diff --git a/.github/workflows/tests.yml b/.github/workflows/tests.yml
new file mode 100644
index 0000000..ec9c6e5
--- /dev/null
+++ b/.github/workflows/tests.yml
@@ -0,0 +1,28 @@
+name: tests
+
+on:
+  push:
+  pull_request:
+
+jobs:
+  pytest:
+    runs-on: ubuntu-latest
+
+    steps:
+      - name: Checkout
+        uses: actions/checkout@v4
+
+      - name: Set up Python
+        uses: actions/setup-python@v5
+        with:
+          python-version: "3.12"
+
+      - name: Install dependencies
+        run: |
+          python -m pip install --upgrade pip
+          pip install -r requirements.txt
+          pip install pytest
+
+      - name: Run tests
+        run: |
+          pytest -q
\ No newline at end of file
diff --git a/pytest.ini b/pytest.ini
new file mode 100644
index 0000000..5ef52eb
--- /dev/null
+++ b/pytest.ini
@@ -0,0 +1,4 @@
+[pytest]
+testpaths = tests
+pythonpath = .
+addopts = -q
\ No newline at end of file
diff --git a/sample_data/eval_cases.json b/sample_data/eval_cases.json
new file mode 100644
index 0000000..65c11b2
--- /dev/null
+++ b/sample_data/eval_cases.json
@@ -0,0 +1,37 @@
+[
+  {
+    "id": "E-001",
+    "title": "Uncapped liability + consequential",
+    "text": "The Vendor shall be liable for all damages including consequential, indirect, special damages. Liability is uncapped.",
+    "expected_risk_level": "High",
+    "expected_categories": ["Liability"]
+  },
+  {
+    "id": "E-002",
+    "title": "Termination for convenience no cure",
+    "text": "Either party may terminate for convenience at any time upon written notice. No cure period is provided.",
+    "expected_risk_level": "Medium",
+    "expected_categories": ["Termination"]
+  },
+  {
+    "id": "E-003",
+    "title": "Neutral baseline",
+    "text": "This agreement describes services and payment terms. Standard confidentiality applies. No special liability clauses are stated.",
+    "expected_risk_level": "Low",
+    "expected_categories": []
+  },
+  {
+    "id": "E-004",
+    "title": "Both liability + termination",
+    "text": "Either party may terminate for convenience at any time upon notice. Vendor is liable for all damages including consequential. Liability is uncapped.",
+    "expected_risk_level": "High",
+    "expected_categories": ["Liability", "Termination"]
+  },
+  {
+    "id": "E-005",
+    "title": "Cure period present (should reduce termination signal)",
+    "text": "Either party may terminate upon material breach if not cured within 30 days after notice. Termination for convenience is not permitted.",
+    "expected_risk_level": "Low",
+    "expected_categories": []
+  }
+]
\ No newline at end of file
diff --git a/src/evaluation_stub.py b/src/evaluation_stub.py
index a152622..967745e 100644
--- a/src/evaluation_stub.py
+++ b/src/evaluation_stub.py
@@ -1,33 +1,152 @@
 from __future__ import annotations
+
+import json
+from dataclasses import dataclass
 from pathlib import Path
+from typing import Any, Dict, List, Set, Tuple
+
+from src.model_adapter import analyze_contract
+
+
+@dataclass
+class EvalCase:
+    id: str
+    title: str
+    text: str
+    expected_risk_level: str
+    expected_categories: List[str]
+
+
+def _normalize_level(level: str) -> str:
+    return str(level).strip().title()
+
+
+def _predicted_categories(result_dict: Dict[str, Any]) -> Set[str]:
+    findings = result_dict.get("findings", [])
+    cats = {f.get("category", "").strip() for f in findings if f.get("category")}
+    return {c for c in cats if c}
+
+
+def run_evaluation(cases: List[EvalCase]) -> Dict[str, Any]:
+    """
+    Runs evaluation cases through the adapter, compares predicted vs expected.
+
+    Returns a dict:
+    - risk_level_accuracy
+    - category_precision/recall/f1 (micro)
+    - mismatches[]
+    """
+    if not cases:
+        return {
+            "risk_level_accuracy": 0.0,
+            "category_precision": 0.0,
+            "category_recall": 0.0,
+            "category_f1": 0.0,
+            "mismatches": [],
+            "n": 0,
+        }
+
+    level_hits = 0
+    tp = fp = fn = 0
+    mismatches: List[Dict[str, Any]] = []
+
+    for c in cases:
+        result = analyze_contract(c.text, title=c.title, source_type="paste").model_dump()
 
-from .model_adapter import analyze_contract
+        pred_level = _normalize_level(result["summary"]["risk_level"])
+        exp_level = _normalize_level(c.expected_risk_level)
 
