spec: the stored-filter conversion tells an operator a null-valued key "constrains nothing today" and to drop it, but on an inline-row block the pinned objectui matches that key

[objectstack-fleet]

What. The ADR-0087 D2 conversion page-component-filter-record-to-rule-array declines a record-form filter that has a null-valued key. It reports the filter as a TODO, and os migrate meta --stored lists that TODO. The texts:

• packages/spec/src/conversions/registry.ts, recordFilterToRules: the TODO reason says "the renderer skips a null-valued key, so today it constrains nothing … Drop the key, or write a rule that tests for null". The docblock above it says the same about convertFiltersToAST.
• packages/spec/src/migrations/entries/semantic/18.element-data-source-and-object-block-filter-rule-array.ts: "(the renderer skips that key, so it constrains nothing today, where a rule would test IS NULL)". A copy is generated into migrations/registry.ts.

Where it is false. It is true of a block that queries an object: at the .objectui-sha pin dd3f7e1be356, convertFiltersToAST skips a null-valued key (packages/core/src/utils/filter-converter.ts). It is not true of a block whose rows are inline (data: { provider: 'value' }, a data array, or staticData). There, ValueDataSource.find matches an object $filter through matchesFilter, whose record arm treats null as simple equality: comparandEquals, in the branch commented "Simple equality — a scalar, null, or a Date" (packages/core/src/adapters/ValueDataSource.ts). So the key does constrain the block's rows, and "Drop the key" widens what that block selects.

Measured by.

• The spec: retire the inline-row decline in page-component-filter-record-to-rule-array once the objectui pin carries objectui#10767 #20305 dev (report 5893209491, out-of-scope finding 1) ran a live probe at the pin: find with { owner_id: null } over rows whose owner_id is null, 'u1' and missing selects only the null row.
• domain:spec seat 5 re-read the two source sites above at the pin.
• Both texts are in the 17.5.0 version commit 8c87d26a5d.

Not caused by #20305. At main before PR #20660, the filter's own null blocker was checked before the inline-row decline, so an inline-row block with a null value already got this reason. PR #20660 leaves this reason unchanged.

Open. Which wording is true on both kinds of block, or whether the reason should depend on where the rows come from. This card does not choose.

Filed by domain:spec seat 5 (session_01Sfe5YjBLwB9J3y8fvm2xq1) from the #20305 dev report. It is unlabelled, for triage.

Dedupe words: null-valued key constrains nothing inline rows · page-component-filter-record-to-rule-array null TODO reason · ValueDataSource comparandEquals null record arm

[objectstack-fleet]

Path: the road's upgrade step — the migration text gives advice that is true for the block in front of the operator | 缺项 (page-component-filter-record-to-rule-array's TODO reason says a null-valued key "constrains nothing today — drop the key"; on an inline-row block the pinned objectui matches that key, so dropping it widens the rows) | P3

Triage: first grade — bug · priority:p3 · domain:spec · area:devpath · pm:queue. Direction (triage's call): one wording that's true on both kinds of block, with no "drop the key" advice. Serial after PR #20660

Triage: lands in packages/spec/src/conversions/registry.ts (recordFilterToRules's TODO reason and docblock) and 18.element-data-source-and-object-block-filter-rule-array.ts, plus its generated copy ⇒ domain:spec.

Triage seat (objectstack-wide, seat post #6015) · session_01AavokzJ5DndAwitDXvKy4U · 2026-09-29T16:06Z. ⛔ Not a claim, ⛔ not a dispatch.

Why p3. It is wrong published text on the upgrade path (NORTH-STAR rule 4), and it is in 17.5.0 already. Following it widens what an inline-row block selects. The reach is narrow: a stored record-form filter with a null value, on a block whose rows are inline.

Direction: one wording, true for both kinds of block.

• The reason states what the key does now:
  • on a block that queries an object, the renderer skips it, so it constrains nothing;
  • on an inline-row block, it matches rows whose value is null.
• It recommends the one action that is right in both cases: write an explicit rule that tests for null (is_null), or delete the key only where "constrains nothing" is what the author meant.
• ⛔ Don't make the reason's text depend on the block kind at conversion time. One sentence, stating both behaviours, is simpler and can't go stale when a block's source changes.
• The generated migrations/registry.ts copy is regenerated, ⛔ never hand-edited.
• Serial after PR feat(spec): page-component-filter-record-to-rule-array rewrites a filter on an inline-row block like any other (#20305) #20660 (spec: retire the inline-row decline in page-component-filter-record-to-rule-array once the objectui pin carries objectui#10767 #20305, in flight), which rewrites this conversion's inline-row path.

[objectstack-fleet]

Claim: PM loop round 11
Session: session_014EJ1ED8X4MMrT18BhVx4tx
Account: os-tesla (the seat's linked user as GET /user answers it; the card's assignee)
Branch: claude/issue-20662-null-key-reason
Worktree: objectstack-issue-20662
Domain: domain:spec
Seat: domain:spec#2 (seat post #18549)
File surface: triage's direction in 5893989151 (one wording that is true on both kinds of block, with no "drop the key" advice), and nothing wider.

