finding(spec): `element:record_picker.filter` still declares the MongoDB-style `FilterConditionSchema` — the last record-form `filter` in `ComponentPropsMap` after the ui#6206-B convergence

[claude]

Observed by the domain:spec seat (session session_01GDA48PuRFrHyRfdkBz8m21) while accepting PR #14404 (#12039 Key 2). Filed unassigned for first-touch grading; ⛔ not folded into that PR (its ruling names one entry).

Measured on origin/main (8d3f0939d)

packages/spec/src/ui/component.zod.ts:2114 — inside ElementRecordPickerPropsSchema:

filter: FilterConditionSchema.optional().describe('Filter criteria for available records'),

Every other filter input in the map declares the platform's array-of-rules orthography, z.array(ViewFilterRuleSchema) (record:related_list, its picker, object-grid, object-metric, plugin-list, data-list, and — since PR #14404 — element:number). This entry is the one remaining record-form declaration.

Why it matters

Maintainer ruling ui#6206-B (2026-08-25, verbatim 「同意」): one filter orthography platform-wide. The element:number convergence was sequenced consumer-first because its runtime read path (analytics) refused raw rule arrays until objectui#6828 made the adapter lower them. This entry's read path has not been measured: the same measurement-first discipline applies before any declaration moves.

Shape of the work (for the card that grades this)

1. Measure the record_picker read path at the current objectui pin: which adapter method it calls with filter, and whether that path lowers a rule array (translateFilterArray) or sends it raw to a door that refuses it. Record the commands and readings.
2. If the runtime takes the array form end to end ⇒ converge the declaration to z.array(ViewFilterRuleSchema).optional() with an ADR-0087 D3 entry (BREAKING accept-set change, launch-window minor), the migration sweep for in-repo authors, and rejection tests asserting invalid_type at filter. If it does not ⇒ file the consumer card first (Option-A sequencing, as spec: two ComponentPropsMap entries contradict their ruled authorities — object-grid.data vs ViewDataSchema, element:number.filter vs the platform filter form (rulings ui#6207-A, ui#6206-B) #12039 Key 2 did), never a declaration the runtime refuses.

Refs: #12039 (Key 2, PR #14404) · objectui#6206 (the ruling) · objectui#6828 (the aggregate-path lowering) · objectui#4971 (registry-inputs parity exemptions, downstream).

---

Generated by Claude Code

[huangyiirene]

Triage — graded. Re-measured at origin/main b8562ff262.

FilterConditionSchema now has exactly one use as a component filter input, and it is this one: packages/spec/src/ui/component.zod.ts:2141 (the card measured :2114 at 8d3f0939d; the line has drifted, the entry has not). The array orthography sits at :1089, :1125, :1819 and elsewhere. The census holds — this is the last record-form declaration.

One thing worth handing the implementer: FilterConditionSchema's import at :4 exists for this site. When the entry converges, check whether that import goes with it — a leftover import of the retired orthography in the file that just finished retiring it is the kind of residue the next census trips over.

Why pm:queue and not a decision

The direction is already ruled — objectui#6206-B, maintainer 2026-08-25 verbatim 「同意」: one filter orthography platform-wide. This card does not reopen it; it executes the last entry. What the card adds, correctly, is the sequencing discipline that the element:number convergence had to learn: measure the consumer before moving the declaration, because #12039 Key 2 found a runtime that refused raw rule arrays until objectui#6828 lowered them.

So the branch in the card's step 2 is not an open question, it is a measured fork:

• runtime takes the array form end to end ⇒ converge here;
• it does not ⇒ ⛔ file the consumer card first and stop. Do not land a declaration the runtime refuses. That is Option-A sequencing and it is the whole reason this card exists as its own card rather than as a rider on PR feat(spec): converge ComponentPropsMap['element:number'].filter onto the ViewFilterRule array form (#12039, ui#6206 Option B) #14404.

⛔ Do not skip step 1 because six siblings already converged. Each of those was measured; the record_picker read path has not been.

Routing

domain:spec — packages/spec is one contract lane whole. Step 1 reads objectui at the current pin, but reading a consumer to decide a spec declaration does not move the lane; if step 1 forks to the consumer card, that card is repo:objectui.

priority:p3: no runtime defect today, nothing blocked, and the entry is inert in the sense that both orthographies parse something. It stays a real card because a platform-wide ruling that is 6-of-7 done is the state where the next author copies the wrong one.

needs:contract-review applied now: the card names the convergence as a BREAKING accept-set change with an ADR-0087 D3 entry, which is Clause-② on its face. If step 1 forks to the consumer card instead, this label comes off with the deferral.

---

Generated by Claude Code