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/domain/schemas/operational_task.py

@@ -5,7 +5,7 @@ import re
 from typing import Annotated, Literal
 from uuid import UUID
 
-from pydantic import Field, JsonValue, model_serializer, model_validator
+from pydantic import Field, JsonValue, field_validator, model_serializer, model_validator
 from app.domain.schemas.answer_ui import (
     CHART_SAMPLE_TIME_PATTERN, ReportUiBlock, validate_chart_sample_time, validate_report_ui_blocks,
 )
@@ -222,25 +222,61 @@ class InvestigationMapTarget(StrictModel):
     feature_id: TaskFeatureId
 
 
-class InvestigationDiagramNodeRef(StrictModel):
+InvestigationNodeAnnotation = Literal["reference", "recorded", "needs_input", "failed", "not_taken"]
+LEGACY_INVESTIGATION_NODE_ANNOTATIONS = {
+    "pending": "reference", "running": "reference", "completed": "recorded",
+    "waiting": "needs_input", "waiting_human": "needs_input",
+    "skipped": "not_taken", "cancelled": "not_taken",
+}
+
+
+class _InvestigationDiagramNode(StrictModel):
     node_id: Annotated[str, Field(pattern=r"^[A-Za-z_][A-Za-z0-9_-]{0,79}$")]
     label: NonEmptyStr
+    detail: Annotated[str, Field(min_length=1, max_length=2000)] | None = None
     evidence_refs: list[NonEmptyStr] = Field(default_factory=list)
     action_refs: list[UUID] = Field(default_factory=list)
-    status: Literal[
-        "pending", "running", "completed", "waiting", "waiting_human", "failed", "skipped", "cancelled"
-    ] | None = Field(default=None, description=(
-        "Evidence-confirmed node state; waiting means prerequisites/external/system result, "
-        "waiting_human requires manual authorization/approval/feedback. "
-        "Unknown stays omitted/null; host controls colors."
-    ))
     map_targets: list[InvestigationMapTarget] = Field(default_factory=list, max_length=100)
 
 
+class InvestigationDiagramNodeRef(_InvestigationDiagramNode):
+    status: InvestigationNodeAnnotation | None = Field(default=None, description=(
+        "A saved process-reference annotation, not live execution state. "
+        "recorded means a finding exists, not that a task or field action completed."
+    ))
+
+    @field_validator("status", mode="before")
+    @classmethod
+    def read_legacy_annotation(cls, value: object) -> object:
+        return LEGACY_INVESTIGATION_NODE_ANNOTATIONS.get(value, value) if isinstance(value, str) else value
+
+
+class InvestigationGraphNodeUpdate(_InvestigationDiagramNode):
+    status: InvestigationNodeAnnotation | None = Field(default=None, description=(
+        "Optional reference annotation: reference, recorded finding, needs_input, failed attempt, or not_taken. "
+        "No live node/task binding or per-step synchronization is required."
+    ))
+
+
+class InvestigationGraphEdge(StrictModel):
+    source: Annotated[str, Field(pattern=r"^[A-Za-z_][A-Za-z0-9_-]{0,79}$")]
+    target: Annotated[str, Field(pattern=r"^[A-Za-z_][A-Za-z0-9_-]{0,79}$")]
+    label: Annotated[str, Field(min_length=1, max_length=200)] | None = None
+
+
+class InvestigationGraphUpdate(StrictModel):
+    title: Annotated[str, Field(min_length=1, max_length=200)] | None = None
+    nodes: list[InvestigationGraphNodeUpdate] = Field(default_factory=list, max_length=100)
+    edges: list[InvestigationGraphEdge] = Field(default_factory=list, max_length=200)
+
+
 class InvestigationDiagram(StrictModel):
     title: NonEmptyStr
     mermaid: Annotated[str, Field(min_length=1, max_length=20000)]
     node_refs: list[InvestigationDiagramNodeRef] = Field(default_factory=list, max_length=100)
+    # Present only on a graph maintained by the host. Legacy Mermaid stays readable.
+    edges: list[InvestigationGraphEdge] | None = Field(default=None, max_length=200)
+    source_mermaid: Annotated[str, Field(min_length=1, max_length=20000)] | None = None
 
     @model_validator(mode="after")
     def validate_description(self):
@@ -323,6 +359,16 @@ class InvestigationProgress(StrictModel):
     diagram: InvestigationDiagram | None = None
 
 
+class InvestigationProgressInput(InvestigationProgress):
+    graph_update: InvestigationGraphUpdate | None = None
+
+    @model_validator(mode="after")
+    def validate_graph_input(self):
+        if self.graph_update is not None and self.diagram is not None:
+            raise ValueError("use graph_update without a generated Mermaid diagram")
+        return self
+
+
 class InvestigationProgressSnapshot(StrictModel):
     revision: PositiveInt
     run_id: UUID
@@ -367,7 +413,7 @@ class AgentArtifactsCommand(StrictModel):
     evidence: list[EvidenceInput] = Field(default_factory=list, max_length=100)
     plan: ExecutionPlanInput | None = None
     report: InvestigationReportInput | None = None
-    progress: InvestigationProgress | None = None
+    progress: InvestigationProgressInput | None = None
 
     @model_validator(mode="after")
     def validate_artifacts(self):
@@ -377,6 +423,8 @@ class AgentArtifactsCommand(StrictModel):
             self.report.diagram if self.report is not None else None,
             self.progress.diagram if self.progress is not None else None,
         ):
+            if diagram is not None and (diagram.edges is not None or diagram.source_mermaid is not None):
+                raise ValueError("structured graph metadata is host-owned; use progress.graph_update")
             missing = _undeclared_diagram_node_refs(diagram)
             if missing:
                 raise ValueError(