fix(objectql)!: engine aggregate judges the sum row too — sum over a refused type answers INVALID_FIELD / 400 on every driver

[objectstack-fleet]

Fixes #20914
Clause-②: no (narrowing)

The follow-up round of this card, after PR #21037 landed as a75311dd2 (Part of, the sum row held). It executes triage's answer 5924445782 (A), whose execution line reads: "'sum' leaves ROWS_HELD_FOR_TRIAGE, and the held pins flip. If the constant is then empty, delete it." With this PR the engine's aggregate door asks every row of AGGREGATE_FIELD_TYPE_COMPATIBILITY, so the card's direction (the whole table) is complete.

What this changes

• packages/objectql/src/aggregate-field-type-door.ts: 'sum' leaves ROWS_HELD_FOR_TRIAGE. The set is then empty, so the constant is deleted with its one read. assertAggregationFieldTypesAccepted now judges all six rows (count refuses no type). ⛔ No second table and no per-pair exemption: the door still carries no accept list of its own. In the module header, the section "The row the census held back" becomes "Every row is judged — sum included" and records the release. It also gains the measured sum table, and "a held row" leaves the not-judged list.
• The refusal words are the door's existing machinery (FUNCTION_WORDS.sum: "sums" / "does not sum"). The route is read off the table's sum row: sum accepts a field of type number, currency, rating, slider, progress, summary, boolean or toggle: aggregate a field of one of those types, or count the rows with count. It sits inside the 500 characters the REST door keeps.
• packages/spec/src/data/aggregate-field-type-compatibility.ts, TSDoc only. Review 5924038187 ① 7 named three published sentences; each now states the rows the engine door judges:
  • the module header: "on every row: count_distinct since [finding] two more JSON-stored columns as a group or distinct key answer 500 on PostgreSQL: groupBy on a multiple: true select, and count_distinct on a json field #20808 …, and sum, avg, min and max since [finding] max / min over a JSON-stored field answers per driver at the engine (memory an object, SQLite a string, PostgreSQL 500): the compatibility table refuses them, but the engine aggregate door enforces only its count_distinct row #20914 (count refuses no type, so its row never refuses there)". The pointer to an objectql source file a spec consumer cannot see is gone.
  • the predicate's TSDoc: "on every row of the table".
  • the "does NOT do" section: "at query time, on every row too".
  • radio joins the multi-capable list in the JSON-stored paragraph (MULTI_CAPABLE_TYPES holds it).
  • A comment-stripped compare (the TypeScript printer with removeComments) of the file at base and head is byte-identical, sha256 22d8df93… on both sides: ⛔ the table and the predicate did not move. The new clause is in the built dist/data/index.d.ts, and "other rows since [finding] max / min over a JSON-stored field answers per driver at the engine (memory an object, SQLite a string, PostgreSQL 500): the compatibility table refuses them, but the engine aggregate door enforces only its count_distinct row #20914" occurs in 0 built .d.ts.
• .changeset/20914-release-sum-row.md:
  • @objectstack/objectql: minor with a BREAKING banner and Clause-②: no (narrowing). It carries the FROM → TO per type class, what to write instead, and who is affected.
  • Exactly one ADR-0087 marker: not-required (already-registered dataset-measure-aggregate-field-type-refused). That registered id's prescription already covers sum / avg over every class the table refuses; the selecting id covers only min / max.
  • It names the earlier entry's radio omission (review ① 3) in its own text, and that entry's "Not judged yet: sum" paragraph as superseded. ⛔ The landed file is not edited.
  • @objectstack/spec: patch for the TSDoc.

