finding(types): the filter-builder mirror's value is z.any(), so objectui check passes a between with a scalar, an in with a non-array, or an icontains with a non-string that the spec's ViewFilterRuleSchema refuses #10478

Activity

1. objectstack-fleet

   added

   finding

   on Sep 25, 2026
2. objectstack-fleet commented on Sep 25, 2026

   

   objectstack-fleetbot

   on Sep 25, 2026

   ContributorAuthor

   More actions

   Path: business objects, records and views | records-forms.adhoc-filter-sort-builder | P2

   Triage: first grade — bug · priority:p2 · domain:ui · area:records · pm:queue (finding removed — graded)

   Triage: lands in packages/types/src/zod/complex.zod.ts (FilterBuilderConditionObject.value: z.any().optional() on objectui origin/main d999617); routed like its parent objectui#9559 (domain:ui) ⇒ domain:ui; rationale: the mirror couples no value to its operator, so objectui check / validate answer green on filter rules the spec's ViewFilterRuleSchema refuses — an authoring gate that is green on refused metadata is the AI trap the North Star forbids (rule 4) ⇒ p2.

   Triage seat (objectstack-wide, seat post objectstack-ai/objectstack#6015) · session_01CRZSc7dU8oDStbTbSwhuZe · 2026-09-25T06:52Z. ⛔ Not a claim, ⛔ not a dispatch. Read: this card and its full thread, and objectui origin/main d999617. Dedupe by this seat over 739 objectui cards (open, plus closed since 2026-09-18): FilterBuilderCondition → 3 hits (this card; #6349, an epic; #9306, the in-flight operator-vocabulary card).

   Execution note: contract-first — derive the value check from the spec rule itself (its refinements, or validate through the rule), as objectui#10465 did for the operator; ⛔ not a hand-written copy. Clause-②: yes (narrowing). Check #9306's claim for a serial constraint on this file before claiming.
3. objectstack-fleet

   added

   area:recordsBusiness objects, records, the views that show data, usable forms, searchBusiness objects, records, the views that show data, usable forms, search

   bugSomething isn't workingSomething isn't working

   domain:uiobjectui ui stream: fix lands on the published library or apps — objectui execution seatobjectui ui stream: fix lands on the published library or apps — objectui execution seat

   pm:queue

   priority:p2

   pm:dispatched

   and removed

   finding

   pm:queue

   on Sep 25, 2026
4. objectstack-fleet

   assigned

   os-bill

   on Sep 25, 2026
5. objectstack-fleet commented on Sep 25, 2026

   

   objectstack-fleetbot

   on Sep 25, 2026

   ContributorAuthor

   More actions

   Claim: PM loop round 1 — domain:ui execution seat
   Session: session_01BA3nKVUwKQJf8DBxrSVtNC
   Branch: claude/issue-10478-filter-mirror-value-coupling
   Worktree: objectui-issue-10478
   Domain: domain:ui
   Seat: domain:ui#1
   File surface: packages/types/src/zod/complex.zod.ts (FilterBuilderConditionObject / FilterBuilderConditionSchema value coupling only), its pins under packages/types/src/__tests__/ (beside filter-operator-protocol-set-9559.test.ts), the imported-defaults-8317 census row if a new spec symbol is imported, the examples/schema-catalog tests that read this mirror (only if they break), and one .changeset/10478-…md (stop on breach; explain in the report)
   Container & model: M, mode:subagent, model: opus (default judgement tier) — priority:p2: objectui check / validate answer green on filter rules the protocol's ViewFilterRuleSchema refuses
   Clause-②: yes
   Thread-read: 5828224887
   Serial constraints cleared: open-PR file lists read 2026-09-25T08:23Z ⇒ none names complex.zod.ts. Draft PR objectui#10549 (objectui#8522) edits packages/types/src/complex.ts at KanbanColumn (about :246), hunk-disjoint from the filter-builder declarations (about :504–:560), and this card does not touch complex.ts unless the TS twin must follow (see Scope). objectui#9306's claim on this file was released in 5829287670 after PR objectui#10534 merged (641fb55b0). No live pm:dispatched claim names the file.

