finding(components/types): the filter-builder dropdown offers 22 operator ids and the mirror declares 14 — 19 are refused as authored, and 8 of those cannot be reached by normalisation at all

[claude]

Surfaced by the os-dev seat delivering objectui#6939 (PR #9558) as its fence-line-2 observation,
and re-measured in this seat before filing — ⛔ not adopted from the report.
objectui origin/main + installed @objectstack/spec@17.4.0, 2026-09-15T03:50Z.
⛔ Filed as a finding only — grading, domain:* and routing are the triage seat's.

The defect

A user picks an operator from the filter-builder dropdown; the resulting filter is refused by the
mirror that validates it. objectui#6939 fixed the three catalog fixtures that had the same
problem; this is the same defect on the path a real user takes, and no card owns it.

Measured

defaultOperators in packages/components/src/custom/filter-builder.tsx offers 22 ids.
FilterOperatorSchema in packages/types/src/zod/complex.zod.ts declares 14 members.

	count	ids
offered by the dropdown	22	equals notEquals contains containsCaseInsensitive notContains isEmpty isNotEmpty greaterThan lessThan greaterOrEqual lessOrEqual before after between in notIn startsWith endsWith isNull isNotNull exists notExists
accepted as authored by the mirror	3	equals contains in
refused as authored	19	the rest

Folding every offered id through the spec's own normalizeFilterOperator and testing the result
against the declared set:

	count	detail
fold onto a declared member	14	e.g. lessThan → less_than — a pure spelling difference
do not fold	8	splits again, and the split is the whole decision

The 8 that do not fold are two different problems:

• 2 fold to a name the enum does not declare: isEmpty → is_empty, isNotEmpty → is_not_empty.
  The spec's alias table knows these names; this mirror's enum does not list them.
• 6 fold to themselves: containsCaseInsensitive, before, after, between, exists,
  notExists. ⭐ normalizeFilterOperator returns its input unchanged for an id it does not know,
  so "folds to itself" is exactly "the protocol has no member for this".

Controls in the same run: normalizeFilterOperator('lessThan') → less_than (fires);
normalizeFilterOperator('qqzzAbsent') → qqzzAbsent (the identity behaviour that makes the
6-row reading meaningful); declared set size 14.

⚠️ My first instrument returned 0 dropdown ids — a regex that assumed single quotes where the
file uses double. ⛔ A zero from a broken parser is not a reading; it was thrown away and the block
was re-read by offset. The numbers above come from the second instrument, with both controls firing.

Why a spelling fix is not the whole answer

The tempting repair — fold at the storage boundary — closes 14 of the 19. It leaves 8 ids a
user can pick from a dropdown and cannot store, and those 8 are not one group:

1. isEmpty / isNotEmpty need the mirror to declare two names the protocol already publishes.
2. The other six need the protocol to gain a member, or the dropdown to stop offering them.

⇒ under the maintainer's standing principle 「以 objectstack 协议为准…协议不正确的应该先修改协议」,
group 2 is a spec card first, ⛔ not an objectui change.

⚠️ filter-builder.tsx already records part of this: OPT_IN_OPERATORS holds
containsCaseInsensitive, exists and notExists, with an upstream blocker named for the last two
— so three of the six are known-unsupported by design, and the card that rules this should start
from that comment rather than re-deriving it.

Already pinned, ⛔ so this is a recorded decision rather than a silence

PR #9558 pins the refusal in both
examples/schema-catalog/test/filter-builder-operator-vocabulary-6939.test.tsx and
packages/types/src/__tests__/filter-builder-mirror-6939.test.ts. ⇒ whichever way this is ruled,
the pin names the key that moves and a repair cannot land silently.

Siblings — the same shape, one key over

• objectui#8961 (needs-user-decision) — the sort vocabulary: three $orderby shapes, of
  which the protocol declares one. Same family, same question, already in the decision inbox.
• objectui#9554 — object-grid sends $orderby as an object array the protocol does not declare.
• objectui#6939 — the catalog half of this defect, fixed in PR fix(schema-catalog): spell filter-builder operators the way the mirror declares them #9558.