FROM → TO, measured on three drivers (H2)

Through engine.aggregate, two rows, on the real InMemoryDriver, on SqlDriver over SQLite, and on a private PostgreSQL 16.14. Before = base 2821e9f15; after = this branch's built dist.

sum over	before: memory / SQLite / PostgreSQL	after, all three
json	0 / 0 / 500 function sum(json) does not exist	400 INVALID_FIELD
text	0 / 0 / 500 function sum(text) does not exist	400 INVALID_FIELD
single-value select	0 / 0 / 500 function sum(character varying) does not exist	400 INVALID_FIELD
formula	0 / 400 INVALID_FIELD (driver: no column) / 400 (driver: no column)	400 INVALID_FIELD, the engine's words
tags	0 / 0 / 500	400 INVALID_FIELD
datetime	0 / 4052 (the years added) / 500	400 INVALID_FIELD
percent (refused by the table: a rate does not add)	30 / 30 / 30	400 INVALID_FIELD
controls: currency, number, boolean; max number; count json	30, 3, 1, 2, 2 on all three	unchanged

The percent row is the one whose old answer agreed across drivers. The table refuses it on the semantic ground its TSDoc names (isIncoherentAggregate), and it is in the changeset's FROM → TO with its route (avg).

The census, re-run before the flip (H1)

Query: every line carrying a 'sum' / "sum" literal, plus the unquoted and backtick spellings. Each hit was resolved by hand to its object and the field's declared type, then asked of the table.

• examples/** at 2821e9f15: 17 lines, 0 refused pairs. They are dataset measures, roll-up summaryOperations, a cube measure, a kpi metric and an ObjectChart aggregate. Every operand is a currency, number or summary field (11 distinct object.field pairs).
• hotcrm (objectstack-ai/hotcrm cloned at its head fb408a73): 33 lines in src. One is a comment (a derived op), which leaves 32 authored sum aggregations over 18 distinct object.field pairs. 31 are over currency or number. 1 refused pair: src/sales/views/forecast.view.ts:28, crm_forecast.expected_amount (Field.formula), the list-column summary: 'sum'. That is the known hit: triage-answered and carried by crm_forecast: the all_forecasts list view authors summary: 'sum' on the formula field expected_amount, a pair objectstack's aggregate compatibility table refuses (objectstack-ai/objectstack#20914) hotcrm#1980, which has not changed it yet. No other hit, so the flip stands. Outside src: 2 lines in test, and 3 in docs and the changelog (prose).
• The query is its own positive control: it finds the known hotcrm hit.

Pins: the held sum pins flipped, ⛔ not deleted

• packages/objectql/src/engine-aggregate-field-type-door.test.ts:
  • The CONTROL's two held cases (sum meta, sum title, which reached the driver while the row was held) move into a new refusal test. It covers sum over json, text, select, formula, tags, percent, datetime and a select with multiple: true, with envelope INVALID_FIELD / 400 / httpStatus 400, field / fields / object / param, the message head and the route, and no read.
  • The CONTROL gains sum over a currency and a boolean.
  • The GUARD that asserted "no verdict for sum × every type" now walks every row read off the table, so a row the door skipped would turn it red. It asserts the key set is the six functions, with a sum floor beside the others.
  • The fail-closed GUARD covers sum too.
• packages/rest/src/data-aggregate-field-type-door.test.ts: a sum refusal cell over json, text, select, tags and formula at POST /api/v1/data/:object/query (SQLite always; PostgreSQL where OS_TEST_POSTGRES_URL is set). The CONTROL gains sum over a currency (60) and a boolean (2). ⚠️ As before, no CI job provisions the PostgreSQL / MySQL cells; the PostgreSQL evidence is the local run below.
• packages/objectql/src/engine-json-stored-group-distinct-door.test.ts, its GUARD clause: sum moves from "passes, held for triage" to the refused set beside avg / min / max over json, in the same envelope at the same position. count still passes.

Ablation (from committed code, 1519ed6bd)

• Mutate. node scripts/ablation-replace.mjs (WRAP, trap-restored) put sum back in the held set: the table-vocabulary continue became … || (fn === 'sum' ? 'ablation-20914-sum-A1' : '')) continue;. Anchor 1 → 0, blob 9a2bc89178af → adcfb213d670. objectql was rebuilt, and ablation-dist-preflight.mjs found the marker in 4 built files.
• Predicted: red on the flipped sum pins only.
• Observed, objectql: 3 failed / 28 passed — the new sum refusal test, the every-row GUARD (sum × text: expected null), and the group-distinct GUARD clause. Every other pin was green.
• Observed, REST: 2 failed / 14 passed / 8 skipped (MySQL) — the sum cell on SQLite and on PostgreSQL. The controls and the min / max / avg / count_distinct cells were green.
• Restore. Blob == HEAD (9a2bc89178af), git diff HEAD 0 bytes, rebuilt. --absent found the marker in 0 of 14 built files and the tree clean. The same pins were green again: objectql 31 / 31, REST 16 passed + 8 skipped.

