Extraction: model in config, chatter gate, geocode cache (server-26#298, server-26#104, server-26#88)

EXTRACTION_MODEL / EMBEDDING_MODEL replace the gpt-4o-mini and
text-embedding-3-small literals. gate_chatter (classify_chatter with roll_call
narrowed to whole-transcript roll calls) skips correction, extraction,
embedding and geocoding for housekeeping, writes skip_reason=non_event_chatter
and still runs the free clearance/plate backstops; upload.py's no-scene
fallback admits it like transcript_too_short. Ships OFF (CHATTER_GATE_ENABLED
=false): a gated call has no units, so it takes the no-fit-check thin attach,
and #127 requires a clean live shadow window first.
In-process TTL cache (24h, negatives 1h, 5000 entries, LRU) in front of the
Geocoding API, keyed by address+bounds; holds coordinates only, inside the
Maps 30-day lat/lng caching limit; errors and key/quota statuses not cached.
#104: the no-scene fallback passes the call doc's severity.
#88: resolved/reassignment parsed by _llm_bool, not bool().

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

drb-c2-core/tests/test_geocode_cache.py

@@ -0,0 +1,162 @@
+"""
+server-26#298 — geocode result cache in front of the Google Geocoding API,
+and server-26#160 — place verification gets an anchor from the talkgroup's
+own name when nobody configured one.
+"""
+from unittest.mock import AsyncMock, patch
+
+import pytest
+
+import app.internal.intelligence as intel
+from app.config import settings
+from app.internal import place_verifier as pv
+
+ANCHOR = {"lat": 41.16, "lng": -73.86, "radius_km": 10.0}
+
+
+class _Counter:
+    def __init__(self, payload=None, exc=None):
+        self.calls = 0
+        self.payload = payload
+        self.exc = exc
+
+    def patch(self):
+        outer = self
+
+        class _Resp:
+            def raise_for_status(self):
+                if outer.exc:
+                    raise outer.exc
+
+            def json(self):
+                return outer.payload
+
+        class _Client:
+            async def __aenter__(self): return self
+            async def __aexit__(self, *a): return False
+
+            async def get(self, *a, **k):
+                outer.calls += 1
+                return _Resp()
+
+        return patch("httpx.AsyncClient", lambda *a, **k: _Client())
+
+
+def _ok(lat=41.161, lng=-73.861):
+    return {"status": "OK", "results": [{"geometry": {
+        "location": {"lat": lat, "lng": lng}, "location_type": "ROOFTOP"}}]}
+
+
+@pytest.fixture(autouse=True)
+def _key():
+    with patch.object(settings, "google_maps_api_key", "test-key"):
+        yield
+
+
+@pytest.mark.asyncio
+async def test_repeat_address_is_one_request():
+    c = _Counter(_ok())
+    with c.patch():
+        a = await intel._geocode_location("12 Main Street, Ossining", anchor=ANCHOR)
+        b = await intel._geocode_location("12 main street, ossining ", anchor=ANCHOR)
+    assert a == b == {"lat": 41.161, "lng": -73.861, "location_type": "ROOFTOP"}
+    assert c.calls == 1
+    assert intel.geocode_cache_stats == {"hits": 1, "misses": 1}
+
+
+@pytest.mark.asyncio
+async def test_cached_answer_is_still_judged_against_the_caller_anchor():
+    c = _Counter(_ok())
+    with c.patch():
+        assert await intel._geocode_location("12 Main Street", anchor=ANCHOR)
+        far = {"lat": 0.0, "lng": 0.0, "radius_km": 10.0}
+        # Different bounds -> different key, so a separate request — never a
+        # cached answer leaking past a different area's distance check.
+        assert await intel._geocode_location("12 Main Street", anchor=far) is None
+    assert c.calls == 2
+
+
+@pytest.mark.asyncio
+async def test_zero_results_is_cached_but_errors_are_not():
+    c = _Counter({"status": "ZERO_RESULTS", "results": []})
+    with c.patch():
+        await intel._geocode_location("Cool Parts, Illinois", anchor=ANCHOR)
+        await intel._geocode_location("Cool Parts, Illinois", anchor=ANCHOR)
