diff --git a/README.md b/README.md
index f5313b9..80b403d 100644
--- a/README.md
+++ b/README.md
@@ -167,6 +167,19 @@ evaluators:
     threshold: 0.7
 ```
 
+Rubric metrics accept named properties in the evaluator config:
+
+```yaml
+evaluators:
+  - name: rubric_based_final_response_quality_v1
+    type: builtin
+    rubrics:
+      - id: latest_correction
+        text: The response uses the user's latest correction.
+```
+
+Rubrics can also come from a matched eval case or its invocations; see [Rubrics](docs/eval-set-format.md#rubrics).
+
 `type: remote` pulls community evaluators from GitHub. Protocol, SDK helpers and runtimes: [Custom Evaluators](docs/custom-evaluators.md).
 
 ## Run it
diff --git a/docs/custom-evaluators.md b/docs/custom-evaluators.md
index 87cb251..0af21d0 100644
--- a/docs/custom-evaluators.md
+++ b/docs/custom-evaluators.md
@@ -89,6 +89,21 @@ agentevals run traces/my_trace.json \
 
 Each evaluator entry in the `evaluators` list uses the following fields. The `type` field determines which other fields are valid.
 
+### `type: builtin` (ADK metrics)
+
+| Field | Required | Default | Description |
+|---|---|---|---|
+| `name` | yes | | The metric, for example `tool_trajectory_avg_score` or `rubric_based_final_response_quality_v1` |
+| `type` | yes | | `builtin` |
+| `threshold` | no | `0.5` | Score at or above this value means PASSED |
+| `judge_model` | no | ADK default | Judge model for LLM-backed metrics |
+| `trajectory_match_type` | no | `EXACT` | `EXACT`, `IN_ORDER` or `ANY_ORDER`, for `tool_trajectory_avg_score` |
+| `rubrics` | for `rubric_based_final_response_quality_v1` and `rubric_based_tool_use_quality_v1` | | A list of `{id, text}` mappings, or plain strings (given positional ids `rubric_0`, `rubric_1`, ...). Ids must be unique. Config rubrics have no `type` field. Other metrics reject `rubrics` |
+| `credential_ref` | no | | Name of a credential the judge key is resolved from (API runs) |
+| `judge_base_url` | no | | Base URL of an OpenAI-compatible judge endpoint |
+
+Each of these metrics requires an applicable rubric for each evaluated invocation, from the config, the matched eval case or that invocation. Its result carries `details.per_invocation[].rubric_scores` and `details.overall_rubric_scores`, one entry per rubric id with its score and the judge's rationale. Untyped case/invocation rubrics are listed in `details.warnings`; other types are ignored as in ADK, so response and tool-use properties can share one eval set. Config rubrics are not filtered by type. Invalid or empty effective rubric sets report an error before any judge call. Scores, aggregation and explanations follow ADK.
+
 ### `type: code` (local scripts)
 
 | Field | Required | Default | Description |
diff --git a/docs/eval-set-format.md b/docs/eval-set-format.md
index b637a76..96cd1d1 100644
--- a/docs/eval-set-format.md
+++ b/docs/eval-set-format.md
@@ -176,6 +176,53 @@ An eval case can have multiple invocations to represent a conversation. Each inv
 | `rubrics` | list[Rubric] | no | Scoring rubrics for this specific invocation |
 | `creation_timestamp` | float | no | Unix timestamp |
 
+### Rubrics
+
+`rubrics` on an eval case apply to every invocation of that case; `rubrics` on an invocation apply to that invocation alone. Each rubric is an ADK `Rubric`: `rubric_id`, `rubric_content.text_property`, and `type`. Response-quality rubrics use `FINAL_RESPONSE_QUALITY`; tool-use rubrics use `TOOL_USE_QUALITY`. As in ADK, each metric ignores properties of another type. Untyped case/invocation properties are ignored with their ids and locations listed in `details.warnings`. The runner preserves authored types. Config-level `{id, text}` rubrics are not filtered by type. Effective rubric ids must be distinct across the metric, case and invocation. A clash or an invocation with no applicable rubrics reports an error before any judge call.
