feat(investigation): preserve cumulative reference graphs

Accumulate investigation paths and preserve failed attempts across rounds. Keep graph annotations separate from execution state, synchronize the API contract, and consolidate documentation.

app/infra/db/postgresql/operational_tasks.py

@@ -628,6 +628,37 @@ class OperationalTaskRepository:
         row = await cursor.fetchone()
         return row["document"] if row else None
 
+    async def get_latest_graph_progress(self, task_id: str) -> dict | None:
+        availability = await self.connection.execute(
+            "SELECT to_regclass('workbench.operational_task_progress') AS relation"
+        )
+        if not (await availability.fetchone())["relation"]:
+            return None
+        cursor = await self.connection.execute(
+            """SELECT document FROM workbench.operational_task_progress
+               WHERE project_id=%s AND task_id=%s
+                 AND jsonb_typeof(document#>'{progress,diagram,edges}')='array'
+               ORDER BY (document->>'execution_generation')::bigint DESC, revision DESC LIMIT 1""",
+            (self.project_id, task_id),
+        )
+        row = await cursor.fetchone()
+        return row["document"] if row else None
+
+    async def get_latest_diagram(self, task_id: str) -> dict | None:
+        cursor = await self.connection.execute(
+            """SELECT diagram FROM (
+                 SELECT document#>'{progress,diagram}' AS diagram, published_at AS occurred_at
+                 FROM workbench.operational_task_progress WHERE project_id=%s AND task_id=%s
+                 UNION ALL
+                 SELECT document#>'{report,diagram}', created_at
+                 FROM workbench.operational_task_reports WHERE project_id=%s AND task_id=%s
+               ) AS diagrams WHERE jsonb_typeof(diagram)='object'
+               ORDER BY occurred_at DESC LIMIT 1""",
+            (self.project_id, task_id, self.project_id, task_id),
+        )
+        row = await cursor.fetchone()
+        return row["diagram"] if row else None
+
     async def save_progress(self, task_id: str, snapshot: dict) -> None:
         cursor = await self.connection.execute(
             """INSERT INTO workbench.operational_task_progress

app/infra/db/postgresql/read_queries.py

@@ -7,7 +7,9 @@ from uuid import UUID
 from psycopg import AsyncConnection
 from psycopg import sql
 from psycopg.rows import dict_row
+from psycopg.types.json import Jsonb
 from fastapi import HTTPException
+from app.domain.schemas.operational_task import LEGACY_INVESTIGATION_NODE_ANNOTATIONS
 from app.infra.db.postgresql.scada import _SCADA_VIEW_SELECT, _device
 from app.infra.db.timescaledb.repositories.realtime import RealtimeRepository
 from app.infra.db.timescaledb.repositories.analysis import AnalysisResultsRepository
@@ -329,6 +331,29 @@ async def task_context_summary(conn: AsyncConnection, *, project_id: UUID, task_
     return result
 
 
+def _reference_diagram_projection(stored_sql: str, stored_params: tuple, section: str) -> tuple[str, tuple]:
+    """Normalize only saved diagram annotations before any data_path or page projection."""
+    if section not in {"progress", "report"}:
+        raise ValueError("reference diagrams only belong to progress or report sections")
+    path = "{data," + section + ",diagram,node_refs}"
+    statement = f"""
+        WITH reference_documents AS (
+            {stored_sql}
+        ), reference_aliases AS (
+            SELECT %s::jsonb AS annotations
+        )
+        SELECT CASE WHEN jsonb_typeof(snapshot#>'{path}')='array' THEN
+            jsonb_set(snapshot,'{path}',COALESCE((
+                SELECT jsonb_agg(CASE WHEN annotations ? (node->>'status')
+                    THEN jsonb_set(node,'{{status}}',annotations->(node->>'status'))
+                    ELSE node END ORDER BY position)
+                FROM jsonb_array_elements(snapshot#>'{path}') WITH ORDINALITY AS nodes(node,position)
+            ),'[]'::jsonb)) ELSE snapshot END AS snapshot
+        FROM reference_documents CROSS JOIN reference_aliases
+    """
+    return statement, (*stored_params, Jsonb(LEGACY_INVESTIGATION_NODE_ANNOTATIONS))
+
+
 async def task_context_page(
     conn: AsyncConnection, *, project_id: UUID, task_id: UUID, section: str,
     report_revision: int | None, data_path: list[str], limit: int, offset: int,
@@ -386,11 +411,16 @@ async def task_context_page(
         stored = """SELECT jsonb_build_object('section','progress','data',p.document) AS snapshot
             FROM workbench.operational_task_progress p
             JOIN workbench.operational_task_runs r ON r.project_id=p.project_id AND r.task_id=p.task_id AND r.run_id=p.run_id
-            JOIN workbench.operational_tasks t ON t.project_id=p.project_id AND t.task_id=p.task_id AND t.execution_generation=r.generation
-            WHERE p.project_id=%s AND p.task_id=%s"""
+            JOIN workbench.operational_tasks t ON t.project_id=p.project_id AND t.task_id=p.task_id
+            WHERE p.project_id=%s AND p.task_id=%s AND r.generation<=t.execution_generation
+              AND (t.execution_generation=r.generation
+                   OR jsonb_typeof(p.document#>'{progress,diagram,edges}')='array')
+            ORDER BY (t.execution_generation=r.generation) DESC,r.generation DESC,p.revision DESC LIMIT 1"""
         params = (project_id, task_id)
     else:
         raise HTTPException(422, "Unknown task context section")
+    if section in {"progress", "report"}:
+        stored, params = _reference_diagram_projection(stored, params, section)
     return await _snapshot_page(conn, stored_sql=stored, stored_params=params,
         data_path=data_path, limit=limit, offset=projection_offset, mcp_sources=[], max_bytes=8000)