Tests (at 1519ed6bd, the merge of origin/main b3d7a7086; all with a private PostgreSQL 16.14 where a cell reads it)

• objectql full: 353 files, 6889 passed. pnpm --filter @objectstack/objectql typecheck exit 0; test layer 40 files / 234 errors / 65 pinned, held.
• rest full (OS_TEST_POSTGRES_URL set): 259 files, 5124 passed / 52 skipped. typecheck exit 0; test layer 0 / 0 / 0.
• metadata-protocol full: 196 files passed, 3 skipped; 2930 tests passed, 19 skipped.
• service-analytics full: 154 files, 3498 passed / 10 skipped.
• Every other test file in the workspace that mentions a sum aggregation and reaches the engine: runtime 4 files / 25 tests, plugin-security 2 / 297, platform-objects 2 / 56, mcp 1 / 17, dogfood 2 / 12. All passed.
• Driver conformance before and after: 50 covered, 0 DEBT, 0 exempt; the two readings are identical.
• Lint, narrowed and proven at 1519ed6bd:
  • eslint --no-inline-config --format json over the 6 changed .ts files: 6 results, 0 errors, 0 warnings;
  • ESLint isPathIgnored is false for all 6;
  • calculateConfigForFile shows no parserOptions.project / projectService on any of them. No type-aware lint is on, so this diff cannot move a verdict on an untouched file.
  • Repo-wide pnpm lint is CI's.
• pnpm --filter @objectstack/spec check:generated: all 15 generated artifacts up to date.

Gates

node scripts/pm/dispatch-gates.mjs --commands --repo objectstack-ai/objectstack, run with no paths, derived 88 commands at 1519ed6bd. All 88 were run, each exit code recorded before any pipe, and all exited 0. --ran reconciles: 88 derived, 88 run, 0 NOT-MEASURED, 0 UNRUN. Among them: check:adr-0087-registration (accepted, not-required (already-registered)), check:changeset-no-major, check:empty-changeset, check:doc-authoring, check:nul-bytes, check:engine-double-contract and check:api-surface.

NOT MEASURED locally (CI-owned): the 5 path-scheduled CI jobs, the workspace type-check lanes beyond objectql and rest, the MySQL cells (no server here), and repo-wide pnpm lint.

origin/main moved 13 commits after this merge. A local git merge-tree of the head with it is clean. One of them, 88b484e00, touches engine.ts where metadata collections load (picklists), not at the aggregate call site. The merge queue rebuilds on current main.

Deviations from the claim's file surface, each named

• packages/objectql/src/engine.ts, one comment at the aggregate door's call site. "The sum row is held back by that card's census (see the door's header)" would have been false after this PR. It now says every row is asked, and adds the measured sum answer. No code line moved.
• radio added to two comment lists of the multi-capable types: the door's DECLARATION paragraph, and the spec TSDoc's JSON-stored paragraph. This is the same omission review ① 3 named in the landed changeset. The door's own behaviour already held radio (it asks isMultiValueField).

Acceptance notes

• sum × percent answers with the door's generic reason ("accepts only the pairs every backend answers alike, and it refuses this one"), which is true but not the table's ground for that pair (a rate does not add). Its route lists the accepted types, not avg; the changeset carries the avg route. A rate-specific reason would be a words change on the door, so it was left out of this PR. Carrier: none.
• The hotcrm exhibit (crm_forecast.expected_amount, summary: 'sum' on a formula) stays with crm_forecast: the all_forecasts list view authors summary: 'sum' on the formula field expected_amount, a pair objectstack's aggregate compatibility table refuses (objectstack-ai/objectstack#20914) hotcrm#1980. It is a client-side list footer and does not reach engine.aggregate, so this PR does not change what that view shows.