+        if pred_level == exp_level:
+            level_hits += 1
 
-SAMPLES_DIR = Path("sample_data")
+        pred_cats = _predicted_categories(result)
+        exp_cats = set(c.expected_categories)
+
+        # micro counts
+        tp += len(pred_cats & exp_cats)
+        fp += len(pred_cats - exp_cats)
+        fn += len(exp_cats - pred_cats)
+
+        if pred_level != exp_level or pred_cats != exp_cats:
+            mismatches.append(
+                {
+                    "id": c.id,
+                    "title": c.title,
+                    "pred_risk_level": pred_level,
+                    "exp_risk_level": exp_level,
+                    "pred_categories": sorted(list(pred_cats)),
+                    "exp_categories": sorted(list(exp_cats)),
+                    "run_id": result.get("run_id"),
+                }
+            )
+
+    accuracy = level_hits / len(cases)
+
+    precision = tp / (tp + fp) if (tp + fp) else 0.0
+    recall = tp / (tp + fn) if (tp + fn) else 0.0
+    f1 = (2 * precision * recall / (precision + recall)) if (precision + recall) else 0.0
+
+    return {
+        "risk_level_accuracy": round(accuracy, 3),
+        "category_precision": round(precision, 3),
+        "category_recall": round(recall, 3),
+        "category_f1": round(f1, 3),
+        "mismatches": mismatches,
+        "n": len(cases),
+    }
+
+
+def load_cases(path: str | Path) -> List[EvalCase]:
+    p = Path(path)
+    payload = json.loads(p.read_text(encoding="utf-8"))
+
+    cases: List[EvalCase] = []
+    for row in payload:
+        cases.append(
+            EvalCase(
+                id=row["id"],
+                title=row.get("title", row["id"]),
+                text=row["text"],
+                expected_risk_level=row["expected_risk_level"],
+                expected_categories=row.get("expected_categories", []),
+            )
+        )
+    return cases
 
 
 def main() -> None:
-    samples = sorted(SAMPLES_DIR.glob("*.txt"))
-    if not samples:
-        print("No sample contracts found in sample_data/. Add .txt files to run evaluation.")
-        return
-
-    scores = []
-    for fp in samples:
-        contract_text = fp.read_text(encoding="utf-8", errors="ignore")
-        result = analyze_contract(contract_text=contract_text, title=fp.stem, source_type="sample_data")
-
-        score = result.summary.overall_risk_score
-        level = result.summary.risk_level
-        findings_count = len(result.findings)
-
-        scores.append(score)
-        print(f"{fp.name}: score={score}, level={level}, findings={findings_count}")
-
-    if scores:
-        print("\nSummary")
-        print(f"count={len(scores)}  min={min(scores)}  max={max(scores)}  avg={sum(scores)/len(scores):.2f}")
+    import argparse
+
+    parser = argparse.ArgumentParser(description="Offline evaluation harness (stub).")
+    parser.add_argument(
+        "path",
+        nargs="?",
+        default="sample_data/eval_cases.json",
+        help="Path to evaluation cases JSON (default: sample_data/eval_cases.json)",
+    )
+    args = parser.parse_args()
+
+    cases = load_cases(args.path)
+    metrics = run_evaluation(cases)
+
+    print("\n=== Evaluation Summary ===")
+    print(f"Cases: {metrics['n']}")
+    print(f"Risk-level accuracy: {metrics['risk_level_accuracy']}")
+    print(
+        "Category micro P/R/F1: "
+        f"{metrics['category_precision']} / {metrics['category_recall']} / {metrics['category_f1']}"
+    )
+
+    if metrics["mismatches"]:
+        print("\n=== Mismatches ===")
+        for m in metrics["mismatches"]:
+            print(
+                f"- {m['id']} | {m['title']} | "
+                f"level {m['pred_risk_level']} vs {m['exp_risk_level']} | "
+                f"cats {m['pred_categories']} vs {m['exp_categories']}"
+            )
+    else:
+        print("\nNo mismatches 🎉")
 
 
 if __name__ == "__main__":
diff --git a/tests/test_evaluation_harness.py b/tests/test_evaluation_harness.py
new file mode 100644
index 0000000..fcff3bf
--- /dev/null
+++ b/tests/test_evaluation_harness.py
@@ -0,0 +1,7 @@
+from src.evaluation_stub import load_cases, run_evaluation
+
+def test_eval_harness_loads_and_runs():
+    cases = load_cases("sample_data/eval_cases.json")
+    metrics = run_evaluation(cases)
+    assert "risk_level_accuracy" in metrics
+    assert metrics["n"] > 0
\ No newline at end of file
diff --git a/tests/test_evidence_rule.py b/tests/test_evidence_rule.py
new file mode 100644
index 0000000..5876307
--- /dev/null
+++ b/tests/test_evidence_rule.py
@@ -0,0 +1,13 @@
+from src.model_adapter import analyze_contract
+
+def test_all_findings_have_evidence():
+    text = "Vendor is liable for all damages including consequential. Liability is uncapped."
+    result = analyze_contract(text, title="Test", source_type="paste").model_dump()
+    findings = result.get("findings", [])
+    assert len(findings) > 0
+
+    for f in findings:
+        ev = f.get("evidence", [])
+        assert isinstance(ev, list)
+        assert len(ev) >= 1
+        assert ev[0].get("snippet")
\ No newline at end of file
diff --git a/tests/test_scoring_monotonic.py b/tests/test_scoring_monotonic.py
new file mode 100644
index 0000000..2929426
--- /dev/null
+++ b/tests/test_scoring_monotonic.py
@@ -0,0 +1,32 @@
+from src.schemas import Finding, Evidence
+from src.scoring import compute_score
+
+def test_high_severity_scores_higher_than_low():
+    low = [
+        Finding(
+            finding_id="T-LOW",
+            category="General",
+            risk_statement="Low risk",
+            severity="Low",
+            confidence=1.0,
+            evidence=[Evidence(clause_ref="X", snippet="...")],
+            recommendation="N/A",
+        )
+    ]
+
+    high = [
+        Finding(
+            finding_id="T-HIGH",
+            category="Liability",
+            risk_statement="High risk",
+            severity="High",
+            confidence=1.0,
+            evidence=[Evidence(clause_ref="Y", snippet="...")],
+            recommendation="N/A",
+        )
+    ]
+
+    s_low, _ = compute_score(low)
+    s_high, _ = compute_score(high)
+
+    assert s_high > s_low
\ No newline at end of file