head ab6fb027, no commits)
Worktree: objectstack-issue-20311
Domain: domain:spec
Seat: domain:spec#1 (seat post #6017)
Ruling-ref: 5861435168 (ruling B on this card) · 5865693155 (ruling A on #20399, the spelling)
File surface:

packages/spec/src/data/filter.zod.ts: $empty: boolean added to FILTER_OPERATORS and FieldOperatorsSchema, its describe carrying ruling B's per-type table; one expansion function keyed on the field definition (type plus multiple), exported for the compile surfaces. ⛔ The is_empty / is_not_empty lowering keeps emitting $null (the flip is a later card).
packages/spec/src/data/field-value.zod.ts only if the table needs a set it lacks, by reference to the existing sets.
Their tests in packages/spec/src/data/. ⛔ The value round-trip fixture untouched; ⛔ ruling 乙 on [finding] the comparand-SHAPE face declares it closes the door "for every driver at once", but an array in the IMPLICIT-EQUALITY slot passes it — and driver-mongodb alone answers it, as an exact-array match #19757 untouched ({ tags: [] } still refused).
The generator-regenerated artefacts, never by hand. .changeset/.
⛔ No compile surface edited (drivers, formula, objectql, service-analytics): each lane's card is filed by this seat at ACCEPT, Blocked-by: #20311, naming the landed function, and the flip card last.
(stop on breach; explain in the report)
Container & model: M, mode:subagent, model: opus (dispatch-gates --tier: no path-derived mandate; clause-② suspect surface). Built at the default tier; the at-tier contract review is owed before enqueue.
Clause-②: yes
Thread-read: 5865751001
Serial constraints cleared: read at 2026-09-28T10:27Z.
No open PR touches filter.zod.ts or field-value.zod.ts (open-PR file lists read at this claim).
PR feat(spec): the number-comparand declared-type door's contract and the platform's numeric grammar #20414 (driver-sql on PostgreSQL answers 500 for a non-numeric string against a number field — where { amount: { $gt: "abc" } } is DATABASE_ERROR / 500 over REST, while InMemoryDriver and SQLite answer 200 with no rows #20336, this seat) adds one export line to the data barrel; this card needs no barrel line (filter.zod.ts is already exported). Disjoint.
packages/spec/src/migrations/registry.ts is carried by open PRs fix(spec)!: refuse a flow node config its executor cannot run — a required key left out, or a decision branch list it cannot read — at all three doors (#20316) #20416, feat(spec)!: retire the list view's own tabs key; named presets are listViews entries #20357, feat(analytics): enforce analytics_cube.public and default it to visible #20348 and feat(automation)!: edge-branched decision is exclusive; mode: 'inclusive' takes every branch (#15429) #20344. This card adds no migration entry (a widening; the lowering is unchanged), so it is expected not to touch it; if the migration gate asks for one, the base merge goes through scripts/pm/os-regen-merge.sh only.

Maintainer ruling on ruling A's placement: `$empty` is staged like `$like` · 2026-09-28T10:30Z

domain:spec seat 1 (session_01B3TqpoQbTAfG7G74GMDWNW). This records a ruling taken before this card's second dispatch.

Who: the maintainer.
Where: this seat's session chat, as an AskUserQuestion answer.
Verbatim: 「照 $like 先例分阶段」.

Why the question arose. Ruling A on #20399 (5865693155) put $empty into FILTER_OPERATORS and FieldOperatorsSchema, on the premise that "an execution surface that has not implemented $empty refuses an unknown operator loudly; nothing passes silently". The seat measured that premise on origin/main 50e273fd before dispatch, and it does not hold:

driver-memory's accepted set is built from FILTER_OPERATORS (filter-refusal.ts:363, SUPPORTED_FIELD_OPERATORS = new Set([...FILTER_OPERATORS, '$like', '$ilike'])).
Its matcher's default: arm breaks and the row passes (memory-matcher.ts, the driver-memory 的实时查询路径根本不支持 $not —— mingo 抛无 code 的 MingoError,CEL !expr 降下来的 RLS scope 在该驱动上直接 500 #5324 note: "ignoring a constraint WIDENS the result set, which on a read scope is a permission bypass").
So once $empty is in FILTER_OPERATORS, any surface that has no arm for it yet would drop a hand-authored { f: { $empty: true } } and return every row.
FILTER_OPERATORS' own docblock, "$like / $ilike are STAGED here, on purpose (Filter AST: like is folded to $contains at the wire — wildcards bind as literals and driver-sql's like/ilike arm is unreachable #7536)", records exactly this mechanism and the staging answer.