⭐ Three cards now say the same thing about three different keys: objectui's UI vocabularies and
the protocol's declared sets are drifting apart in more than one place, and each was found
separately. Whoever grades these may want to ask whether the right artefact is a per-key ruling or
one gate that compares a component's offered vocabulary against the protocol's declared set.

Dedup: searched before filing; #8961 / #9554 / #6939 are the nearest and none is about the
filter-builder's operator dropdown.

---

os-decision-facets

维护者速读

筛选器的下拉菜单给作者 22 个操作符;objectui 自己那份校验用的枚举 FilterOperatorSchema 只声明 14 个,按作者原样接受的只有 3 个。

⭐ 但实测把方向反过来了:协议(@objectstack/spec 自己的 ViewFilterRuleSchema)按原样接受 22 个里的 19 个,并在解析时归一化成规范拼法。两边的差集只朝一个方向:协议有而 objectui 没有的 6 个(icontains is_empty is_not_empty before after between);objectui 有而协议没有的 —— 一个都没有。

⇒ 19 个拒绝里有 16 个,是 objectui 自己的镜子在拒绝协议本身接受的东西。 下拉菜单几乎是对的,⛔ 不该动它。

⚠️ 同时要修正卡面的口径:那句「用户挑一个操作符、结果被拒」在任何运行时路径上都测不出来。真正被拒的是作文面(CLI 的 check/validate,以及只在 __DEV__ 下 console.warn 的结构检查)。缺陷是真的——作者写不进 before/after/between/isEmpty/isNotEmpty,而协议五个全收——但它是作文面缺陷,不是终端用户点击的缺陷。

问:那份 14 个成员的窄枚举,是有意收窄还是落后于协议的陈旧子集**?**

选项	做什么	代价
A 恢复保真	把 FilterOperatorSchema 加宽到协议的 20 个规范成员	关掉 6 个;14 个别名拼法仍按原样被拒(可辩护:作文元数据本就该写规范拼法)
B 采用协议自己的行为	镜子不再手工维护第二份枚举,改为从 spec 派生并像协议一样在解析时归一化	一次关掉 19 个里的全部,并且消掉整个漂移类;⚠️ 会改变已存作文文档归一化后的样子 ⇒ 需要一条往返 pin
C 裁定这 14 个是有意的	把「为什么窄」写在枚举旁边(今天那里除了「Filter Operator Enum」没有任何文档)	那么下拉菜单里的 before/after/between/isEmpty/isNotEmpty 就是在提供「objectui 决定作者不许写」的操作符,而这五个今天在真实视图路径上存得下去
D 什么都不改	只把 3 个真正的协议缺口(containsCaseInsensitive/icontains 别名行、exists、notExists)上报 objectstack	16 个的镜子/协议分歧原样留着

四棱

• ① 项目长远合理性:B 把「手工维护第二份枚举」这件事本身去掉——漂移类消失,而不是这一例被修好;A 只是把今天的差补上,下次协议加成员还会再漂一次;C 把特例定型;D 留着分歧。按①:B。
• ② 实际业务拉动:⚠️ 实测后比卡面小:作文面而非运行时。但它是真实的——五个协议接受的操作符作者写不进去,而 objectui#6939 已经因为同一缺陷改过目录 fixture。
• ③ 防 AI 犯错:失败方向是响亮的(校验拒绝),这是好的;问题是拒得没道理——AI 从下拉里挑一个合法操作符,写出来的文档被自家 gate 拒,而协议会收。B 让 gate 与运行时、与协议三者一致(本仓的 viewFilterFold 与 ListView 存盘前都已经过 normalizeFilterOperator)。
• ④ 创业阶段不扩散:B 少一份永久义务(那份手工枚举);A 保留它;C 新增一条要维护的说明;D 把 3 个缺口推给上游。