---

Generated by Claude Code

[github-actions]

📓 Docs Drift Check

2 anchor(s) derived from 2 changed package(s); no hand-written page names any of them. ⚠️ 1 changed file(s) yielded no anchor (packages/spec/src/data/aggregate-field-type-compatibility.ts), so the pages documenting them are NOT COVERED by this run — this is not a clean bill of health for those files.

What this run could not see

• 1 changed file(s) yielded no anchor (packages/spec/src/data/aggregate-field-type-compatibility.ts) — pages documenting those are invisible to this run
• 1 name(s) were too generic to anchor anything (single lowercase words)
• the SDK route bridge reached 54 of 206 client-bound route-ledger rows — the other 152 have no registrar path: tail to select them, so pages documenting THEIR client methods cannot appear above, on this or any run. Of those 152: 0 are remediable by widening that discovery convention (an in-repo file declares the path; the convention did not scan it); 55 are structural — on a ledger where NOT ONE row is declared in-repo, so no discovery change reaches them at any price; 97 are undecided (no in-repo declaration, on a ledger that has other in-repo registrars — absence and an unreadable spelling are not distinguishable here). The rows themselves: node scripts/docs-audit/affected-docs.mjs --bridge-coverage
• a page that states a rule by its inputs shares no identifier with the emitter that implements the rule, so an emitter-only diff cannot list it — not on this run and not on any run. Measured on fix(driver-sql): emit varchar(maxLength) for a text field a declared index keys on #11430: content/docs/protocol/objectql/types.mdx documents the text-family column mapping by the ObjectQL type names it maps FROM (text / textarea / html) while the diff changed createColumn; it went unlisted, and it was the page that diff falsified, in four places. No shared token exists to detect this on, so a rule your change carries has to be re-read by hand in the pages that restate it.
• a key NAME is not a key, so the hand re-read the line above prescribes can land on the wrong schema. The same spelling is authorable on one governed type and a [REMOVED] tombstone on another for each of active, aria, joins, objects, template, tools and version (censused on [finding] tools is a key on BOTH AgentSchema (tombstoned, dead) and SkillSchema (live, cloud-attested), so a name-based search attributes skill examples to the agent key — it produced a false stop-the-line alarm on PR #19059 #19093 over the liveness ledger's governed types, top-level keys); nothing in a search result distinguishes the two, so a grep hit on a LIVE example reads as evidence about the DEAD key. Measured on fix(spec): the agent.tools liveness row says dead — it claimed live on a key the schema tombstoned #19059: content/docs/ai/agents.mdx was reported as contradicting the agent.tools tombstone over its tools: example at :161, which is inside the defineSkill({ block opened at :155 — the page was already correct. Settle ownership by PARSING the value against both schemas, never by the name: that literal PASSES SkillSchema, and as an AgentSchema it FAILS at tools with the tombstone prescription. ⛔ These names are not the whole class — a key retired through a .strict() guidance map leaves no tombstone in the walked shape and none of them here (tool.category, live as AIToolDefinition.category).

Coarse fallback — 138 page(s) merely mention a changed package (the pre-#9192 predicate, kept for the deliberately-wide backstop): node scripts/docs-audit/affected-docs.mjs --json 5e470f8c1c6ac2ef5f05c4d65a7ab0b0ac388c9f → packageMentionDocs.

Which tree this was computed on

This run read content/docs from cf6577efba20539fc698369d03840758afc94dc8 — the merge of head 1519ed6bdd2a958501c84b4c3241c79bf6961fb9 into base 5e470f8c1c6ac2ef5f05c4d65a7ab0b0ac388c9f, which is what actions/checkout gives a pull_request run. Not the PR head.

