Repository navigation
console filter builder: a progress field is offered text operators (Contains, Starts with…) that the server refuses, and no numeric Greater/Less than #27497
Description
domain:devx#1 · sunhaolin · session_017xRBcSwDXZrZg7nGjP6P3U
Filing gate: ① a user-visible product defect with a reproduction, found by an off-clause probe of QA run #27438 (wave 3 of the 18.3.0 sweep, anchor #27377) and reproduced twice. reach: measured in the console's list filter builder. Every operator it offers for the field is refused by the data door with a 400.
Business reason: a user filtering a list by a progress field is offered only text operators (Contains, Starts with…), which the server refuses, and no Greater-than or Less-than, which the server accepts. The filter the user most likely wants ("progress above 50%") cannot be built in the UI.
Reader who acts: the triage seat, then the implementer for domain:ui (ui/packages/components/src/custom/filter-builder.tsx).
Dedupe (open AND closed): the query "filter builder progress field operators text contains instead of greater than" returned 0 hits.
Batch-of: 16 · QA run #27438
Symptom
The list filter builder offers text operators for a progress field and no numeric ones, and the server refuses those text operators.
Reproduction (subject c4c515a, stock showcase)
- As admin, open
/_console/apps/com.example.showcase/showcase_task/view/showcase_task.grid, click the toolbar filter, then Add filter, and pick the field Progress (served typeprogress,task.object.ts:70).- The operator list is: Equals, Does not equal, Contains, Contains (ignore case), Does not contain, Is empty, Is not empty, Starts with, Ends with. There is no Greater than or Less than.
- For comparison, Estimate (h), a
numberfield, gets Equals, Does not equal, Is empty, Is not empty, Greater than, Less than, ≥, ≤.
- On the server,
GET /api/v1/data/showcase_task?filter=["progress","contains","4"]→ 400 ("aims the text operator $contains at a declared progress field"). GET /api/v1/data/showcase_task?filter=["progress",">",50]→ 200, total 5.
Expected: progress is offered the numeric operators the server accepts.
Mechanism
In ui/packages/components/src/custom/filter-builder.tsx at about line 1207, the numeric type list is [number, currency, percent, rating] and leaves out progress, so the field falls through to the text operators.
QA-source: #27438 · records-forms.adhoc-filter-sort-builder · observed beside the clauses
Generated by Claude Code
Activity
objectstack-fleet commentedon Oct 29, 2026
triage#1 · os-zhuang · session_01AavokzJ5DndAwitDXvKy4U
Path: 看数据的视图 | records-forms.adhoc-filter-sort-builder | P②
Triage: p2 ·
domain:ui·area:records· 2026-10-29T11:23Z⛔ Not a claim.
bug,pm:queue.Generated by Claude Code
objectstack-fleet commentedon Oct 29, 2026
domain:ui#1 · os-sales · session_01L6bhgXU26UBEjd1KKHgNEx
Claim: PM loop round 2 · 2026-10-29T12:55Z
Session:
session_01L6bhgXU26UBEjd1KKHgNEx(seat; in-seatmode:subagentdev, same session, identity by branch)Account:
os-sales(the seat's linked user asGET /useranswers it; the card's assignee)Branch:
claude/issue-27497-progress-filter-number-opsWorktree:
objectstack-issue-27497Domain:
domain:uiSeat:
domain:ui#1File surface:
ui/packages/types/src/filter-field-types.ts, its mirror pins (ui/packages/types/src/__tests__/filter-builder-mirror-6939.test.ts,examples/schema-catalog/test/filter-builder-mirror-6939.test.tsx), generated artefacts the list feeds (regenerated), one patch changeset (stop on breach; explain in the report)Container & model:
S,mode:subagent,model: TIER_DEFAULT(seat-rundispatch-gates --tieron the paths: no path-derived mandate),effort: mediumClause-②: no
Metadata-protocol: none
Responsibility: console filter builder (this lane) | the one field-type list
FILTER_NUMBER_LIKE_TYPES(#27400) | every console user filtering a list on a progress fieldThread-read: 6495326385
Premise: holds — on
origin/mainad3a640e,FILTER_NUMBER_LIKE_TYPESis number/currency/percent/rating andprogressis absent fromFILTER_FIELD_TYPES, so a progress field falls to text operators.Serial constraints cleared: open PR #27340 (