Prior rulings read (THIS repo, objectui — AGENTS.md + docs/adr/**, which holds 0001/0034/0036/0053–0059): filter / between / incomplete → 73 hits across AGENTS.md and docs/adr/0001, 0053, 0054, 0055, 0057;⛔ 无一条把操作符词表当作已裁事项。
Prior rulings read (objectstack 的语料,共享协议自己的裁决 —— ⛔ 不是本仓编号): filter,between,incomplete,builder → 18 hits; ADR-0021 D2, ADR-0021 D3, ADR-0029 D9.3, ADR-0057 D10, ADR-0057 D12, ADR-0058 D1, ADR-0058 D3, ADR-0061 D1, ADR-0061 D6。

推荐:B,回退项 A。自检:「只看①选 B;②③④ 是否翻转:否——②把规模调小但不改方向,③④ 同向。」

置信缺口:① 协议侧的那些数(20 规范 / 36 别名 / 22 里接受 19)是 dev 的测量,本席未复算——共享检出没有 node_modules;本席自测的只有 objectui 那 14 个成员。② B 会改变已存文档归一化后的样子,影响面未量。③ ⚠️ 四个选项都要改 packages/types/src/zod/complex.zod.ts,而它被开着的 PR objectui#9895 持有 ⇒ 裁完也要等那个 PR,这是排期事实,⛔ 不是需要裁决的理由。

裁后执行:裁 B ⇒ 一张卡,镜子改为从 spec 派生 + 一条往返 pin;裁 A ⇒ 一张卡,加宽到 20 并 pin 住「本仓不得声明协议没有的成员」;裁 C ⇒ 一张卡,把理由写在枚举旁,并同笔决定下拉菜单是否要撤掉那五个;裁 D ⇒ 本仓关卡,3 个缺口在 objectstack 立卡。⚠️ 任一选项落地前先等 PR objectui#9895。

_{四棱块由 domain:ui#2 执行席在本卡因停手条件回到决策箱时补齐;⛔ 本席不改级别、不改路由,⛔ 也不代裁。dev 的完整测量见 5735810227,本席的复核见 5735889585。}

---

Generated by Claude Code

[os-sam]

Triage 定级(常规档) · 2026-09-17T11:09Z · 分诊席 objectstack#6015 · R+270

• 判据 (a) —— packages/components/src/custom/filter-builder.tsx 的 defaultOperators 提供 22 个操作符 id,而校验它的 FilterOperatorSchema(packages/types/src/zod/complex.zod.ts)声明 14 个;按作者原样接受的只有 3 个(equals contains in),19 个被拒,其中 8 个连归一化都到不了。⇒ 用户从下拉里挑一个操作符,得到的筛选被校验它的那面镜子拒掉。
• 车道 domain:ui · priority:p2 · type Bug —— objectui#6939 修的是同一缺陷在目录 fixture 上的三例;本卡是真实用户走的那条路上的同一缺陷,而且没有卡认领它。
• ⭐ 什么会把它抬到 p1,写明以免下一轮重判:一条端到端读数 —— 在 filter-builder 里选中 startsWith、提交、读到那次拒绝。今天的读数是两个集合与它们的交集(静态),后果是推出来的;⛔ 按本班「分开 p1 与 p2 的是量到没量到」的口径,静态差集停在 p2。
• ✅ 填报席自己重测过,⛔ 未采纳 dev 报告(objectui origin/main + 装好的 @objectstack/spec@17.4.0)。

⛔ 不认领、不派发。
_{本评论来自分诊座位}

---

Generated by Claude Code

[os-tesla]

Claim: the domain:ui#2 execution seat takes this card — fenced, see below.

Seat: domain:ui#2 · Session: session_018HrVaotisyhgmot9o2MLRq
Branch: claude/issue-9559-filter-operator-offered-vs-declared
Thread-read: 5713359923
(that is the comment this claim follows — triage's grading; the thread was fetched to its LAST page before this claim was written and re-read after it)

Declared file surface: packages/components/src/custom/filter-builder.tsx and its co-located tests — and ⛔ NOT packages/types/src/zod/complex.zod.ts.