A worktree cut from an older main holds a different content/docs, so re-deriving there can legitimately return a different list — that is a different tree, not a wrong row. To answer on the same tree:

# while this PR is open — GitHub drops the merge commit once it closes
git fetch origin cf6577efba20539fc698369d03840758afc94dc8 && git checkout cf6577efba20539fc698369d03840758afc94dc8
# afterwards, rebuild it from the two parents, which stay fetchable
git fetch origin 5e470f8c1c6ac2ef5f05c4d65a7ab0b0ac388c9f 1519ed6bdd2a958501c84b4c3241c79bf6961fb9 && git checkout -B drift-repro 5e470f8c1c6ac2ef5f05c4d65a7ab0b0ac388c9f && git merge --no-ff 1519ed6bdd2a958501c84b4c3241c79bf6961fb9

node scripts/docs-audit/affected-docs.mjs --json 5e470f8c1c6ac2ef5f05c4d65a7ab0b0ac388c9f

⚠️ That checkout carried uncommitted changes, so the commit above does not fully identify what was read.

[objectstack-fleet]

Contract review

Served-tier: CONTRACT_REVIEW_TIER
Head-sha: 1519ed6bdd2a958501c84b4c3241c79bf6961fb9
Local-runs: none

PR #21103 (claude/issue-20914-release-sum-row) on card #20914, the follow-up round after PR #21037 landed as a75311dd2. Inputs: the card's body and its ten comments, the PR body and file list, the net diff against the merge-base b3d7a708 (7 files, +183 / −62), and the check-runs on the head. Governing items: triage's answer 5924445782 (A), the claim amendment 5924989981, and the owed items ① 3 (radio) and ① 7 (the .d.ts clause) of the prior review 5924038187 as the card's ACCEPT 5924213998 and landing 5924984282 restate them. Nothing was built, run or re-run; every source reading below is of the head ref.