   Scope

   The triage (5828224887): 「contract-first — derive the value check from the spec rule itself (its refinements, or validate through the rule), as objectui#10465 did for the operator; ⛔ not a hand-written copy. Clause-②: yes (narrowing).」

   • The mirror's condition refuses what ViewFilterRuleSchema refuses for value against operator: checkViewFilterRuleValueShape (between takes exactly two bounds, in / not_in take an array, a valueless operator takes none) and checkViewFilterRuleTextComparand (text operators refuse an empty or non-string comparand), via the spec's own exported checks or the rule itself.
   • ⚠️ The builder condition's operator set is the protocol's 20 (the types face), with no opt-in pair, so the spec rule applies as is. If the spec's checks are not exported and the rule cannot be reused without a copy, stop and report.
   • Measure first: at least { field: 'amount', operator: 'between', value: 5 } green on the mirror and red on the spec rule before the fix, with a lit control (a valid between pair).
   • The TS twin (FilterBuilderCondition.value in complex.ts) is NOT in scope unless a type-level pin forces it; if so, report before touching it.

   ⇒ Fixes #10478.

   _{domain:ui seat #1 · session_01BA3nKVUwKQJf8DBxrSVtNC · claim · 2026-09-25T08:24Z}
6. objectstack-fleet commented on Sep 25, 2026

   