What the ruling changes. Only the membership timing moves. The substance of ruling A stands: the $empty spelling, ruling B's per-type table as its describe, one expansion function keyed on the field definition, the typeless lowering, and ruling 乙 on #19757 untouched.

This card ([Decision] what 「is empty」 means on a text column and on a multi-value column: null only (the spec's lowering today), or null OR '' / [] — three objectui builders disagree, and a stored sharing rule's rows depend on the answer #20311): $empty is declared in FieldOperatorsSchema (the describe carries the ruled table), together with the exported expansion function. It is ⛔ NOT added to FILTER_OPERATORS, so every surface that derives acceptance from that array keeps refusing it loudly (INVALID_FILTER / 400). The is_empty / is_not_empty lowering keeps emitting $null.
The compile-surface lane cards (filed by this seat at ACCEPT, Blocked-by: #20311): each gives its surface a $empty arm, the way driver-memory widens its own set for $like.
The flip card (last): it adds $empty to FILTER_OPERATORS once every surface has its arm, and flips the lowering.

os-dev-report
{
"issue": 20311,
"status": "done",
"branch": "claude/issue-20311-is-empty-per-type-lowering",
"pr": "#20442",
"session": "session_01B3TqpoQbTAfG7G74GMDWNW — subagent run; this is the parent seat's harness-stamped id",
"premise_still_valid": true,
"summary": "$empty: boolean is declared in FieldOperatorsSchema (the enforced copy) and SpecialOperatorSchema (the documentation copy). The two share one description string, which is ruling B's per-type table. The one expansion ruling A asks for ships as expandEmptyOperator(field), keyed on type plus multiple, returning a frozen EMPTY_OPERATOR_ARMS row { arm, emptyString, emptyList }. isEmptyFilterValue(value, expansion?) ships beside it; called with no expansion it gives the by-value reading for formula and having. Both live in a new spec module, data/filter-empty-operator.ts, exported from the data entry. It is staged per the amendment: absent from FILTER_OPERATORS (the docblock gains a $empty paragraph carrying the measured per-face table), STAGED_AHEAD_OF_BACKENDS = [$empty, $ilike, $like], the is_empty lowering is unchanged ($null), and ruling 乙 is untouched. Premise A1 is partly falsified. No surface drops the predicate. Every driver, having, mongodb and the analytics where compile refuse with INVALID_FILTER 400. read-scope-sql refuses with READ_SCOPE_COMPILE_FAILED 500. formula matchesFilterCondition answers a silent false. That false is its fail-closed posture for any operator it has no arm for, and this PR does not change it. It does, however, falsify formula's own docblock claim that a DECLARED operator never gets the silent false (open_questions[0]).",
"premise_evidence": {
"tree": "objectstack base 50e273f (origin/main at dispatch), head 32e926d; worktree ../objectstack-issue-20311, kept branch fast-forwarded from ab6fb02 by merge (no force)",
"A1_per_surface": [
"driver-sql SqlDriver.find (better-sqlite3): $empty true / false / nested in $and all REFUSED INVALID_FILTER 400; control $null answered [2]; control $bogus REFUSED 400. driver-sqlite-wasm and driver-turso local inherit this compiler (not driven separately)",
"driver-turso RemoteTransport.find: REFUSED INVALID_FILTER 400 (all three); control $null compiled IS NULL",
"service-analytics compileScopedFilterToSql: REFUSED READ_SCOPE_COMPILE_FAILED 500 fail-closed (all three, and $bogus alike)",
"service-analytics lowerAnalyticsWhere: passes the condition through unchanged (a lowering); normalizeAnalyticsFilterTree right after it REFUSES INVALID_FILTER 400",
"formula matchesFilterCondition: NOT LOUD. It returns false for every record (flag true and flag false alike), the same as for undeclared $bogus; control $null answered the null row",
"objectql applyHaving and matchesHaving: REFUSED INVALID_FILTER 400",
"driver-memory find and match (checkCondition): REFUSED INVALID_FILTER 400",
"driver-mongodb translateFilter: REFUSED INVALID_FILTER 400"
],
"A1_how": "scratch tsx probe in the scratchpad (not committed), run after this change was built; SQL faces through built dist, the rest through source by absolute path",
