add prototype/test_cluster.py

prototype/test_cluster.py

@@ -0,0 +1,119 @@
+"""Tests for the story-layer clustering.
+
+Acceptance (per plan): two reworded copies of the same story -> ONE cluster.
+Also: exact-duplicate fast path, and distinct stories stay separate.
+"""
+import time
+
+import pytest
+
+from cluster import Clusterer, Item
+
+
+def _item(title, url, source="src", summary="body", published=None):
+    return Item(guid=url, url=url, title=title, summary=summary,
+                source=source, published=published or time.time())
+
+
+def _clustered(items):
+    return Clusterer().cluster(items)
+
+
+def test_two_rewrites_merge_into_one_cluster():
+    """The core accept: BBC + NPR reworded coverage of the same crash -> 1 story.
+
+    Wire-syndication pattern: divergent lede/headline, but the body largely
+    re-uses the same wire text (4 shared sentences).
+    """
+    core = ("The two crew members on board were killed. "
+            "Investigators said the plane lost contact with the tower shortly "
+            "before impact. Authorities have opened an inquiry into the cause. "
+            "Air traffic around Miami Beach was halted for hours after the crash.")
+    items = [
+        _item("Cargo Plane Crashes Near Miami Beach Killing Two People",
+              "https://bbc.example/cargo-crash",
+              source="BBC",
+              summary="A cargo aircraft went down close to Miami Beach on Tuesday "
+                      "morning. " + core),
+        _item("Two dead after cargo jet crashes outside Miami",
+              "https://npr.example/miami-crash",
+              source="NPR",
+              summary="Emergency crews responded to the wreckage near Miami Beach. "
+                      + core),
+    ]
+    stories = _clustered(items)
+    assert len(stories) == 1, f"expected 1 cluster, got {len(stories)}"
+    assert stories[0].coverage_count == 2
+
+
+def test_exact_duplicate_title_dedupes_without_similarity():
+    items = [
+        _item("OpenAI Sued Over Training Data", "https://tc.example/openai-suit",
+              source="TechCrunch"),
+        _item("OpenAI Sued Over Training Data", "https://verge.example/openai-suit-2",
+              source="The Verge"),
+        _item("OpenAI Sued Over Training Data", "https://mirror.example/openai-suit-3",
+              source="Mirror"),
+    ]
+    stories = _clustered(items)
+    assert len(stories) == 1
+    assert stories[0].coverage_count == 3
+
+
+def test_unrelated_stories_stay_separate():
+    items = [
+        _item("Apple unveils new iPhone at launch event", "https://a.example/1",
+              summary="Apple took the stage in Cupertino to reveal the next iPhone."),
+        _item("Hurricane forecast calls for record storm surge", "https://a.example/2",
+              summary="Meteorologists upgraded the storm to a category five system."),
+        _item("Local bakery wins national award for sourdough", "https://a.example/3",
+              summary="Judges praised the long fermentation and crust texture."),
+    ]
+    stories = _clustered(items)
+    assert len(stories) == 3
+
+
+def test_chain_merging_transitive():
+    """A≈B and B≈C => all three one cluster, even if A≉C."""
+    core = ("The two crew members on board were killed. Authorities have opened "
+            "an inquiry into the cause. Air traffic around Miami Beach was halted "
+            "for hours after the crash.")
+    items = [
+        _item("Cargo plane crash near Miami leaves two dead", "https://bbc.example/x",
+              source="BBC", summary="The aircraft went down on Tuesday morning. " + core),
+        _item("Two dead as cargo aircraft crashes by Miami beach", "https://npr.example/x",
+              source="NPR", summary="Emergency crews reached the wreckage. " + core),
+        _item("Miami cargo crash kills both pilots on board", "https://wp.example/x",
+              source="WaPo", summary="The plane came down just off the beach. " + core),
+    ]
+    stories = _clustered(items)
+    assert len(stories) == 1
+    assert stories[0].coverage_count == 3
+
+
+def test_ranking_prefers_high_coverage_and_freshness():
+    from cluster import compute_scores
+    now = time.time()
+    items = [
+        _item("Viral story covered everywhere", "https://a.example/hot", published=now - 1000),
+        _item("Viral story covered everywhere", "https://b.example/hot-2", published=now - 900),
+        _item("Viral story covered everywhere", "https://c.example/hot-3", published=now - 800),
+        _item("Niche old post", "https://d.example/niche", published=now - 86400 * 3),
+    ]
+    stories = _clustered(items)
+    assert len(stories) == 2
+    scored = compute_scores(stories, time.time())
+    # 3-source fresh story outranks the single-source old one
+    assert scored[0][1].coverage_count == 3
+
+
+def test_summary_field_capped_input():
+    """LLM lead text is headline + ~300 chars, never the long tail (defended in code)."""
+    from cluster import llm_summarize_topk
+    import os
+    os.environ.pop("LLM_BASE_URL", None)
+    # passes without error even with no LLM backend (network -> skipped/strange localhost)
+    # we only assert the display cap helper field exists on Story
+    items = [_item("Test", "https://a.example/t")]
+    s = _clustered(items)[0]
+    assert hasattr(s, "summary")