fix(i18n): stop the Review Queue tab claiming a 180-day window its filter never applied (#769)

.changeset/stale-articles-review-queue-label.md

@@ -0,0 +1,29 @@
+---
+"hotcrm": patch
+---
+
+Knowledge base: the Review Queue tab no longer claims a 180-day window it never applied.
+
+In English, Chinese, Spanish and Japanese the knowledge-article review tab was named
+"Stale (>180d)" (过期 (>180 天) / Obsoletos (>180d) / 古い (>180日)), but its filter
+selected only `status = published`. The tab therefore returned **every** published
+article — including one reviewed minutes ago, merely sorted to the bottom — under a
+heading that promised a six-month cut. Anyone who read the tab as a worklist was
+reading a list of the whole knowledge base.
+
+The four names now match the view's own metadata label, "Review Queue · Oldest First",
+which is what the view actually does: every published article, least-recently-reviewed
+first. Which rows the tab returns is unchanged.
+
+The 180-day window was measured before being ruled out rather than assumed impossible.
+The date macro does resolve on the read path now, and lands on the start of the calendar
+day 180 days back — but a comparison against it does not match articles that have never
+been reviewed at all, and those are exactly the rows a review queue most needs to show.
+Saying "older than 180 days **or** never reviewed" needs a disjunction, and a view filter
+is a flat list of conditions combined with AND, so the honest window cannot currently be
+written. Whether this queue should become a cut is now a separate, open product question.
+
+The house-rule guard that catches this defect class (a view name promising a time scope
+its filter does not express) previously recognised only current-calendar-period phrases
+like "This Quarter", so it could not see "> 180d". It now also reads parameterised day
+windows in all four locales.

src/translations/en.ts

@@ -263,7 +263,7 @@ export const en: TranslationData = {
         all_articles: { label: 'All Articles' },
         published_articles: { label: 'Published' },
         my_drafts: { label: 'My Drafts' },
-        stale_articles: { label: 'Stale (>180d)' },
+        stale_articles: { label: 'Review Queue · Oldest First' },
       },
       _sections: {
         basic: { label: 'Article Information' },

src/translations/es-ES.ts

@@ -282,7 +282,7 @@ export const esES: TranslationData = {
         all_articles: { label: 'Todos los artículos' },
         published_articles: { label: 'Publicados' },
         my_drafts: { label: 'Mis borradores' },
-        stale_articles: { label: 'Obsoletos (>180d)' },
+        stale_articles: { label: 'Cola de revisión · Más antiguos primero' },
       },
       _sections: {
         basic: { label: 'Información del Artículo' },

src/translations/ja-JP.ts

@@ -290,7 +290,7 @@ export const jaJP: TranslationData = {
         all_articles: { label: 'すべての記事' },
         published_articles: { label: '公開済み' },
         my_drafts: { label: '自分の下書き' },
-        stale_articles: { label: '古い (>180日)' },
+        stale_articles: { label: 'レビューキュー · 古い順' },
       },
       _sections: {
         basic: { label: '記事情報' },

src/translations/zh-CN.ts

@@ -256,7 +256,7 @@ export const zhCN: TranslationData = {
         all_articles: { label: '全部文章' },
         published_articles: { label: '已发布' },
         my_drafts: { label: '我的草稿' },
-        stale_articles: { label: '过期 (>180 天)' },
+        stale_articles: { label: '复核队列 · 最久未复核在前' },
       },
       _sections: {
         basic: { label: '文章信息' },

