update

tests/test_metric_presenter.py

@@ -0,0 +1,68 @@
+from __future__ import annotations
+
+import subprocess
+from pathlib import Path
+
+
+REPO_ROOT = Path(__file__).resolve().parents[1]
+
+
+def _run_node(script: str) -> str:
+    """Execute a short Node.js script and return stdout."""
+    completed = subprocess.run(
+        ["node", "-e", script],
+        cwd=REPO_ROOT,
+        capture_output=True,
+        text=True,
+        encoding="utf-8",
+        check=True,
+    )
+    return completed.stdout.strip()
+
+
+def test_metric_presenter_applies_thresholds_and_noise_direction() -> None:
+    """MetricPresenter should centralize thresholds and inverse noise semantics."""
+    metric_js = (REPO_ROOT / "webapp" / "static" / "js" / "metric_presenter.js").as_posix()
+    script = f"""
+const fs = require("fs");
+const vm = require("vm");
+const code = fs.readFileSync("{metric_js}", "utf8");
+const sandbox = {{ window: {{}}, console }};
+vm.runInNewContext(code, sandbox);
+const p = sandbox.window.MetricPresenter;
+const result = {{
+  faith085: p.scoreClass("faithfulness", 0.85),
+  faith070: p.scoreClass("faithfulness", 0.70),
+  faith064: p.scoreClass("faithfulness", 0.64),
+  noise010: p.scoreClass("noise_sensitivity", 0.10),
+  noise030: p.scoreClass("noise_sensitivity", 0.30),
+  noise050: p.scoreClass("noise_sensitivity", 0.50),
+  desc: p.describeMetric("faithfulness"),
+  noiseDesc: p.describeMetric("noise_sensitivity"),
+  noiseBin: p.binColor("noise_sensitivity", 0.0),
+  faithBin: p.binColor("faithfulness", 0.8)
+}};
+console.log(JSON.stringify(result));
+"""
+    output = _run_node(script)
+    assert '"faith085":"good"' in output
+    assert '"faith070":"warn"' in output
+    assert '"faith064":"bad"' in output
+    assert '"noise010":"good"' in output
+    assert '"noise030":"warn"' in output
+    assert '"noise050":"bad"' in output
+    assert '"desc":"' in output
+    assert '"noiseDesc":"' in output
+    assert '"noiseBin":"#16a34a"' in output
+    assert '"faithBin":"#16a34a"' in output
+
+
+def test_report_and_index_load_metric_presenter_helper() -> None:
+    """The report page should use the shared helper for card descriptions and colors."""
+    index_html = (REPO_ROOT / "webapp" / "static" / "index.html").read_text(encoding="utf-8")
+    report_js = (REPO_ROOT / "webapp" / "static" / "js" / "report.js").read_text(encoding="utf-8")
+    app_js = (REPO_ROOT / "webapp" / "static" / "js" / "app.js").read_text(encoding="utf-8")
+
+    assert "js/metric_presenter.js" in index_html
+    assert "MetricPresenter.describeMetric" in report_js
+    assert "MetricPresenter.scoreClass" in app_js

tests/test_webapp_report_builder.py

@@ -88,3 +88,30 @@ def test_infer_metrics_excludes_weight_columns_without_snapshot(tmp_path: Path)
     )
 
     assert _infer_metrics_from_scores(run_dir) == ["faithfulness"]
+
+
+def test_build_report_ranks_noise_sensitivity_with_lower_values_as_better(tmp_path: Path) -> None:
+    """Lowest-sample review should treat higher noise sensitivity as worse."""
+    run_dir = tmp_path / "run"
+    run_dir.mkdir(parents=True, exist_ok=True)
+    (run_dir / "scores.csv").write_text(
+        "\n".join(
+            [
+                "sample_id,question,noise_sensitivity",
+                "s-good,q1,0.10",
+                "s-warn,q2,0.30",
+                "s-bad,q3,0.90",
+            ]
+        ),
+        encoding="utf-8",
+    )
+    (run_dir / "summary.md").write_text("summary", encoding="utf-8")
+    (run_dir / "optimization_advice.md").write_text("", encoding="utf-8")
+
+    report = build_report(run_dir, ["noise_sensitivity"])
+
+    assert [sample.sample_id for sample in report.lowest_samples[:3]] == [
+        "s-bad",
+        "s-warn",
+        "s-good",
+    ]

tests/webapp/test_profile_manager.py

@@ -1,4 +1,6 @@
 import pytest
+from unittest.mock import sentinel
+
 from webapp.models import LLMProfile, ProfileApplyRequest, ProfileApplyResponse
 
 def test_llm_profile_defaults():
@@ -147,3 +149,57 @@ def test_resolve_openai_client_kwargs_falls_back_to_env(tmp_path, monkeypatch):
     assert kwargs["api_key"] == "sk-env"
     assert kwargs["base_url"] == "http://env-base/v1"
     assert kwargs["timeout"] == 45.0
+
+
+def test_build_models_uses_high_default_max_tokens_for_structured_judge(monkeypatch):
+    """Structured RAGAS judge calls should use a larger completion budget by default."""
+    import rag_eval.metrics.factory as factory
+    from rag_eval.settings import EvaluationSettings
+
+    captured: dict[str, object] = {}
+
+    def fake_llm_factory(model, client=None, **kwargs):
+        captured["model"] = model
+        captured["client"] = client
+        captured["kwargs"] = kwargs
+        return sentinel.llm
+
+    monkeypatch.setattr(factory, "AsyncOpenAI", lambda **kwargs: sentinel.client)
+    monkeypatch.setattr(factory, "llm_factory", fake_llm_factory)
+    monkeypatch.setattr(factory, "embedding_factory", lambda **kwargs: sentinel.embeddings)
+
+    llm, embeddings = factory.build_models(
+        "gpt-5",
+        "text-embedding-3-small",
+        EvaluationSettings(),
+    )
+
+    assert llm is sentinel.llm
+    assert embeddings is sentinel.embeddings
+    assert captured["model"] == "gpt-5"
+    assert captured["client"] is sentinel.client
+    assert captured["kwargs"] == {"max_tokens": 4096}
+
+
+def test_build_models_allows_env_override_for_judge_max_tokens(monkeypatch):
+    """Operators should be able to raise the judge completion budget via settings."""
+    import rag_eval.metrics.factory as factory
+    from rag_eval.settings import EvaluationSettings
+
+    captured: dict[str, object] = {}
+
+    def fake_llm_factory(model, client=None, **kwargs):
+        captured["kwargs"] = kwargs
+        return sentinel.llm
+
+    monkeypatch.setattr(factory, "AsyncOpenAI", lambda **kwargs: sentinel.client)
+    monkeypatch.setattr(factory, "llm_factory", fake_llm_factory)
+    monkeypatch.setattr(factory, "embedding_factory", lambda **kwargs: sentinel.embeddings)
+
+    factory.build_models(
+        "gpt-5",
+        "text-embedding-3-small",
+        EvaluationSettings(RAGAS_LLM_MAX_TOKENS=8192),
+    )
+
+    assert captured["kwargs"] == {"max_tokens": 8192}