+
+The `rubric_based_*` metrics grade each invocation on its own, with that invocation's prompt, steps and final response, so a case-level rubric is judged once per turn. In a multi-turn case a rubric that describes the end state ("produces the completed draft") is therefore judged against the intermediate turns as well, where it fails, and lowers the case's mean. Write such a rubric on the final invocation instead, and keep config/case-level rubrics for properties every turn must have. Every evaluated turn still needs an applicable property. A terminal response may be a clarification request when that is the expected outcome of the case. Earlier conversational context and a dedicated final-turn expectations format are separate from this per-invocation behavior.
+
+```json
+{
+  "eval_id": "booking-clarification-and-completion",
+  "rubrics": [
+    {
+      "rubric_id": "clear-outcome",
+      "type": "FINAL_RESPONSE_QUALITY",
+      "rubric_content": {"text_property": "Clearly distinguishes a clarification question from a completed draft."}
+    }
+  ],
+  "conversation": [
+    {
+      "invocation_id": "clarification",
+      "user_content": {"role": "user", "parts": [{"text": "Draft a booking for Paris."}]},
+      "final_response": {"role": "model", "parts": [{"text": "What date should the booking use?"}]},
+      "rubrics": [
+        {
+          "rubric_id": "asks-date",
+          "type": "FINAL_RESPONSE_QUALITY",
+          "rubric_content": {"text_property": "Asks for the missing date without claiming a completed booking."}
+        }
+      ]
+    },
+    {
+      "invocation_id": "completion",
+      "user_content": {"role": "user", "parts": [{"text": "Friday; actually use Lyon."}]},
+      "final_response": {"role": "model", "parts": [{"text": "Draft booking for Lyon on Friday."}]},
+      "rubrics": [
+        {
+          "rubric_id": "uses-latest-destination",
+          "type": "FINAL_RESPONSE_QUALITY",
+          "rubric_content": {"text_property": "Uses Lyon and Friday in the completed draft."}
+        }
+      ]
+    }
+  ]
+}
+```
+
+Here `clear-outcome` applies to both invocations, `asks-date` only to the clarification, and `uses-latest-destination` only to the completed draft. A case intended to stop at clarification can contain just the first invocation and its properties; it does not need to require a completed answer. This uses the existing invocation rubric field without adding a new scope option or changing ADK semantics.
+
 ### Content
 
 Uses the Google GenAI `Content` format:
diff --git a/gen/go/agentevals/api/v1alpha1/evaluation.pb.go b/gen/go/agentevals/api/v1alpha1/evaluation.pb.go
index 03757e6..0e020b8 100644
--- a/gen/go/agentevals/api/v1alpha1/evaluation.pb.go
+++ b/gen/go/agentevals/api/v1alpha1/evaluation.pb.go
@@ -174,7 +174,7 @@ type Metric struct {
 	RequiresLlm bool `protobuf:"varint,5,opt,name=requires_llm,json=requiresLlm,proto3" json:"requires_llm,omitempty"`
 	// Needs Google Cloud credentials.
 	RequiresGcp bool `protobuf:"varint,6,opt,name=requires_gcp,json=requiresGcp,proto3" json:"requires_gcp,omitempty"`
-	// Needs rubrics in the eval set.
+	// Needs rubrics, from the evaluator or the eval set.
 	RequiresRubrics bool `protobuf:"varint,7,opt,name=requires_rubrics,json=requiresRubrics,proto3" json:"requires_rubrics,omitempty"`
 	// Usable with recorded traces.
 	Working       bool `protobuf:"varint,8,opt,name=working,proto3" json:"working,omitempty"`
@@ -323,7 +323,9 @@ type BuiltinEvaluator struct {
 	// Name of a credential_refs entry whose value is the judge API key.
 	CredentialRef string `protobuf:"bytes,5,opt,name=credential_ref,json=credentialRef,proto3" json:"credential_ref,omitempty"`
 	// Base URL of an OpenAI compatible judge endpoint.