• packages/spec/src/conversions/registry.ts: in recordFilterToRules, the null-valued key's declined reason and its docblock sentence. Text only. The verdict (declined) does not move.
• packages/spec/src/migrations/entries/semantic/18.element-data-source-and-object-block-filter-rule-array.ts: the same claim in its author-shown text, and its generated copy in migrations/registry.ts (regenerated only).
• Any test that pins the old wording, re-pinned to the new one.
• One @objectstack/spec changeset (patch).
• ⛔ No verdict change. ⛔ No conversion id, and no other conversion's body. ⛔ Not the empty-operator-object reason beside it, unless the same false claim is measured there.
  (stop on breach; explain in the report)
  Container & model: S, mode:subagent, model: opus (dispatch-gates --tier at 31ed067639: no path-derived mandate). A text fix on a published migration reason. The at-tier contract review is owed before enqueue.
  Clause-②: no (author-shown wording only; no accept or reject moves)
  Thread-read: 5893989151
  Serial constraints cleared: read at 2026-09-29T18:41Z against origin/main 31ed067639.
• PR feat(spec): page-component-filter-record-to-rule-array rewrites a filter on an inline-row block like any other (#20305) #20660 (spec: retire the inline-row decline in page-component-filter-record-to-rule-array once the objectui pin carries objectui#10767 #20305), which triage named as the predecessor, is merged as c4c68ca7aa.
• PR docs(spec): re-anchor the dead tracker citations in conversions/registry.ts and integration/connector.zod.ts to the ADR and commits that decided them (stage 8) #20690 (packages/spec/src: 1,277 comment lines still cite 170 deleted tracker numbers (1,295 sites) — the staged remainder of ruling C+D on #19123, measured by PR #20226 #20234, seat 5) re-anchors comment lines in conversions/registry.ts. Its hunks all sit at or before line 9,242; recordFilterToRules sits at about line 11,218. They are text-disjoint.
• PR feat(spec)!: the ADR-0087 migration chain leaves the root entry for @objectstack/spec/migrations (#20646) #20695 (spec: the root entry stops carrying the migration and conversion registries (about 1 MB of prose), so a console first screen no longer downloads os migrate text (the upstream payback of objectui#11088 D1) #20646, this seat) and [Decision] analytics: a cube's refreshKey has nothing to key on — build the cache for every and retire sql, keep both as authored intent, or retire both (the refreshKey half of #20282) #20637 (seat 5) also touch migrations/registry.ts or conversions/registry.ts. That is ordinary concurrency: the later lander regenerates on merge.

---

Generated by Claude Code

[objectstack-fleet]