"A2": "two in-package pins derive from FieldOperatorsSchema keys: filter-comparand-type.test.ts (judged scalar set; source SCALAR_COMPARAND_OPERATORS gains $empty) and filter-save-door-face-parity.test.ts (BOOLEAN_SLOTS; source BOOLEAN_FLAG_OPERATORS gains $empty). Suites enumerating FILTER_OPERATORS are untouched. No FieldOperatorsSchema-key enumeration outside packages/spec (grep: prose only). No gate needed an edit outside packages/spec.",
"A3": "STRING_VALUE_TYPES = 14 text types; isMultiValueField = MULTI_OPTION_TYPES or MULTI_CAPABLE_TYPES with multiple:true; the sets are disjoint (pinned); lookup and lookup multiple land on null_only and multi_value",
"A4": "isEmptyFilterValue(value) with no expansion: null, undefined, empty string and empty list count as empty; it diverges from the declared table only on a non-text column holding an empty string (pinned)",
"stored_rules": "objectstack 50e273f examples criteria sharing rules using emptiness: 0. objectui b45d463: 0 stored rules; the 22 emptiness hits in sharing-rule-related files are builder code and tests. No lowering changes, so no rule changes result. Production NOT MEASURED.",
"dollar_empty_corpus": "$empty occurrences before this change: objectstack 0, objectui 0"
},
"tests": "All at head 32e926d unless noted. (1) spec full local project: vitest run --project local, 563 files, 16536 passed, 1 todo, exit 0. (2) pnpm --filter @objectstack/spec typecheck: exit 0 (tsc, scripts typecheck and test typecheck). (3) Targeted spec filter suites (10 files incl. filter-empty-operator, vocabulary, comparand-type, save-door parity, filter, comparand-shape, equality/ne array doors, view-operator parity, page-component conversion): 675 passed, 1 todo. (4) Consumer filter/operator suites, downstream direction (consumers of @objectstack/spec), after rebuilding spec and each upstream closure from this branch: formula 16 files/570 passed; driver-memory 10/392; driver-mongodb 6+1 skipped/219 passed, 38 skipped; objectql (filter/operator/having) 16/624; driver-sql 14/331 passed, 4 skipped; driver-turso 4/170; metadata-protocol 4/51; lint 2/76; service-analytics (filter/operator/read-scope) 42/951; rest 6/109; plugin-sharing (criteria/sharing-rule) 6/146. All exit 0. (5) Ablation via scripts/ablation-replace.mjs on the committed tree, adding $empty to FILTER_OPERATORS: anchor 1 to 0, blob 0912c2776e5e to 74b0e4d23fe0. Exactly the two staging pins went red (2 failed, 23 passed; the expected direction). Restored: blob == HEAD and git diff HEAD empty. The spec tests import src relatively, so no dist is involved. A second ablation (dropping $empty from the save-door flag set) is NOT MEASURED: the verify lock queue timed out (exit 99). (6) eslint --no-inline-config --format json over the 9 changed TS files: 9 files, 0 errors, 0 warnings. Population is the eslint.config.mjs all-ts block; that config never enables type-aware linting, so untouched files cannot move. Full pnpm lint is CI's.",
"gates": "node scripts/pm/dispatch-gates.mjs --commands --repo objectstack-ai/objectstack at 32e926d derived 108 commands (a superset of the dispatch list's 72). All 108 were run with exit codes recorded; --ran reconciled with exit 0: 106 run at exit 0, 2 NOT MEASURED, 0 unrun. The two NOT MEASURED are check:dual-build-cjs-loads and check:type-check-debt, exit 3 PREREQUISITE NOT MET: both need every package built, and CI runs them. check:skill-examples exited 3 until client-react was built, then 0. spec check:generated: all current. check:adr-0087-registration: 0 (no declared-breaking changeset). check:changeset-no-major: 0. Derivation warning: the tree was 7 commits behind origin/main dbddf02; one derivation input changed there (scripts/docs-audit/handwritten-docs.json). None of those commits touch this diff's files, and the branch was not re-merged.",
"line_budget": "582 changed lines (+575 / -7) in 14 files, under the 5000 human-merge line; no governed surface (no skills/, .claude/, docs/adr/**) in the diff",
"files_changed": [
".changeset/20311-empty-filter-operator-staged.md",
"content/docs/references/data/filter.mdx (generated by gen:docs)",
"packages/spec/api-surface/data.json (generated)",
"packages/spec/authorable-surface/data.json (generated by the build)",
"packages/spec/export-origins/data.json (generated)",
"packages/spec/src/data/filter.zod.ts",
"packages/spec/src/data/filter-empty-operator.ts (new)",
"packages/spec/src/data/filter-empty-operator.test.ts (new)",
"packages/spec/src/data/index.ts (one export line)",
"packages/spec/src/data/filter-comparand-type.ts",
"packages/spec/src/data/filter-comparand-type.test.ts",
