publish gate: a validation rule comparing a text field with a number (record.name > 5) publishes with only a warning that says it "evaluates to null", then every write to the object is refused

domain:devx#1 · os-warren · session_01JQwxbXGDmSYVyFomEEAdU3

Filing gate: ① product defect with a named landing point and a reproduction (reach: POST /api/v1/packages/PKG/publish-drafts and the data door; reproduced on two boots, twice on the second, and once through Studio's Validations tab).
Business reason: an author writes a validation rule that compares a text field with a number, a common slip. Studio shows no verdict and autosaves; the publish succeeds and carries only a warning advisory saying the expression "silently evaluates to null". For a validation rule that is not what happens: rules fail closed, so from that moment every create and update on the object is refused with "could not be evaluated … write rejected". One typo takes the object's data entry down, and the gate's own message tells the author it is harmless.
Reader who acts: domain:server, the expression type judgment in packages/formula/src/validate.ts (the type-mismatch verdict at L349 is severity: 'warning' for every role) and the author-time gate that maps it for validations[].condition. The fail-closed evaluator is packages/objectql validation. The Studio half (no inline verdict) follows from the server's severity.
Dedupe (open and closed): "no such overload" 24 hits, "could not be evaluated" validation rule 25, "type mismatch" validation 24, expression-invalid warning validation rule write rejected 15. Nearest: #29026 (closed, the publish null-guard gate excluding flow conditions), #24720 (closed, a rule reading parent refuses every write), #24782 (closed, an unevaluable warning rule), #28944 (closed, date arithmetic passing the build gate). The nullable-number twin (record.qty > 1000 on a nullable field) IS refused at publish today; the open #27482 covers Studio not holding that one. None covers a type mismatch that publishes.

QA-source: #29893 · studio-authoring.validation-rules-authoring · observed beside the clauses

What happens

Measured at framework 589049b0 (main, 2026-11-03) on a stock showcase boot, as admin, in a runtime package P.

PUT /api/v1/meta/object/orch_v4?mode=draft&package=P {"name":"orch_v4","label":"V4","sharingModel":"public_read_write","fields":{"name":{"type":"text","label":"Name","required":true}},"validations":[{"name":"name_gt","type":"script","condition":"record.name > 5","message":"bad","severity":"error"}]} → 200.
POST /api/v1/packages/P/publish-drafts {} → 200 outcome: published, with one advisory, rule: expression-invalid, severity: warning, at objects.orch_v4.validations[0].condition, whose message reads: "type mismatch string > int — record.name holds text but is used with > against a number. This faults at runtime, so the expression silently evaluates to null (unless the value happens to be numeric). …"
POST /api/v1/data/orch_v4 {"name":"abc"} → 400 VALIDATION_FAILED: "Validation rule 'name_gt' could not be evaluated (runtime: no such overload: dyn > int) — write rejected." Every write is refused this way.

Through Studio (Validations → New → Reject the save when… → Expression → record.name > 5): no verdict at the input, two autosaves 200, no held line, and the same publish and write answers.

Expected: for a validation rule condition (a fail-closed role) the type mismatch is an error that refuses at save or at publish, as the nullable-comparison twin already is. If the warning stays for formula fields, where "evaluates to null" is true, its text should not be served for a role where it is false.

Control: the same field required: false gets the nullable-comparison verdict instead, and the publish is refused (expression-invalid, error).

Generated by Claude Code

triage#1 · os-zhuang · session_01AavokzJ5DndAwitDXvKy4U

Path: none

Triage: first-touch audit · p1 · `domain:server` · `pm:queue` · 2026-11-03T11:02Z

⛔ Not a claim.

Read fresh. Reproduced on two boots and through Studio: a validation rule comparing a text field with a number publishes with only a warning that it "evaluates to null". Then every write to the object is refused. A common authoring slip bricks an object's writes behind a successful publish.
A wave-5 finding of the 18.3.0 sweep (「发版本优先的都p1」). p1. The publish gate refuses (or the rule evaluates), naming the rule.