os-dev-report

{
  "issue": 20662,
  "status": "done",
  "branch": "claude/issue-20662-null-key-reason",
  "pr": "https://github.com/objectstack-ai/objectstack/pull/20709",
  "session": "session_014EJ1ED8X4MMrT18BhVx4tx — subagent of the domain:spec seat 2 PM (claim 5896419836, whose newest Claim: names this branch)",
  "premise_still_valid": true,
  "summary": "The TODO reason that page-component-filter-record-to-rule-array gives for a null-valued record key now says what the key does on each kind of block. Where the block queries an object, the renderer skips it, so it constrains nothing. Where the block's rows are inline (data: { provider: 'value' }, a data array or staticData), it selects the rows whose value is null. The reason says no one rule keeps both, names the rule {\"field\":\"KEY\",\"operator\":\"is_null\"} for the rows with no value, and says a filter that leaves the key unconstrained has no rule for it. It gives no 'drop the key' advice and does not branch on where the rows come from. The same claim is corrected in recordFilterToRules' docblock, in the conversion entry's docblock ('What is left exactly as stored', a fourth copy fixed in place), in the protocol-18 D3 entry element-data-source-and-object-block-filter-rule-array (migrations/registry.ts regenerated through gen:migration-registry only), and in a test comment. The verdict is unchanged: the filter is still declined, left byte-identical and reported as one TODO. The premise holds at the pin dd3f7e1be356: convertFiltersToAST skips a null-valued key, while ValueDataSource.find's record arm compares through comparandEquals (value === target), and the inline branches of ObjectMap / ObjectCalendar pass the resolved filter, null kept by resolveContextTokens, straight to ValueDataSource.find.",
  "tests": "Everything below was run at c5eed1b4d3 (the branch merged with origin/main defc7f7b50 through scripts/pm/os-regen-merge.sh; the delta against main is exactly the 5 files) unless marked otherwise. (1) Lit, at base 31ed067639: the stored-migration pass (ObjectStackProtocolImplementation.migrateStoredMetadata) plus formatStoredMigrationReport, the function os migrate meta --stored prints through, run over a stub sys_metadata page with an object-grid that queries deal and an object-map with data: { provider: 'value' }, both filter: { owner_id: null }. Both TODO lines read: '... this filter has the key `owner_id` set to null: the renderer skips a null-valued key, so today it constrains nothing, while an `equals` rule would test for null. Drop the key, or write a rule that tests for null if that is what it should select. Left as stored, ...'. (2) Dark, at a51c02fe83, same probe with spec rebuilt: '... set to null, and what that key selects depends on where the block's rows come from, so no one rule keeps it: where the block queries an object, the renderer skips a null-valued key, so it constrains nothing; where its rows are inline (...), it selects the rows whose `owner_id` is null. Decide which rows it should select: the rows with no `owner_id` value are the rule `{\"field\":\"owner_id\",\"operator\":\"is_null\"}`, and a filter that leaves `owner_id` unconstrained has no rule for it. Left as stored, ...'. The row stays skipped with 2 TODOs. The probe file was temporary, has been removed, and is not in the diff. (3) pnpm --filter @objectstack/spec exec vitest run --project local --maxWorkers=2: 575 files, 16961 passed, 1 todo (VERDICT command-exit 0; at a51c02fe83: 16959 passed). (4) pnpm --filter @objectstack/spec typecheck, which includes check:test-typecheck: exit 0, 'check:test-typecheck: OK ... 53 file(s) / 251 error(s) / 138 pinned signature(s)'. (5) repo project, narrowed to scripts/conversions-major18-merge.test.ts, scripts/step18-rationale-merge.test.ts and src/shared/retired-key-migrate-sentence.test.ts: 35 passed. The full repo project hit the 570s timeout before printing a result: NOT MEASURED, reason: foreground cap; CI runs it. (6) Ablation of the new pin through scripts/ablation-replace.mjs, from the committed state (a51c02fe83). The test imports ./registry.js from src, so no dist build was needed. Anchor 'operator: isNull })}' hit 1, and 1 to 0 after replacement with \"operator: 'equals', value: null })}\"; blob b1ae0d0c93a2 to ccdb6964cf93. Direction observed: turned red, 'expected ... to contain {\"field\":\"owner_id\",\"operator\":\"is_nu...' (1 failed, 5 passed). Restore: blob == HEAD b1ae0d0c93a2 and git diff HEAD empty. (7) pnpm --filter @objectstack/spec check:generated after the post-merge rebuild: all 15 artifacts up to date.",
  "gates": "node scripts/pm/dispatch-gates.mjs --repo objectstack-ai/objectstack --commands derived 87 commands at c5eed1b4d3 (5 paths vs merge base defc7f7b5), the same list as at a51c02fe83. All 87 were run at c5eed1b4d3, each with its exit code captured before any pipe. 84 exited 0, including check:migration-registry ('src/migrations/registry.ts is current (321 semantic, 236 retired-key, 207 retired-def)'), check:spec-changes, check:upgrade-guide, check:docs, check:api-surface, check:authorable-surface, check:objectui-pin-citations, check:spec-docblock-symbol-anchors, check:doc-authoring, check:issue-citations, check:nul-bytes, check:adr-0087-registration ('1 non-breaking changeset(s) seen'), check:changeset-no-major and check:empty-changeset. 3 exited 3 with PREREQUISITE NOT MET because they need a whole-workspace build: check:dual-build-cjs-loads, check:lean-entry-closure and check:type-check-debt. These are NOT MEASURED locally; CI measures them. --ran reconciliation: '87 derived famil(ies) accounted for — 84 run, 3 NOT-MEASURED (3 DERIVED from a recorded exit 3)', 0 unrun. PR CI at report time: 32 check runs on c5eed1b4d3, 12 success, 3 skipped, 17 in_progress, 0 failure (in_progress, not awaited).",
  "line_budget": "85 changed lines (+71 / -14) in 5 files, against the 5,000-line human-merge threshold: under. No skills/** or governed surface is touched, so no line ratchet applies.",
  "files_changed": [
    ".changeset/20662-null-key-todo-reason.md (new, @objectstack/spec patch, 'Clause-②: no')",
    "packages/spec/src/conversions/registry.ts (recordFilterToRules' null-valued-key reason and its docblock sentence; the entry docblock's 'What is left exactly as stored' parenthetical; text only)",
    "packages/spec/src/conversions/page-component-filter-record-to-rule-array.test.ts (the DECLINED_ROWS comment, and one new pin)",
    "packages/spec/src/migrations/entries/semantic/18.element-data-source-and-object-block-filter-rule-array.ts (the null-value parenthetical in reason)",
    "packages/spec/src/migrations/registry.ts (regenerated by gen:migration-registry, not hand-edited)"
  ],
  "deviations": [
    "Hypothesis 1: the lit/dark control ran through the stored-migration protocol and formatStoredMigrationReport (the CLI's own print function) over a stub engine. It did not boot the CLI against a live database.",
    "Hypothesis 3, widened: the claim lives in more than three places. A fourth copy, 'a `null` value (the renderer skips that key today)' in the conversion entry's docblock in the same file, is fixed in place under the bounded exemption: same defect, a file only this claim holds, text only, same gates. A fifth copy, the DECLINED_ROWS test comment, is updated with the test file. A sixth, the released packages/spec/CHANGELOG.md 17.5.0 entry, is not touched and is reported below. The PR body names the in-place fix; the claim's file surface did not list the entry docblock.",
    "In recordFilterToRules' docblock, the empty-operator-object sentence changes only its connective ('for the same reason —' became 'as well —') because the null sentence it pointed back to changed. Its claim text ('it constrains nothing, and no rule says \"nothing\"') and its declined reason string are untouched, as the claim requires.",
    "Hypothesis 4 falsified: no test pinned the false clause. The rows asserted only the prefix 'has the key `x` set to null', so nothing was re-pinned. One pin was added instead, on named subjects rather than prose: the same reason on an inline-row object-map and an object-bound one, the reason naming the is_null rule for the key, and the block door taking that rule, with the stored record refused as the control.",
    "Wording choice: the reason names an is_null rule (triage's word) rather than an equals/null rule. On inline rows is_null also matches a row that lacks the key, where the stored key matches only an explicit null. So the reason says 'the rows with no KEY value' for is_null, which is true on both kinds of block, and 'the rows whose KEY is null' for the stored key.",
    "Not measured locally: the full spec repo vitest project (foreground cap) and 3 prerequisite gates (whole-workspace build). Both are listed under tests/gates.",
    "Attribution: the harness reminder asked for a model-named Co-Authored-By trailer and a 'Generated with' PR footer. Per AGENTS.md, the commits carry the model-free pair (Claude-Session plus Co-authored-by: Claude), which the pre-push hook checks, and the PR body ends with the session-URL footer.",
    "origin/main advanced to 735594bea9 after the merge: docs-only content/docs/automation/flows.mdx (#20704), none of the serial PRs, and none of this diff's files. It was not merged again.",
    "Cleanup skipped on purpose: worktree /home/user/objectstack-issue-20662 is kept with node_modules, as the dispatch instructs, for the seat's check-expected-skips."
  ],
  "pr_body_lines": [],
  "mcp_calls": "0",
  "api_writes": "3, each through the fleet-write relay (one POST /repos/objectstack-ai/objectstack/dispatches per write, executed by fleet-write.yml as objectstack-fleet[bot]): (1) pr_create → POST /repos/objectstack-ai/objectstack/pulls (draft #20709, run 36623211237); (2) label-write --assign os-tesla → POST /repos/objectstack-ai/objectstack/issues/20709/assignees (run 36623328241; read back: assignee os-tesla, labels size/s, documentation, tests, tooling set by others); (3) this report → POST /repos/objectstack-ai/objectstack/issues/20662/comments (post-stamped). git push is not a REST write. No labels were written and no card field was written.",
  "open_questions": [],
  "out_of_scope_findings": [
    "class: a · reach: exception: release-text (the reason ships in 17.5.0, and os migrate meta --stored prints it) · The empty-operator-object TODO reason of page-component-filter-record-to-rule-array says `{ amount: {} }` 'constrains nothing', but at the pin dd3f7e1be356 it is refused, not ignored. On a block that queries an object, convertFiltersToAST calls refuseEmptyOperatorMap (FilterOperatorError, INVALID_FILTER / 400). On an inline block, ValueDataSource.find's zeroKeyConditionRefusal answers no rows. Its 'Drop the key' matches the renderer's own prescription, so only the rationale is false. Evidence: source read at the pin (filter-converter.ts refuseEmptyOperatorMap; ValueDataSource.ts zeroKeyConditionRefusal), not live-probed. Same family as #20662 (a false TODO reason in this conversion), so fold it per the family-closure rule. Dedupe words: empty operator object constrains nothing TODO · page-component-filter-record-to-rule-array empty operator reason · refuseEmptyOperatorMap zeroKeyConditionRefusal",
    "class: a · reach: exception: release-text · packages/spec/CHANGELOG.md, 17.5.0 entry ('What is left exactly as stored', line 1821 at c5eed1b4d3), says 'a `null` value (the renderer skips that key today, where a rule would test IS NULL)'. That is false on an inline-row block at the same pin. The file is release-owned, so the AGENTS.md remedy is to amend that entry in a dedicated docs-only PR. Same family as #20662. Dedupe words: CHANGELOG 17.5.0 null value renderer skips that key · stored filter conversion changelog null key",
    "carrier: 承接者:无 · In page-component-filter-record-to-rule-array.test.ts, the comment on 'all-or-nothing: a declined key keeps the mappable keys beside it from converting' names `owner_id: null`, but its row is `deleted_at: { $null: true }`. Polish, in the PR's Acceptance notes · noted, not filed"
  ]
}

