fix(quotas): list unpriced models by recorded cost; integer token limits in status

The unpriced-model notice asked the live registry whether a model has a
price, so a priced model whose provider was later disabled showed up as
unpriced. It now lists models whose calls this period were all recorded at
$0. Token limits are serialized as integers.

docsgpt/api/admin/quotas.py

@@ -151,13 +151,15 @@ def _delete_policy(scope: str, subject_id: Optional[str]):
 
 
 def _unpriced_models(conn) -> list[dict]:
-    """Catalog models used this period that no cost limit can see."""
+    """Models used this period whose calls were all recorded at $0 for want of a price."""
     start, _ = window_bounds(settings.QUOTA_PERIOD)
     return [
         row
         for row in TokenUsageRepository(conn).tokens_by_model(start=start)
-        # BYOM ids are UUIDs; those calls are $0 by design, not by omission.
-        if not looks_like_uuid(row["model_id"]) and not is_priced(row["model_id"])
+        # Judged by what was recorded, so a priced model whose provider has since
+        # been disabled is not listed. BYOM ids are UUIDs and $0 by design; a
+        # model explicitly priced at $0 is free, not unpriced.
+        if row["cost"] == 0 and not looks_like_uuid(row["model_id"]) and not is_priced(row["model_id"])
     ]
 
 

docsgpt/quotas/service.py

@@ -43,18 +43,19 @@ class BucketStatus:
     def to_dict(self) -> dict:
         """Return the JSON shape shared by the admin and user quota endpoints."""
 
-        def budget(limit: ResolvedLimit, used: float) -> dict:
+        def budget(limit: Optional[float], resolved: ResolvedLimit, used: float) -> dict:
             return {
-                "limit": limit.limit,
+                "limit": limit,
                 "used": used,
-                "source": limit.source,
-                "source_id": limit.source_id,
+                "source": resolved.source,
+                "source_id": resolved.source_id,
             }
 
+        tokens, cost = self.limits.tokens, self.limits.cost
         return {
             "bucket": self.bucket,
-            "tokens": budget(self.limits.tokens, self.tokens_used),
-            "cost": budget(self.limits.cost, round(self.cost_used, 6)),
+            "tokens": budget(None if tokens.unlimited else int(tokens.limit), tokens, self.tokens_used),
+            "cost": budget(cost.limit, cost, round(self.cost_used, 6)),
             "resets_at": self.resets_at.isoformat(),
         }
 

tests/api/test_quota_endpoints.py

@@ -222,7 +222,8 @@ class TestUserPolicy:
             body = _body(client.get("/api/admin/quotas/users/u1"))
         overall = body["effective"][0]
         assert overall["bucket"] == "all"
-        assert overall["tokens"] == {"limit": 900.0, "used": 40, "source": "team", "source_id": big}
+        assert overall["tokens"] == {"limit": 900, "used": 40, "source": "team", "source_id": big}
+        assert isinstance(overall["tokens"]["limit"], int)
         assert overall["cost"] == {"limit": 1.0, "used": 0.25, "source": "instance", "source_id": None}
         assert body["policies"] == []
 
@@ -243,12 +244,14 @@ class TestUserPolicy:
 
 
 class TestUnpricedModels:
-    def test_lists_used_catalog_models_without_a_price(self, client, db):
+    def test_lists_models_recorded_at_zero_for_want_of_a_price(self, client, db):
         usage = TokenUsageRepository(db)
         usage.insert(user_id="u1", prompt_tokens=10, model_id="local-llama")
-        usage.insert(user_id="u1", prompt_tokens=5, model_id="claude-haiku-4-5", cost=0.1)
+        # Priced when called; its provider may be disabled by now.
+        usage.insert(user_id="u1", prompt_tokens=5, model_id="retired-priced-model", cost=0.1)
+        usage.insert(user_id="u1", prompt_tokens=3, model_id="free-model")
         usage.insert(user_id="u1", prompt_tokens=7, model_id="7d0c1a52-2f5e-4c53-9a0e-111111111111")
-        with _admin(), patch("docsgpt.api.admin.quotas.is_priced", lambda m: m == "claude-haiku-4-5"):
+        with _admin(), patch("docsgpt.api.admin.quotas.is_priced", lambda m: m == "free-model"):
             unpriced = _body(client.get("/api/admin/quotas"))["unpriced_models"]
         assert unpriced == [{"model_id": "local-llama", "tokens": 10, "cost": 0.0}]
 
@@ -266,7 +269,7 @@ class TestMyQuota:
             body = _body(client.get("/api/user/quota"))
         (bucket,) = body["buckets"]
         assert bucket["bucket"] == "all"
-        assert bucket["tokens"] == {"limit": 100.0, "used": 30}
+        assert bucket["tokens"] == {"limit": 100, "used": 30}
         assert bucket["cost"] == {"limit": None, "used": 0.0}
         assert "source" not in json.dumps(body) and "secret" not in json.dumps(body)
 

tests/quotas/test_service.py

@@ -210,7 +210,7 @@ class TestStatusAndPayload:
         (status,) = QuotaService.status("u1", now=NOW)
         assert status.to_dict() == {
             "bucket": "all",
-            "tokens": {"limit": 100.0, "used": 40, "source": "instance", "source_id": None},
+            "tokens": {"limit": 100, "used": 40, "source": "instance", "source_id": None},
             "cost": {"limit": 2.0, "used": 0.5, "source": "instance", "source_id": None},
             "resets_at": "2026-10-01T00:00:00+00:00",
         }