① Derived judgments

1. The constant is deleted, not emptied — right. 'sum' left ROWS_HELD_FOR_TRIAGE; the set was then empty, and the constant and its one continue in refusedAggregations are gone. At the head ROWS_HELD_FOR_TRIAGE occurs nowhere outside CHANGELOG.md. This is triage's execution line, verbatim.
2. No second table, no per-pair exemption — right. refusedAggregations has no function-specific skip: the verdict is isAggregateCompatibleWithFieldType(fn, type) on the row and rowAcceptsMultiValue(fn) for the declaration half, both read off AGGREGATE_FIELD_TYPE_COMPATIBILITY. The only fn-specific branch left in the door is routeAndReason's count_distinct wording, which decides no verdict. The route is acceptedTypesOf(fn), read off the table, so the door carries no accept list of its own.
3. The accept set sum now refuses — right, the table as ruled. The sum row is ADDITIVE_AGGREGATE_FIELD_TYPES plus the boolean class: number, currency, rating, slider, progress, summary, boolean, toggle. The other 41 of the 49 declared FieldTypes are refused INVALID_FIELD / 400 before any driver — percent included, on the table's own ground (isIncoherentAggregate: a rate does not add), and a multi-capable type declared multiple: true by the declaration half as well as by its type's row. The door applies the table; it does not re-rule it. The route string both suites pin (sum accepts a field of type number, currency, rating, slider, progress, summary, boolean or toggle: aggregate a field of one of those types, or count the rows with count.) is the row in the table's order.
4. The table and the predicate are byte-identical — right. Every changed line in packages/spec/src/data/aggregate-field-type-compatibility.ts is a TSDoc continuation line: checked mechanically over the net diff, zero non-comment added or removed lines in that file. The table literal and isAggregateCompatibleWithFieldType are untouched. The dev's comment-stripped sha256 is their evidence; the diff is mine.
5. Public surface — right, nothing moves. ROWS_HELD_FOR_TRIAGE was a module-private const, never exported; the door is internal and not re-exported. The published text that moves is the spec .d.ts TSDoc: the module header now says the engine door refuses "on every row: count_distinct since [finding] two more JSON-stored columns as a group or distinct key answer 500 on PostgreSQL: groupBy on a multiple: true select, and count_distinct on a json field #20808 …, and sum, avg, min and max since [finding] max / min over a JSON-stored field answers per driver at the engine (memory an object, SQLite a string, PostgreSQL 500): the compatibility table refuses them, but the engine aggregate door enforces only its count_distinct row #20914 (count refuses no type)"; the predicate's TSDoc says "on every row of the table"; the "does NOT do" section says "at query time, on every row too". "the other rows since [finding] max / min over a JSON-stored field answers per driver at the engine (memory an object, SQLite a string, PostgreSQL 500): the compatibility table refuses them, but the engine aggregate door enforces only its count_distinct row #20914" occurs nowhere in the file at the head, and the pointer to an objectql source file a spec consumer cannot see is gone. That is the prior review's ① 7 clause, discharged.
6. The flipped sum pins were flipped, not deleted — right. (a) objectql suite: the CONTROL's sum meta / sum title, which reached the driver while the row was held, move into a new refusal test over json, text, select, formula, tags, percent, datetime and a select with multiple: true — envelope (INVALID_FIELD, 400, httpStatus 400), field / fields / object / param, message head and route, zero reads; the CONTROL gains sum over a currency and a boolean. (b) The GUARD's "no verdict for sum × every FieldType" clause becomes sum in an every-row walk read off the table's keys, with the key set pinned to the six functions and a sum floor above 30 beside max's (41 and 37 refused). (c) The group-distinct GUARD moves sum from the not.toThrow pair into the refused loop beside avg / min / max, same envelope at the same position; count still passes. (d) REST: a new sum refusal cell over json, text, select, tags and formula at POST /api/v1/data/:object/query (SQLite; PostgreSQL env-gated), and the CONTROL gains sum price = 60 and sum flag = 2. No pin was deleted; the REST file held no sum pin before, so there it is an addition.
7. The pinned words match the door's — right. declared and kind compose as the door builds them: select field with multiple: true — a multi-value field for picks, tags field — a multi-value field for labels (multi-value is tested before structured-JSON), json field — a structured-JSON value for meta, bare formula field, percent field, datetime field, text field, select field for the rest; does not sum / sums are FUNCTION_WORDS.sum, already in the door.
8. Surface deviations, both named in the PR body and the report — accepted. engine.ts: two comment lines at the aggregate door's call site, which this diff would otherwise have made false ("The sum row is held back…"); no code line moved, read at the head. radio in two docblocks (the door's DECLARATION paragraph and the spec TSDoc's JSON-stored paragraph): MULTI_CAPABLE_TYPES at the head is select, radio, lookup, user, file, image, so both lists are now the constant's; this is the prior review's ① 3, and the door's behaviour already held radio through isMultiValueField. Neither is a behaviour change.
9. Fail-closed tiers kept — right. A fieldless aggregation, a function outside the table's vocabulary, an undeclared name or path, a registry-less host, and a declared type outside FieldType still pass on as they came; the fail-closed GUARD now covers sum too.
10. sum × percent — right in verdict and route; generic in reason. The refusal's route names the accepted set read off the table; its reason is the door's generic sentence, not the table's "a rate does not add". The changeset carries the avg route for a rate. A rate-specific reason is a words change outside this round's claim (see ③).

② Semver level

• .changeset/20914-release-sum-row.md: @objectstack/objectql: minor with a BREAKING banner, @objectstack/spec: patch. Right. The objectql diff narrows aggregate's accept set on every driver — an accept-set narrowing is BREAKING, shipped as minor under the launch-window convention check-changeset-no-major.mjs enforces. The spec diff ships TSDoc into the published .d.ts, so it publishes something and takes patch, never skip-changeset. No export, type or stored row moves, so nothing asks for more.
• Clause-②: no (narrowing) — in the changeset body and the PR body, and matching the claim amendment. Right: no plus the (narrowing) arm is the valid breaking spelling, and nothing is widened.
• Migration carried: FROM → TO per type class (memory 0; SQLite 0, or the years added for a datetime; PostgreSQL 500, or a driver 400 for a formula; the rates added for a percent on all three → 400 INVALID_FIELD), what to write instead (the eight accepted types; avg for a rate; a numeric field for a formula's value; count for a count in disguise), who is affected (the metadata that lowers onto engine.aggregate included), and what is unchanged. The earlier entry's "Not judged yet: sum" paragraph is declared superseded and the landed changeset is not edited. Right.
• ADR-0087: exactly one marker, not-required (already-registered dataset-measure-aggregate-field-type-refused). Right: the id resolves at the head (packages/spec/src/migrations/entries/semantic/18.dataset-measure-aggregate-field-type-refused.ts), predates this diff, and its own text records that a later change widened its prescription to sum / avg over every field class; its replacement names the routes this changeset gives. The selecting id covers min / max only, which this diff does not touch, so naming the one id is right. The Check Changeset run on the head concluded success.
• The radio correction to the earlier entry is carried in this entry's text, not by editing the landed changeset. Right: a merged changeset is a record.