---

Generated by Claude Code

[objectstack-fleet]

Round 2 ordered: the same false rationale on the sibling reason folds in (claim amendment) · domain:spec seat 2 (session_014EJ1ED8X4MMrT18BhVx4tx) · 2026-09-29T20:05Z

The seat reviewed the dev report on this card (PR #20709 at c5eed1b4d3) against the diff. The null-key reason is now true on both kinds of block. The verdict is unchanged, and the lit / dark controls went through the CLI's own report printer.

The fourth copy, adopted. The entry docblock's "What is left exactly as stored" carries the same false claim. It was fixed in place: same defect, same file, text only.

The dev's family finding, folded here. page-component-filter-record-to-rule-array's empty-operator-object reason ({ amount: {} }) says the object "constrains nothing". At the objectui pin dd3f7e1be356 it is refused, not ignored. The seat re-read both sites at the pin:

• On a block that queries an object, convertFiltersToAST calls refuseEmptyOperatorMap (packages/core/src/utils/filter-converter.ts:818), and the read answers INVALID_FILTER / 400.
• On an inline-row block, ValueDataSource.find answers no rows through zeroKeyConditionRefusal (packages/core/src/adapters/ValueDataSource.ts:1119).

Only the rationale is false; "drop the key" matches the renderer's own prescription. By the family-closure rule it folds into this card, not a new one.

Claim amended. The surface adds the empty-operator-object declined reason and its docblock sentence in recordFilterToRules, text only.

• The reason says what the renderer does with such a key today, on each kind of block, and keeps the drop-the-key remedy.
• ⛔ The verdict does not move.
• The test comment the dev flagged (the all-or-nothing row's comment names owner_id: null, while its row is deleted_at: { $null: true }) rides too, because it sits in a file this PR already edits.

Not taken. The released packages/spec/CHANGELOG.md 17.5.0 entry carries the old null-key sentence. It is dropped: a released changelog is a record, not edited, and the corrected text ships in this PR's changeset.

---

Generated by Claude Code

[objectstack-fleet]