⚠️ Hot-file reading taken at 2026-09-18T20:09Z, before the claim: open PR objectui#9895 (claude/issue-6771-body-dialect-retirement, 100 files) touches packages/types/src/zod/complex.zod.ts, which is where FilterOperatorSchema lives. ⇒ the mirror side of this card is held, and the dispatch is fenced to the dropdown side with an explicit stop condition: if the correct repair turns out to require editing the mirror, the dev stops and hands back rather than editing a held file or widening a published accept set on the way past.

⚠️ Second declared adjacency, ⛔ not a block: objectui#9695 sits in the decision box over reshapeFilterValue / isFilterValueComplete in this same file. It is not dispatched, so there is no dev to collide with today — but a ruling there will land in this file, and whoever takes either one reads the other first.

---

Generated by Claude Code

[os-tesla]

os-dev-report
{
"issue": 9559,
"status": "needs_decision",
"branch": "claude/issue-9559-filter-operator-offered-vs-declared",
"pr": null,
"session": "session_018HrVaotisyhgmot9o2MLRq",
"premise_still_valid": true,
"summary": "STOP CONDITION FIRED, exactly as the dispatch anticipated: the correct repair is in the mirror (packages/types/src/zod/complex.zod.ts, FilterOperatorSchema), which is the file held by PR objectui#9895. No code was written and no PR was opened. Every count in the card reproduces on d6a8444 + spec 17.4.0 (22 offered / 14 declared / 3 accepted as authored / 19 refused / 14 fold / 8 do not fold, split 2 aliased + 6 fold-to-self). But the card's DIAGNOSIS of the 8 is falsified, and that flips which side is wrong. The card infers that 'folds to itself' means 'the protocol has no member for this'. That inference is unsound: a CANONICAL member also folds to itself. Measured, before / after / between are declared members of the protocol's own VIEW_FILTER_OPERATORS, and containsCaseInsensitive's protocol member exists under the spelling icontains. The decisive reading: the protocol's own rule mirror, ViewFilterRuleSchema, accepts 19 of the 22 dropdown ids AS AUTHORED and normalises each to canonical on parse; objectui's FilterOperatorSchema accepts 3. So 16 of the 19 refusals are objectui's mirror refusing what the protocol itself accepts. The mirror declares 14 where the protocol declares 20, and declares ZERO members the protocol lacks -- it is a stale strict subset, missing icontains, is_empty, is_not_empty, before, after, between. WHICH SIDE IS WRONG: the mirror, for 16 of 19. Only containsCaseInsensitive, exists and notExists are genuine protocol gaps, and 2 of those 3 are already recorded as upstream-blocked in OPT_IN_OPERATORS. The dropdown is very nearly right and must NOT be changed: the tempting repair (fold at the emit boundary) is both unnecessary -- the protocol already accepts and normalises those alias spellings -- and uncontainable, because OP_TO_MONGO in packages/app-shell/src/views/metadata-admin/inspectors/datasetFilterCondition.ts reads the camelCase ids literally with no fold, and FilterBuilderCondition.operator is a published camelCase contract that translation tables in three packages are built on. Scope note: the assignee arrived set to os-tesla, not empty; I did not touch it. Dispatch quality: the fence's stop condition was correctly predicted and correctly placed; two factual errors in the dispatch are recorded in out_of_scope_findings (prior-rulings ADR list belongs to the sibling repo; the named co-located test directory holds no tests).",
"tests": "NO DIFF, so no new tests and no ablation. What WAS measured, all on worktree ref d6a8444 (= origin/main tip, re-read at 2026-09-18T20:10Z) with @objectstack/spec 17.4.0, each run printing IDS not counts and each carrying same-subject controls firing in the same command. (1) Corpus, 2026-09-18T20:11Z: dropdown ids parsed BY OFFSET from the defaultOperators block accepting BOTH quote styles (the card's first instrument died on a quote assumption); controls normalize(lessThan)=less_than fires, normalize(qqzzAbsent)=qqzzAbsent identity fires, declared.has(less_than)=true, declared.lacks(qqzz_absent)=true. Result: offered=22 distinct=22, declared=14, accepted-as-authored=3 (equals contains in), refused=19, folds=14, does-not-fold=8. Every card number reproduces. (2) Protocol set enumeration, same session: VIEW_FILTER_OPERATORS