Filed by the domain:ui execution seat (PM session session_01EMrWaQw3XS5DxTHxp4yRyC) on the finding of an in-seat contract review at CONTRACT_REVIEW_TIER (claude-fable-5) of PR #7400. ⛔ Not graded and no domain:* — routing, type and priority are the triage seat's.
⭐ This is the reviewer's Q1, raised explicitly as "the maintainer's question, not the PM's or mine". The seat is not answering it, and the reviewer deliberately did not either. It is separate from #7399 (that one is the byte ceiling); this one is about ruling scope and a published capability.
The question
#7194 ruled B — a scatter with two or more series refuses. Verbatim, from the ruling (comment 5507673384):
Ruled: B. A chartType: 'scatter' with two or more series renders the file's existing refusal surface (the #7146 / #7147 / #7148 shell) stating that a scatter binds one measure and naming the authored fix; a single-series scatter is unchanged. No multi-measure projection (option A) is built now — zero authored scatters in-repo, so the capability waits for a real caller and reopens as A, one payment, no transition.
⚠️Neither the ruling nor the decision card it adopted mentions compareTo or variant.
But a compare-to scatter is a two-series scatter in the product, so the refusal swallows it.
Verified on PR #7400's head eacc559
| fact | evidence |
|---|
| scatter is on the published compare-to list | packages/plugin-charts/CHANGELOG.md:1727 — "overlay a muted comparison-period series… Pie / donut / funnel ignore compareTo" ⇒ scatter does not |
| it was designed, not incidental | comparisonStyle(s, kind) (AdvancedChartImpl.tsx:92) has a scatter-specific opacity branch, and the base scatter arm already calls comparisonStyle(s, 'scatter') |
| the overlay is synthesised as a second series | ObjectChart.tsx:338supportsCompareTo excludes only pie/donut/funnel; augmentedSeries (815) appends { dataKey: '…__comparison', variant: 'comparison' } |
| the dashboard path has no chart-type exclusion either | DatasetWidget.tsx:759 — widget types scatter and bubble both map to chartType: 'scatter' (377) |
| ⚠️ today's picture on that path is already wrong | the overlay reads y through the single YAxis dataKey={scatterYKey} = series[0].dataKey, so "previous period" is painted exactly on "current" — the same defect #7194 fixes |
⇒ Exempting the overlay in the renderer would restore that wrong picture. The PR is right about that. The question is whether refusing is what B meant.
Why the ruling's own text does not settle it
The literal predicate ("two or more series") does reach it. But the ruling's stated premise — "zero authored scatters in-repo, so the capability waits for a real caller" — was measured on chartType: 'scatter' literals. A compare-to scatter is one authored series plus one synthesised overlay from a documented product feature. It is not an authored two-series scatter, and the census that justified B cannot see it.
⚠️ The PR currently resolves that gap by itself: its pin comment reads "The ruling counts series, not variants" — the implementer's reading, not the ruling's words — and its changeset records it as settled ("This applies to a compareTo overlay too"). That is what the review requires be undone until this card rules.
⭐ A second defect on that path, independent of which way Q1 goes
Ruling B requires the refusal name "the authored fix". On the compare-to path the tile says:
Keep exactly one series: amount, amount__comparison
The author has exactly one series in their spec and never wroteamount__comparison. The named fix is not applicable, and the real fix ("remove compareTo") is unknowable at the renderer, which only sees series. ⇒ Even a confirmed refusal needs its copy or its refusal site changed on this path. A refusal that misdirects the author is worse than the wrong picture it replaces, because it costs a session to disprove.
The shapes (recorded, not recommended)
(a) Confirm — a compare-to scatter refuses. Then the copy or the refusal site must name the actual authored fix there — e.g. raise it upstream in ObjectChart / DatasetWidget where compareTo is known, or add a compare-to-aware sentence. ⚠️ Note the primary series, which drew correctly, is also lost from that tile.
(b) Exclude scatter from compareTo, exactly as pie / donut / funnel already are and as the CHANGELOG already publishes for them: the primary draws, the overlay is never synthesised, the renderer's guard is never reached in product, and the PR's pin inverts. ⚠️ Touches packages/plugin-dashboard — outside PR #7400's declared file surface — and changes a published statement.
(c) Project the overlay only — option A territory, which the ruling explicitly declined.
Reachability, stated honestly
0 in-repo instances. The reviewer widened the census beyond the PR's own regex to the dashboard-widget spelling (type: 'scatter' | 'bubble'), which the PR's pattern cannot see: still 0, with a control (type: 'bar') firing at 10 files including five examples/schema-catalog/**/plugin-dashboard/*.json. So the zero holds on a stronger basis than the PR claimed. ⚠️ But it is authorable today through both surfaces, and incidence in cloud tenant metadata is unmeasurable — the decision card for #7194 said so. ⛔ Unmeasured incidence is not low incidence.
