fix: prevent truncated Ollama answers

CHANGELOG.md

@@ -22,6 +22,8 @@
   refuses to infer unavailable values such as water volume.
 - Added editable Unraid environment/template settings and a Docker host-gateway
   mapping for the Ollama service running on the server.
+- Added an explicit 16,384-token Ollama context window and reject truncated
+  `done_reason=length` responses instead of showing an incomplete answer.
 
 ## Sprint 201 Semantic area-selection metrics (2026-07-15)
 

backend/README.md

@@ -1100,6 +1100,7 @@ OLLAMA_BASE_URL=http://host.docker.internal:11434
 OLLAMA_DEFAULT_MODEL=qwen3.5:9b
 OLLAMA_TIMEOUT_SECONDS=120
 OLLAMA_MAX_OUTPUT_TOKENS=700
+OLLAMA_CONTEXT_TOKENS=16384
 ```
 
 The Unraid deployment adds `host.docker.internal:host-gateway` automatically.
@@ -1107,7 +1108,9 @@ Verify the connection with `GET /api/v1/assistant/status`, inspect installed
 models with `GET /api/v1/assistant/models` and ask a grounded question through
 `POST /api/v1/projects/{project_id}/assistant/query`. A requested model must be
 present in Ollama `/api/tags`. Missing water depth/bathymetry remains explicit;
-the assistant cannot turn 2D water geometry into volume.
+the assistant cannot turn 2D water geometry into volume. GeoIntel rejects an
+answer when Ollama reports `done_reason=length`, so a visibly truncated sentence
+is never presented as a complete result.
 
 ## Helpful repository scripts
 

backend/app/core/config.py

@@ -51,6 +51,7 @@ class Settings(BaseSettings):
     ollama_default_model: str = Field(default="qwen3.5:9b", validation_alias="OLLAMA_DEFAULT_MODEL")
     ollama_timeout_seconds: int = Field(default=120, ge=5, le=600, validation_alias="OLLAMA_TIMEOUT_SECONDS")
     ollama_max_output_tokens: int = Field(default=700, ge=100, le=4_000, validation_alias="OLLAMA_MAX_OUTPUT_TOKENS")
+    ollama_context_tokens: int = Field(default=16_384, ge=4_096, le=131_072, validation_alias="OLLAMA_CONTEXT_TOKENS")
     cors_origins: list[str] | str = Field(
         default=["http://localhost:5173", "http://127.0.0.1:5173"],
         validation_alias="CORS_ORIGINS",

backend/app/services/geo_assistant_service.py

@@ -367,9 +367,23 @@ class GeoAssistantService:
                 "stream": False,
                 "think": False,
                 "keep_alive": "10m",
-                "options": {"temperature": 0.1, "num_predict": self.settings.ollama_max_output_tokens},
+                "options": {
+                    "temperature": 0.1,
+                    "num_ctx": self.settings.ollama_context_tokens,
+                    "num_predict": self.settings.ollama_max_output_tokens,
+                },
             },
         )
+        if response.get("done_reason") == "length":
+            raise AppError(
+                code="OLLAMA_RESPONSE_TRUNCATED",
+                message="Ollama kon geen volledig antwoord binnen de ingestelde contextlimiet genereren.",
+                details={
+                    "context_tokens": self.settings.ollama_context_tokens,
+                    "max_output_tokens": self.settings.ollama_max_output_tokens,
+                },
+                status_code=502,
+            )
         message = response.get("message") if isinstance(response.get("message"), dict) else {}
         answer = str(message.get("content") or "").strip()
         if not answer:

backend/tests/test_sprint202_temporal_metrics_and_ollama.py

@@ -155,11 +155,57 @@ def test_geo_assistant_sends_grounded_context_without_thinking_trace(monkeypatch
     assert captured["path"] == "/api/chat"
     assert captured["payload"]["stream"] is False
     assert captured["payload"]["think"] is False
+    assert captured["payload"]["options"]["num_ctx"] == 16_384
     assert "Gebruik uitsluitend feiten en cijfers uit CONTEXT_JSON" in captured["payload"]["messages"][0]["content"]
     assert "scope.label is het exact geanalyseerde gebied" in captured["payload"]["messages"][0]["content"]
     assert "water_volume_available" in captured["payload"]["messages"][0]["content"]
 
 
+def test_geo_assistant_rejects_truncated_ollama_answer(monkeypatch) -> None:
+    service = GeoAssistantService(ollama_settings())
+    monkeypatch.setattr(service, "list_models", lambda: [AssistantModelRead(name="qwen3.5:9b")])
+    monkeypatch.setattr(
+        service,
+        "_build_context",
+        lambda *args, **kwargs: (
+            {"scope": {"label": "Gemeente Mol"}},
+            [],
+            [],
+            [],
+            [],
+            "Gemeente Mol",
+        ),
+    )
+    monkeypatch.setattr(
+        service,
+        "_request_json",
+        lambda path, payload=None: {
+            "done": True,
+            "done_reason": "length",
+            "message": {"role": "assistant", "content": "Een onvolledige zin"},
+        },
+    )
+
+    with pytest.raises(AppError) as exc_info:
+        service.query(
+            object(),
+            project_id=uuid4(),
+            payload=AssistantQueryRequest(question="Hoe evolueerde Mol?"),
+        )
+
+    assert exc_info.value.code == "OLLAMA_RESPONSE_TRUNCATED"
+
+
+def test_unraid_ollama_context_window_is_configurable() -> None:
+    compose = (ROOT / "docker-compose.unraid.yml").read_text(encoding="utf-8")
+    template = (ROOT / "deploy/unraid/geointel-unraid-template.xml").read_text(encoding="utf-8")
+    env_example = (ROOT / "deploy/unraid/geointel.env.example").read_text(encoding="utf-8")
+
+    assert "OLLAMA_CONTEXT_TOKENS: ${OLLAMA_CONTEXT_TOKENS:-16384}" in compose
+    assert 'Target="OLLAMA_CONTEXT_TOKENS"' in template
+    assert "OLLAMA_CONTEXT_TOKENS=16384" in env_example
+
+
 def test_temporal_comparison_preserves_all_compatible_semantic_metrics() -> None:
     earlier = {
         "metrics": [

deploy/unraid/README.md

@@ -201,6 +201,7 @@ OLLAMA_BASE_URL=http://host.docker.internal:11434
 OLLAMA_DEFAULT_MODEL=qwen3.5:9b
 OLLAMA_TIMEOUT_SECONDS=120
 OLLAMA_MAX_OUTPUT_TOKENS=700
+OLLAMA_CONTEXT_TOKENS=16384
 ```
 
