[WIP] Track redesign of issue definitions for Thread Observability

addons/thread-observability/app/src/thread_observability/api/dashboard.html

@@ -973,7 +973,7 @@ <h2 id="nd-title">…</h2>
 let graphRefreshInFlight = false;
 const chatState = { conversationId: null, messages: [] };
 let nodesSort = { key: null, dir: 'asc' };  // null => server "worst first" default
-const SEV_RANK = { critical: 0, error: 0, warning: 1, warn: 1, info: 2, notice: 3 };
+const SEV_RANK = { critical: 0, crit: 0, error: 0, warning: 1, warn: 1, info: 2, notice: 3 };
 const ASSESSMENT_STATE_CLASS = {
   probation: 'good',
   relaxing: 'good',
@@ -1698,16 +1698,6 @@ <h2 id="nd-title">…</h2>
   const pillsHost = $('#issues-sev-pills');
   issuesBody.innerHTML = '';
   pillsHost.innerHTML = '';
-  // Issue detection is paused pending redesign (#5). The API returns
-  // `status: "placeholder"` until new rules ship. Render the note
-  // verbatim so users (and any AI consumer scraping the dashboard)
-  // don't read an empty list as "all clear".
-  if (s.issues && s.issues.status === 'placeholder') {
-    const note = s.issues.note
-      || 'Issue detection is paused pending redesign.';
-    issuesBody.appendChild(el('div', {class: 'muted', style: 'padding: 6px 0;'}, note));
-    return;
-  }
   const issues = ((s.issues && s.issues.issues) || []).slice();
   if (!issues.length) {
     issuesBody.appendChild(el('div', {class:'empty'}, 'No active issues.'));
@@ -1719,10 +1709,10 @@ <h2 id="nd-title">…</h2>
     const sev = (i.severity || 'info').toLowerCase();
     counts[sev] = (counts[sev] || 0) + 1;
   }
-  const order = ['critical', 'error', 'warning', 'warn', 'info', 'notice'];
+  const order = ['critical', 'crit', 'error', 'warning', 'warn', 'info', 'notice'];
   for (const k of order) {
     if (!counts[k]) continue;
-    const cls = (k === 'critical' || k === 'error') ? 'bad' : (k.startsWith('warn') ? 'warn' : 'muted');
+    const cls = (k === 'critical' || k === 'crit' || k === 'error') ? 'bad' : (k.startsWith('warn') ? 'warn' : 'muted');
     pillsHost.appendChild(el('span', {
       class: 'pill ' + cls,
       style: 'margin-left:4px;',

addons/thread-observability/app/src/thread_observability/api/http_api.py

@@ -866,20 +866,6 @@ def health_snapshot() -> dict[str, object]:
 
     @app.get("/v1/issues/active")
     def list_active_issues() -> dict[str, object]:
-        # Issue detection is paused pending redesign — see tracking
-        # issue #5 and placeholder issue #4. We deliberately return an
-        # empty list with an explicit ``status: "placeholder"`` so
-        # consumers (dashboard, MCP, AI reasoners) don't misread the
-        # absence of issues as "all clear".
-        from ..pipeline.reasoner import ISSUES_PAUSED, ISSUES_PAUSED_NOTE
-        if ISSUES_PAUSED:
-            return {
-                "count": 0,
-                "issues": [],
-                "status": "placeholder",
-                "note": ISSUES_PAUSED_NOTE,
-                "computed_at": _utc_now(),
-            }
         try:
             issues = get_store().list_active_issues()
             return {"count": len(issues), "issues": issues, "computed_at": _utc_now()}