"packages/spec/src/data/filter-save-door-refusals.ts",
"packages/spec/src/data/filter-save-door-face-parity.test.ts",
"packages/spec/src/data/filter-operator-vocabulary.test.ts"
],
"deviations": [
"File surface grew inside packages/spec. The expansion sits in a new module, filter-empty-operator.ts, with one line in the data barrel, and not in filter.zod.ts. Reason: importing field-value.zod into filter.zod.ts made gen:skill-refs rewrite skills/objectstack-query and skills/objectstack-api reference indexes, which would have made the PR Tier H; the two files also meet in the field.zod import cycle. The precedent is filter-text-operator-declared-type.ts. The barrel line sits after that module's export and is disjoint from open PR 20414's hunk.",
"filter-comparand-type.ts and filter-save-door-refusals.ts (plus their pins) were edited to keep the FieldOperatorsSchema-key-derived reconciliation green (A2).",
"Consequence of the save-door flag set: a NON-boolean $empty is now refused at FilterConditionSchema, as $null and $exists are. Strictly this narrows the save door for a key that was never declared and that nothing writes (0 occurrences in either repo). It is declared Clause-② yes (widening) per ruling A and stated in the changeset.",
"$empty is also declared on SpecialOperatorSchema (the documentation copy), as $like is on StringOperatorSchema; the amendment names only FieldOperatorsSchema.",
"The description string is pinned verbatim. The role file says not to pin prose, but ruling A demands exactly this pin, and the ruling governs.",
"The PR body carries Clause-② yes, copied from the claim; the changeset carries Clause-② yes (widening).",
"A status reply went to the seat mid-run, via SendMessage to main."
],
"mcp_calls": "0",
"api_writes": "3, each sent as one fleet-write relay stroke (POST /repos/objectstack-ai/objectstack/dispatches, executed as objectstack-fleet[bot]): pr_create = POST /repos/objectstack-ai/objectstack/pulls (PR 20442, draft); assign = POST /repos//issues/20442/assignees (hotlong, via scripts/pm/label-write.mjs; read-back matched, no labels written); comment = POST /repos//issues/20311/comments (this report, via scripts/pm/post-stamped.mjs). git push is not a REST write.",
"open_questions": [
{
"question": "Should PR 20442 land before formula's lane card? Once $empty is declared, formula's matchesFilterCondition gives it the silent fail-closed false that formula's own docblock says a DECLARED operator must never get (the $like precedent: formula got its arm in the declaring PR). Every other face refuses loudly.",
"options": [
"A — land as is. The formula lane card gives the arm (by value, isEmptyFilterValue(value)) and corrects the docblock. The window is fail-closed: a write-side check denies, nothing widens. The description and changeset tell authors not to use $empty yet.",
"B — hold 20442's merge until the formula lane card is ready, and land the two together.",
"C — widen this card to add formula's arm now. This breaches the order's file bound and the amendment's placement of arms in lane cards, so it needs a ruling."
],
"recommendation": "A. Long-term: the maintainer ruled the arms into lane cards, and the staging keeps every surface from widening. Real need: $empty has zero producers today, so the window reaches no one. AI-error: the description says staged, and the query faces refuse loudly. Startup scope: A adds no extra sequencing. B is the safe alternative if the seat wants the formula docblock claim never false on main."
}
],
"out_of_scope_findings": [
"carrier: formula's $empty lane card (domain:engine, filed by the seat at ACCEPT) · packages/formula/src/matches-filter.ts header claims the silent false is reachable only for a name the protocol does not declare; with $empty declared that is false (measured: false for $empty true and false alike) · noted, not filed",
"carrier: service-analytics' lane card (domain:services) · read-scope-sql compileScopedFilterToSql refuses every unknown operator (also $bogus) with READ_SCOPE_COMPILE_FAILED 500, where the other faces answer INVALID_FILTER 400; loud and fail-closed, pre-existing · noted, not filed",
"carrier: the flip card · packages/spec/src/ui/view-grouping-query.ts says the empty-group predicate and the view filter's is_empty agree on what empty means; true while the lowering stays $null, false once it flips · noted, not filed"
]
}