-	JudgeBaseUrl  string `protobuf:"bytes,6,opt,name=judge_base_url,json=judgeBaseUrl,proto3" json:"judge_base_url,omitempty"`
+	JudgeBaseUrl string `protobuf:"bytes,6,opt,name=judge_base_url,json=judgeBaseUrl,proto3" json:"judge_base_url,omitempty"`
+	// Rubrics for the rubric_based_* metrics that take them. Ids are unique.
+	Rubrics       []*Rubric `protobuf:"bytes,7,rep,name=rubrics,proto3" json:"rubrics,omitempty"`
 	unknownFields protoimpl.UnknownFields
 	sizeCache     protoimpl.SizeCache
 }
@@ -400,6 +402,66 @@ func (x *BuiltinEvaluator) GetJudgeBaseUrl() string {
 	return ""
 }
 
+func (x *BuiltinEvaluator) GetRubrics() []*Rubric {
+	if x != nil {
+		return x.Rubrics
+	}
+	return nil
+}
+
+// One property a rubric metric's judge checks.
+type Rubric struct {
+	state         protoimpl.MessageState `protogen:"open.v1"`
+	Id            string                 `protobuf:"bytes,1,opt,name=id,proto3" json:"id,omitempty"`
+	Text          string                 `protobuf:"bytes,2,opt,name=text,proto3" json:"text,omitempty"`
+	unknownFields protoimpl.UnknownFields
+	sizeCache     protoimpl.SizeCache
+}
+
+func (x *Rubric) Reset() {
+	*x = Rubric{}
+	mi := &file_agentevals_api_v1alpha1_evaluation_proto_msgTypes[4]
+	ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
+	ms.StoreMessageInfo(mi)
+}
+
+func (x *Rubric) String() string {
+	return protoimpl.X.MessageStringOf(x)
+}
+
+func (*Rubric) ProtoMessage() {}
+
+func (x *Rubric) ProtoReflect() protoreflect.Message {
+	mi := &file_agentevals_api_v1alpha1_evaluation_proto_msgTypes[4]
+	if x != nil {
+		ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
+		if ms.LoadMessageInfo() == nil {
+			ms.StoreMessageInfo(mi)
+		}
+		return ms
+	}
+	return mi.MessageOf(x)
+}
+
+// Deprecated: Use Rubric.ProtoReflect.Descriptor instead.
+func (*Rubric) Descriptor() ([]byte, []int) {
+	return file_agentevals_api_v1alpha1_evaluation_proto_rawDescGZIP(), []int{4}
+}
+
+func (x *Rubric) GetId() string {
+	if x != nil {
+		return x.Id
+	}
+	return ""
+}
+
+func (x *Rubric) GetText() string {
+	if x != nil {
+		return x.Text
+	}
+	return ""
+}
+
 // An evaluator file on the server, run as a subprocess.
 type CodeEvaluator struct {
 	state protoimpl.MessageState `protogen:"open.v1"`
@@ -419,7 +481,7 @@ type CodeEvaluator struct {
 
 func (x *CodeEvaluator) Reset() {
 	*x = CodeEvaluator{}
-	mi := &file_agentevals_api_v1alpha1_evaluation_proto_msgTypes[4]
+	mi := &file_agentevals_api_v1alpha1_evaluation_proto_msgTypes[5]
 	ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
 	ms.StoreMessageInfo(mi)
 }
@@ -431,7 +493,7 @@ func (x *CodeEvaluator) String() string {
 func (*CodeEvaluator) ProtoMessage() {}
 
 func (x *CodeEvaluator) ProtoReflect() protoreflect.Message {
-	mi := &file_agentevals_api_v1alpha1_evaluation_proto_msgTypes[4]
+	mi := &file_agentevals_api_v1alpha1_evaluation_proto_msgTypes[5]
 	if x != nil {
 		ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
 		if ms.LoadMessageInfo() == nil {
@@ -444,7 +506,7 @@ func (x *CodeEvaluator) ProtoReflect() protoreflect.Message {
 
 // Deprecated: Use CodeEvaluator.ProtoReflect.Descriptor instead.
