fix(investigation): persist large results and require graph links

Full-network pressure comparisons exceeded the worker limit before saved-result previews could run. Preserve complete calculation results and keep read-time summaries and pagination.

Incremental graph publication accepted isolated nodes after flow-authoring rules were lost. Require new steps to connect explicitly while retaining historical snapshots and status-only patches. Regression tests reproduce both failures on the previous code.

app/services/calculation_worker.py

@@ -93,8 +93,8 @@ async def execute_child(input_path: str, output_path: str):
             category = 'EXECUTION_CONTEXT_UNAVAILABLE'
         payload = {'status': 'failed', 'error': {'code': category}}
     encoded = json.dumps(payload, allow_nan=False)
-    if len(encoded.encode()) > 8 * 1024 * 1024:
-        encoded = json.dumps({'status': 'failed', 'error': {'code': 'RESULT_TOO_LARGE'}})
+    # Keep the complete calculation for persistence. MCP result reads apply
+    # their own summary/page budgets after the result has been saved.
     Path(output_path).write_text(encoded)
 
 
@@ -212,8 +212,6 @@ class CalculationWorker:
                         terminal = {'status': 'timed_out', 'error': {'code': 'EXECUTION_TIMEOUT'}}
                     elif process.returncode is not None:
                         try:
-                            if output_path.stat().st_size > 8 * 1024 * 1024:
-                                raise ValueError('Result exceeds persistence limit')
                             terminal = json.loads(output_path.read_text())
                         except (OSError, ValueError):
                             terminal = {'status': 'failed', 'error': {'code': 'CHILD_PROCESS_FAILED'}}

app/services/investigation_graph.py

@@ -39,6 +39,7 @@ def update_investigation_graph(
         ))
 
     nodes = {node.node_id: node.model_copy(deep=True) for node in previous.node_refs} if previous else {}
+    saved_ids = set(nodes)
     require(len({node.node_id for node in update.nodes}) == len(update.nodes),
             "GRAPH_NODE_DUPLICATE", "Graph update contains duplicate node ids", status=422)
     for node in update.nodes:
@@ -81,6 +82,26 @@ def update_investigation_graph(
             edges.append(edge.model_copy(deep=True))
     require(len(nodes) <= 100 and len(edges) <= 200, "GRAPH_CAPACITY_EXCEEDED",
             "Investigation graph exceeds 100 nodes or 200 edges; existing paths were retained", status=422)
+    # Validate only newly added steps. Historical snapshots and status-only
+    # patches stay valid, even if an earlier publisher omitted relationships.
+    new_ids = set(nodes) - saved_ids
+    if new_ids and len(nodes) > 1:
+        neighbours = {node_id: set() for node_id in nodes}
+        for edge in edges:
+            neighbours[edge.source].add(edge.target)
+            neighbours[edge.target].add(edge.source)
+        connected = set(saved_ids) or {next(iter(nodes))}
+        pending = list(connected)
+        while pending:
+            for node_id in neighbours[pending.pop()] - connected:
+                connected.add(node_id)
+                pending.append(node_id)
+        missing = sorted(new_ids - connected)
+        require(not missing, "GRAPH_RELATION_REQUIRED",
+                "Add explicit edges connecting these new steps to the investigation narrative: "
+                + ", ".join(missing)
+                + ". Describe branches, parallel findings or retries; edges do not schedule execution.",
+                status=422)
     lines = [source or "flowchart TD"]
     for node in nodes.values():
         if in_source(node.node_id):