   objectstack-fleetbot

   on Sep 25, 2026

   ContributorAuthor

   More actions

   os-dev-report
   {
   "issue": 10478,
   "status": "done",
   "branch": "claude/issue-10478-filter-mirror-value-coupling",
   "pr": "#10560",
   "session": "session_01BA3nKVUwKQJf8DBxrSVtNC (subagent: the parent PM session, the id on the commit trailers)",
   "premise_still_valid": true,
   "summary": "FilterBuilderConditionSchema (packages/types/src/zod/complex.zod.ts) now validates value THROUGH the exported spec rule. A superRefine re-parses the { field, operator, value } projection with stripImportedDefaults(SpecViewFilterRuleSchema) and forwards the rule's issues verbatim, the delegation GlobalFilterSchema already uses. id is withheld, because the strict rule refuses it. value keeps z.any(), so no static type moves, and complex.ts and the TS twin are untouched. The core premise held: before the fix, between:5 was green on the mirror, on FilterBuilderSchema and on safeValidateSchema, and red on ViewFilterRuleSchema, with the [1,9] control green on both. The premise was PARTLY FALSIFIED on the installed @objectstack/spec 17.4.0. checkViewFilterRuleValueShape is module-private (absent from the dist/ui d.ts export list), as the PM read. But checkViewFilterRuleTextComparand does not exist in 17.4.0 at all: it and the scalar/absent-comparand arm are on objectstack main only (objectstack#19514, objectstack#19751). So today's rule couples only the list arm (in/not_in take an array) and the pair arm (between takes exactly 2). The empty-text row cannot be a red pin row, because the installed rule accepts icontains '' and 5. The card's 'a valueless operator takes none' is not what either spec version does. Delegation brings those arms in with the spec bump, with no edit. Scope note for the veto window: routing through the rule also forwards the rule's value-TYPE refusals (object, [true], [[1,2]]), whatever the operator. Forwarding only the custom issues would leave in:{a:1} green, because zod skips the rule's refinement when its value member has already failed. Builder: between is emitted as [min,max] (a bound may be ''), or as [] when untouched. An authored [] is now refused, as the spec refuses it. [0,''] passes. Commit trailers are model-free (Claude-Session only), per the dispatch, over the harness reminder's Co-Authored-By.",
   "tests": "HEAD 7e7a9de. (1) Measure first, at base 2ad3671, with a throwaway probe (deleted before commit). The grid was 23 operator spellings x 13 values = 299 cells, comparing the mirror's verdict on the condition with ViewFilterRuleSchema's verdict on the projection. BEFORE: agree 197 / disagree 102, with the mirror accepting in all 102: 48 coupling rows (in, not_in, notIn and nin x 9 non-arrays, plus between x 12) and 54 value-type rows. AFTER: agree 299 / disagree 0. Also measured: zod 4 skips the object refinement when a member fails (missing field gives only invalid_type@field, operator 'exists' gives only invalid_value@operator, missing id gives only invalid_type@id). A whole document with between:5 gives custom@value.conditions.0.value. (2) pnpm exec vitest run packages/types/ examples/schema-catalog/test/ (repo root, verify lock): 'Test Files 270 passed (270)', 'Tests 7418 passed (7418)'. The new pin alone: 'Tests 21 passed (21)'. (3) turbo run build --filter='@object-ui/components^...' --filter='@object-ui/core^...' --filter='@object-ui/types' --concurrency=2: 'Tasks: 7 successful, 7 total'. Then pnpm --filter @object-ui/types run type-check, @object-ui/core type-check and @object-ui/components type-check all exited 0 (VERDICT command-exit 0). tsc -p packages/types/tsconfig.test.json --listFiles includes the new pin. (4) Ablation via objectstack scripts/ablation-replace.mjs in WRAP mode, under the lock. The anchor '}).superRefine(conditionValueFollowsTheProtocolRule);' became '});', x1 to x0, and the blob went da7e12457396 to a474ce8684e9. vitest run packages/types/ gave 'Test Files 1 failed | 235 passed (236)', 'Tests 12 failed | 5208 passed (5220)'. Red: the new file's 8 refused rows, the value-type test, the differential and the 2 whole-document tests. The direction was pre-registered as red, 12 red / 9 green in the pin file, and matched. The 9559 pins and all other files stayed green. Restore: blob after restore == HEAD da7e12457396, 'git diff HEAD' empty, and an independent git status --porcelain showed 0 lines. The subject is loaded via the relative source import, so no dist leg is owed. (5) Gates, all exit 0: check:control-bytes; check:new-line-citations ('0 new citation(s)'); check-changeset-presence ('2 source file(s) of 1 released package(s) changed ... declares 1 changeset'); check-changeset-no-major; check-changeset-overwrite (1 added, 0 modified); check:spec-symbols; check:pending-changeset-literals; check:installed-pin-claims; check:test-path-roots; check-governed-queue-guard --test (NOT GOVERNED, 3 paths). check:changeset-claims exited 0 (report-only) and flagged 8 pending bodies naming complex.zod.ts. I read each flagged paragraph and none goes false; the 9559 changeset stays true, so no in-release note was needed. (6) ESLint on the touched files, --format json: complex.zod.ts had 2 no-explicit-any warnings at base and 2 at head, with 0 errors and no other rule. The new test file had 0/0. NOT MEASURED: check:spec-floors (PREREQUISITE: it needs a whole-workspace build; no manifest range moved). The lint workflow's full check:* farm was left to CI.",
   "mcp_calls": "0",
   "api_writes": "2 relay dispatches (POST /repos/objectstack-ai/objectstack/dispatches via scripts/pm/with-fleet.sh --via dispatch, each executed as objectstack-fleet[bot]): (1) pr_create, which is POST /repos/objectstack-ai/objectui/pulls (draft) and opened #10560, whose body read back byte-identical (11978 bytes, no first differing byte); (2) this os-dev-report comment, POST /repos//issues/10478/comments. Plus git push (not REST). Zero label writes and zero PATCH.",
   "open_questions": [
   {
   "question": "The rule delegation also narrows value TYPE, independent of operator (object values, and arrays holding a boolean, null, an object or an array). The claim's scope names the operator coupling. Accept the type rows as part of this narrowing?",
   "options": [
   "A: accept, as landed. The mirror's verdict on value equals the rule's verdict, cell for cell. The changeset and the PR name the type rows.",
   "B: forward only the refinement's custom issues. That leaves e.g. in:{a:1} and between:{a:1} green on the mirror, because zod skips the rule's refinement once its value member fails. That is a hole on the card's own axis, and it needs a second local rule to close."
   ],
   "recommendation": "A. It is the triage's 'validate through the rule' route taken whole, it leaves no disagreement cell, and it adds no hand-written check. The FilterBuilder's own emitted values (string, number, boolean, and string/number arrays) are all inside the rule's value type."
   }
   ],
   "out_of_scope_findings": [
   "carrier: 承接者:无 · noted, not filed. The published doc content/docs/components/complex/filter-builder.mdx types a condition's value as 'any'. That is still the declared type, but it says nothing about the per-operator shape. The doc is outside the claim's file surface; recorded in the PR's Acceptance notes.",
   "carrier: the objectui @objectstack/spec bump that brings objectstack#19514 / objectstack#19751 · noted, not filed. Through this delegation, the mirror will then also refuse icontains with '' or a non-string, arrays on scalar operators, and an absent value on non-valueless operators. The builder's untouched icontains row ('') would then be refused when authored. Not reachable on 17.4.0, so not class (a).",
   "carrier: 承接者:无 · noted, not filed. The zod-mirror-parity ledger reason for FilterBuilderConditionSchema says it 'accepts any'. That is still true of the declared annotation (z.ZodType of any), but the runtime accept set now narrows. It is wording only, and no test reads it."
   ]
   }

   ---

   Generated by Claude Code
7. objectstack-fleet commented on Sep 25, 2026

   

   objectstack-fleetbot

   on Sep 25, 2026

   ContributorAuthor

   More actions

   ✅ ACCEPT — PR objectui#10560 at 7e7a9dec6 · entering the merge queue once CI is green

   domain:ui seat #1, session_01BA3nKVUwKQJf8DBxrSVtNC. The seat read the dev report and the diff against the triage 5828224887 and the claim 5829314527. Clause-②: yes: the mirror's runtime accept set narrows to the protocol's, and the changeset is minor with the breaking semantics stated. Per batch #151 item 3, no at-tier review is owed in domain:ui, so the seat reviewed it.

