jobq graph: state filter on /api/jobq/graph + multi-select checkboxes

GET /api/jobq/graph gains a states query param (comma-separated
hive_jobq::State names): narrows the served root groups to the named
states, keeping a group whole (filtering by a root's own state, which
is already its subtree's rolled-up answer). Absent, empty, or fully
unrecognised is the identity filter, matching prior behaviour.

hive-jobq-graph.js gains a row of per-state checkboxes above the tree,
re-fetching the endpoint with the selection on toggle. Default
selection hides Done and Skipped.

Server-side filtering (not client-side hiding) so hive-jobq-graph-update's
node list, and everything downstream of it in builds.js (count pill,
live-log panel), only ever sees what's actually shown.

docs/web-ui/dashboard.md

@@ -255,6 +255,24 @@ Settled entries render their **full step tree**, not just a bare
 summary — unlike the old `DagView` projection, this wire does not
 filter `Done` nodes out.
 
+**State filter (hyperhive#2606).** A row of per-state checkboxes above
+the tree — one per lifecycle state, matching the row glyphs — lets the
+operator narrow which root groups render; unchecking a state re-fetches
+`GET /api/jobq/graph?states=<checked, comma-joined>` rather than
+hiding rows client-side, so `hive-jobq-graph-update`'s node list (and
+everything downstream of it — the count pill, the live-log panel) only
+ever sees what's actually shown. Filtering is by a **root's own**
+state, which is already its subtree's rolled-up answer, so a group is
+kept or dropped whole, never split mid-tree. Default selection is
+every state **except** `Done`/`Skipped` — a fresh queue view leads
+with what's still moving or needs attention, not the settled tail; the
+`states` param is omitted entirely (identical request to before this
+filter existed) when every state is checked. Server-side: the query
+narrows [`Queue::graph_snapshot`]'s already-bounded (`MAX_HISTORY_DAGS`)
+root set — the history cap and the state filter are independent
+concerns, so a narrow filter never reaches further back in time to
+compensate.
+
 Below the queue, a **live build-log panel** (`#rebuild-live-log`,
 `renderRebuildLiveLog`) shows the currently-running rebuild's output
 inline — collapsible, with a live/ok/fail badge and a `↓ raw`

frontend/packages/shared/src/jobq-graph/hive-jobq-graph.css

@@ -10,6 +10,26 @@
   color: var(--fg);
 }
 
+.jg-filter {
+  display: flex;
+  flex-wrap: wrap;
+  gap: 0.15em 0.9em;
+  margin: 0 0 0.5em;
+  font-size: 0.85em;
+}
+
+.jg-filter-label {
+  display: inline-flex;
+  align-items: center;
+  gap: 0.3em;
+  color: var(--muted);
+  cursor: pointer;
+  user-select: none;
+}
+.jg-filter-label:has(input:checked) {
+  color: var(--fg);
+}
+
 .jg-body {
   display: flex;
   flex-direction: column;

frontend/packages/shared/src/jobq-graph/hive-jobq-graph.js

@@ -14,20 +14,20 @@
 // `buildTree`'s `_waitsOn` resolution. `cancellable` attribute adds a
 // per-node cancel button dispatching `hive-jobq-graph-cancel` instead of
 // POSTing anything itself — see `CANCELLABLE_STATES` + the click-delegate
-// in `connectedCallback` below for both.
+// in `connectedCallback` below for both. A per-state checkbox row above
+// the tree re-fetches `endpoint` with `?states=<checked>` on toggle — see
+// `_buildFilterBar`/`_fetchUrl` below and docs/web-ui/dashboard.md's
+// R3BU1LD QU3U3 section for the full filter rationale.
 //
 // Usage: <hive-jobq-graph endpoint="/api/jobq/graph"></hive-jobq-graph> —
 // self-fetches on connect. `.refresh()` re-fetches + re-renders;
 // `.render(nodes)` renders host-pushed data directly, no fetch. Every
 // render dispatches a bubbling/composed `hive-jobq-graph-update` event
-// (`detail: { nodes }`) so a host needing the raw list for something the
-// tree doesn't show (a count badge, a live-log panel) listens instead of
-// running its own parallel fetch.
+// (`detail: { nodes }`) so a host needing the raw list (a count badge, a
+// live-log panel) listens instead of running its own parallel fetch.
 //
-// Shadow DOM + own styles, per instruction — unlike light-DOM
-// <hive-tab-strip>, this renders a whole subtree nothing else needs to
-// select into. Theme custom properties (--fg, --red, ...) still pierce
-// the shadow boundary by inheritance; only plain class rules are local.
+// Shadow DOM + own styles: theme custom properties (--fg, --red, ...)
+// pierce the shadow boundary by inheritance; only plain class rules local.
 