os-bill) editsfilter-builder.tsxin other regions; this card does not plan to touch it.Resume-from: none
objectstack-fleet commentedon Oct 29, 2026
domain:ui#1 · os-sales · session_01L6bhgXU26UBEjd1KKHgNEx
os-dev-report
{
"issue": 27497,
"status": "needs_decision",
"branch": "claude/issue-27497-progress-filter-number-ops",
"pr": null,
"session": "session_01L6bhgXU26UBEjd1KKHgNEx",
"premise_still_valid": true,
"summary": "Premise holds on origin/main ad3a640: FILTER_NUMBER_LIKE_TYPES in ui/packages/types/src/filter-field-types.ts is number/currency/percent/rating, progress is absent from FILTER_FIELD_TYPES, so operatorsForFieldType gives a progress column the text bucket. Stopped before any edit on the dispatch's own tripwire: FILTER_FIELD_TYPES spreads FILTER_NUMBER_LIKE_TYPES and FilterFieldSchema.type in ui/packages/types/src/zod/complex.zod.ts is z.enum(FILTER_FIELD_TYPES); FilterFieldSchema and FilterBuilderSchema (fields: z.array(FilterFieldSchema)) are exported from the './zod' entry of the published package @object-ui/types (18.2.0, not private, exports map lists './zod' as dist/zod/index.zod), and FilterFieldType types FilterField['type'] on the main entry. Adding progress to the list therefore widens a published accept set by one member, which contradicts the claimed Clause-②: no / Metadata-protocol: none. Empty branch was pushed (no commits beyond ad3a640); the worktree was removed clean; no PR opened.",
"tests": "none run: stopped at the measurement step before any edit, per the dispatch's stop-and-report instruction.",
"gates": "NOT MEASURED: no diff exists, so dispatch-gates --commands has nothing to derive and --ran nothing to reconcile.",
"dogfood_battery": "n/a — no PR",
"mcp_calls": "0",
"api_writes": "1 — POST /repos//issues/27497/comments (this report, via scripts/pm/post-stamped.mjs). git push of the empty branch is not a REST write.",
"open_questions": [
{
"question": "Adding progress to FILTER_NUMBER_LIKE_TYPES (the triage direction) also widens the published FilterFieldSchema.type enum and the FilterFieldType union in @object-ui/types by one member, progress. Land it with the widening declared, or keep the accept set unchanged?",
"options": [
"A — Follow the triage direction as written: add progress to FILTER_NUMBER_LIKE_TYPES, update the mirror pins (BUCKETED_TYPES row; the spec-FieldType check already accepts it, since progress is a spec FieldType), and re-claim with Clause-②: yes (widening) and Metadata-protocol: added, so the changeset is minor on @object-ui/types and the PR owes a Skills-docs line. About 20 changed lines; the review tier follows the Clause-② yes declaration (TIER_DEFAULT lane-seat record per contract-review.md).",
"B — Keep the published enum as it is: map progress to the number family only inside the renderer, or at the console producer that builds the filter fields. That is a per-field override, which the triage comment rules out, and a filter-builder node authored with type progress would still be refused by the validator while the console draws it. Not recommended."
],
"recommendation": "A, because progress is already a spec FieldType, so the widening brings the renderer's vocabulary into line with the spec rather than inventing a dialect; it keeps one shared list, as the triage asks; and the declaration lines then match the diff. Only the claim's Clause-② and Metadata-protocol lines and the changeset level change."
}
],
"out_of_scope_findings": [],
"deviations": [
"Conflict named: os-dev.md puts the bare marker os-dev-report on line one, but post-stamped.mjs refuses any body whose first non-empty line is not the seat signature (dry run exit 2, 'the body opens with no signature line'). The dispatch's order is used: signature on line one, marker alone on line two.",
"Conflict named, not exercised (no PR): the dispatch asks for the bare footer in the PR body; os-dev.md and AGENTS.md give PR bodies the session-URL footer form."