+    assert c.calls == 1
+
+    denied = _Counter({"status": "REQUEST_DENIED"})
+    with denied.patch():
+        await intel._geocode_location("Snowden Avenue", anchor=ANCHOR)
+        await intel._geocode_location("Snowden Avenue", anchor=ANCHOR)
+    assert denied.calls == 2, "a key/quota failure is not an answer about the address"
+
+    boom = _Counter(_ok(), exc=RuntimeError("503"))
+    with boom.patch():
+        await intel._geocode_location("Croton Point", anchor=ANCHOR)
+        await intel._geocode_location("Croton Point", anchor=ANCHOR)
+    assert boom.calls == 2
+
+
+@pytest.mark.asyncio
+async def test_ttl_expiry_and_size_bound():
+    c = _Counter(_ok())
+    with c.patch(), patch.object(settings, "geocode_cache_ttl_seconds", 0):
+        await intel._geocode_location("12 Main Street", anchor=ANCHOR)
+        await intel._geocode_location("12 Main Street", anchor=ANCHOR)
+    assert c.calls == 2, "TTL 0 disables the cache"
+
+    c = _Counter(_ok())
+    with c.patch(), patch.object(settings, "geocode_cache_max_entries", 2):
+        for street in ("A Street", "B Street", "C Street"):
+            await intel._geocode_location(street, anchor=ANCHOR)
+    assert len(intel._geocode_cache) == 2
+
+
+# ── server-26#160 ───────────────────────────────────────────────────────────
+
+@pytest.mark.asyncio
+async def test_unconfigured_channel_gets_an_anchor_from_its_talkgroup_name():
+    resolved = {"center": {"lat": 40.71, "lng": -74.0}, "radius_km": 25.0}
+    seen = {}
+
+    async def _geo(query, anchor):
+        seen["query"], seen["anchor"] = query, anchor
+        return {"lat": 40.7, "lng": -74.0}
+
+    with patch.object(pv.settings, "place_verification_enabled", True), \
+         patch.object(intel.fstore, "doc_get_cached", AsyncMock(return_value={"lat": 41.2, "lon": -73.8})), \
+         patch.object(intel, "_get_node_state", AsyncMock(return_value="New York")), \
+         patch.object(pv.area_context, "resolve_anchor", AsyncMock(return_value=resolved)) as res, \
+         patch.object(pv, "_geocode_in_anchor", _geo):
+        await pv.verify("c1", "Respond to Lexington Avenue", None, ["Lexington Avenue"],
+                        {}, {}, talkgroup_name="New York City PD", node_id="node-1")
+    res.assert_awaited_once_with({"municipality": "New York City", "state": "New York"})
+    assert seen["anchor"] == {"lat": 40.71, "lng": -74.0, "radius_km": 25.0}
+    assert seen["query"] == "Lexington Avenue, New York City, New York"
+
+
+@pytest.mark.asyncio
+async def test_no_state_means_no_derived_anchor():
+    """'Springfield PD' alone could anchor anywhere — skip, never guess."""
+    with patch.object(pv.settings, "place_verification_enabled", True), \
+         patch.object(intel.fstore, "doc_get_cached", AsyncMock(return_value={"lat": 41.2, "lon": -73.8})), \
+         patch.object(intel, "_get_node_state", AsyncMock(return_value="")), \
+         patch.object(pv, "_geocode_in_anchor") as geo:
+        out = await pv.verify("c1", "Respond to Main Street", None, ["Main Street"],
+                              {}, {}, talkgroup_name="Springfield PD", node_id="node-1")
+    assert out == (None, None)
+    geo.assert_not_called()
+
+
+@pytest.mark.asyncio
+async def test_configured_place_without_anchor_still_skips():
+    """A place that is set but too wide (no anchor stored) keeps the
+    load-bearing skip; the talkgroup-name fallback is only for unset areas."""
+    with patch.object(pv.settings, "place_verification_enabled", True), \
+         patch.object(pv, "_talkgroup_anchor", AsyncMock()) as derived, \
+         patch.object(pv, "_geocode_in_anchor") as geo:
+        out = await pv.verify("c1", "Main Street", None, ["Main Street"],
+                              {"state": "Colorado"}, {}, talkgroup_name="Denver PD", node_id="n")
+    assert out == (None, None)
+    derived.assert_not_awaited()
+    geo.assert_not_called()