Generated by Claude Code

Review: ACCEPT · PR #20442 at head `32e926db1eab` · 2026-09-28T13:11Z

domain:spec seat 1 (session_01B3TqpoQbTAfG7G74GMDWNW), reviewer of record under claim 5868117891. Dev report on this card (os-dev-report). Rulings: B 5861435168, A 5865693155 (#20399), and the maintainer's staging amendment 5868169573. The seat took the readings below on GitHub or on the branch.

check	reading
Shape	Draft, base `main`, first body line `Fixes #20311`, the only closing keyword. `mergeable_state` reads `blocked`, which is the draft state, not a conflict.
Scope	14 files, +575 / −7, all in `packages/spec` plus generated output and one changeset. The expansion lives in a new module, `data/filter-empty-operator.ts`, rather than in `filter.zod.ts`: importing `field-value.zod` into `filter.zod.ts` made `gen:skill-refs` rewrite two `skills/**` reference indexes, which would have made the PR Tier H. The precedent is `filter-text-operator-declared-type.ts`. Accepted.
Staging (the amendment)	`$empty` is declared in `FieldOperatorsSchema` (and `SpecialOperatorSchema`, as `$like` is on `StringOperatorSchema`), with ruling B's per-type table as its describe, pinned verbatim. It is absent from `FILTER_OPERATORS`: the Special row still ends at `'$null', '$exists'` on the branch. The lowering is unchanged, and ruling 乙 is untouched.
Premise A1	Measured per face. No surface drops the predicate. Every driver, `having`, mongodb and the analytics where refuse with `INVALID_FILTER` / 400. `compileScopedFilterToSql` refuses with `READ_SCOPE_COMPILE_FAILED` / 500, as it does for any unknown operator. formula `matchesFilterCondition` answers a silent `false` (fail-closed).
Changeset	`@objectstack/spec` `minor`; the body carries `Clause-②: yes`, the changeset `Clause-②: yes (widening)`.
Governed / size	Not governed; 582 changed lines.
Evidence	The report's ablation: adding `$empty` to `FILTER_OPERATORS` turns exactly the two staging pins red.

Flags carried to the at-tier review (③):

open_questions[0]: formula's silent false. It falsifies formula's docblock claim that a DECLARED operator never gets that answer. The dev recommends A (land as is; the arm and the docblock fix ride the engine lane card). The seat agrees: the amendment placed the arms in lane cards; the window is fail-closed (a write-side check denies, and nothing widens); and $empty has 0 producers in either repo.
A non-boolean $empty is now refused at FilterConditionSchema, because it joins the boolean-flag set with $null / $exists. This narrows a key that was never declared and that nothing writes.
The describe string is pinned verbatim. The role file discourages pinning prose, but ruling A asks for exactly this pin.

Filed at this ACCEPT, as ruling A orders:

filter: the engine's compile surfaces answer $empty by the field's declared type (driver-sql and heirs, turso remote, driver-memory, driver-mongodb, formula, objectql having) — ruling A on #20399 #20444: the domain:engine arms. It also carries the formula docblock fix.
analytics: service-analytics' two filter faces answer $empty by the field's declared type (read-scope SQL, the analytics where) — ruling A on #20399 #20445: the domain:services arms. It also carries the read-scope 400-vs-500 question.
filter: flip is_empty / is_not_empty to $empty and add it to FILTER_OPERATORS, once every compile surface answers it (the last step of ruling A on #20399) #20446: the flip (FILTER_OPERATORS and the lowering), Blocked-by: both lane cards. It also carries the view-grouping-query.ts note.

needs:contract-review goes on PR #20442 in this act. The isolated at-tier review starts now.

Close-out by the domain:spec seat 1 (session_01B3TqpoQbTAfG7G74GMDWNW) at 2026-09-28T17:09Z.

Delivered. PR #20442 merged as b810ddb6f through the merge queue, and this card closed completed on its Fixes line. pm:dispatched is removed.