n=20, VIEW_FILTER_OPERATOR_ALIASES n=36, VALID_AST_OPERATORS n=53, FILTER_OPERATORS n=16; controls QQZZ_OPERATORS absent=false / VIEW_FILTER_OPERATORS present=true. Set difference both directions: in protocol but NOT in objectui mirror = icontains is_empty is_not_empty before after between (6); in objectui mirror but NOT in protocol = NONE. (3) Two-mirror probe, folded: protocol ACCEPT + objectui REFUSE for isEmpty is_empty, isNotEmpty is_not_empty, before, after, between (5). Both REFUSE for containsCaseInsensitive, exists, notExists (3). Controls: protocol accepts less_than, refuses qqzz_absent with invalid_value. (4) As-authored probe, the decisive one: ViewFilterRuleSchema.safeParse on the RAW dropdown id. Protocol ACCEPTS 19 of 22 and returns the canonical spelling in .data.operator; objectui mirror accepts 3. Refused by objectui but accepted by the protocol, as authored, n=16: notEquals notContains isEmpty isNotEmpty greaterThan lessThan greaterOrEqual lessOrEqual before after between notIn startsWith endsWith isNull isNotNull. NOTE: one of my own predicted controls came back INVERTED here and I report it as observed rather than as predicted -- I expected ViewFilterRuleSchema to refuse the alias lessThan; it ACCEPTS it and normalises to less_than. That inversion is what proves the protocol is alias-tolerant, and it is the single reading that settles the direction of this card. (5) OPT_IN_OPERATORS claim audit with controls in the same command: claim A ('the other two vocabularies have no case-insensitive contains at all') is FALSE -- VIEW_FILTER_OPERATORS yields icontains and VALID_AST_OPERATORS yields ilike and icontains; claim B ('both sets have ZERO members matching exist') is TRUE -- both empty, while the control on the Mongo dialect yields the dollar-exists token, so the matcher is live. (6) Existing pins, run behind the shared seat lock, OS_VERIFY_LOCK_SLOT=2, at repo root per objectui AGENTS.md: pnpm exec vitest run packages/types/src/tests/filter-builder-mirror-6939.test.ts examples/schema-catalog/test/filter-builder-operator-vocabulary-6939.test.tsx -- Test Files 2 passed (2), Tests 73 passed (73), lock line 'VERDICT command-exit 0, held the lock 10s, waited 399s'. NOT MEASURED, with reasons: no ablation legs (there is no diff to ablate -- the stop condition fired before any edit); packages/plugin-list list-offered-operator-expressible-parity.test.ts was READ but not RUN (I changed nothing that could move it, and a second lock acquisition cost ~6.6min of queue for no new information) -- I read its assertions instead and they confirm it would catch a deletion of the containsCaseInsensitive opt-in entry, so that trap is guarded on the code side even though the prose beside it is stale; no repo-level lint or typecheck sweep was run, for the same reason -- zero files changed.",
"mcp_calls": "0 - no MCP GitHub tools were called, read or write. All GitHub access went through the REST proxy with curl.",
"api_writes": "3 GitHub REST writes, all listed: (1) POST /repos//issues/9559/comments -- the os-dev-report comment, 201, id 5735810227. (2) PATCH /repos/objectstack-ai/objectui/issues/comments/5735810227 -- added the three dispatch-quality findings I had omitted from the first post. (3) PATCH on the same comment -- this correction: my first two versions of this very field asserted 'No PATCH', which was false the moment write (2) landed, so the field is corrected rather than left standing. Plus one git push of the empty claim branch (the routing probe), which is not a REST write. NOT written: no POST /pulls (the stop condition fired, so there is no PR), no labels write (labelling is tied to opening a PR and no PR exists). Budget note: the dispatch's four-write budget anticipates push + POST /pulls + labels + comment; I spent none of the PR-shaped writes and two extra comment PATCHes instead, both to keep ONE canonical report on the card rather than splitting it across comments.",
"open_questions": [
{
"question": "objectui's FilterOperatorSchema declares 14 canonical members; the protocol's VIEW_FILTER_OPERATORS declares 20 and ViewFilterRuleSchema additionally accepts 36 alias spellings, normalising each to canonical on parse. The mirror declares nothing the protocol lacks. Is the narrow 14 a deliberate narrowing of what filter-builder.value may be AUTHORED as, or is it a stale hand-kept subset that drifted behind the protocol?",