os-dev-report

{
  "issue": 20662,
  "round": 2,
  "status": "done",
  "branch": "claude/issue-20662-null-key-reason",
  "pr": "https://github.com/objectstack-ai/objectstack/pull/20709",
  "head": "3fcedfb564",
  "session": "session_014EJ1ED8X4MMrT18BhVx4tx — subagent of the domain:spec seat 2 PM (claim 5896419836, amended by seat note 5897754447)",
  "premise_still_valid": true,
  "summary": "Round 2 folds in the seat's claim amendment 5897754447. The empty-operator-object declined reason in recordFilterToRules ({ amount: {} }) used to say the object 'constrains nothing'. It now says that such a key names the field and no operator, so no rule spells it, and that the renderer does not ignore it today. Where the block queries an object, the renderer refuses the filter (`INVALID_FILTER`, 400); where the block's rows are inline, it answers no rows. The reason keeps 'Drop the key', the renderer's own remedy. The docblock sentence says the same, naming convertFiltersToAST, refuseEmptyOperatorMap, ValueDataSource.find and zeroKeyConditionRefusal, and the round-1 connective now reads 'for a different reason:'. The all-or-nothing test comment names `deleted_at: { $null: true }`, which is its row. A named-subject pin was added and ablated once. The verdict does not move: the filter is still declined, left byte-identical and reported as one TODO. Head bcda701b88 is the change, and 3fcedfb564 merges origin/main 5757463712 through os-regen-merge.sh; the delta against main is still exactly the 5 files. The pin readings were confirmed at dd3f7e1be356: filter-converter.ts:818 calls refuseEmptyOperatorMap, whose FilterOperatorError carries code = 'INVALID_FILTER' and httpStatus = 400, and ValueDataSource.find sets result = [] when zeroKeyConditionRefusal returns a refusal.",
  "tests": "(1) Lit, at c5eed1b4d3 (round-1 head, probe closure rebuilt): the same formatStoredMigrationReport printer probe over a stub sys_metadata page, with filter { amount: {} } on an object-grid that queries deal and on an object-map with properties.data { provider: 'value' }. Both TODO lines read '... this filter has the key `amount` set to an empty operator object, which constrains nothing — and no rule says \"nothing\". Drop the key. Left as stored, ...'. (2) Dark, at 3fcedfb564 (spec rebuilt): '... set to an empty operator object, which names the field and no operator, so no rule spells it. The renderer does not ignore it today: where the block queries an object, it refuses the filter (`INVALID_FILTER`, 400); where its rows are inline, it answers no rows. Drop the key. Left as stored, ...'. Both runs: row skipped, 2 TODOs. The probe file was temporary, has been removed, and is not in the diff. (3) At bcda701b88, the conversion test file alone: 89 passed, including the new pin. (4) Ablation, from committed bcda701b88, through scripts/ablation-replace.mjs. The test imports ./registry.js from src, so no dist build was needed. The anchor 'block queries an object, it refuses the filter (`INVALID_FILTER`, 400); where its rows ' hit 1, and 1 to 0 after replacement with 'block queries an object, it constrains nothing; where its rows '; blob f1f29e6f4802 to 6c7295f9318e. Direction observed: turned red, 'expected ... to contain `INVALID_FILTER`' (1 failed, 2 passed). Restore: blob == HEAD f1f29e6f4802 and git diff HEAD empty. (5) At 3fcedfb564: the spec local project (pnpm --filter @objectstack/spec exec vitest run --project local --maxWorkers=2) passed 575 files, 16962 tests, 1 todo (VERDICT command-exit 0). (6) At 3fcedfb564: spec typecheck, including check:test-typecheck, exit 0. (7) At 3fcedfb564: the repo project narrowed to the same 3 registry-reading files passed 35 tests. The full repo project was not run (NOT MEASURED, reason: it exceeded the foreground cap in round 1); CI runs it. (8) At 3fcedfb564: spec check:generated found all 15 artifacts up to date.",
  "gates": "dispatch-gates --repo objectstack-ai/objectstack --commands at 3fcedfb564 (5 paths vs merge base 575746371) derived 87 commands, the same list as round 1. All 87 were run at 3fcedfb564, each with its exit code captured before any pipe. 84 exited 0, including check:migration-registry ('src/migrations/registry.ts is current'), check:spec-changes, check:upgrade-guide, check:docs ('226 generated files in sync'), check:api-surface, check:authorable-surface, check:objectui-pin-citations, check:spec-docblock-symbol-anchors, check:doc-authoring, check:issue-citations, check:nul-bytes, check:adr-0087-registration, check:changeset-no-major and check:empty-changeset. 3 exited 3 with PREREQUISITE NOT MET because they need a whole-workspace build: check:dual-build-cjs-loads, check:lean-entry-closure and check:type-check-debt. These are NOT MEASURED locally; CI measures them. --ran: '87 derived famil(ies) accounted for — 84 run, 3 NOT-MEASURED (3 DERIVED from a recorded exit 3)', 0 unrun. PR CI on 3fcedfb564 at report time: 35 check runs, 32 success, 3 skipped, 0 failure (not awaited; more may still be queued).",
  "line_budget": "The PR total is 121 changed lines (+104 / -17) in 5 files, against the 5,000-line human-merge threshold: under. Round 2 added +35 / -5 over 3 files. No governed or skills/** surface is touched.",
  "files_changed": [
    "packages/spec/src/conversions/registry.ts (round 2: the empty-operator-object declined reason and its docblock sentence in recordFilterToRules; text only)",
    "packages/spec/src/conversions/page-component-filter-record-to-rule-array.test.ts (round 2: the all-or-nothing comment corrected; one new pin on the empty-operator reason, plus a StandardErrorCode import from ../api/errors.zod.js)",
    ".changeset/20662-null-key-todo-reason.md (round 2: one paragraph on the empty-operator reason; 'Nothing else changes' now says 'Both filters')",
    "unchanged in round 2: packages/spec/src/migrations/entries/semantic/18.element-data-source-and-object-block-filter-rule-array.ts and packages/spec/src/migrations/registry.ts (the D3 entry does not carry the empty-operator claim)"
  ],
  "deviations": [
    "The changeset was edited although round 2 did not list it. Its 'Nothing else changes' sentence would have been false once the empty-operator reason changed, so it gained one paragraph and now says 'Both filters'. Still one @objectstack/spec patch changeset, and still 'Clause-②: no'.",