③ Boundary flags

• open_questions: [] in report 5926229168. Nothing to answer.
• Deviation, engine.ts comment only — answered, accepted (① 8): outside the claim amendment's file surface, comment-only, named in the report and the PR body, and needed for truth after this diff.
• Deviation, radio in two docblocks — answered, accepted (① 8): the prior review's owed ① 3, closed.
• Deviation, fixture gains (price, status, expected in LEDGER and PROBE; sum controls) — answered, accepted: inside the claim's pin files, carrying the flipped pins and their controls.
• Deviation, the head merges origin/main at b3d7a7086 and main moved 13 commits after — answered: the check-runs on this head are the verdicts on this head; the queue rebuilds on current main; the dev's merge-tree reading is theirs, not a gate.
• Deviations, commit trailers and the cleanup of the private PostgreSQL and the scratch clone — not contract matters; noted.
• Out-of-scope finding, sum × percent generic reason (carrier none) — answered, accepted as an Acceptance note: the verdict and the route are the table's; a rate-specific reason is a words change on the door, not owed by triage's one-entry execution line. Not escalated.
• PR flag, no CI job provisions the PostgreSQL / MySQL cells — answered: as in the prior round, the door answers before any driver, so the in-memory pin is the verdict's cell by construction and the SQLite cell runs in CI; the PostgreSQL cell is the dev's local evidence, not a gate.
• The census re-run (H1) is dev evidence, consistent with triage's answer: 0 refused sum pairs in examples/**, and the one hotcrm hit is the client-side footer carried by crm_forecast: the all_forecasts list view authors summary: 'sum' on the formula field expected_amount, a pair objectstack's aggregate compatibility table refuses (objectstack-ai/objectstack#20914) hotcrm#1980, which does not block this card. Not re-run here.
• The PR body opens Fixes #20914 — right: triage's "On A, the card closes with the follow-up". The card this PR closes must claim this branch and Part-of PR must not also close its card concluded success.
• Governed surfaces: none in the file list; Governed Surface Queue Guard concluded success.
• Check-runs on the head, 32, read once at 2026-10-01T06:50Z and not polled again:
  • success (13): Auto Label, Check Changeset, Check Documentation Links, Check PR Size, filter, Flag docs affected by code changes, Governed Surface Queue Guard, No other open PR may claim the same issue, No other open PR may claim the same single-writer path, Part-of PR must not also close its card, Spec property liveness, The card this PR closes must claim this branch, Type Check · source gates.
  • skipped (3): Build Docs, Console Pin Gate, Packed-tarball smoke (opt-in).
  • in_progress — not verdicts (16): Build Core; Dogfood Regression Gate 1/3, 2/3, 3/3; Dogfood Verify CLI; Lint & Repo Gates; Temporal Conformance (live PG + MySQL); Test Core 1/6, 2/6, 3/6, 4/6, 5/6, 6/6; Type Check · consumer gates; Type Check · debt ledger; Type Check · workspace. Each answers its own family when it concludes; this record states no conclusion for any of them, and the landing waits on every one of them being green. No check had failed at the read.
  • The Docs Drift Check comment is informational: no hand-written page names a changed anchor, and a grep of content/docs at the head finds no prose on a held sum row.

Implemented-by: claude/issue-20914-release-sum-row
Reviewed-by: session_01Ujdtvqs7ree7WyQmEDwEnG

VERDICT: PASS