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.

tests/unit/test_calculation_worker.py

@@ -1,5 +1,6 @@
 """Worker boundaries use real managed subprocesses, without external databases."""
 import asyncio
+import json
 from functools import wraps
 from types import SimpleNamespace
 from contextlib import asynccontextmanager
@@ -27,6 +28,66 @@ def run_async(function):
     return wrapper
 
 
+@run_async
+async def test_child_keeps_complete_result_above_eight_mib(monkeypatch, tmp_path):
+    from app.mcp import calculation_jobs
+    from app.services import calculation_worker
+
+    data = {"comparisons": "x" * (8 * 1024 * 1024), "map_payload": {"features": ["coverage"]}}
+
+    @asynccontextmanager
+    async def scope(row):
+        yield object()
+
+    async def calculate(scope, row):
+        return SimpleNamespace(model_dump=lambda **kwargs: data)
+
+    monkeypatch.setattr(calculation_worker, "execution_scope", scope)
+    monkeypatch.setattr(calculation_jobs, "execute_calculation_job", calculate)
+    input_path, output_path = tmp_path / "input.json", tmp_path / "result.json"
+    input_path.write_text("{}")
+    await calculation_worker.execute_child(str(input_path), str(output_path))
+    assert output_path.stat().st_size > 8 * 1024 * 1024
+    assert json.loads(output_path.read_text()) == {"status": "succeeded", "result": data}
+
+
+@run_async
+async def test_supervisor_persists_complete_result_above_eight_mib(monkeypatch, tmp_path):
+    monkeypatch.setenv("CALCULATION_WORKER_STATE_DIR", str(tmp_path))
+    original = asyncio.create_subprocess_exec
+
+    async def spawn(*args, **kwargs):
+        # A real child writes the same file protocol as the calculation process.
+        code = ('import json,pathlib,sys; pathlib.Path(sys.argv[1]).write_text('
+                'json.dumps({"status":"succeeded","result":{"comparisons":"x"*(8*1024*1024),'
+                '"map_payload":{"features":["coverage"]}}}))')
+        return await original(sys.executable, "-c", code, args[-1], start_new_session=True)
+
+    monkeypatch.setattr(asyncio, "create_subprocess_exec", spawn)
+    results = []
+
+    class Repository:
+        async def heartbeat(self, *args):
+            return "running"
+
+        async def finish(self, row, owner, **terminal):
+            results.append(terminal)
+            return True
+
+    @asynccontextmanager
+    async def repository(project):
+        yield Repository()
+
+    worker = CalculationWorker(WorkerSettings(heartbeat_seconds=.02), repository=repository)
+    await asyncio.wait_for(worker.run_job({"project_id": "project", "job_id": "large-result",
+                                          "timeout_seconds": 10}), 15)
+    assert len(results) == 1
+    assert results[0]["status"] == "succeeded"
+    assert results[0]["result"] == {
+        "comparisons": "x" * (8 * 1024 * 1024), "map_payload": {"features": ["coverage"]},
+    }
+
+
 def test_worker_settings_require_valid_concurrency(monkeypatch):
     settings = WorkerSettings.from_environment()
     assert settings.max_concurrency == 2