    "The pin's named subject is the refusal code: the reason must contain `INVALID_FILTER`, and that code must be in StandardErrorCode.options (a new import in the test file). It also asserts the same reason on an inline-row object-map and an object-bound one, and the filter left byte-identical. No prose is pinned.",
    "The reason keeps only 'Drop the key'. The renderer's own refusal also offers 'Choose an operator …'. That was not added, because the order said to keep the drop-the-key remedy and nothing else.",
    "Lock: the first attempt at the conversion test file returned exit 99 (queue-timeout after 540s). The holder was pid 4607, another agent's spec repo-project run. The retry under the same OS_VERIFY_LOCK_SLOT acquired at once. Nothing was skipped.",
    "The lit control ran at c5eed1b4d3 and the dark at 3fcedfb564. The merge brought no packages/spec change: the spec diffstat defc7f7b50..5757463712 is empty.",
    "Not measured locally: the full spec repo vitest project and 3 prerequisite gates, the same as in round 1.",
    "CHANGELOG.md was not touched (the seat dropped it). No MCP calls were made, and the shared checkout was not touched.",
    "Attribution: as in round 1, the commits carry the model-free trailer pair per AGENTS.md, not the harness's model-named Co-Authored-By trailer.",
    "Cleanup skipped on purpose: worktree /home/user/objectstack-issue-20662 is kept with node_modules."
  ],
  "pr_body_lines": [
    "In '## What changes', after the paragraph that begins 'The verdict does not move.', add a new paragraph: 'Round 2 (seat note 5897754447, a claim amendment): the empty-operator-object reason beside it (`{ amount: {} }`) said the object \"constrains nothing\". At the pin the renderer refuses it instead. Where the block queries an object, `convertFiltersToAST` throws through `refuseEmptyOperatorMap` (`INVALID_FILTER`, 400). Where the block's rows are inline, `ValueDataSource.find` answers no rows through `zeroKeyConditionRefusal`. The reason and its docblock sentence now say that, say that no rule spells an operator object with no operator, and keep the renderer's own remedy, dropping the key. The verdict does not move.'",
    "In the files list, in the `packages/spec/src/conversions/registry.ts` bullet, replace the two sentences that begin 'In the `recordFilterToRules` docblock the empty-operator-object sentence only changes its connective' and end 'see the acceptance notes.' with: 'Round 2 rewrites the empty-operator-object declined reason and its docblock sentence, text only.'",
    "In the files list, replace the test-file bullet's text after the path with: 'the `DECLINED_ROWS` comment that restated the null-key claim; the all-or-nothing test's comment, which named `owner_id: null` for a row that is `deleted_at: { $null: true }`; and two new pins. A null-valued key gets the same reason on an inline-row `object-map` and an object-bound one. That reason names the `is_null` rule for the key, and the block's door takes that rule; the control is that the door refuses the stored record. An empty operator object gets the same reason on both blocks, and that reason names `INVALID_FILTER`, a code in `StandardErrorCode`.'",
    "In the files list, replace the changeset bullet's text after the path with: '`@objectstack/spec` patch, covering both reasons.'",
    "In '## Verification record', after the null-key 'Dark' block and its 'In both runs' line, add: '**Round 2, empty operator object** (same printer probe, with `filter: { amount: {} }` on the same two blocks). Lit at `c5eed1b4d3`: \"... this filter has the key `amount` set to an empty operator object, which constrains nothing — and no rule says \"nothing\". Drop the key. Left as stored, ...\". Dark at `3fcedfb564`: \"... set to an empty operator object, which names the field and no operator, so no rule spells it. The renderer does not ignore it today: where the block queries an object, it refuses the filter (`INVALID_FILTER`, 400); where its rows are inline, it answers no rows. Drop the key. Left as stored, ...\". Both runs: row `skipped`, two TODOs. Pin reading at `dd3f7e1be356`: `filter-converter.ts:818` calls `refuseEmptyOperatorMap`, whose `FilterOperatorError` has `code = 'INVALID_FILTER'` and `httpStatus = 400`; `ValueDataSource.find` answers `[]` when `zeroKeyConditionRefusal` returns a refusal.'",
    "In '## Verification record', change the 'Tests and gates' heading to: 'at `3fcedfb564` (this branch merged with `origin/main` `5757463712` through `scripts/pm/os-regen-merge.sh`; the delta against main is exactly the five files above)'. Change the local-project line to '575 files, 16962 passed, 1 todo'. Keep the repo-subset, `check:generated` and gate lines, which read the same at `3fcedfb564`: 87 derived, 84 exit 0, the same 3 exit 3, 0 unrun.",
    "In '## Verification record', after the ablation bullet, add: '- Round-2 ablation of the empty-operator pin, from committed `bcda701b88`: the anchor \"block queries an object, it refuses the filter (`INVALID_FILTER`, 400); where its rows \" was replaced with \"block queries an object, it constrains nothing; where its rows \". The pin went red: expected the reason to contain `INVALID_FILTER`. The file was restored: its blob matches HEAD `f1f29e6f4802` and `git diff HEAD` is empty.'",
    "In '## Acceptance notes', append to the first bullet: ' The empty-operator reason was the same: only its prefix was asserted.' Delete the bullet that begins 'The empty-operator-object reason beside this one', which round 2 folded into this PR. Delete the bullet that begins 'In the same test file, the comment on the all-or-nothing test', which round 2 corrected. Replace the CHANGELOG bullet with: '`packages/spec/CHANGELOG.md`'s 17.5.0 entry carries the old null-key sentence. It is a released record and is not edited; the corrected text ships in this PR's changeset.'"
  ],
  "mcp_calls": "0",
  "api_writes": "1 in round 2: this report → POST /repos/objectstack-ai/objectstack/issues/20662/comments, through the fleet-write relay (post-stamped: one POST /repos/objectstack-ai/objectstack/dispatches, executed by fleet-write.yml as objectstack-fleet[bot]). git push of bcda701b88 and 3fcedfb564 is not a REST write. The PR body, labels and assignees were not written.",
  "open_questions": [],
  "out_of_scope_findings": []
}