addons/thread-observability/app/src/thread_observability/api/mcp_tools.py

@@ -146,11 +146,12 @@ class ToolCallRequest(BaseModel):
     {
         "name": "list_active_issues",
         "description": (
-            "Return all currently-open Thread network issues. "
-            "NOTE: Issue detection is currently paused pending a redesign of the rule set "
-            "(see tracking issue #5). Until new rules ship, this tool returns an empty list "
-            "with `status: \"placeholder\"`. Do NOT infer \"all clear\" from the empty list — "
-            "instead, reason from the raw data (topology, partitions, links, nodes)."
+            "Return all currently-open Thread network issues computed by deterministic rules. "
+            "Each issue includes the affected EUI64 (or null for mesh-wide issues), "
+            "`first_seen_at`, `last_seen_at`, a severity that reflects actionability × freshness, "
+            "and an evidence payload that includes the EUIs involved and the observation that triggered it. "
+            "Current rule taxonomy: "
+            "`real_partition_split`, `dead_link_reference`, `route_to_otbr_unreachable`."
         ),
         "inputSchema": {"type": "object", "properties": {}, "required": []},
     },    {
@@ -1031,17 +1032,6 @@ async def _dispatch_tool(name: str, arguments: dict[str, Any]) -> dict[str, Any]
         except Exception as exc:  # noqa: BLE001
             return {"error": str(exc)}
     if name == "list_active_issues":
-        # Mirrors /v1/issues/active. Issue detection is paused
-        # pending redesign (#5); return an explicit placeholder so AI
-        # consumers don't infer "all clear" from an empty list.
-        from ..pipeline.reasoner import ISSUES_PAUSED, ISSUES_PAUSED_NOTE
-        if ISSUES_PAUSED:
-            return {
-                "count": 0,
-                "issues": [],
-                "status": "placeholder",
-                "note": ISSUES_PAUSED_NOTE,
-            }
         try:
             issues = get_store().list_active_issues()
             return {"count": len(issues), "issues": issues}

addons/thread-observability/app/src/thread_observability/pipeline/analyze_node.py

@@ -69,6 +69,9 @@ def _evidence_implicates_eui(evidence: Any, eui64: str) -> bool:
     members = evidence.get("members")
     if isinstance(members, list) and eui64 in members:
         return True
+    involved = evidence.get("involved_eui64s")
+    if isinstance(involved, list) and eui64 in involved:
+        return True
     # Nested partitions[].members (partition_split shape).
     partitions = evidence.get("partitions")
     if isinstance(partitions, list):
@@ -78,6 +81,14 @@ def _evidence_implicates_eui(evidence: Any, eui64: str) -> bool:
             part_members = part.get("members")
             if isinstance(part_members, list) and eui64 in part_members:
                 return True
+            sample = part.get("members_sample")
+            if isinstance(sample, list) and eui64 in sample:
+                return True
+    recent_changes = evidence.get("recent_partition_changes")
+    if isinstance(recent_changes, list):
+        for row in recent_changes:
+            if isinstance(row, dict) and row.get("eui64") == eui64:
+                return True
     return False
 
 

addons/thread-observability/app/src/thread_observability/pipeline/device_discovery.py

@@ -1044,24 +1044,6 @@ async def _persist_matter_diagnostics(
     """
     rich = _LAST_MATTER_RICH_INFO
     if not rich:
-        # v0.9.43: even with no rich info this cycle (matter-server WS hiccup
-        # or a single empty poll), we MUST still reconcile the
-        # ``partition_split`` issue. Otherwise an issue opened on a prior
-        # cycle becomes immortal — it never sees a non-split observation
-        # again because the empty-rich early-return below would skip the
-        # close branch. Latent bug observed live as issue #54 hanging open
-        # after the partition had long since healed.
-        try:
-            active = [
-                i for i in s.list_active_issues()
-                if i.get("kind") == "partition_split"
-            ]
-            for issue in active:
-                s.close_issue(int(issue["id"]))
-        except Exception as exc:  # noqa: BLE001
-            log.warning(
-                "partition_split close-on-empty failed: %s", exc,
-            )
         return {
             "nodes_with_diagnostics": 0,
             "links_recorded": 0,
@@ -1590,33 +1572,6 @@ async def _persist_matter_diagnostics(
         for pid, members in sorted(live_partitions.items())
     ]
 