 import { el } from '../dom.js';
 import { attachShadowCss } from '../shadow-css.js';
@@ -43,6 +43,14 @@ const STATE_GLYPH = {
   Skipped: '·',
 };
 
+// Declaration order doubles as render order for the filter checkboxes —
+// matches `hive_jobq_wire::ALL_STATES` on the wire, so the row reads in the
+// same lifecycle order the rollup endpoint counts in.
+const ALL_STATES = Object.keys(STATE_GLYPH);
+
+// Product call: "default selection filters out skipped and done."
+const DEFAULT_HIDDEN_STATES = new Set(['Done', 'Skipped']);
+
 // Non-terminal states a cancel button makes sense on. Finishing is
 // included — "own logic done, children still running" is still a subtree
 // worth stopping early.
@@ -133,9 +141,14 @@ class HiveJobqGraph extends HTMLElement {
     // <hive-agent-menu> hit when a row cache moves an already-built
     // element without a real detach.
     if (this._root) return;
+    // Set before the first `refresh()` call below, so the very first fetch
+    // already carries the default filter rather than flashing every state
+    // and re-fetching a moment later.
+    this._selectedStates = new Set(ALL_STATES.filter((s) => !DEFAULT_HIDDEN_STATES.has(s)));
     this._root = attachShadowCss(this, graphCss);
+    this._filterBar = this._buildFilterBar();
     this._body = el('div', { class: 'jg-body' });
-    this._root.append(this._body);
+    this._root.append(this._filterBar, this._body);
     // One delegated listener rather than a per-button one — cancel buttons
     // come and go on every re-render, a delegated listener on the stable
     // container doesn't need rebinding.
@@ -151,15 +164,51 @@ class HiveJobqGraph extends HTMLElement {
     this.refresh();
   }
 
-  // Re-fetch `endpoint` and re-render. Public so a host page can call it
-  // on its own refresh cadence (SSE tick, poll, whatever fits the page) —
-  // this element intentionally owns no transport of its own.
-  async refresh() {
+  // One checkbox per `ALL_STATES` entry, pre-ticked per `_selectedStates`.
+  // Built once at connect — toggling a box mutates `_selectedStates` and
+  // re-fetches rather than rebuilding the row, so focus/scroll position in
+  // the row itself is never disturbed by a data refresh.
+  _buildFilterBar() {
+    const bar = el('div', { class: 'jg-filter' });
+    for (const state of ALL_STATES) {
+      const id = 'jg-filter-' + state.toLowerCase();
+      const cb = el('input', { type: 'checkbox', id });
+      cb.checked = this._selectedStates.has(state);
+      cb.addEventListener('change', () => {
+        if (cb.checked) this._selectedStates.add(state);
+        else this._selectedStates.delete(state);
+        this.refresh();
+      });
+      const label = el('label', { for: id, class: 'jg-filter-label jg-state-' + state.toLowerCase() },
+        cb, ' ', STATE_GLYPH[state] + ' ' + state);
+      bar.append(label);
+    }
+    return bar;
+  }
+
+  // `endpoint` plus the current filter selection as a `states=` query
+  // param — omitted entirely when every state is checked, so the
+  // unfiltered default case sends the exact same request as before this
+  // filter existed.
+  _fetchUrl() {
     const endpoint = this.getAttribute('endpoint');
-    if (!endpoint || !this._body) return;
+    if (!endpoint) return null;
+    if (!this._selectedStates || this._selectedStates.size >= ALL_STATES.length) return endpoint;
+    const url = new URL(endpoint, window.location.origin);
+    url.searchParams.set('states', Array.from(this._selectedStates).join(','));
+    return url.pathname + url.search;
+  }
+
+  // Re-fetch `endpoint` (filtered by the current checkbox selection) and
+  // re-render. Public so a host page can call it on its own refresh
+  // cadence (SSE tick, poll, whatever fits the page) — this element
+  // intentionally owns no transport of its own.
+  async refresh() {
+    const url = this._fetchUrl();
+    if (!url || !this._body) return;
     let nodes;
     try {
-      const r = await fetch(endpoint);
+      const r = await fetch(url);
       if (!r.ok) throw new Error('http ' + r.status);
       nodes = await r.json();
     } catch (err) {

hive-c0re/src/dashboard/state_snapshot.rs

@@ -651,24 +651,57 @@ fn build_approval_views(approvals: Vec<Approval>) -> Vec<ApprovalView> {
     out
 }
 