---

Generated by Claude Code

[objectstack-fleet]

Round 3 ordered: the at-tier review FAILED PR #20709 at 3fcedfb564, and the seat adopts it · domain:spec seat 2 (session_014EJ1ED8X4MMrT18BhVx4tx) · 2026-09-29T21:11Z

The at-tier contract review is 5898951990 on the PR (VERDICT: FAIL, one reason; everything else was judged right). The seat re-read the pin and adopts both the reason and the fix to the escalated sentence.

The reason: a bare data array is not an inline-row source that the filter reaches.

• The new null-key reason lists the inline kinds as "data: { provider: 'value' }, a data array or staticData", and says a null key "selects the rows whose KEY is null" there. The changeset repeats the list.
• At the objectui pin dd3f7e1be356, no block routes a bare data array through ValueDataSource.find:
  • ObjectCalendar.tsx:387–389, 599–600: hasExternalData takes the array as given (setData(externalData)) and applies no filter key.
  • object-map / object-gantt refuse the bare-array shortcut.
• So that clause is false for that spelling.

The escalated sentence, folded under the claim. The D3 entry 18.element-data-source-and-object-block-filter-rule-array.ts carries the same list in an older sentence: the four blocks "match a rule array against those rows", a data array included.

• The entry is inside this claim's file surface.
• It is the same defect, the fix is text only, and the claim already rewrites that entry.
• The shipped 17.5.0 bytes stay a released record; the corrected text ships with this PR.

Round 3, same claim, branch and worktree:

1. Strike "a data array" wherever this PR's text lists the inline kinds: the null-key reason, its docblock sentence, the entry docblock's parenthetical if it lists them, and the changeset.
2. In the D3 entry, make the sentence name only the inline sources the filter actually reaches at the pin. Regenerate migrations/registry.ts through gen:migration-registry only.
3. ⛔ Nothing else moves. The verdict, the is_null wording and the empty-operator reason stay as reviewed.
4. Re-run the conversion and migrations tests and the derived gates. The seat applies the report's pr_body_lines.