四维分析
长期健全性 —— 一个已发布的能力(scatter 上的 compareTo,CHANGELOG 明写只有 pie/donut/funnel 忽略它,且代码里有 scatter 专属的 comparisonStyle 分支)被一条从未点名它的拒绝裁决顺带吃掉,这正是"裁决"二字被稀释的方式:下一次读这条裁决的人会以为它裁过 compareTo,而它没有。⚠️ 反向也成立:今天那条路径画的是错的图(叠加层被画在主序列的 y 上),(b) 保留能力就等于知情地继续画错。⇒ 本轴不指向某个选项,它指向"必须显式裁",而不是由实现者的一句 pin 注释代裁。
真实业务需求 —— 仓内 0 例(复审员用更强的对照重测过,对照命中 10 个文件,所以这个 0 是读数)。⚠️ 但两条产品路径今天都能写出来,云端租户元数据里的实际发生率测不到。⇒ 需求侧给不出方向,只能说明"不紧急",不能说明"可以默认"。
让 AI 写的代码不易出错 —— 这一轴最尖锐,且与 Q1 怎么裁无关:当前拒绝面对作者说「Keep exactly one series: amount, amount__comparison」,而作者的 spec 里只有一个 series,amount__comparison 是系统合成的、他从没写过。⇒ 这是一条指向不存在的写法的拒绝 —— 比它所替代的错图更贵,因为错图至少能被看出来,而错误的修复指引会把人送去改一个不存在的字段。无论 (a) 还是 (b),这一条都必须修。
创业公司范围纪律 —— (a) 最省,只动本 PR 的文案/拒绝位;(b) 要动 packages/plugin-dashboard,超出本 PR 已申报的文件面,并且要改一句已发布的 CHANGELOG 陈述;(c) 是裁决已明确不做的 option A。⇒ 本轴偏向 (a),但它是四轴里最弱的一条,⛔ 不足以单独决定一个已发布能力的去留。
Status
PR #7400 keeps needs:contract-review and stays draft. The review's verdict is PASS WITH REQUIRED AMENDMENTS; the amendment tied to this card is "do not land the compareTo behaviour as decided until Q1 is answered". #7194 is now blocked on this card in addition to #7399 — ⚠️two independent decisions, one about bytes and one about scope; either alone does not release it.
Related
#7194 (the card and ruling B) · PR #7400 · #7399 (the other, unrelated decision blocking the same PR) · #7113 (the authoring-time zod refusal this deliberately does not pre-empt) · #7146 / #7147 / #7148 (the refusal shell precedents)
Filed by the
domain:uiexecution seat (PM sessionsession_01EMrWaQw3XS5DxTHxp4yRyC) on the finding of an in-seat contract review atCONTRACT_REVIEW_TIER(claude-fable-5) of PR #7400. ⛔ Not graded and nodomain:*— routing,typeand priority are the triage seat's.⭐ This is the reviewer's Q1, raised explicitly as "the maintainer's question, not the PM's or mine". The seat is not answering it, and the reviewer deliberately did not either. It is separate from #7399 (that one is the byte ceiling); this one is about ruling scope and a published capability.
The question
#7194 ruled B — a scatter with two or more
seriesrefuses. Verbatim, from the ruling (comment 5507673384):compareToorvariant.But a compare-to scatter is a two-series scatter in the product, so the refusal swallows it.
Verified on PR #7400's head
eacc559packages/plugin-charts/CHANGELOG.md:1727— "overlay a muted comparison-period series… Pie / donut / funnel ignorecompareTo" ⇒ scatter does notcomparisonStyle(s, kind)(AdvancedChartImpl.tsx:92) has a scatter-specific opacity branch, and the base scatter arm already callscomparisonStyle(s, 'scatter')ObjectChart.tsx:338supportsCompareToexcludes only pie/donut/funnel;augmentedSeries(815) appends{ dataKey: '…__comparison', variant: 'comparison' }DatasetWidget.tsx:759— widget typesscatterandbubbleboth map tochartType: 'scatter'(377)YAxis dataKey={scatterYKey}=series[0].dataKey, so "previous period" is painted exactly on "current" — the same defect #7194 fixes⇒ Exempting the overlay in the renderer would restore that wrong picture. The PR is right about that. The question is whether refusing is what B meant.