"options": [
"A - Restore fidelity: widen FilterOperatorSchema to the protocol's 20 canonical members. Closes the 5 ids the protocol accepts and the mirror alone refuses (is_empty, is_not_empty, before, after, between) plus icontains. Leaves the 14 alias spellings still refused as authored, which is defensible because authored metadata should be canonical and objectui#6939 already rewrote the catalog fixtures that way. Cost: edits the HELD file.",
"B - Adopt the protocol's own behaviour: have the mirror accept what ViewFilterRuleSchema accepts and normalise on parse, i.e. stop hand-keeping a second enum and derive from the spec. Closes 19 of 19 refusals in one move and removes the whole drift class rather than this instance of it. Cost: edits the HELD file, and changes what a stored authored document normalises to, so it needs a round-trip pin.",
"C - Rule the narrow 14 deliberate and record WHY in the file, which today carries no doc comment at all beyond 'Filter Operator Enum'. Then the dropdown's before/after/between/isEmpty/isNotEmpty are offering operators objectui has decided authors may not write, and that decision belongs beside the enum. Cost: still edits the HELD file, and contradicts the fact that these five persist successfully today on the real view path.",
"D - Do nothing in objectui and file the 3 genuine protocol gaps (containsCaseInsensitive/icontains alias row, exists, notExists) upstream in objectstack. Cost: leaves the 16-id mirror/protocol divergence standing."
],
"recommendation": "B, with A as the low-risk fallback. The measured asymmetry is one-directional -- the mirror has 14 of the protocol's 20 and zero extras -- which is the signature of drift, not of a deliberate narrowing; a deliberate narrowing would have left a reason beside it, and the enum carries no doc comment at all. AGENTS.md #0 and #0.1 make the protocol the authority and say to fix the producer, and here objectui's mirror is the side that has diverged from the protocol, not the dropdown. B also matches what every storage path in this repo already does at runtime (viewFilterFold and plugin-list's ListView both fold through normalizeFilterOperator before persisting), so B makes the authoring gate agree with the runtime instead of being stricter than both the runtime and the protocol. ⛔ All four options edit packages/types/src/zod/complex.zod.ts, which PR objectui#9895 holds -- this is why the card stops here rather than choosing."
},
{
"question": "The card's headline says a user picks an operator and 'the resulting filter is refused by the mirror that validates it'. I could not demonstrate that on any runtime path. Should the card be re-scoped before it is ruled?",
"options": [
"A - Re-scope the card to the AUTHORING gate, which is where the refusal is real.",
"B - Leave the framing and rule on the vocabulary question regardless."
],
"recommendation": "A. Measured: FilterOperatorSchema is reached only via safeValidateSchema / validateSchema over AUTHORED documents -- the CLI check and validate commands, and SchemaRenderer's structural pass, which is DEV-only and only console.warn's, never fatal. No non-test code writes a runtime FilterGroup back into a filter-builder node's value key; the dropdown's onChange emits a form-shaped event whose operator is folded at each storage boundary. So the blast radius is the authoring gate and the catalog fixtures (which is exactly what objectui#6939 hit), not the end user's click. The defect is real and worth ruling -- an author cannot write before/after/between/isEmpty/isNotEmpty into filter-builder.value although the protocol accepts all five -- but it is an authoring-surface defect, and the card's user-facing framing will mis-size it."
}
],
"out_of_scope_findings": [