-    # Open/close partition_split issue (now reasoning over live partitions only).
-    try:
-        active = [i for i in s.list_active_issues() if i.get("kind") == "partition_split"]
-        if split:
-            distinct_epids = sorted({
-                p["extended_pan_id"] for p in partition_summary
-                if p.get("extended_pan_id")
-            })
-            s.open_issue(
-                kind="partition_split",
-                severity="warning",
-                evidence={
-                    "partitions": partition_summary,
-                    "partition_count": len(live_partitions),
-                    # If partitions report different extended_pan_ids,
-                    # this is a credentials-mismatch (stale dataset on
-                    # one device) not an RF-fragmentation issue.
-                    "distinct_extended_pan_ids": distinct_epids,
-                    "credentials_mismatch_suspected": len(distinct_epids) > 1,
-                },
-            )
-        else:
-            for issue in active:
-                s.close_issue(int(issue["id"]))
-    except Exception as exc:  # noqa: BLE001
-        log.warning("Failed to update partition_split issue: %s", exc)
-
     log.info(
         "diagnostics persisted: nodes=%d links=%d partitions=%d split=%s "
         "changes=%d phantoms_marked=%d phantoms_cleared=%d excluded_partitions=%d "

addons/thread-observability/app/src/thread_observability/pipeline/playbooks.json

@@ -129,6 +129,57 @@
       ],
       "references": []
     },
+    {
+      "id": "real_partition_split",
+      "title": "Real partition split (evidence-backed)",
+      "applies_to": ["real_partition_split"],
+      "summary": "Multiple live partitions plus evidence of a device transitioning between partitions. This narrows diagnosis vs. a bare partition-count: it suggests an unstable boundary device or RF gap rather than stale IDs.",
+      "evidence_to_collect": [
+        "Issue evidence.partitions and evidence.recent_partition_changes",
+        "get_mesh_state to confirm current partition membership",
+        "diff_topology across the split window (if snapshots exist)"
+      ],
+      "remediation_steps": [
+        "Identify the device(s) in evidence.recent_partition_changes; check their placement and power stability.",
+        "Bridge the RF gap between partitions (add/relocate a router) and re-run ingest.",
+        "If partitions disagree on extended_pan_id, resolve credentials drift (re-commission the minority leader)."
+      ],
+      "references": []
+    },
+    {
+      "id": "dead_link_reference",
+      "title": "Dead-link reference (unknown neighbor EUI64)",
+      "applies_to": ["dead_link_reference"],
+      "summary": "A router references an EUI64 that is not present in the registry-backed nodes table, persisted across multiple ingestion ticks. This usually indicates a recommissioned device leaving stale neighbor/route cache entries.",
+      "evidence_to_collect": [
+        "Issue evidence.references[] (reporter, unknown neighbor, source, seen_count)",
+        "get_neighbors for the reporter router to see link_established/allocated state",
+        "list_all_nodes to confirm the neighbor truly is absent"
+      ],
+      "remediation_steps": [
+        "If the neighbor is a known physical device, re-commission it so HA registry matches the mesh identity.",
+        "Restart the reporter router to flush stale neighbor cache entries.",
+        "If references persist and the neighbor is truly gone, inspect for duplicate identities or a stuck child table."
+      ],
+      "references": []
+    },
+    {
+      "id": "route_to_otbr_unreachable",
+      "title": "Route to OTBR unreachable (loop/unknown next hop)",
+      "applies_to": ["route_to_otbr_unreachable"],
+      "summary": "Walking the next-hop chain from a router to the OTBR terminates in a loop, unknown next hop, or missing route-table entry. This points at routing-table corruption, partition mismatch, or a missing upstream neighbor relationship.",
+      "evidence_to_collect": [
+        "Issue evidence.route_walk (hops + issues)",
+        "get_mesh_state to confirm partitions and current links",
+        "get_neighbors for the router and its first upstream hop"
+      ],
+      "remediation_steps": [
+        "If the route walk reports different_partition, resolve the partition split first.",
+        "Restart the affected router to repopulate its route table.",
+        "If unknown_next_hop persists, inspect the router's neighbor table for missing upstream links and RF quality."
+      ],
+      "references": []
+    },
     {
       "id": "sed_battery_drain",
       "title": "Sleepy-end-device unexpected battery drain",