Why the ruling's own text does not settle it
The literal predicate ("two or more
series") does reach it. But the ruling's stated premise — "zero authored scatters in-repo, so the capability waits for a real caller" — was measured onchartType: 'scatter'literals. A compare-to scatter is one authored series plus one synthesised overlay from a documented product feature. It is not an authored two-series scatter, and the census that justified B cannot see it.compareTooverlay too"). That is what the review requires be undone until this card rules.⭐ A second defect on that path, independent of which way Q1 goes
Ruling B requires the refusal name "the authored fix". On the compare-to path the tile says:
The author has exactly one series in their spec and never wrote
amount__comparison. The named fix is not applicable, and the real fix ("removecompareTo") is unknowable at the renderer, which only sees series. ⇒ Even a confirmed refusal needs its copy or its refusal site changed on this path. A refusal that misdirects the author is worse than the wrong picture it replaces, because it costs a session to disprove.The shapes (recorded, not recommended)
(a) Confirm — a compare-to scatter refuses. Then the copy or the refusal site must name the actual authored fix there — e.g. raise it upstream in⚠️ Note the primary series, which drew correctly, is also lost from that tile.
ObjectChart/DatasetWidgetwherecompareTois known, or add a compare-to-aware sentence.(b) Exclude scatter from⚠️ Touches
compareTo, exactly as pie / donut / funnel already are and as the CHANGELOG already publishes for them: the primary draws, the overlay is never synthesised, the renderer's guard is never reached in product, and the PR's pin inverts.packages/plugin-dashboard— outside PR #7400's declared file surface — and changes a published statement.(c) Project the overlay only — option A territory, which the ruling explicitly declined.
Reachability, stated honestly
0 in-repo instances. The reviewer widened the census beyond the PR's own regex to the dashboard-widget spelling (⚠️ But it is authorable today through both surfaces, and incidence in cloud tenant metadata is unmeasurable — the decision card for #7194 said so. ⛔ Unmeasured incidence is not low incidence.
type: 'scatter' | 'bubble'), which the PR's pattern cannot see: still 0, with a control (type: 'bar') firing at 10 files including fiveexamples/schema-catalog/**/plugin-dashboard/*.json. So the zero holds on a stronger basis than the PR claimed.四维分析
长期健全性 —— 一个已发布的能力(scatter 上的 compareTo,CHANGELOG 明写只有 pie/donut/funnel 忽略它,且代码里有 scatter 专属的⚠️ 反向也成立:今天那条路径画的是错的图(叠加层被画在主序列的 y 上),(b) 保留能力就等于知情地继续画错。⇒ 本轴不指向某个选项,它指向"必须显式裁",而不是由实现者的一句 pin 注释代裁。
comparisonStyle分支)被一条从未点名它的拒绝裁决顺带吃掉,这正是"裁决"二字被稀释的方式:下一次读这条裁决的人会以为它裁过 compareTo,而它没有。真实业务需求 —— 仓内 0 例(复审员用更强的对照重测过,对照命中 10 个文件,所以这个 0 是读数)。⚠️ 但两条产品路径今天都能写出来,云端租户元数据里的实际发生率测不到。⇒ 需求侧给不出方向,只能说明"不紧急",不能说明"可以默认"。
让 AI 写的代码不易出错 —— 这一轴最尖锐,且与 Q1 怎么裁无关:当前拒绝面对作者说「Keep exactly one series:
amount,amount__comparison」,而作者的 spec 里只有一个 series,amount__comparison是系统合成的、他从没写过。⇒ 这是一条指向不存在的写法的拒绝 —— 比它所替代的错图更贵,因为错图至少能被看出来,而错误的修复指引会把人送去改一个不存在的字段。无论 (a) 还是 (b),这一条都必须修。创业公司范围纪律 —— (a) 最省,只动本 PR 的文案/拒绝位;(b) 要动
packages/plugin-dashboard,超出本 PR 已申报的文件面,并且要改一句已发布的 CHANGELOG 陈述;(c) 是裁决已明确不做的 option A。⇒ 本轴偏向 (a),但它是四轴里最弱的一条,⛔ 不足以单独决定一个已发布能力的去留。Status
PR #7400 keeps⚠️ two independent decisions, one about bytes and one about scope; either alone does not release it.
needs:contract-reviewand stays draft. The review's verdict is PASS WITH REQUIRED AMENDMENTS; the amendment tied to this card is "do not land thecompareTobehaviour as decided until Q1 is answered". #7194 is now blocked on this card in addition to #7399 —Related
#7194 (the card and ruling B) · PR #7400 · #7399 (the other, unrelated decision blocking the same PR) · #7113 (the authoring-time zod refusal this deliberately does not pre-empt) · #7146 / #7147 / #7148 (the refusal shell precedents)