 func (*CodeEvaluator) Descriptor() ([]byte, []int) {
-	return file_agentevals_api_v1alpha1_evaluation_proto_rawDescGZIP(), []int{4}
+	return file_agentevals_api_v1alpha1_evaluation_proto_rawDescGZIP(), []int{5}
 }
 
 func (x *CodeEvaluator) GetName() string {
@@ -510,7 +572,7 @@ type RemoteEvaluator struct {
 
 func (x *RemoteEvaluator) Reset() {
 	*x = RemoteEvaluator{}
-	mi := &file_agentevals_api_v1alpha1_evaluation_proto_msgTypes[5]
+	mi := &file_agentevals_api_v1alpha1_evaluation_proto_msgTypes[6]
 	ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
 	ms.StoreMessageInfo(mi)
 }
@@ -522,7 +584,7 @@ func (x *RemoteEvaluator) String() string {
 func (*RemoteEvaluator) ProtoMessage() {}
 
 func (x *RemoteEvaluator) ProtoReflect() protoreflect.Message {
-	mi := &file_agentevals_api_v1alpha1_evaluation_proto_msgTypes[5]
+	mi := &file_agentevals_api_v1alpha1_evaluation_proto_msgTypes[6]
 	if x != nil {
 		ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
 		if ms.LoadMessageInfo() == nil {
@@ -535,7 +597,7 @@ func (x *RemoteEvaluator) ProtoReflect() protoreflect.Message {
 
 // Deprecated: Use RemoteEvaluator.ProtoReflect.Descriptor instead.
 func (*RemoteEvaluator) Descriptor() ([]byte, []int) {
-	return file_agentevals_api_v1alpha1_evaluation_proto_rawDescGZIP(), []int{5}
+	return file_agentevals_api_v1alpha1_evaluation_proto_rawDescGZIP(), []int{6}
 }
 
 func (x *RemoteEvaluator) GetName() string {
@@ -601,7 +663,7 @@ type Evaluator struct {
 
 func (x *Evaluator) Reset() {
 	*x = Evaluator{}
-	mi := &file_agentevals_api_v1alpha1_evaluation_proto_msgTypes[6]
+	mi := &file_agentevals_api_v1alpha1_evaluation_proto_msgTypes[7]
 	ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
 	ms.StoreMessageInfo(mi)
 }
@@ -613,7 +675,7 @@ func (x *Evaluator) String() string {
 func (*Evaluator) ProtoMessage() {}
 
 func (x *Evaluator) ProtoReflect() protoreflect.Message {
-	mi := &file_agentevals_api_v1alpha1_evaluation_proto_msgTypes[6]
+	mi := &file_agentevals_api_v1alpha1_evaluation_proto_msgTypes[7]
 	if x != nil {
 		ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
 		if ms.LoadMessageInfo() == nil {
@@ -626,7 +688,7 @@ func (x *Evaluator) ProtoReflect() protoreflect.Message {
 
 // Deprecated: Use Evaluator.ProtoReflect.Descriptor instead.
 func (*Evaluator) Descriptor() ([]byte, []int) {
-	return file_agentevals_api_v1alpha1_evaluation_proto_rawDescGZIP(), []int{6}
+	return file_agentevals_api_v1alpha1_evaluation_proto_rawDescGZIP(), []int{7}
 }
 
 func (x *Evaluator) GetKind() isEvaluator_Kind {
@@ -701,7 +763,7 @@ type EvalParams struct {
 
 func (x *EvalParams) Reset() {
 	*x = EvalParams{}
-	mi := &file_agentevals_api_v1alpha1_evaluation_proto_msgTypes[7]
+	mi := &file_agentevals_api_v1alpha1_evaluation_proto_msgTypes[8]
 	ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
 	ms.StoreMessageInfo(mi)
 }
@@ -713,7 +775,7 @@ func (x *EvalParams) String() string {
 func (*EvalParams) ProtoMessage() {}
 
 func (x *EvalParams) ProtoReflect() protoreflect.Message {
-	mi := &file_agentevals_api_v1alpha1_evaluation_proto_msgTypes[7]
+	mi := &file_agentevals_api_v1alpha1_evaluation_proto_msgTypes[8]
 	if x != nil {
 		ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
 		if ms.LoadMessageInfo() == nil {
@@ -726,7 +788,7 @@ func (x *EvalParams) ProtoReflect() protoreflect.Message {
 
 // Deprecated: Use EvalParams.ProtoReflect.Descriptor instead.