 The model dropdown comes from Ollama `/api/tags`, so changing the installed

deploy/unraid/geointel-unraid-template.xml

@@ -38,4 +38,5 @@
   <Config Name="Ollama Base URL" Target="OLLAMA_BASE_URL" Default="http://host.docker.internal:11434" Mode="" Description="Ollama API reachable from the container. The deployment maps host.docker.internal to the Unraid host gateway." Type="Variable" Display="always" Required="true" Mask="false">http://host.docker.internal:11434</Config>
   <Config Name="Default Ollama Model" Target="OLLAMA_DEFAULT_MODEL" Default="qwen3.5:9b" Mode="" Description="Preferred locally installed Ollama model. Users can select another installed model in GeoIntel." Type="Variable" Display="always" Required="true" Mask="false">qwen3.5:9b</Config>
   <Config Name="Ollama Timeout Seconds" Target="OLLAMA_TIMEOUT_SECONDS" Default="120" Mode="" Description="Maximum wait for one local assistant response." Type="Variable" Display="advanced" Required="true" Mask="false">120</Config>
+  <Config Name="Ollama Context Tokens" Target="OLLAMA_CONTEXT_TOKENS" Default="16384" Mode="" Description="Context window reserved for grounded GIS measurements and the generated answer." Type="Variable" Display="advanced" Required="true" Mask="false">16384</Config>
 </Container>

deploy/unraid/geointel.env.example

@@ -56,3 +56,4 @@ OLLAMA_BASE_URL=http://host.docker.internal:11434
 OLLAMA_DEFAULT_MODEL=qwen3.5:9b
 OLLAMA_TIMEOUT_SECONDS=120
 OLLAMA_MAX_OUTPUT_TOKENS=700
+OLLAMA_CONTEXT_TOKENS=16384

deploy/unraid/run-dockerman-container.sh

@@ -45,6 +45,7 @@ OLLAMA_BASE_URL="${OLLAMA_BASE_URL:-http://host.docker.internal:11434}"
 OLLAMA_DEFAULT_MODEL="${OLLAMA_DEFAULT_MODEL:-qwen3.5:9b}"
 OLLAMA_TIMEOUT_SECONDS="${OLLAMA_TIMEOUT_SECONDS:-120}"
 OLLAMA_MAX_OUTPUT_TOKENS="${OLLAMA_MAX_OUTPUT_TOKENS:-700}"
+OLLAMA_CONTEXT_TOKENS="${OLLAMA_CONTEXT_TOKENS:-16384}"
 