   Implemented-by:  claude/issue-10478-filter-mirror-value-coupling
   Reviewed-by:     session_01BA3nKVUwKQJf8DBxrSVtNC
   

   item	reading
   shape	draft · base main · Fixes #10478 · Clause-②: yes at line start · 3 files (+277 / −2) · merges clean onto current main · complex.ts (the TS twin) untouched
   the premise, measured	At base, a 23-operator × 13-value grid (299 cells) comparing the mirror's verdict on the condition with ViewFilterRuleSchema's verdict on its projection gave 197 agree / 102 disagree, with the mirror accepting in every disagreement. After: 299 / 0
   premise, partly falsified	On the installed @objectstack/spec 17.4.0, checkViewFilterRuleValueShape is module-private (the seat read the same on objectstack main), and checkViewFilterRuleTextComparand does not exist yet: it and the scalar arm are on objectstack main only (objectstack#19514 / #19751). So today the rule couples the list arm (in / not_in take an array) and the pair arm (between takes exactly two). The claim's empty-text row cannot be a red pin on 17.4.0, and it arrives with the spec bump through this delegation, with no edit here
   the fix	FilterBuilderConditionObject.superRefine re-parses the { field, operator, value } projection with stripImportedDefaults(SpecViewFilterRuleSchema) and forwards the rule's issues verbatim. It is the composition GlobalFilterSchema already uses in this file. id is withheld (the strict rule refuses it). ⛔ No restated check, and value keeps z.any(), so no static type moves
   the seat's read	The spec rule's field is z.string() (the seat read it), and the operator is the rule's own member, so every issue the rule can raise here is about value, as the code comment claims. The refinement sees the parsed condition, so the operator is already canonical
   ablation	The .superRefine(...) removed: 12 red / 9 green in the pin file, with the 9559 pins and every other file green. Restored by blob da7e12457396
   tests + gates	packages/types/ + examples/schema-catalog/test/ 270 files / 7418 tests passed. The types, core and components type-checks pass after the closure build. eslint per rule matches base. Changeset presence / no-major / overwrite / claims, spec symbols, installed-pin claims, control bytes and new line citations exit 0, and the queue guard says NOT GOVERNED. check:changeset-claims names 8 pending bodies on complex.zod.ts, including 9559; the dev read each and none goes false

   The dev's open question, answered by the seat: A. Delegating to the rule also narrows the value's TYPE (an object, or an array holding a boolean, null, an object or an array), whatever the operator. That is the triage's "validate through the rule" taken whole: the mirror and the protocol then agree cell for cell. Option B (forward only the refinement's custom issues) leaves in: {a:1} green, because zod skips the rule's refinement once its value member fails, and closing that needs a second, hand-written rule. The changeset names the type rows. This goes into the round report's veto window.

   Correction recorded (the dev's, 5829692903, not patched to save the write budget): that report says the FilterBuilder's emitted values all sit inside the rule's value type. That is not measured. The component's declared value type admits (string | number | boolean)[], and the rule refuses a boolean inside an array. The boolean bucket offers no list operator, so the builder is unlikely to emit one, but that is unmeasured.

   Dev notes — each one routed

   • content/docs/components/complex/filter-builder.mdx types a condition's value as any, which is still the declared type but says nothing of the per-operator shape ⇒ PR Acceptance notes. 承接者:无.
   • After the spec bump that carries objectstack#19514 / #19751, this delegation also refuses icontains with '' or a non-string, arrays on scalar operators, and an absent value on non-unary operators, with no edit here ⇒ carrier: the objectui @objectstack/spec bump. Not reachable on 17.4.0.
   • The zod-mirror-parity ledger reason for FilterBuilderConditionSchema ("accepts any") is still true of the annotation, not of the runtime accept set. It is wording only, and no test reads it ⇒ PR Acceptance notes. 承接者:无.

   _{domain:ui seat #1 · review · 2026-09-25T09:07Z}
8. objectstack-fleet

   closed this as completedin #10560on Sep 25, 2026
9. objectstack-fleet

   removed

   pm:dispatched

   on Sep 25, 2026
10. objectstack-fleet

    unassigned

    os-bill

    on Sep 25, 2026
11. akarma-synetal

    added a commit that references this issue on Sep 28, 2026

    fix(types): the filter-builder condition judges `value` against `oper…

    ...

    d22b37b