 func (*EvalParams) Descriptor() ([]byte, []int) {
-	return file_agentevals_api_v1alpha1_evaluation_proto_rawDescGZIP(), []int{7}
+	return file_agentevals_api_v1alpha1_evaluation_proto_rawDescGZIP(), []int{8}
 }
 
 func (x *EvalParams) GetEvaluators() []*Evaluator {
@@ -769,7 +831,7 @@ type InvocationResult struct {
 
 func (x *InvocationResult) Reset() {
 	*x = InvocationResult{}
-	mi := &file_agentevals_api_v1alpha1_evaluation_proto_msgTypes[8]
+	mi := &file_agentevals_api_v1alpha1_evaluation_proto_msgTypes[9]
 	ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
 	ms.StoreMessageInfo(mi)
 }
@@ -781,7 +843,7 @@ func (x *InvocationResult) String() string {
 func (*InvocationResult) ProtoMessage() {}
 
 func (x *InvocationResult) ProtoReflect() protoreflect.Message {
-	mi := &file_agentevals_api_v1alpha1_evaluation_proto_msgTypes[8]
+	mi := &file_agentevals_api_v1alpha1_evaluation_proto_msgTypes[9]
 	if x != nil {
 		ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
 		if ms.LoadMessageInfo() == nil {
@@ -794,7 +856,7 @@ func (x *InvocationResult) ProtoReflect() protoreflect.Message {
 
 // Deprecated: Use InvocationResult.ProtoReflect.Descriptor instead.
 func (*InvocationResult) Descriptor() ([]byte, []int) {
-	return file_agentevals_api_v1alpha1_evaluation_proto_rawDescGZIP(), []int{8}
+	return file_agentevals_api_v1alpha1_evaluation_proto_rawDescGZIP(), []int{9}
 }
 
 func (x *InvocationResult) GetScore() float64 {
@@ -830,7 +892,7 @@ type MetricResult struct {
 
 func (x *MetricResult) Reset() {
 	*x = MetricResult{}
-	mi := &file_agentevals_api_v1alpha1_evaluation_proto_msgTypes[9]
+	mi := &file_agentevals_api_v1alpha1_evaluation_proto_msgTypes[10]
 	ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
 	ms.StoreMessageInfo(mi)
 }
@@ -842,7 +904,7 @@ func (x *MetricResult) String() string {
 func (*MetricResult) ProtoMessage() {}
 
 func (x *MetricResult) ProtoReflect() protoreflect.Message {
-	mi := &file_agentevals_api_v1alpha1_evaluation_proto_msgTypes[9]
+	mi := &file_agentevals_api_v1alpha1_evaluation_proto_msgTypes[10]
 	if x != nil {
 		ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
 		if ms.LoadMessageInfo() == nil {
@@ -855,7 +917,7 @@ func (x *MetricResult) ProtoReflect() protoreflect.Message {
 
 // Deprecated: Use MetricResult.ProtoReflect.Descriptor instead.