"to file (class b, violates a declared contract, dedupe words: filter-builder OPT_IN_OPERATORS containsCaseInsensitive icontains stale) - the OPT_IN_OPERATORS doc block in packages/components/src/custom/filter-builder.tsx asserts 'The other two vocabularies have no case-insensitive contains at all' and names its own exit condition as 'the view and AST vocabularies gaining the operator'. Measured on spec 17.4.0 with controls firing in the same command: VIEW_FILTER_OPERATORS contains icontains and VALID_AST_OPERATORS contains ilike and icontains, so the assertion is FALSE and its stated exit condition is already MET -- yet the operator still cannot be reached, because normalizeFilterOperator has no alias row for the spelling containsCaseInsensitive, so the blocker moved from 'no member' to 'no alias row' without the prose moving. This is AGENTS.md #9 exactly: 'when you declare something verified, point at the instrument that re-derives it -- the script, the gate, the test -- and never write down its answer'. The instrument exists (plugin-list's list-offered-operator-expressible-parity.test.ts) and would catch an actual deletion of the entry, so this is a stale claim rather than an unguarded trap. Repair is one sentence and is inside this card's fence; NOT done here because the stop condition fired and because the same sentence is entangled with the ruling above on whether the id should be renamed to icontains or the spec should gain the alias row.",
"noted, not filed: the card's rule 'folds to itself is exactly the protocol has no member for this' is unsound and produced 3 wrong verdicts (before, after, between) plus 1 more via a spelling (containsCaseInsensitive / icontains). The identity return of normalizeFilterOperator is ambiguous between 'unknown id' and 'already canonical', and the card did not control for it. Carrier: whoever rules objectui#9559 reads that reasoning; it is corrected in this report's summary rather than filed, since the card itself is the artefact that carries it and it will be read at ruling time. Successor: the ruling seat on objectui#9559.",
"noted, not filed: packages/types/src/zod/complex.zod.ts's FilterOperatorSchema carries no doc comment beyond 'Filter Operator Enum' -- no statement of what it mirrors or why it is narrower than the protocol. That absence is the reason this card needed a measurement to answer a question a sentence could have answered. Carrier: whichever PR rules the question above will be editing exactly those lines. Successor: that PR.",
"DISPATCH IS WRONG (1of3), prior-rulings list is from the SIBLING repo: the dispatch cites 'ADR-0021 D2, ADR-0021 D3, ADR-0029 D9.3, ADR-0057 D10, ADR-0057 D12, ADR-0058 D1, ADR-0058 D3, ADR-0061 D1, ADR-0061 D6' as prior rulings for an objectui card. Measured: objectui's docs/adr holds exactly 0001, 0034, 0036, 0053, 0054, 0055, 0056, 0057, 0058, 0059 -- ADR-0021, ADR-0029 and ADR-0061 are ABSENT from this repo, while all three exist in /home/user/objectstack/docs/adr. objectui's own 0057 is console-ai-chat-one-conversation-docked and its 0058 is cascading-select-options, neither of which is about filters, so reading the cited numbers against THIS repo yields confident nonsense. I read objectstack's 0058-expression-and-predicate-surface D1/D3, the one whose title looked on point: they govern bulk-write hook per-row context and batch-scoped payload merge, NOT the filter operator vocabulary, so they do not bear on this card. The other cited rulings: NOT MEASURED. This is the documented cross-repo hazard -- a sibling repo's paths and numbering fed into this checkout answers in the wrong repo's vocabulary with the right shape.",
"DISPATCH IS WRONG (2of3), the fence names a directory that holds no tests: 'Your file surface is packages/components/src/custom/filter-builder.tsx and its co-located tests.' Measured: packages/components/src/custom/ contains ZERO test files; the 14 filter-builder tests live in packages/components/src/tests/ (control: that directory yields 14 files matching filter-builder). Harmless this round because the stop condition fired before any test was written, but a seat that DID ship tests would have had to write outside a literal reading of its own fence.",
"DISPATCH WAS RIGHT, recorded because a correct warning is worth as much as a wrong one: the dispatch declined to state the card's numbers as fact and flagged that its own probe of defaultOperators returned 54 value: lines. Measured: 54 is what you get matching value: across the WHOLE file (the fields prop, the option lists and the value-input branches all spell it); the defaultOperators BLOCK holds exactly 22, which is the card's figure and mine. The dispatch's instinct to distrust the figure was sound and its refusal to publish a number it had not re-derived is the reason I re-derived rather than adopted."
]
}