src/views/knowledge_article.view.ts

@@ -5,9 +5,13 @@ import { defineView } from '@objectstack/spec/ui';
 /**
  * Knowledge Article Views
  *
- *   • grid     — agent-facing article queue (status / category / audience)
+ *   • grid      — agent-facing article queue (status / category / audience)
  *   • published — public articles only
- *   • stale     — published > 180 days without review
+ *   • stale     — every published article, least-recently-reviewed first
+ *
+ * The third one is a RANKING, not a cut: `stale_articles` is the metadata
+ * item's name, not a claim about which rows come back. See its own note below
+ * for why it stays a ranking (#769).
  */
 export const KnowledgeArticleViews = defineView({
   list: {
@@ -73,32 +77,65 @@ export const KnowledgeArticleViews = defineView({
     stale_articles: {
       name: 'stale_articles',
       type: 'grid',
-      // Operator-only filter + oldest-review-first sort, and the reason
-      // recorded here has expired (#744).
+      // Operator-only filter + least-recently-reviewed-first sort. This view
+      // is a RANKING and the label says so; it deliberately applies no
+      // 180-day cut. Both halves of that sentence were settled by measurement
+      // (#769), so the reasoning is recorded here rather than re-derived.
+      //
+      // 1. The old note's premise expired (#744, #773). It read: the list data
+      //    path does NOT resolve date macros, so `last_reviewed_at <
+      //    '{180_days_ago}'` compared the literal string, and since '2026-…' <
+      //    '{…' is lexicographically true it matched EVERY published article.
+      //    True on @objectstack 16.1.0; false from 17.0.0-rc.0, when
+      //    `resolveFilterTokens()` was wired into the ObjectQL read path
+      //    (objectql #3582) ahead of the middleware chain —
+      //    `find`/`findOne`/`count`/`aggregate`, saved-view filters included.
+      //    Measured on the pinned 17.0.0-rc.2: `{180_days_ago}` resolves to
+      //    the START of the calendar day 180 days ago (00:00:00.000Z), so as
+      //    an EXCLUSIVE upper bound under `less_than` it excludes that whole
+      //    day — the #3777 day-boundary convention, and the right sense for a
+      //    label reading "> 180d".
+      //
+      // 2. Expressible is not the same as expressible WITHOUT LOSS. A review
+      //    queue's most overdue population is the never-reviewed one, and
+      //    `last_reviewed_at` is nullable in exactly the case that matters.
+      //    On the SINGLE-record path `knowledge_article_publish_timestamps`
+      //    keeps it stamped — the engine's `sys_fetch_previous_update` builtin
+      //    hands the hook a `previous`, so publishing and every later edit
+      //    re-stamp. On the BULK path (`multi: true`) there is no
+      //    `input.id`, that builtin cannot fetch anything, the hook sees no
+      //    status at all and stamps nothing — so a bulk load or mass edit,
+      //    which is how an org imports an existing knowledge base, leaves
+      //    published articles with no review timestamp (measured; filed
+      //    separately). `$lt` matches neither null nor an absent key, so a
+      //    bare window deletes precisely those rows — the imported ones, the
+      //    least reviewed of all. The honest condition is "earlier than
+      //    {180_days_ago} OR empty" — and a view `filter` cannot say that. It
+      //    is a FLAT, strict array of `{field, operator, value}` rules
+      //    (`ViewFilterRuleSchema`) combined with AND; there is no `or`, no
+      //    nesting, no logic key. Spelled as the two rules the grammar does
+      //    allow, the window and the emptiness test AND together and the view
+      //    returns ZERO rows. (The engine itself has `$or` and answers the
+      //    question correctly — this is a view-authoring limit, not an engine
+      //    limit.)
       //
-      // It read: the list data path does NOT resolve date macros, so
-      // `last_reviewed_at < '{180_days_ago}'` compared the literal string, and
-      // since '2026-…' < '{…' is lexicographically true it matched EVERY
-      // published article — including ones reviewed minutes ago (verified
-      // against the running console). That was measured on @objectstack 16.1.0
-      // and stopped being true at 17.0.0-rc.0, when `resolveFilterTokens()` was
-      // wired into the ObjectQL read path (objectql #3582) ahead of the
-      // middleware chain — `find`/`findOne`/`count`/`aggregate`, saved-view
-      // filters included. `{180_days_ago}` matches the parameterised token
-      // grammar `DATE_MACRO_PARAM_RE`, so on the pinned 17.0.0-rc.2 it reaches
-      // the driver as a resolved instant and the lexicographic inversion above
-      // is unreachable on this path. `test/forecast-current-quarter-view.test.ts`
-      // pins that seam against a real engine for the view filters that use it.
+      // So the label is what changed in #769, not the filter: the four locale
+      // packs had translated this view as 'Stale (>180d)' / '过期 (>180 天)' /
+      // 'Obsoletos (>180d)' / '古い (>180日)', promising a window the filter
+      // never expressed — the #730 defect class, living (as it did there) only
+      // in the translated half. They now render the metadata label faithfully,
+      // and every user-visible name for this view again promises exactly the
+      // ordering it delivers. No `emptyState` is authored because the change
+      // introduces no new empty state: "no published articles" is the same
+      // zero this tab has always had, unlike `closing_this_quarter` (#746),
+      // where scoping made "empty" newly reachable.
       //
-      // So the window is expressible now, and this view still does not express
-      // it — which matters more here than it does for `stale_opportunities`,
-      // because the four locale labels DO promise one ('Stale (>180d)' and its
-      // translations) while the metadata label promises only an ordering. That
-      // gap is #769: adding the bound is a behaviour change with two things to
-      // measure first (`last_reviewed_at` is `Field.datetime`, so a bare
-      // calendar-day comparand carries the #3777 upper-bound convention, and
-      // never-reviewed articles have no value at all), so it is filed rather
-      // than done here. Nothing below was touched.
+      // `test/forecast-current-quarter-view.test.ts` owns the house rule and
+      // pins all of the above against a real engine, including the day-window
+      // claim vocabulary that would fail this view the day a label reclaims a
+      // window. Whether the queue SHOULD become a 180-day cut is a product
+      // question that needs the null population handled first — filed
+      // separately, not decided by a translator.
       label: 'Review Queue · Oldest First',
       data: { provider: 'object', object: 'crm_knowledge_article' },
       columns: ['article_number', 'title', 'category', 'owner_id', 'last_reviewed_at'],