 func (*MetricResult) Descriptor() ([]byte, []int) {
-	return file_agentevals_api_v1alpha1_evaluation_proto_rawDescGZIP(), []int{9}
+	return file_agentevals_api_v1alpha1_evaluation_proto_rawDescGZIP(), []int{10}
 }
 
 func (x *MetricResult) GetMetricName() string {
@@ -924,7 +986,7 @@ type TurnRef struct {
 
 func (x *TurnRef) Reset() {
 	*x = TurnRef{}
-	mi := &file_agentevals_api_v1alpha1_evaluation_proto_msgTypes[10]
+	mi := &file_agentevals_api_v1alpha1_evaluation_proto_msgTypes[11]
 	ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
 	ms.StoreMessageInfo(mi)
 }
@@ -936,7 +998,7 @@ func (x *TurnRef) String() string {
 func (*TurnRef) ProtoMessage() {}
 
 func (x *TurnRef) ProtoReflect() protoreflect.Message {
-	mi := &file_agentevals_api_v1alpha1_evaluation_proto_msgTypes[10]
+	mi := &file_agentevals_api_v1alpha1_evaluation_proto_msgTypes[11]
 	if x != nil {
 		ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
 		if ms.LoadMessageInfo() == nil {
@@ -949,7 +1011,7 @@ func (x *TurnRef) ProtoReflect() protoreflect.Message {
 
 // Deprecated: Use TurnRef.ProtoReflect.Descriptor instead.
 func (*TurnRef) Descriptor() ([]byte, []int) {
-	return file_agentevals_api_v1alpha1_evaluation_proto_rawDescGZIP(), []int{10}
+	return file_agentevals_api_v1alpha1_evaluation_proto_rawDescGZIP(), []int{11}
 }
 
 func (x *TurnRef) GetTraceId() string {
@@ -995,7 +1057,7 @@ type TraceResult struct {
 
 func (x *TraceResult) Reset() {
 	*x = TraceResult{}
-	mi := &file_agentevals_api_v1alpha1_evaluation_proto_msgTypes[11]
+	mi := &file_agentevals_api_v1alpha1_evaluation_proto_msgTypes[12]
 	ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
 	ms.StoreMessageInfo(mi)
 }
@@ -1007,7 +1069,7 @@ func (x *TraceResult) String() string {
 func (*TraceResult) ProtoMessage() {}
 
 func (x *TraceResult) ProtoReflect() protoreflect.Message {
-	mi := &file_agentevals_api_v1alpha1_evaluation_proto_msgTypes[11]
+	mi := &file_agentevals_api_v1alpha1_evaluation_proto_msgTypes[12]
 	if x != nil {
 		ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
 		if ms.LoadMessageInfo() == nil {
@@ -1020,7 +1082,7 @@ func (x *TraceResult) ProtoReflect() protoreflect.Message {
 
 // Deprecated: Use TraceResult.ProtoReflect.Descriptor instead.
 func (*TraceResult) Descriptor() ([]byte, []int) {
-	return file_agentevals_api_v1alpha1_evaluation_proto_rawDescGZIP(), []int{11}
+	return file_agentevals_api_v1alpha1_evaluation_proto_rawDescGZIP(), []int{12}
 }
 
 func (x *TraceResult) GetTraceId() string {
@@ -1127,7 +1189,7 @@ type RunResult struct {
 
 func (x *RunResult) Reset() {
 	*x = RunResult{}
-	mi := &file_agentevals_api_v1alpha1_evaluation_proto_msgTypes[12]
+	mi := &file_agentevals_api_v1alpha1_evaluation_proto_msgTypes[13]
 	ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
 	ms.StoreMessageInfo(mi)
 }
@@ -1139,7 +1201,7 @@ func (x *RunResult) String() string {
 func (*RunResult) ProtoMessage() {}
 
 func (x *RunResult) ProtoReflect() protoreflect.Message {
-	mi := &file_agentevals_api_v1alpha1_evaluation_proto_msgTypes[12]
+	mi := &file_agentevals_api_v1alpha1_evaluation_proto_msgTypes[13]
 	if x != nil {
 		ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
 		if ms.LoadMessageInfo() == nil {
@@ -1152,7 +1214,7 @@ func (x *RunResult) ProtoReflect() protoreflect.Message {
 
 // Deprecated: Use RunResult.ProtoReflect.Descriptor instead.
