feat(config): temperature=0 for reproducible output (Phase 0, T-082)

config.example.yaml

@@ -16,6 +16,10 @@ api:
   # and has a vision-capable model. REQUIRED — edit before first run.
   base_url: "https://your-llm-endpoint.example/v1"
   model: "your-vision-model-name"
+  # Sampling temperature for every model call. 0 (the default) gives reproducible
+  # output, which is what makes "did this change help?" answerable. Raise only if
+  # you deliberately want variation.
+  temperature: 0
   # OPTIONAL cost estimation. Token usage is always reported; a monetary estimate
   # is only added when BOTH prices below are set (per single token — the unit most
   # OpenAI-compatible /v1/models endpoints use). Provider-neutral: look up your
@@ -90,7 +94,7 @@ diagrams:
 
     Strukturera svaret så här:
     1. Vad diagrammet visar: titel/ämne och vad x- respektive y-axeln mäter (inklusive enheter).
-    2. Vilka serier/data: nämn alla serier som visas (linjer, staplar, etc.) och hur de skiljs åt visuellt.
+    2. Vilka serier/data: nämn alla serier som visas (linjer, staplar, etc.) och hur de skiljs åt visuellt. Beskriv varje serie med hur den utvecklas, helst med konkreta siffror.
     3. De viktigaste observationerna: trender, toppar, bottnar, brytpunkter, divergenser. Var konkret med siffror och årtal.
     4. Eventuella anmärkningar och källor om de finns med i bilden.
 

src/figmark/config.py

@@ -29,6 +29,9 @@ class ApiConfig:
     input_token_price: float | None = None
     output_token_price: float | None = None
     currency: str | None = None
+    # Sampling temperature for every model call. Defaults to 0 for reproducible
+    # output — the prerequisite for measuring whether a change helped (T-082).
+    temperature: float = 0.0
 
 
 @dataclass
@@ -214,6 +217,7 @@ def load_config(config_path: str | Path = "config.yaml") -> Config:
         input_token_price=in_price,
         output_token_price=out_price,
         currency=(str(api_raw["currency"]).strip() if api_raw.get("currency") else None),
+        temperature=float(api_raw.get("temperature", 0.0) or 0.0),
     )
 
     input_raw = raw.get("input") or {}

src/figmark/describe.py

@@ -198,6 +198,7 @@ def describe_image(
         try:
             response = client.chat.completions.create(
                 model=cfg.api.model,
+                temperature=cfg.api.temperature,
                 max_tokens=MAX_TOKENS,
                 messages=[
                     {

src/figmark/diagrams.py

@@ -466,6 +466,7 @@ def describe_diagram(
 
     response = client.chat.completions.create(
         model=cfg.api.model,
+        temperature=cfg.api.temperature,
         max_tokens=MAX_TOKENS,
         messages=[
             {

src/figmark/ocr.py

@@ -192,6 +192,7 @@ def ocr_page_with_vision(
     try:
         response = client.chat.completions.create(
             model=cfg.api.model,
+            temperature=cfg.api.temperature,
             max_tokens=OCR_MAX_TOKENS,
             messages=[
                 {

src/figmark/summarize.py

@@ -48,6 +48,7 @@ def detect_language(client, pages, cfg: Config, cache_path: Path) -> str:
 
     response = client.chat.completions.create(
         model=cfg.api.model,
+        temperature=cfg.api.temperature,
         max_tokens=LANG_DETECT_MAX_TOKENS,
         messages=[{"role": "user", "content": f"{LANG_DETECT_PROMPT}\n\nText:\n{sample}"}],
     )
@@ -115,6 +116,7 @@ def summarize_document(client, pages, cfg: Config, cache_path: Path, language: s
     )
     response = client.chat.completions.create(
         model=cfg.api.model,
+        temperature=cfg.api.temperature,
         max_tokens=SUMMARY_MAX_TOKENS,
         messages=[{"role": "user", "content": user_text}],
     )

tests/test_summarize.py

@@ -59,7 +59,7 @@ def _stub_client(text: str, finish_reason: str = "stop"):
 
 
 _CFG = SimpleNamespace(
-    api=SimpleNamespace(model="test-model"),
+    api=SimpleNamespace(model="test-model", temperature=0.0),
     document_summary=SimpleNamespace(sample_words=50, prompt="Summarise."),
 )