+/// `/api/jobq/graph` query string. Today's only field is `states`: a
+/// comma-separated allow-list of `hive_jobq::State` names (`"Pending"`,
+/// `"Running"`, ...). Empty / absent ⇒ no filter (current behaviour, every
+/// visible root). Set ⇒ only **root** groups whose own state is named are
+/// served — a root's state is already its subtree's rolled-up answer (see
+/// `hive_jobq_wire`'s doc), so filtering the root filters the whole group.
+/// Unknown tokens are silently ignored (an unrecognised name matches
+/// nothing rather than erroring the whole request), mirroring
+/// `DashboardStreamQuery::kinds` above.
+#[derive(Deserialize, Default, IntoParams)]
+pub(super) struct JobqGraphQuery {
+    states: Option<String>,
+}
+
+/// Parses [`JobqGraphQuery::states`] into the list
+/// [`crate::job_queue::JobQueue::graph_snapshot`] wants. `None` when absent
+/// or when every token failed to parse — both mean "no filter" rather than
+/// "match nothing", so an empty/garbled query reads as the unfiltered call
+/// it replaces rather than an empty result set.
+fn parse_states(raw: Option<&str>) -> Option<Vec<hive_jobq::State>> {
+    let states: Vec<hive_jobq::State> = raw?
+        .split(',')
+        .map(str::trim)
+        .filter(|s| !s.is_empty())
+        .filter_map(|s| serde_json::from_value(serde_json::Value::String(s.to_owned())).ok())
+        .collect();
+    (!states.is_empty()).then_some(states)
+}
+
 #[utoipa::path(
     get,
     path = "/api/jobq/graph",
+    params(JobqGraphQuery),
     responses(
         (status = 200, description = "every node of every retained job group, \
             as generic `hive_jobq` graph nodes: identity, the parent tree, \
             dependency edges with their accepted-outcome sets, lifecycle, and \
             one opaque per-node payload. Group roots ride as ordinary nodes \
-            (`parent: null`) and `Done` nodes are not filtered — a consumer \
-            renders the graph without knowing what any node means.",
+            (`parent: null`) and `Done` nodes are not filtered by default — a \
+            consumer renders the graph without knowing what any node means. \
+            `?states=` narrows to root groups in the named states.",
             body = Vec<hive_jobq_wire::GraphNode>),
     ),
     tag = "state_snapshot"
 )]
 pub(super) async fn jobq_graph(
     State(state): State<AppState>,
+    axum::extract::Query(q): axum::extract::Query<JobqGraphQuery>,
 ) -> axum::Json<Vec<hive_jobq_wire::GraphNode>> {
-    axum::Json(state.coord.job_queue.graph_snapshot())
+    let states = parse_states(q.states.as_deref());
+    axum::Json(state.coord.job_queue.graph_snapshot(states.as_deref()))
 }
 
 #[utoipa::path(

hive-c0re/src/job_queue/mod.rs

@@ -326,19 +326,32 @@ impl JobQueue {
 
     /// Every node of every visible group, as generic graph nodes.
     ///
-    /// **Nothing is hidden.** Group roots ride as ordinary nodes (so a
-    /// consumer needs no special case for "the container" and reads the
-    /// root's own `state` as the group's answer), and `Done` nodes stay (so a
-    /// finished step is visible rather than vanishing from the payload, which
-    /// is what makes a fast rebuild render as a single node).
+    /// **Nothing is hidden**, other than an explicit `states` ask. Group
+    /// roots ride as ordinary nodes (so a consumer needs no special case for
+    /// "the container" and reads the root's own `state` as the group's
+    /// answer), and `Done` nodes stay by default (so a finished step is
+    /// visible rather than vanishing from the payload, which is what makes a
+    /// fast rebuild render as a single node).
+    ///
+    /// `states`, when given, keeps only the **root** groups whose own state
+    /// is named — a root's state is already its subtree's rolled-up answer
+    /// (see `hive_jobq_wire`'s doc), so filtering the root filters the whole
+    /// group without needing to also filter descendants. `None` (or the
+    /// full state set) is the unfiltered call, matching prior behaviour.
+    /// A slice rather than a set: `State` (`hive_jobq`'s) derives `Eq` but
+    /// not `Hash`, and the whole vocabulary is 7 variants — a linear
+    /// membership check per root costs nothing at that size, so there is no
+    /// reason to widen `hive_jobq::State`'s derive list just for this.
     ///
     /// The projection itself is [`hive_jobq_wire`]'s; all this layer supplies
     /// is *which* groups to show — see [`visible_roots`] for why the graph
     /// can't decide that for itself.
     #[must_use]
-    pub fn graph_snapshot(&self) -> Vec<GraphNode> {
+    pub fn graph_snapshot(&self, states: Option<&[State]>) -> Vec<GraphNode> {
         let inner = self.lock();
-        inner.graph().wire_snapshot(visible_roots(&inner))
+        let roots = visible_roots(&inner);
+        let roots = filter_roots_by_state(&inner, roots, states);
+        inner.graph().wire_snapshot(roots)
     }
 