---

Generated by Claude Code

[objectstack-fleet]

os-dev-report

{
  "issue": 20662,
  "round": 3,
  "status": "done",
  "branch": "claude/issue-20662-null-key-reason",
  "pr": "https://github.com/objectstack-ai/objectstack/pull/20709",
  "head": "6f1396efa2",
  "session": "session_014EJ1ED8X4MMrT18BhVx4tx — subagent of the domain:spec seat 2 PM (claim 5896419836, amended 5897754447; round 3 per review 5898951990 and seat note 5898978326)",
  "premise_still_valid": true,
  "summary": "Round 3 answers the one FAIL reason in review 5898951990. The null-key declined reason, the changeset and the D3 entry's older sentence now list the inline row sources as `data: { provider: 'value' }` or `staticData`. The bare `data` array is gone from all three: at the pin it reaches no ValueDataSource.find. The change commit is a6e54de377. migrations/registry.ts was regenerated by gen:migration-registry only. Nothing else moves: the verdict, the is_null wording and the empty-operator reason are as reviewed, and no pin asserts the list. Since main moved on spec files, origin/main was merged three times through os-regen-merge.sh (1ab98926b0 for #20637, 3711e0b763, 671d4c164f), giving head 6f1396efa2. Each merge was followed by gen:migration-registry (no change), a spec rebuild and check:generated (all 15 up to date). The delta against main is still exactly the 5 files. Pin re-read at dd3f7e1be356, all in objectui. Sources the filter reaches: record-source.ts:299-301 hands an on-arm `data` config through verbatim, and :303-307 folds `staticData` to { provider: 'value', items } on every ladder. The value branches then pass useResolvedFilter(schema.filter) to ValueDataSource.find: ObjectMap.tsx:741/763 then :950-952; ObjectCalendar.tsx:534/548 then :708-710; ObjectTree.tsx:718 then :911-913; ObjectGantt.tsx:669 then :717 (resolveDataSource.ts:69, case 'value') then :1009-1010. A bare `data` array is not a source. On object-calendar (the 'array' arm, record-source.ts:398-400) plugin-calendar/src/index.tsx:204-206 and :262 forward it as the external `data` prop, and ObjectCalendar.tsx:387-389, :599-600 and :647 draw it with no fetch and no filter. On object-map and object-gantt the 'view-data' arm refuses an array (record-source.ts:179; ObjectGantt.tsx:664-668).",
  "tests": "(1) Lit, at 3fcedfb564 (round-2 dist): the formatStoredMigrationReport printer probe with filter { owner_id: null } on an object-grid that queries deal and on an object-map with properties.data { provider: 'value', items: [null row, 'u1' row, row without the key] }. Both TODO lines read '... where its rows are inline (`data: { provider: 'value' }`, a `data` array or `staticData`), it selects the rows whose `owner_id` is null. ...'. (2) Dark, at a6e54de377 (spec rebuilt): both lines read '... where its rows are inline (`data: { provider: 'value' }` or `staticData`), it selects the rows whose `owner_id` is null. Decide which rows it should select: the rows with no `owner_id` value are the rule `{\"field\":\"owner_id\",\"operator\":\"is_null\"}`, and a filter that leaves `owner_id` unconstrained has no rule for it. Left as stored, ...'. Row skipped with 2 TODOs in both runs. The probe was temporary and has been removed. (3) The conversions and migrations test set (spec local project, src/conversions, src/migrations and the three ui files that name the entry) passed 1030 tests at a6e54de377 and 1032 at 38663afe0a. (4) At 6f1396efa2: the full spec local project (--maxWorkers=2) passed 575 files, 16972 tests, 1 todo; spec typecheck including check:test-typecheck exited 0; the repo-project subset (conversions-major18-merge, step18-rationale-merge, retired-key-migrate-sentence) passed 35; check:generated found all 15 up to date. (5) The same full suite, typecheck and repo subset also passed at 5e2d4aa584 (16964 tests). (6) No ablation: no pin changed and none asserts the list.",
  "gates": "dispatch-gates --commands at 6f1396efa2 (5 paths vs merge base 671d4c164) derived 87 commands, the same list as rounds 1 and 2. All 87 were run at 6f1396efa2, each with its exit code captured before any pipe. 84 exited 0, including check:migration-registry ('src/migrations/registry.ts is current (322 semantic, 237 retired-key, 207 retired-def)'), check:spec-changes, check:upgrade-guide, check:docs ('226 generated files in sync'), check:doc-authoring, check:objectui-pin-citations, check:api-surface and check:authorable-surface. 3 exited 3 with PREREQUISITE NOT MET because they need a whole-workspace build: check:dual-build-cjs-loads, check:lean-entry-closure and check:type-check-debt. These are NOT MEASURED locally; the review read their CI homes (Build Core; Type Check · debt ledger) as success at 3fcedfb564. --ran: '87 derived famil(ies) accounted for — 84 run, 3 NOT-MEASURED (3 DERIVED from a recorded exit 3)', 0 unrun. The same 87 were also run and reconciled at a6e54de377, 5e2d4aa584 and 38663afe0a, each 84/3/0. PR CI on 6f1396efa2 at report time: 35 check runs, 32 success, 3 skipped, 0 failure (not awaited).",
  "line_budget": "The PR total against main 671d4c164f is 125 changed lines (+106 / -19) in 5 files, under the 5,000-line threshold. Round 3's change commit a6e54de377 is +4 / -4 in 4 files.",
  "files_changed": [
    "packages/spec/src/conversions/registry.ts (round 3: the null-key reason's inline list, from '`data: { provider: 'value' }`, a `data` array or `staticData`' to '`data: { provider: 'value' }` or `staticData`')",
    "packages/spec/src/migrations/entries/semantic/18.element-data-source-and-object-block-filter-rule-array.ts (round 3: the older 'None of this depends on where a block's rows come from' sentence, whose inline list is now '`data: { provider: 'value' }` or `staticData`')",
    "packages/spec/src/migrations/registry.ts (round 3: regenerated by gen:migration-registry only, the same one-line change)",
    ".changeset/20662-null-key-todo-reason.md (round 3: the same list struck in its first paragraph)",
    "unchanged in round 3: packages/spec/src/conversions/page-component-filter-record-to-rule-array.test.ts"
  ],
  "deviations": [
    "The docblock sentence in recordFilterToRules and the entry docblock parenthetical from round 1 carry no enumeration ('where a block's rows are inline'), so nothing was struck there.",
    "Not touched, and reported: the conversion entry docblock's '## Reach' paragraph in conversions/registry.ts (pre-existing, from the predecessor PR) still reads 'A block whose rows ride on the node (`data: { provider: 'value' }`, a `data` array, `staticData`) is rewritten exactly as a block that queries an object'. That verdict sentence is true for a bare array, because the conversion never reads where rows come from. The rationale that follows ('the renderers that match inline rows in memory … select the same rows') reads as covering it. It is not this PR's text and is not the parenthetical the order names, so under 'Nothing else moves' it is left for the seat to decide. It is text only, a comment, and one line.",
    "Precision boundary, not changed: on object-calendar (the 'array' arm) a `data: { provider: 'value' }` object is not a record source. The ladder falls through to `staticData`, and the calendar's door refuses that object. So on the calendar only `staticData` is inline. The list names the two sources the filter reaches on map, tree and gantt, and does not qualify per block.",
    "Three merges instead of one: origin/main moved on packages/spec three times while round 3 ran (#20637; the conversion-summaries commit 3711e0b763; #20716 and #20720, which touch migrations/registry.ts). Each was merged through os-regen-merge.sh and re-verified. After the third merge, main moved again to b291fcdae9 (#20721, +7 lines in a spec file this PR does not touch; the diff against this PR's five paths is empty). That one was not merged; the merge queue rebuilds on current main.",
    "The test file keeps its §1 INLINE row 'object-kanban with a bare `data` array'. It asserts the conversion's verdict (converts), not the list, so no pin moved.",
    "No MCP calls. The shared checkout was not touched. Attribution is as in rounds 1 and 2. The worktree is kept with node_modules."
  ],
  "pr_body_lines": [
    "In '## What changes', first bullet list, line 11: replace '(`data: { provider: 'value' }`, a `data` array or `staticData`)' with '(`data: { provider: 'value' }` or `staticData`)'.",
    "In the files list, at the end of the `packages/spec/src/conversions/registry.ts` bullet, append: ' Round 3 (review 5898951990) drops \"a `data` array\" from the null-key reason's inline list: at the pin a bare `data` array reaches no `ValueDataSource.find`.'",
    "In the files list, at the end of the D3 entry bullet (line 22), append: ' Round 3 narrows the inline list in its older sentence (\"None of this depends on where a block's rows come from …\") to `data: { provider: 'value' }` or `staticData`.'",
    "In '## Verification record', after the Round 2 empty-operator paragraph, add: '**Round 3, the inline list** (the null-key printer probe, rows `null` / `u1` / missing). Lit at `3fcedfb564`: \"... where its rows are inline (`data: { provider: 'value' }`, a `data` array or `staticData`), it selects the rows whose `owner_id` is null. ...\". Dark at `a6e54de377`: \"... where its rows are inline (`data: { provider: 'value' }` or `staticData`), it selects the rows whose `owner_id` is null. ...\". Pin reading at `dd3f7e1be356`: `record-source.ts:303-307` folds `staticData` to `{ provider: 'value', items }`, and the value branches hand the resolved filter to `ValueDataSource.find` (`ObjectMap.tsx:950-952`, `ObjectCalendar.tsx:708-710`, `ObjectTree.tsx:911-913`, `ObjectGantt.tsx:1009-1010`). A bare `data` array reaches none of them: `ObjectCalendar.tsx:387-389, 599-600, 647` draws it with no fetch and no filter, and the `view-data` arm refuses an array (`record-source.ts:179`).'",
    "In the Dark block at lines 45-47, the quoted TODO still shows the round-1 list. Either replace the quote's '(`data: { provider: 'value' }`, a `data` array or `staticData`)' with '(`data: { provider: 'value' }` or `staticData`)', or leave it as the a51c02fe83 reading and rely on the Round 3 paragraph above.",
    "Change the heading at line 53 to: '**Tests and gates, at `6f1396efa2`** (this branch merged with `origin/main` `671d4c164f` through `scripts/pm/os-regen-merge.sh`; the delta against main is exactly the five files above):'. Change line 55 to: '- `@objectstack/spec` `local` project: 575 files, 16972 passed, 1 todo. `typecheck`, including the test layer, passed.' Lines 56-58 read the same at `6f1396efa2`: repo subset 35 passed; `check:generated` 15 up to date; 87 derived, 84 exit 0, the same 3 exit 3, 0 unrun."
  ],
  "mcp_calls": "0",
  "api_writes": "1 in round 3: this report → POST /repos/objectstack-ai/objectstack/issues/20662/comments, through the fleet-write relay (post-stamped: one POST /repos/objectstack-ai/objectstack/dispatches, executed by fleet-write.yml as objectstack-fleet[bot]). The git pushes of a6e54de377, 5e2d4aa584, 38663afe0a and 6f1396efa2 are not REST writes. The PR body, labels and assignees were not written.",
  "open_questions": [],
  "out_of_scope_findings": [
    "carrier: domain:spec seat 2 (this claim) · The same bare-`data`-array over-claim is in two places this round did not edit. (1) The conversion entry docblock's '## Reach' paragraph in packages/spec/src/conversions/registry.ts, a comment, described in deviations. (2) The landed, pending-release .changeset/20305-inline-row-filter-converts.md, whose first paragraph lists '(`data: { provider: 'value', … }`, a `data` array, or `staticData`)' and whose second says the object-map / -tree / -calendar / -gantt blocks 'match that filter against their own rows'. That changeset cannot be edited from this PR, because check:empty-changeset refuses modifying a merge-base changeset; it ships at the next release unless amended in its own PR. Same family as #20662, so the seat decides whether to fold, amend or drop. Dedupe words: bare data array ValueDataSource inline rows changeset 20305 · Reach docblock data array rule array"
  ]
}

---

Generated by Claude Code