 install_dockerman_metadata() {
   if [ -d /boot/config/plugins/dockerMan ]; then
@@ -120,6 +121,7 @@ docker run -d \
   -e OLLAMA_DEFAULT_MODEL="$OLLAMA_DEFAULT_MODEL" \
   -e OLLAMA_TIMEOUT_SECONDS="$OLLAMA_TIMEOUT_SECONDS" \
   -e OLLAMA_MAX_OUTPUT_TOKENS="$OLLAMA_MAX_OUTPUT_TOKENS" \
+  -e OLLAMA_CONTEXT_TOKENS="$OLLAMA_CONTEXT_TOKENS" \
   -v "${GEOINTEL_POSTGIS_DATA_PATH}:/var/lib/postgresql/data" \
   -v "${GEOINTEL_STORAGE_PATH}:/app/storage" \
   -v "${GEOINTEL_MODELS_PATH}:/app/models" \

docker-compose.unraid.yml

@@ -43,6 +43,7 @@ services:
       OLLAMA_DEFAULT_MODEL: ${OLLAMA_DEFAULT_MODEL:-qwen3.5:9b}
       OLLAMA_TIMEOUT_SECONDS: ${OLLAMA_TIMEOUT_SECONDS:-120}
       OLLAMA_MAX_OUTPUT_TOKENS: ${OLLAMA_MAX_OUTPUT_TOKENS:-700}
+      OLLAMA_CONTEXT_TOKENS: ${OLLAMA_CONTEXT_TOKENS:-16384}
     ports:
       - "${GEOINTEL_FRONTEND_PORT:-1202}:80"
     volumes:

docker-compose.yml

@@ -48,6 +48,7 @@ services:
       OLLAMA_DEFAULT_MODEL: ${OLLAMA_DEFAULT_MODEL:-qwen3.5:9b}
       OLLAMA_TIMEOUT_SECONDS: ${OLLAMA_TIMEOUT_SECONDS:-120}
       OLLAMA_MAX_OUTPUT_TOKENS: ${OLLAMA_MAX_OUTPUT_TOKENS:-700}
+      OLLAMA_CONTEXT_TOKENS: ${OLLAMA_CONTEXT_TOKENS:-16384}
     ports:
       - "${GEOINTEL_BACKEND_PORT:-8000}:8000"
     volumes:

docs/API_CONTRACTS.md

@@ -1637,4 +1637,7 @@ metrics from PostGIS and includes dated observations only for persisted
 temporal series. Geometry is not sent to Ollama. The response contains the
 answer, used model, scope label, context metrics, discovered temporal series,
 source dataset ids and warnings. Missing measurements remain unavailable;
-specifically, no water volume is inferred from 2D water geometry.
+specifically, no water volume is inferred from 2D water geometry. The backend
+sets an explicit Ollama context window and returns
+`OLLAMA_RESPONSE_TRUNCATED` instead of accepting a response with
+`done_reason=length` as a complete answer.

docs/CODEX_EXECUTION_LOG.md

@@ -8389,7 +8389,7 @@ Validation evidence:
 - Focused temporal, source, Ollama, navigation and Unraid regression tests
   passed, including direct coverage for multi-metric history and DockerMan host
   mapping.
-- Full readiness passed 615 backend tests, backend compilation, 91 documented
+- Full readiness passed 617 backend tests, backend compilation, 91 documented
   API routes, one Alembic head, frontend TypeScript typecheck/build and all
   shell syntax gates.
 - Tower deployment passed the all-in-one container health check, live PostGIS
@@ -8403,6 +8403,15 @@ Validation evidence:
   message keys now use a session-local monotonic id generator; no persisted or
   security-sensitive identity depends on it. Typecheck, production build and
   full readiness reran successfully.
+- The regional operator then reused the five governed forest snapshots and
+  imported 15 new authoritative snapshots: water, built functions and
+  transport for 2013, 2016, 2019, 2022 and 2025. A live exact-Area Mol water
+  comparison returned 606.16 ha in 2013 and 641.75 ha in 2025 across all five
+  observations; the API retained the 10 m raster and object-lineage warnings.
+- Browser QA also reproduced an Ollama `done_reason=length`: the previous
+  implicit 4096-token context left only 86 tokens after a 4010-token grounded
+  prompt. GeoIntel now requests a configurable 16,384-token context and rejects
+  any future length-truncated response instead of presenting a partial answer.
 
 Known limitations:
 - Water volume remains unavailable until a governed depth/bathymetry source is