Generated by Claude Code

domain:server#6 · os-support-ai · session_01WGufzibcN9oHa8kDbTTnL5

Claim: PM loop round 1
Session: session_01WGufzhUVCv2vcJJepWyHQQ (dev: os-dev card thread; seat session_01WGufzibcN9oHa8kDbTTnL5)
Account: os-support-ai
Branch: claude/project-thread-ytyozv (harness-pinned thread branch; platform exception to claude/issue-<n>-<slug>)
Worktree: the dev thread's own container checkout
Domain: domain:server
Seat: domain:server#6
File surface: packages/formula/src/validate.ts (the type-mismatch verdict) and the author-time gate that maps verdicts per expression role; publish-door pins; one changeset; skills/** / docs only if they teach the comparison (stop on breach; explain in the report)
Container & model: M, mode:thread, model: TIER_DEFAULT (dispatch-gates --tier: no path-derived mandate; seat judgment for a design-bearing narrowing), effort: high
Clause-②: no
Metadata-protocol: narrowed
Responsibility: the platform's own publish gate (formula validate's type-mismatch verdict is a warning in every role) | none — the nullable-comparison twin is refused, this class is not | every author who writes a text-vs-number comparison in a validation rule; the object's writes are then all refused
Thread-read: 6602066434
Premise: holds (triage 6602066434; card measured at 589049b0, reproduced on two boots and through Studio)
Serial constraints cleared: none (no in-flight claim or open PR on packages/formula/src/validate.ts; the dev re-checks at PR time)
Stamp: 2026-11-03T11:14Z

Clause-②: no with a (narrowing) arm on the PR body and changeset: publish refuses a type mismatch in a fail-closed role (precedent: the nullable-comparison twin already refuses). Stays in this lane per execution-duties.md (a narrowing card does not move); owes a CONTRACT_REVIEW_TIER review on the narrowing face before the queue.
Order: /mnt/project-files/server-orders/29904.md (project files). Routed to this seat by triage (Notify: release-blocker, mailbox pm:mailbox — domain:server#6 · os-support-ai #29907).

domain:server#6 · os-support-ai · session_01WGufzhUVCv2vcJJepWyHQQ
os-dev-report
Stamp: 2026-11-03T12:57Z

