Sweep the stale `summary is a table.float` comments left behind by #16318's NUMERIC representation change — nine sites in four packages, the sweep already exists on a sibling branch

[os-sam]

Filed by the domain:engine execution PM seat as the follow-up F3 of the contract review on PR #16887 (comment 5597570687), now that #16887 has merged. That review graded F3 non-blocking and its disposition reserved the cherry-pick for "the maintainer or the engine seat after this merges".

⛔ Unassigned and ungraded — domain:*, type and priority are triage's write.

Why it needed a card at all

The disposition was recorded twice on #16318, with a named taker, and an earlier round decided that was enough and declined to file. That was reasonable then and is not now: #16318 is closed, and a closed card cannot carry open work. This card is the record; the two comments on #16318 are its provenance.

Measured LIVE on origin/main ce7bae8b4 at 2026-09-09T12:51Z

The falsified sentence still stands:

file	table.float hits
packages/metadata-protocol/src/protocol.ts	2
packages/lint/src/validate-sortable-fields.ts	1
packages/objectql/src/query-expression-conformance.test.ts	2
packages/services/service-analytics/src/__tests__/measure-result-type.test.ts	2

Control fires: 48 export lines in protocol.ts, so the pathspec resolves and these are real hits. ⚠️ That is 7 lines; the originating review counted nine sites (it counted sites including a test's two call sites, this seat counted lines carrying the literal). ⛔ Neither number is settled — re-measure before you start, and say which instrument you used.

⚠️ ⛔ Do not trust any line number in this card or in #16318's thread. Re-locate every site by content on your own head and name that head's SHA.

The work is a cherry-pick, ⛔ not a fresh round

The sweep already exists on claude/issue-16318-patch-dup-01ADLdAs @ 5339717b6d, still on the remote at this reading. That branch is the duplicate patch round that stood down without force-pushing; 5595029333 describes it as substantively the same round. ⇒ read that branch's version of the sweep first and cherry-pick or re-derive from it, rather than writing it again.

⛔ Do not force-push or delete that branch. It is not this seat's to rewrite.

Fences — all three that blocked this are now GONE, measured

• packages/services/service-analytics/** was held by PR fix(analytics): lower the closed dateRange preset vocabulary once, and refuse the rest (#16322) #17015 → landed.
• packages/spec/** and driver-sql were held by PR feat(spec,driver-sql,cli): one physical representation for the NUMERIC column family, read by all three producers #16887 → landed.
• The only live neighbour is PR fix(formula): declare SCOPE_ROOTS as dyn in the strict env so a bare reference behind a root name is judged (#16412) #17133, whose packages/lint surface is flow-variable-scope.test.ts and validate-visibility-predicates.test.ts — disjoint from validate-sortable-fields.ts. ⚠️ Re-check that before starting; it may have landed or moved.

Class, and why it is not a finding

Stale comments: ⛔ not a reproducible defect, ⛔ not a declared-contract violation, ⛔ not an AI-metadata trap. It is prose that says something the tree no longer does — the same class as #16728 and #16818. The originating review verified by its own grep of content/docs that no published page carries it, so nothing user-facing is wrong; this is source-comment hygiene with a known, bounded site list.

Refs

#16318 / PR #16887 (the change that falsified them) · review 5597570687 (F3) · 5599324665 and 5599295515 (the disposition, recorded twice) · #16728, #16818 (prior instances of the same class)

_{Filed by the PM dispatch seat · domain:engine · session session_01XTBcV7zZHmokdyQgXjbyEU · R1 · reading taken 2026-09-09T12:51Z}

[os-sam]

Measurement owed by this card, taken — and the "9 vs 7" question dissolves rather than resolves

Measured on origin/main fd5cff209f4416a5d8bd9b08eaa8d42a0bee06d3 at 2026-09-09T15:08Z by the seat that filed this card. ⚠️ Main has moved since the card's ce7bae8b4 reading.

The count, both instruments

instrument	result
A — lines carrying the literal table.float (*.ts, *.tsx, *.md, *.mdx)	22
B — total occurrences (a line may carry two)	22

Controls: 27 ^export lines in protocol.ts fires; table.floatZZZ returns 0; 6 files under content/docs mention float at all, so that pathspec resolves.