],
"line_budget": "0 / 150 — no diff",
"files_changed": []
}
objectstack-fleet commentedon Oct 29, 2026
domain:ui#1 · os-sales · session_01L6bhgXU26UBEjd1KKHgNEx
待裁决：筛选器的数字字段清单要不要扩到 progress（会放宽一个已发布的接受集）· 2026-10-29T13:05Z
一句话问题：控制台列表筛选里，「进度」类字段只给出服务端会拒收的文本条件（包含、开头是…），没有「大于 / 小于」；修它就得让一个已发布包（
@object-ui/types）多接受一种字段类型，这是放宽契约，按规矩要你定方向。背景：开发子代理按分诊方向（把 progress 加进现有数字类型清单，不做逐字段覆盖）开工前实测停手，报告见 6497372664。这份清单
FILTER_NUMBER_LIKE_TYPES自 #27400 起是唯一来源，同时喂三处：校验器FilterFieldSchema.type（z.enum(FILTER_FIELD_TYPES)，从已发布的@object-ui/types/zod入口导出）、类型FilterField['type']、渲染器的分桶。所以加一个成员就是三处一起放宽，认领评论里的Clause-②: no不成立，应为yes。Governing text：
ui/packages/types/src/__tests__/filter-builder-mirror-6939.test.ts文件头记录了此接受集历次变动都是裁决驱动：objectui#6939 维护者裁决（评论 5510084784）、决策批 #88 放宽到十四个（objectui#7562）、#27369 收成单一来源且「接受集未动」。spec 侧：packages/spec/src/data/field-value.zod.ts的NUMERIC_VALUE_TYPES= number / currency / percent / rating / slider / progress / summary，服务端按它判数字字段。协议声明：
Clause-②: yes；Metadata-protocol: added（元数据作者可写的filter-builder字段type多一个值）；changeset 对@object-ui/types为 minor，PR 正文欠一行Skills-docs:。前提（附复测命令）：
origin/main上清单为 number/currency/percent/rating，progress 不在FILTER_FIELD_TYPES：git show origin/main:ui/packages/types/src/filter-field-types.ts | grep FILTER_NUMBER_LIKE_TYPES。服务端：卡面复现filter=["progress",">",50]→ 200、["progress","contains","4"]→ 400。顺带实测：slider、summary 同为 spec 数字类型，也不在清单里，是同一类缺口（服务端对它们的数字条件未实测）。选项
FILTER_NUMBER_LIKE_TYPES，申报改为Clause-②: yes/Metadata-protocol: added，镜像测试补一行。客户看到：进度字段能筛「大于 50%」，文本条件消失。工作量：ui 车道 · 1 PR · 约 20 行；无直接放宽先例（最近同文件 refactor(ui-types): one source for the filter-builder field-type list; close the three K4 review should-fixes #27400 是重构，+368/−249）。slider、summary 另立卡。type: progress的筛选器节点仍被校验器拒、控制台却画得出来，形成「声明与实际不符」。工作量：约 15 行。分诊已明确排除逐字段覆盖。NUMERIC_VALUE_TYPES（本包已依赖@objectstack/spec），一次补齐 progress、slider、summary，以后 spec 加数字类型自动跟上。工作量：ui 车道 · 1 PR · 约 40–60 行，外加先实测服务端对 slider、summary 数字条件的接受情况；无先例。业务含义直译：A 是「修好这一个按钮」；B 是「在前台贴一张便条，后台账本不改」；C 是「让前台的商品分类表直接抄总部目录，不再自己维护一份」。
四轴（业务立场）
type: progress被校验器接受且渲染器兑现，出错时响亮；B 让校验器拒、渲染器收，AI 写元数据时被静默误导。Prior rulings read:
FILTER_FIELD_TYPES、Ruled、rulingin filter-builder-mirror-6939.test.ts → 4 hits（objectui#6939 5510084784、决策批 #88、#27369、objectui#9559 ruling B）; ADR none; thread: none推荐
推荐 C。终态句：两年后筛选器的字段分类应与平台字段类型的值族一一对应，像 Salesforce / Airtable 那样按字段类型给条件，不逐字段配置，也不在 UI 另抄一份清单；C 就是这个终态，A 朝同一方向但仍是手抄清单。自检：只看①选 C；②③④ 是否翻转：是（④：slider、summary 今天零实测拉动）。按规则答是不改字母，④只调执行：开发先实测服务端对 slider、summary 的数字条件，服务端拒的不入清单并写进报告。工作量 C 约为 A 的 2–3 倍，差距不到 3 倍。
裁后执行
Clause-②: yes/Metadata-protocol: added，派席内开发，PR 带 minor changeset 与Skills-docs:行，ACCEPT 前做TIER_DEFAULT契约复核记录；另立 slider/summary 卡交分诊。pm:queue，下一位 ui 席照裁执行。Generated by Claude Code
objectstack-fleet commentedon Oct 29, 2026
domain:ui#1 · os-sales · session_01L6bhgXU26UBEjd1KKHgNEx
Ruling: maintainer letter C on analysis 6497480039 · 2026-10-29T14:23Z
The maintainer (Sales) picked option C on the seat thread's decision card (「C 跟 spec 数字类型对齐」), 2026-10-29T14:22Z. Recorded by the domain:ui seat 1; reversible on one word from the maintainer.
C: the filter builder's number bucket follows the spec's numeric value family,
NUMERIC_VALUE_TYPESinpackages/spec/src/data/field-value.zod.ts(number, currency, percent, rating, slider, progress, summary), taken from that export rather than copied, so a spec numeric type added later follows automatically.Execution parameters:
FILTER_NUMBER_LIKE_TYPESinui/packages/types/src/filter-field-types.tsderives fromNUMERIC_VALUE_TYPES(@objectstack/spec/data, already a dependency of@object-ui/types);FILTER_FIELD_TYPES,FilterFieldSchema.typeandFilterFieldTypefollow from it as today.sliderandsummary, measure the server's operator acceptance for each (GET /api/v1/data/<object>?filter=["<field>",">",N]on a stock showcase field, or the server's operator-set code cited). A type the server refuses numeric operators for stays out, and the report says so with the reading.Clause-②: yes (widening),Metadata-protocol: added; changesetminoron@object-ui/types,patchon any other package it moves; PR body carriesSkills-docs:.TIER_DEFAULTcontract-review record by this seat before ACCEPT (declaration face).Ruling-ref for the next claim and report: this comment.
Generated by Claude Code
objectstack-fleet commentedon Oct 29, 2026
domain:ui#1 · os-sales · session_01L6bhgXU26UBEjd1KKHgNEx
Claim: PM loop round 3 (supersedes 6497251607 on maintainer ruling C, Ruling-ref: 6499302510) · 2026-10-29T14:25Z
Session:
session_01L6bhgXU26UBEjd1KKHgNEx(seat; in-seatmode:subagentdev, same session, identity by branch)Account:
os-sales(the seat's linked user asGET /useranswers it; the card's assignee)Branch:
claude/issue-27497-progress-filter-number-opsWorktree:
objectstack-issue-27497Domain:
domain:uiSeat:
domain:ui#1File surface:
ui/packages/types/src/filter-field-types.ts(number bucket derived from@objectstack/spec/dataNUMERIC_VALUE_TYPES), its mirror pins (ui/packages/types/src/__tests__/filter-builder-mirror-6939.test.ts,ui/examples/schema-catalog/test/filter-builder-mirror-6939.test.tsx),ui/packages/components/src/custom/filter-builder.tsxonly if a renderer branch is measured missing, generated artefacts the list feeds (regenerated), one changeset; ⛔packages/spec/**(stop on breach; explain in the report)Container & model:
S,mode:subagent,model: TIER_DEFAULT(seat-rundispatch-gates --tieron the paths: no path-derived mandate),effort: mediumClause-②: yes (widening:
FilterFieldSchema.type/FilterFieldTypeon the published@object-ui/typesgain the spec numeric types the server reads numerically)Metadata-protocol: added
Responsibility: console filter builder (this lane) | the one field-type list
FILTER_NUMBER_LIKE_TYPES(#27400) | every console user filtering a list on a progress fieldThread-read: 6499302510
Premise: holds — on
origin/mainad3a640e,FILTER_NUMBER_LIKE_TYPESis number/currency/percent/rating andprogressis absent fromFILTER_FIELD_TYPES, so a progress field falls to text operators.Serial constraints cleared: open PR #27340 (
os-bill) editsfilter-builder.tsxin other regions; this card does not plan to touch it.Resume-from: none
objectstack-fleet commentedon Oct 29, 2026
domain:ui#1 · os-sales · session_01L6bhgXU26UBEjd1KKHgNEx
os-dev-report
{
"issue": 27497,
"status": "done",
"branch": "claude/issue-27497-progress-filter-number-ops",
"pr": "#27631",
"session": "session_01L6bhgXU26UBEjd1KKHgNEx",
"premise_still_valid": true,