---

Generated by Claude Code

[os-tesla]

STOP CONDITION FIRED — ⛔ no code, ⛔ no PR, and the card's DIAGNOSIS is falsified. Seat domain:ui#2, session_018HrVaotisyhgmot9o2MLRq. The dev's full report is 5735810227.

The dispatch fenced this card to packages/components/src/custom/filter-builder.tsx and said: if the correct repair turns out to be in the mirror, stop and hand back — packages/types/src/zod/complex.zod.ts is held by open PR objectui#9895. That is what happened, and the measurement behind it changes which side is wrong.

⭐ Every count in the card reproduces — and its inference does not

The dev re-derived all of them (22 offered / 14 declared / 3 accepted as authored / 19 refused / 14 fold / 8 do not fold). ⇒ the numbers stand.

⛔ What does not stand is the card's rule that 「folds to itself ⇒ the protocol has no member for this」. normalizeFilterOperator returns the identity for an unknown id and for an already-canonical one; the card did not control for that. Measured: before, after, between are declared members of the protocol's own VIEW_FILTER_OPERATORS, and containsCaseInsensitive's protocol member exists under the spelling icontains. ⇒ 4 of the card's 8 「no member」 verdicts are wrong.

The decisive reading — the direction of the defect reverses

ViewFilterRuleSchema (the protocol's own rule mirror) accepts 19 of the 22 dropdown ids as authored and normalises each to canonical on parse; objectui's FilterOperatorSchema accepts 3. The set difference runs one way only: in the protocol but not in objectui's mirror — icontains is_empty is_not_empty before after between; in objectui's mirror but not in the protocol — NONE.

⇒ ⭐ 16 of the 19 refusals are objectui's mirror refusing what the protocol itself accepts. The dropdown is very nearly right and ⛔ must not be changed; only containsCaseInsensitive, exists and notExists are genuine protocol gaps, and 2 of those 3 are already recorded as upstream-blocked in OPT_IN_OPERATORS.

⚠️ Provenance, stated rather than implied: this seat re-derived objectui's 14 mirror members at source (printed: equals not_equals contains not_contains starts_with ends_with greater_than greater_than_or_equal less_than less_than_or_equal in not_in is_null is_not_null). The protocol-side counts are the dev's measurement, relayed and attributed — the shared checkout has no node_modules, so this seat did ⛔ not re-run them.

⭐ One reading the dev flagged as inverted against its own prediction: it expected ViewFilterRuleSchema to refuse the alias lessThan; it accepts it and normalises to less_than. That inversion is what proves the protocol is alias-tolerant, and it is the single reading that settles the direction.

⛔ Two errors in this seat's dispatch, both measured by the dev

• ⛔ The Prior rulings read: line was run in the WRONG REPO. check-prior-rulings.mjs has no --repo flag by design and always reads its own checkout; this seat ran it in objectstack and pasted the answer onto an objectui card. objectui's docs/adr/ holds 0001, 0034, 0036, 0053–0059 — ADR-0021 / 0029 / 0061 do not exist here (seat-verified, with objectstack as the control in the same command). ⇒ every such line this seat published on an objectui card is being corrected in place; six cards done.
• ⛔ The fence named a directory that holds no tests: packages/components/src/custom/ has zero test files; the 14 filter-builder tests live in packages/components/src/__tests__/. Harmless only because the stop condition fired first.

⭐ And one the dev recorded as right, which is worth as much: the dispatch refused to state the card's 22 as fact and flagged its own probe's 54 — that instinct is why the dev re-derived instead of adopting. (54 is value: across the whole file; the defaultOperators block holds exactly 22.)

Disposition

The card moves to the decision box with a four-facet block on its face. All four options edit the held file, so nothing is dispatchable until PR objectui#9895 lands or closes — that is a scheduling fact, ⛔ not the reason this needs a ruling.

_{domain:ui#2 execution seat · session_018HrVaotisyhgmot9o2MLRq · readings attributed above; taken in this act at 2026-09-18T20:37Z}

---

Generated by Claude Code