     /// Per-state counts over the **same** groups [`Queue::graph_snapshot`]
@@ -393,6 +406,33 @@ fn find_node(sched: &Sched, id: u64) -> Option<NodeId> {
         .find_map(|n| (n.id.get() == id).then_some(n.id))
 }
 
+/// [`Queue::graph_snapshot`]'s `states` ask, applied to an already-bounded
+/// root set: keeps only the roots whose own `state` is named.
+///
+/// Applied *after* [`visible_roots`] rather than folded into it — the
+/// history cap bounds how much settled work is retained at all, which is a
+/// different question from which of the retained (and live) groups the
+/// caller wants shown right now. `None` (or an unrecognised/absent query)
+/// is the identity filter.
+fn filter_roots_by_state(
+    sched: &Sched,
+    roots: Vec<NodeId>,
+    states: Option<&[State]>,
+) -> Vec<NodeId> {
+    let Some(states) = states else {
+        return roots;
+    };
+    roots
+        .into_iter()
+        .filter(|&id| {
+            sched
+                .graph()
+                .node(id)
+                .is_some_and(|n| states.contains(&n.state))
+        })
+        .collect()
+}
+
 /// The visible **group** set for [`Queue::graph_snapshot`]: every live group
 /// root, plus the newest [`MAX_HISTORY_DAGS`] settled ones.
 ///

hive-c0re/src/job_queue/tests.rs

@@ -306,7 +306,7 @@ fn state_of(q: &JobQueue, dag_id: u64) -> State {
     // rolled-up outcome), so there is nothing to derive here any more.
     // A root aged out of the retained history reads `Done`: it settled, or
     // it would still be live.
-    q.graph_snapshot()
+    q.graph_snapshot(None)
         .iter()
         .find(|n| n.id == dag_id)
         .map_or(State::Done, |n| n.state)
@@ -318,7 +318,7 @@ fn state_of(q: &JobQueue, dag_id: u64) -> State {
 /// asking "how many DAGs" counts the parentless ones. Counting rows would
 /// count steps, which is a different number: one rebuild is ~7 nodes.
 fn dag_count(q: &JobQueue) -> usize {
-    q.graph_snapshot()
+    q.graph_snapshot(None)
         .iter()
         .filter(|n| n.parent.is_none())
         .count()
@@ -1471,6 +1471,55 @@ fn history_retains_live_dags_and_the_newest_terminals() {
     );
 }
 
+/// `graph_snapshot`'s `states` ask filters by **root** state — a whole group
+/// is kept or dropped together, never split mid-subtree. `None` is the
+/// identity filter (every visible root, current default behaviour).
+#[test]
+fn graph_snapshot_states_filters_whole_groups_by_root_state() {
+    let q = JobQueue::new(2);
+    let roots_a = insert_named(&q, |builder| restart_online(builder, &["agent-a"], false));
+    let roots_b = insert_named(&q, |builder| restart_online(builder, &["agent-b"], false));
+    let [head_a] = roots_a.as_slice() else {
+        panic!("a one-agent restart names one root, got {roots_a:?}")
+    };
+    let [head_b] = roots_b.as_slice() else {
+        panic!("a one-agent restart names one root, got {roots_b:?}")
+    };
+
+    assert!(q.cancel(*head_a), "queued dag cancels");
+    assert_eq!(state_of(&q, *head_a), State::Cancelled);
+    assert_eq!(state_of(&q, *head_b), State::Pending, "untouched sibling");
+
+    let roots_of = |states: Option<&[State]>| -> Vec<u64> {
+        q.graph_snapshot(states)
+            .into_iter()
+            .filter(|n| n.parent.is_none())
+            .map(|n| n.id)
+            .collect()
+    };
+
+    assert_eq!(
+        roots_of(Some(&[State::Cancelled])),
+        vec![*head_a],
+        "only the cancelled group's root rides"
+    );
+
+    assert_eq!(
+        roots_of(Some(&[State::Pending])),
+        vec![*head_b],
+        "only the pending group's root rides"
+    );
+
+    let mut both = roots_of(None);
+    both.sort_unstable();
+    let mut expected = vec![*head_a, *head_b];
+    expected.sort_unstable();
+    assert_eq!(
+        both, expected,
+        "no filter shows every visible root, as before"
+    );
+}
+
 #[test]
 fn error_truncation_cuts_on_a_char_boundary() {
     // `truncate_error` is a pure `&str -> String`. This used to submit a DAG,