"summary": "Ruling C executed. FILTER_NUMBER_LIKE_TYPES in ui/packages/types/src/filter-field-types.ts is now taken from NUMERIC_VALUE_TYPES in @objectstack/spec/data instead of a hand-kept four, so progress, slider and summary join the number bucket, FILTER_FIELD_TYPES, FilterFieldSchema.type and FilterFieldType (a widening by three members). Server acceptance was measured live on a fresh stock showcase built from this branch: numeric comparisons answer 200 and contains answers 400 for all three types, so nothing was excluded. filter-builder.tsx was not touched: every renderer branch reads the bucket by membership, and the render-half per-member measurement covers the new members.",
"tests": "pnpm --dir ui exec vitest run packages/types/ packages/components/src/tests/filter-builder packages/plugin-list/src/tests/list-offered-operator-expressible-parity.test.ts examples/schema-catalog/ :: lock VERDICT command-exit 0, Test Files 453 passed (453), Tests 13312 passed (13312). pnpm --filter @object-ui/types run type-check :: exit 0 (tsconfig.test.json --listFiles includes the mirror test, 1 hit). pnpm --filter @object-ui/example-schema-catalog --filter @object-ui/components run type-check :: exit 0. Ablation via scripts/ablation-replace.mjs (subjects resolve to src through the ui vitest alias for @object-ui/types and the relative import, so no dist leg): the derivation was mutated to filter out slider (anchor 1 to 0, blob 2af20a77c320 to c231fbe12b53), and the mirror run went red with Test Files 1 failed and 2 tests failed (the bucket-equals-spec-family pin and the slider accept row); the restore matched the HEAD blob and git diff HEAD was empty. Reverse type verification: adding formula to the exhaustive type-face record fails tsc -p tsconfig.test.json with TS2353, and the file was restored to its HEAD blob. A first attempt used an anchor contained in its replacement; the tool refused it as a no-op and the run was redone with a disjoint anchor. Server measurement on showcase (fresh DB, port 44186, this branch's build): progress gt 50 gives 200 (5 rows), gt '50' 200, gte 200, lte 200, contains 400; f_slider gt 10 200, lte 10 200, contains 400; showcase_invoice.total (summary) gt 0 200 (4 rows), lt 1e8 200 (12 rows), contains 400; showcase_expense_report.total_amount (summary) gte 0 200.",
"gates": "On head 0c17b16, dispatch-gates --commands --repo objectstack-ai/objectstack derived 34 commands. All 34 were run, each with its exit captured before any pipe, and all 34 exited 0 (33 lock-free; pnpm --dir ui exec vitest run scripts/ under the lock: 169 passed, 2 skipped). dispatch-gates --ran: 34 derived, 34 run, 0 NOT-MEASURED, a derived zero. pnpm check:new-line-citations (ui): 0 new citations, exit 0. PR check-runs on 0c17b16 at report time: 33 total, 4 success, 11 skipped, 16 in_progress, 2 queued, 0 failed, so CI is in_progress. The type-check lanes and artifact-roster families are left to CI.",
"dogfood_battery": "n/a — no ! in the PR title and Clause-② carries the (widening) arm, not (narrowing)",
"mcp_calls": "0",
"api_writes": "2 — POST /repos/objectstack-ai/objectstack/dispatches (fleet-write relay pr_create, request fw-20261029T152456Z-6443c8, run 40901342135 success, created draft PR #27631 with no assignees; the read-back says 9551 bytes sent and 9551 stored, identical); POST /repos//issues/27497/comments (this report, via scripts/pm/post-stamped.mjs). git push is not a REST write.",
"open_questions": [],
"out_of_scope_findings": [
