The queue note named the wrong reason the isnull filter was dead

openspec/changes/add-work-queue/proposal.md

@@ -46,12 +46,17 @@ additive: one page, one menu entry, one relocation.
 
 The base filter is `{"assignee": "IS NULL", "isFinalStatus": false}`.
 
-`assignee: "IS NULL"` is the literal sentinel `MariaDbSearchHandler::applyNullCheck`
-matches. The documented `assignee_isnull=true` spelling does **not** work over HTTP:
-`SearchQueryHandler::cleanQuery` tests the value with `=== true`, and a query string
-can only ever deliver the string `"true"`, so the operator silently degrades to
-`IS NOT NULL` and the page renders empty. Verified against a live instance: 17
-unassigned cases with the sentinel, 20 assigned, 2 unassigned and open.
+`assignee: "IS NULL"` is the literal sentinel every OpenRegister condition builder
+matches by value. Verified against a live instance: 17 unassigned cases with the
+sentinel, 20 assigned, 2 unassigned and open.
+
+The suffix spelling `assignee_isnull=true` was dead when this page shipped, and the
+reason is deeper than it first looked: the only code mentioning the operator had no
+production callers, and `isnull` was absent from
+`MagicSearchHandler::COMPARISON_OPERATORS`, so the filter contributed no condition at
+all rather than an inverted one. Fixed in openregister `isnull-filter-operator`. The
+sentinel is kept here because it also works on instances that do not yet carry that
+change, and the two are two spellings of one predicate.
 
 `isFinalStatus` is a real boolean on every case row (measured: 37 of 37 carry it, none
 absent), so plain equality reaches it and no derived predicate is needed.

openspec/changes/add-work-queue/specs/add-work-queue/spec.md

@@ -8,10 +8,11 @@ dossiq SHALL provide a Queue page at `/queue`, an index over the `case` schema
 filtered to cases with no `assignee` and `isFinalStatus` false. It SHALL sit first in
 the My work group, above Assigned to me.
 
-The base filter SHALL be `assignee: "IS NULL"` and `isFinalStatus: false`. The
-`assignee_isnull=true` spelling SHALL NOT be used: OpenRegister's
-`SearchQueryHandler::cleanQuery` compares the value with `=== true`, so a query string
-degrades it to `IS NOT NULL` and the page renders empty with no error.
+The base filter SHALL be `assignee: "IS NULL"` and `isFinalStatus: false`.
+`assignee: "IS NULL"` is the literal sentinel every OpenRegister condition builder
+matches by value, and it SHALL be preferred over the `assignee_isnull=true` suffix,
+which was unimplemented when this page shipped and works only on instances carrying
+openregister `isnull-filter-operator`.
 
 #### Scenario: The queue holds unassigned open cases
 

src/manifest.json

@@ -1025,7 +1025,7 @@
       "route": "/queue",
       "type": "index",
       "title": "Queue",
-      "_note": "The one queue: every case that is open and that nobody has picked up. It is a FILTER over `case`, not a collection of its own, so this is a plain type:index rather than a bespoke view. `assignee: \"IS NULL\"` is the literal sentinel MariaDbSearchHandler::applyNullCheck matches; the documented `assignee_isnull=true` spelling does NOT survive a query string, because SearchQueryHandler::cleanQuery compares the value with `=== true` and degrades it to IS NOT NULL, which renders an empty page with no error. `isFinalStatus: false` keeps closed cases out; it is a real boolean on every case row, not a derived one, so plain equality reaches it. Assigning a case removes it from here and puts it on that person's Assigned to me; All cases shows it either way. The caseType folder sidebar is the same control the Cases index uses, so the queue narrows by case type without a second page. WHY THIS IS A PAGE AND NOT A `menu[].query` PRESET, which is what gate-68 (ADR-097 Decision 5) proposes for a second index over an already-indexed schema: a preset is only a link into the target page, where the query filters and the reader's own facet filters share one map, so a facet interaction can widen the queue back to every case. A base filter on the page itself cannot be cleared by the person reading it, and that is the property that makes the queue trustworthy. Assigned to me is a page for the same reason.",
+      "_note": "The one queue: every case that is open and that nobody has picked up. It is a FILTER over `case`, not a collection of its own, so this is a plain type:index rather than a bespoke view. `assignee: \"IS NULL\"` is the literal sentinel every OpenRegister condition builder matches by value. The suffix spelling `assignee_isnull=true` was dead when this page shipped: the only code that mentioned it had no callers, and `isnull` was missing from MagicSearchHandler::COMPARISON_OPERATORS, so the filter contributed no condition at all. Fixed in openregister `isnull-filter-operator`; the sentinel is kept because it also works on instances that do not yet carry that change. `isFinalStatus: false` keeps closed cases out; it is a real boolean on every case row, not a derived one, so plain equality reaches it. Assigning a case removes it from here and puts it on that person's Assigned to me; All cases shows it either way. The caseType folder sidebar is the same control the Cases index uses, so the queue narrows by case type without a second page. WHY THIS IS A PAGE AND NOT A `menu[].query` PRESET, which is what gate-68 (ADR-097 Decision 5) proposes for a second index over an already-indexed schema: a preset is only a link into the target page, where the query filters and the reader's own facet filters share one map, so a facet interaction can widen the queue back to every case. A base filter on the page itself cannot be cleared by the person reading it, and that is the property that makes the queue trustworthy. Assigned to me is a page for the same reason.",
       "config": {
         "register": "dossiq",
         "schema": "case",