The emptiness operator $empty: boolean is declared on FieldOperatorsSchema and on the documentation copy. Its description is ruling B's per-type table.
It is staged the way $like was: absent from FILTER_OPERATORS, and is_empty still lowers to $null.
The one expansion every compile surface will call, expandEmptyOperator(field), keyed on the field definition (ruling A), is exported from @objectstack/spec/data with isEmptyFilterValue and EMPTY_OPERATOR_ARMS.
At-tier record 5874422033 PASS at d581aed71, after the patch round that reconciled PR feat(spec): the number-comparand declared-type door's contract and the platform's numeric grammar #20414's FieldOperatorsSchema partition pin (audit 5871530372).
On main after the merge, both PRs' exports are present once each in api-surface/data.json and export-origins/data.json.

What remains, and who carries it. The gap ruling B closes for users stays open until the flip, by design of the staging.

filter: the engine's compile surfaces answer $empty by the field's declared type (driver-sql and heirs, turso remote, driver-memory, driver-mongodb, formula, objectql having) — ruling A on #20399 #20444 (domain:engine): formula's $empty arm and its docblock correction.
analytics: service-analytics' two filter faces answer $empty by the field's declared type (read-scope SQL, the analytics where) — ruling A on #20399 #20445 (domain:services): the analytics faces, including read-scope's 500.
filter: flip is_empty / is_not_empty to $empty and add it to FILTER_OPERATORS, once every compile surface answers it (the last step of ruling A on #20399) #20446 (domain:spec): the flip. It adds $empty to FILTER_OPERATORS and moves the is_empty lowering. It must also re-read the describe's staging sentence, view-grouping-query.ts, and driver-sql on PostgreSQL answers 500 for a non-numeric string against a number field — where { amount: { $gt: "abc" } } is DATABASE_ERROR / 500 over REST, while InMemoryDriver and SQLite answer 200 with no rows #20336's module sentence about is_empty on a number field (record 5874422033 §③.3–4).
All three carry Blocked-by: #20311, so the unblock sweep returns them to the queue now.
A sequencing note for objectql: refuse a non-numeric string compared against a number field at the engine's field-aware filter door (400, every driver and position) — the door half of #20336 #20351's engine suite (the $empty unjudged case row) is posted on objectql: refuse a non-numeric string compared against a number field at the engine's field-aware filter door (400, every driver and position) — the door half of #20336 #20351.
The queue-flake anchor Queue-flake anchor: src/data/filter-number-comparand-declared-type.test.ts #20455 is closed with this landing.

Generated by Claude Code

feat(spec): the number-comparand declared-type door's contract and th…

feat(spec): declare the staged $empty filter operator and its per-typ…

feat(drivers,formula,objectql): the engine's filter faces answer the …

docs(driver-turso): re-anchor the dead tracker citations to the commi…

docs(driver-mongodb): re-anchor the dead tracker citations to the com…

test(spec): the third data/ file group's test titles state each cited…

Navigation Menu

Repository navigation

[Decision] what 「is empty」 means on a text column and on a multi-value column: null only (the spec's lowering today), or null OR `''` / `[]` — three objectui builders disagree, and a stored sharing rule's rows depend on the answer #20311

维护者速读

After the ruling

Maintainer ruling on ruling A's placement: `$empty` is staged like `$like` · 2026-09-28T10:30Z

Review: ACCEPT · PR #20442 at head `32e926db1eab` · 2026-09-28T13:11Z

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

[Decision] what 「is empty」 means on a text column and on a multi-value column: null only (the spec's lowering today), or null OR '' / [] — three objectui builders disagree, and a stored sharing rule's rows depend on the answer #20311

Description

维护者速读

After the ruling

Activity

objectstack-fleet commented on Sep 28, 2026

objectstack-fleet commented on Sep 28, 2026

objectstack-fleet commented on Sep 28, 2026

Maintainer ruling on ruling A's placement: $empty is staged like $like · 2026-09-28T10:30Z

objectstack-fleet commented on Sep 28, 2026

objectstack-fleet commented on Sep 28, 2026

Review: ACCEPT · PR #20442 at head 32e926db1eab · 2026-09-28T13:11Z

objectstack-fleet commented on Sep 28, 2026

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions

[Decision] what 「is empty」 means on a text column and on a multi-value column: null only (the spec's lowering today), or null OR `''` / `[]` — three objectui builders disagree, and a stored sharing rule's rows depend on the answer #20311

Maintainer ruling on ruling A's placement: `$empty` is staged like `$like` · 2026-09-28T10:30Z

Review: ACCEPT · PR #20442 at head `32e926db1eab` · 2026-09-28T13:11Z