"carrier: none · noted in PR Acceptance notes, not filed: ui/packages/plugin-view/src/config/view-config-utils.ts normalizeFieldType keeps its own numeric spelling list (integer/int/float/double/number/currency/money/percent/rating), which also leaves out progress, slider and summary. This is read-only inference, not reproduced through a public door, so it is not a filing class. Dedupe words: normalizeFieldType progress, view config filter numeric type, plugin-view field type list",
"carrier: none · noted in PR Acceptance notes: ui/packages/components/docs/FilterBuilder.md lists type as five spellings and was already stale before this change (polish)"
],
"deviations": [
"Conflict named: os-dev.md puts the bare marker os-dev-report on line one of the report comment, but post-stamped.mjs refuses a body that does not open with the seat signature. The dispatch order was used: signature on line one, marker on line two.",
"Conflict named: the dispatch asked for the bare Generated-by footer in the PR body, but os-dev.md and AGENTS.md give PR bodies the session-URL form. The session-URL form was used (os-dev.md wins on conflict), and the dispatch's trailing harness lines were kept after it.",
"Conflict named: the harness attribution reminder asked for a model-named Co-Authored-By commit trailer, but AGENTS.md requires the model-free pair and the pre-push hook refuses a model identifier. The commit carries Claude-Session plus Co-authored-by: Claude, model-free.",
"Interpretation of 'derived, not copied': the runtime value is derived from NUMERIC_VALUE_TYPES. That export is typed as a read-only set of strings, so it yields no literals at compile time, and packages/spec is out of the file surface. The type face therefore declares a literal union (SpecNumericValueType), pinned equal to the spec set by an exhaustive record plus a runtime key comparison. A spec addition reaches the runtime list, the zod enum and the renderer at once, and reddens the pin until the union names it.",
"Spelling only: the dispatch names the render-half pin as examples/schema-catalog/test/filter-builder-mirror-6939.test.tsx (relative to ui/), and the claim names it ui/examples/schema-catalog/...; these are the same file.",
"The changeset is minor on @object-ui/types only. No other published package's files moved (filter-builder.tsx untouched; schema-catalog is an example).",
"Ablation was run once, on the derivation (dropping slider), and it reddened the types-half pins. The render-half operator block was not ablated separately: NOT MEASURED, reason: one ablation per tested behaviour was judged sufficient, since the block iterates the same bucket."
],
"line_budget": "100 / 200 (+95 / -5, 4 files, per dispatch-gates)",
"files_changed": [
".changeset/27497-filter-number-bucket-spec-numeric.md",
"ui/examples/schema-catalog/test/filter-builder-mirror-6939.test.tsx",
"ui/packages/types/src/tests/filter-builder-mirror-6939.test.ts",
"ui/packages/types/src/filter-field-types.ts"
]
}
objectstack-fleet commentedon Oct 29, 2026
domain:ui#1 · os-sales · session_01L6bhgXU26UBEjd1KKHgNEx
落地 · 2026-10-29T19:55Z
9ab75eea1b在origin/main上git grep -c "NUMERIC_VALUE_TYPES" origin/main -- ui/packages/types/src/filter-field-types.tsexport= 6）状态：按裁决 C（6499302510）落地，卡已由 PR 关闭，摘
pm:dispatched。