feat(matching): implement two-stage matching with DeepSeek arbiter, fail-stop guard, and zero-probability filter

bot/tools/test_fallback_matching.py

@@ -116,7 +116,7 @@ async def test_jev_failure_triggers_fallback() -> None:
     matching.jev_ask = lambda state, qs: asyncio.sleep(0.001, result=None)
 
     # Mock DeepSeek fallback matcher
-    async def fake_fallback(raw_input, mentions, candidates_by_field, patch_context, trace_id):
+    async def fake_fallback(*args, **kwargs):
         return {
             normalize_text("میلگرد ۱۴"): {
                 "field": "material",
@@ -181,12 +181,84 @@ async def test_jev_failure_triggers_fallback() -> None:
         matching.match_with_deepseek = orig_deepseek
 
 
+async def test_two_stage_adjudication_flow() -> None:
+    """2. When Jev succeeds, outputs are NOT directly committed; they are sent to DeepSeek Arbiter."""
+    ctx = SimpleNamespace(resolver=FakeResolver())
+    saved_key = settings.TYPESAFE_API_KEY
+    settings.TYPESAFE_API_KEY = "test-typesafe-key"
+
+    orig_jev_ask = matching.jev_ask
+    matching.jev_ask = lambda state, qs: asyncio.sleep(0.001, result={
+        "answers": {
+            "mention_0": {
+                "choice": "میلگرد ۱۴ آجدار",
+                "confidence": 0.95,
+                "probabilities": {"میلگرد ۱۴ آجدار": 0.95, "تسمه فلزی": 0.0, "_none": 0.05},
+            },
+            "mention_1": {
+                "choice": "لوله پی‌وی‌سی ۴ اینچ",
+                "confidence": 0.82,
+                "probabilities": {"لوله پی‌وی‌سی ۴ اینچ": 0.82, "لوله پلی‌اتیلن": 0.0, "_none": 0.18},
+            },
+        }
+    })
+
+    arbiter_called_with = []
+
+    async def fake_arbiter(*args, **kwargs):
+        arbiter_called_with.append((args, kwargs))
+        items = kwargs.get("items") or (args[2] if len(args) > 2 else [])
+        res = {}
+        for it in items:
+            v0 = it["variants"][0]
+            if v0 == "میلگرد ۱۴":
+                # Endorsed
+                res[normalize_text(v0)] = {
+                    "field": it["field"], "status": "exact", "id": "m1", "name": "میلگرد ۱۴ آجدار",
+                    "variants": it["variants"], "source": "deepseek-arbiter", "adjudication_action": "endorsed_jev",
+                }
+            elif v0 == "لوله ۶ اینچ":
+                # Vetoed Jev (Jev wrongly picked 4-inch pipe)
+                res[normalize_text(v0)] = {
+                    "field": it["field"], "status": "new", "id": None, "name": None,
+                    "variants": it["variants"], "source": "deepseek-arbiter", "adjudication_action": "vetoed_to_new",
+                }
+        return res
+
+    orig_arbiter = matching.adjudicate_with_deepseek
+    matching.adjudicate_with_deepseek = fake_arbiter
+
+    try:
+        mentions = [
+            {"field": "material", "variants": ["میلگرد ۱۴"]},
+            {"field": "material", "variants": ["لوله ۶ اینچ"]},
+        ]
+        res = await matching.resolve_mentions(
+            ctx, {"items": []}, mentions, trace_id="t_arb",
+            raw_input="استعلام میلگرد ۱۴ و لوله ۶ اینچ",
+        )
+
+        check("Two-stage arbitration called when Jev succeeds", len(arbiter_called_with) == 1)
+        passed_items = arbiter_called_with[0][1].get("items") or arbiter_called_with[0][0][2]
+        m0_probs = passed_items[0]["jev_probabilities"]
+        check("Zero probability items filtered out", "تسمه فلزی" not in m0_probs)
+        check("Non-zero probability preserved exactly", m0_probs.get("میلگرد ۱۴ آجدار") == 0.95)
+        check("_none with non-zero probability preserved", m0_probs.get("_none") == 0.05)
+        check("Catalog candidates list complete and untouched", len(passed_items[0]["candidates"]) > 0)
+        check("Arbiter endorsed valid Jev choice as exact", res[normalize_text("میلگرد ۱۴")]["status"] == "exact")
+        check("Arbiter vetoed wrong Jev choice to new", res[normalize_text("لوله ۶ اینچ")]["status"] == "new" and res[normalize_text("لوله ۶ اینچ")]["id"] is None)
+    finally:
+        settings.TYPESAFE_API_KEY = saved_key
+        matching.jev_ask = orig_jev_ask
+        matching.adjudicate_with_deepseek = orig_arbiter
+
+
 async def test_defensive_failure_guard() -> None:
-    """2. If DeepSeek fallback fails or returns unparseable JSON, MatchingFailureError is raised."""
+    """3. Universal Fail-Stop Guard: If DeepSeek (fallback or arbiter or NLU) fails, MatchingFailureError is raised."""
     ctx = SimpleNamespace(resolver=FakeResolver())
     matching.jev_ask = lambda state, qs: asyncio.sleep(0.001, result=None)
 
-    async def boom_fallback(raw_input, mentions, candidates_by_field, patch_context, trace_id):
+    async def boom_fallback(*args, **kwargs):
         raise MatchingFailureError("DeepSeek API 500 error")
 
     orig_deepseek = matching.match_with_deepseek
@@ -203,6 +275,27 @@ async def test_defensive_failure_guard() -> None:
     finally:
         matching.match_with_deepseek = orig_deepseek
 
+    # Also test Arbiter failure raises MatchingFailureError
+    matching.jev_ask = lambda state, qs: asyncio.sleep(0.001, result={"answers": {"mention_0": {"choice": "m1", "confidence": 0.9}}})
+    async def boom_arbiter(*args, **kwargs):
+        raise MatchingFailureError("Arbiter DeepSeek timeout")
+
+    orig_arbiter = matching.adjudicate_with_deepseek
+    matching.adjudicate_with_deepseek = boom_arbiter
+    try:
+        failed_arb = False
+        try:
+            await matching.resolve_mentions(ctx, {"items": []}, mentions, trace_id="t2_arb")
+        except MatchingFailureError:
+            failed_arb = True
+        check("Arbiter failure raises MatchingFailureError (stops cleanly)", failed_arb)
+    finally:
+        matching.adjudicate_with_deepseek = orig_arbiter
+
+    # Test NLUFailureError is subclass of Exception
+    from app.nlu import NLUFailureError
+    check("NLUFailureError is subclass of Exception", issubclass(NLUFailureError, Exception))
+
 
 async def test_quote_resolve_payload_zero_difflib() -> None:
     """3. resolve_payload in quote.py respects Jev/fallback map and never runs difflib."""
@@ -292,6 +385,7 @@ async def main() -> None:
     fallback_matcher.obs.event = lambda *a, **k: asyncio.sleep(0.0001)
     try:
         await test_jev_failure_triggers_fallback()
+        await test_two_stage_adjudication_flow()
         await test_defensive_failure_guard()
         await test_quote_resolve_payload_zero_difflib()
         await test_quote_commit_defensive_get_or_create()