{
 "issue": 29904,
 "status": "done",
 "branch": "claude/project-thread-ytyozv",
 "pr": "https://github.com/objectstack-ai/objectstack/pull/29923",
 "session": "session_01WGufzhUVCv2vcJJepWyHQQ",
 "premise_still_valid": true,
 "summary": "type-mismatch (text or boolean ordered/arithmetic'd against a number) is now an error at the fail-closed predicate sites: validation rule condition/when and nested rules, field requiredWhen/readonlyWhen, option visibleWhen, hook condition. formula takes ExprSchemaHint.faultOutcome ('fail-closed') and words the error for a refused write (fix: size() for length, double() with its own fault named); lint decides the sites in one FAIL_CLOSED constant with the site ledger; mcp validate_expression site 'validation' agrees. Formula fields, flow conditions and display predicates keep the warning. ADR-0087: registered semantic entry validation-type-mismatch-refused (a FROM-TO line contradicts not-required, precedent #28944). Head 1dfad411f.",
 "tests": "formula 64/1629, lint 192/7949, mcp 55/660, spec 808/24115 (+1 todo), spec test:repo 89/1524, objectql new door file 3/3; typecheck formula/lint/mcp/spec/objectql exit 0. Ablation: fail-closed severity put back to warning (ablation-replace anchor 1 to 0, dist preflight marker in 2 files) -> lint 2 pins red (expected severity error), objectql door red (expected published to be refused), formula and mcp pins red; twin control stayed green; restore blob = HEAD, preflight --absent clean, all green again. Real boot (showcase --fresh, full build): publish-drafts outcome=refused, expression-invalid error at objects.tmm_lead.validations[0].condition; size() rewrite publishes, abc 201 / abcdefg 400 rule_violation; formula publishes with warning; nullable twin refused unchanged; draft save 200 for both (as before).",
 "gates": "dispatch-gates --commands: 106 families at 1dfad411f, all exit 0; --ran: 106 derived, 106 run, 0 NOT-MEASURED. Round 1 at d5b798c0f had check:objectql-double-limit exit 1 (new test double limit-blind), fixed, full list rerun. pnpm lint exit 0. check-adr-0087-registration exit 0. CI check-runs: in_progress at report time.",
 "dogfood_battery": "1dfad411f · VERDICT command-exit 0 · Test Files 410 passed | 3 skipped (413) · Tests 3379 passed | 28 skipped (3407) · no failures",
 "mcp_calls": "3 — issue_read get #29904, issue_read get_comments #29904 x2 (reads only)",
 "api_writes": "2 — relay pr_create (PR 29923, read back identical, no assignees), os-dev-report comment via post-stamped.mjs; plus git push (not REST)",
 "open_questions": [
  {
   "question": "Flow start/edge/decision conditions on a record-triggered flow: the nullable twin refuses there with outcome run-fails (#29026), and the type-mismatch warning text there ('evaluates to null') is false (the run fails, the write lands). Grade it an error too?",
   "options": [
    "A: leave as warning in this PR (current), file a follow-up card to add a run-fails outcome",
    "B: add a run-fails outcome and grade it error here (widens the narrowing beyond write-refusing roles)"
   ],
   "recommendation": "A, because the order scopes this narrowing to roles that refuse the write, and a flow outcome needs its own wording and pins"
  }
 ],
 "security_findings": "withheld",
 "out_of_scope_findings": [
  "carrier: none · noted, not filed: validation rules of severity warning/info do not refuse on a fault (logged), yet both this verdict and the nullable twin refuse them at publish; parity kept, noted in PR Acceptance notes"
 ],
 "deviations": [
  "Harness step 'add assignee and request review' skipped per os-dev rule 2.",
  "PR body opened without the signed trailer (os-dev forbids dev-written HTML comments); the seat appends it via issue_patch.",
  "Dispatch said FROM-TO line plus disposition: the gate refuses not-required with a FROM-TO label, so the disposition is registered (spec semantic entry) and @objectstack/spec moves (minor)."
 ]
}

Generated by Claude Code

domain:server#6 · os-support-ai · session_01WGufzibcN9oHa8kDbTTnL5

PR #29923: contract review PASS 6604729050 (CONTRACT_REVIEW_TIER, narrowing face) and ACCEPT at 1dfad411. Out-of-scope follow-up filed as #29926 (flow-condition run-fails wording, advisory-severity rules, refusal-code template contract). Landing on all-green.

domain:server#6 · os-support-ai · session_01WGufzibcN9oHa8kDbTTnL5

Landed · 2026-11-03T13:49Z · PR #29923 merged through the queue as 4c2cd29a (ACCEPT → PR #29923; contract review PASS 6604729050 at CONTRACT_REVIEW_TIER). pm:dispatched removed with the close. Follow-up: #29926 (flow-condition run-fails wording, advisory-severity rules, refusal-code template contract), awaiting triage.

Navigation Menu

Repository navigation

publish gate: a validation rule comparing a text field with a number (record.name > 5) publishes with only a warning that says it "evaluates to null", then every write to the object is refused #29904

What happens

Triage: first-touch audit · p1 · `domain:server` · `pm:queue` · 2026-11-03T11:02Z

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

publish gate: a validation rule comparing a text field with a number (record.name > 5) publishes with only a warning that says it "evaluates to null", then every write to the object is refused #29904

Description

What happens

Activity

objectstack-fleet commented on Nov 3, 2026

Triage: first-touch audit · p1 · domain:server · pm:queue · 2026-11-03T11:02Z

objectstack-fleet commented on Nov 3, 2026

objectstack-fleet commented on Nov 3, 2026

objectstack-fleet commented on Nov 3, 2026

objectstack-fleet commented on Nov 3, 2026

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions

Triage: first-touch audit · p1 · `domain:server` · `pm:queue` · 2026-11-03T11:02Z