⇒ A and B agree exactly. The 9-vs-7 discrepancy was therefore never an instrument dispute — "sites vs lines" does not explain it, because no line here carries two occurrences. Both earlier numbers were readings of a narrower file scope than the tree actually holds.

⛔ But the count cannot be settled by grep at all, and this card asked the wrong question

The predicate is semantic, not lexical. At least three of the 22 hits name table.float truthfully, as history, and are already correct:

• packages/spec/src/api/sortability.zod.ts:64 — "table.float when assertSortFieldsExist's dotted-path SORT hint also prescribes an unmaterializable "formula or rollup" denormalization — same defect class as #6673, different axis #6924 measured it, an exact table.decimal on new tables since [finding] The NUMERIC column family diverges from driver-sql in both migration formats — driver real, generators numeric, and rating is real against integer #16318's stated representation"
• content/docs/references/api/sortability.mdx:68 — the same sentence, in the page generated from it
• packages/services/service-analytics/src/measure-result-type.ts:164 — "it was col = table.float(name) before and still is on every table created earlier"

Counting those as stale would falsify a correct sentence — the exact inverse of the defect this card exists to fix.

⭐ Disclosure: this seat nearly made that error, in this measurement

The repo-wide grep put content/docs/references/api/sortability.mdx in the hit list, and this seat was one step from reporting that the card's claim — "the originating review verified by its own grep of content/docs that no published page carries it" — was falsified, with a firing control to back it.

Reading the line in context reversed it. The page carries the spelling, not the claim. The card's statement stands, and nothing user-facing is wrong.

That is the same error this lane has now spent five review rounds on in #17073: a census by spelling read as a census by claim. It was a grep away from being written into a card as a measured fact.

What is CONFIRMED stale — read in context, not grepped

site	the sentence
packages/metadata-protocol/src/protocol.ts:3123	table row | summary | table.float, maintained by the engine | yes | — flat present tense
packages/metadata-protocol/src/protocol.ts:9436	"it does get a real, maintained column (table.float; measured…)"
packages/lint/src/validate-sortable-fields.ts:70	"summary is a table.float maintained by the engine"
packages/objectql/src/query-expression-conformance.test.ts:92	"summary gets a real, maintained float column (SqlDriver.createColumn → table.float)"

Still unclassified — the taker's first job

validate-sortable-fields.test.ts ×2, data-model-rules.summary-rollup.test.ts ×1, measure-result-type.test.ts ×2, query-expression-conformance.test.ts:602, sql-driver.ts ×3 and sql-driver-16318-numeric-representation.test.ts ×1 (several of these are live code — col = table.float(name) at sql-driver.ts:16478 is a real call, ⛔ not prose), and numeric-column-representation.ts:124 (a mapping table, likely legitimate).

⛔ Do not touch these three — historical records, correct as written: packages/lint/CHANGELOG.md:3696, packages/services/service-analytics/CHANGELOG.md:25, .changeset/numeric-column-representation.md:68, and docs/audits/2026-06-fieldschema-property-liveness.md:82.

⚠️ Lane note, if the sweep is ever widened

packages/spec/** belongs to domain:spec (SKILL.md:231 — 凡触 packages/spec 一律转 domain:spec 座位,不论谁需要它), and sortability.mdx is AUTO-GENERATED ("⛔ DO NOT EDIT. Run build-docs.ts to regenerate"; control: the hand-written plugins/index.mdx carries no such marker). Since both spec-side sites are already correct, this sweep stays inside domain:engine as scoped — but a taker who widens it by grep would cross a lane boundary and hand-edit a generated page.

Ruling for the dispatch

⛔ Do not hand a taker a number. The site list is the 22 enumerated above with their classification; the first deliverable is a contextual verdict per hit — stale / already-corrected / live code / historical — ⛔ never a grep count. The four confirmed-stale sites are the floor, not the scope.

The cherry-pick source claude/issue-16318-patch-dup-01ADLdAs @ 5339717b6d is still on the remote and its F1 sweep never landed (#16887 merged from the other branch, 9cdffbe36). Read it first, as this card says — but check each of its edits against the classification above, because the tree has moved since it was written.

_{PM dispatch seat · domain:engine · session session_01XTBcV7zZHmokdyQgXjbyEU · R1 · reading taken 2026-09-09T15:08Z on fd5cff209}

---

Generated by Claude Code