 func (*RunResult) Descriptor() ([]byte, []int) {
-	return file_agentevals_api_v1alpha1_evaluation_proto_rawDescGZIP(), []int{12}
+	return file_agentevals_api_v1alpha1_evaluation_proto_rawDescGZIP(), []int{13}
 }
 
 func (x *RunResult) GetTraceResults() []*TraceResult {
@@ -1203,7 +1265,7 @@ type EvaluateTracesRequest struct {
 
 func (x *EvaluateTracesRequest) Reset() {
 	*x = EvaluateTracesRequest{}
-	mi := &file_agentevals_api_v1alpha1_evaluation_proto_msgTypes[13]
+	mi := &file_agentevals_api_v1alpha1_evaluation_proto_msgTypes[14]
 	ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
 	ms.StoreMessageInfo(mi)
 }
@@ -1215,7 +1277,7 @@ func (x *EvaluateTracesRequest) String() string {
 func (*EvaluateTracesRequest) ProtoMessage() {}
 
 func (x *EvaluateTracesRequest) ProtoReflect() protoreflect.Message {
-	mi := &file_agentevals_api_v1alpha1_evaluation_proto_msgTypes[13]
+	mi := &file_agentevals_api_v1alpha1_evaluation_proto_msgTypes[14]
 	if x != nil {
 		ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
 		if ms.LoadMessageInfo() == nil {
@@ -1228,7 +1290,7 @@ func (x *EvaluateTracesRequest) ProtoReflect() protoreflect.Message {
 
 // Deprecated: Use EvaluateTracesRequest.ProtoReflect.Descriptor instead.
 func (*EvaluateTracesRequest) Descriptor() ([]byte, []int) {
-	return file_agentevals_api_v1alpha1_evaluation_proto_rawDescGZIP(), []int{13}
+	return file_agentevals_api_v1alpha1_evaluation_proto_rawDescGZIP(), []int{14}
 }
 
 func (x *EvaluateTracesRequest) GetTraces() *TraceInput {
@@ -1288,7 +1350,7 @@ type EvaluationProgress struct {
 
 func (x *EvaluationProgress) Reset() {
 	*x = EvaluationProgress{}
-	mi := &file_agentevals_api_v1alpha1_evaluation_proto_msgTypes[14]
+	mi := &file_agentevals_api_v1alpha1_evaluation_proto_msgTypes[15]
 	ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
 	ms.StoreMessageInfo(mi)
 }
@@ -1300,7 +1362,7 @@ func (x *EvaluationProgress) String() string {
 func (*EvaluationProgress) ProtoMessage() {}
 
 func (x *EvaluationProgress) ProtoReflect() protoreflect.Message {
-	mi := &file_agentevals_api_v1alpha1_evaluation_proto_msgTypes[14]
+	mi := &file_agentevals_api_v1alpha1_evaluation_proto_msgTypes[15]
 	if x != nil {
 		ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
 		if ms.LoadMessageInfo() == nil {
@@ -1313,7 +1375,7 @@ func (x *EvaluationProgress) ProtoReflect() protoreflect.Message {
 
 // Deprecated: Use EvaluationProgress.ProtoReflect.Descriptor instead.
 func (*EvaluationProgress) Descriptor() ([]byte, []int) {
-	return file_agentevals_api_v1alpha1_evaluation_proto_rawDescGZIP(), []int{14}
+	return file_agentevals_api_v1alpha1_evaluation_proto_rawDescGZIP(), []int{15}
 }
 
 func (x *EvaluationProgress) GetMessage() string {
@@ -1336,7 +1398,7 @@ type GroupPerformance struct {
 
 func (x *GroupPerformance) Reset() {
 	*x = GroupPerformance{}
-	mi := &file_agentevals_api_v1alpha1_evaluation_proto_msgTypes[15]
+	mi := &file_agentevals_api_v1alpha1_evaluation_proto_msgTypes[16]
 	ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
 	ms.StoreMessageInfo(mi)
 }
@@ -1348,7 +1410,7 @@ func (x *GroupPerformance) String() string {
 func (*GroupPerformance) ProtoMessage() {}
 
 func (x *GroupPerformance) ProtoReflect() protoreflect.Message {
-	mi := &file_agentevals_api_v1alpha1_evaluation_proto_msgTypes[15]
+	mi := &file_agentevals_api_v1alpha1_evaluation_proto_msgTypes[16]
 	if x != nil {
 		ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
 		if ms.LoadMessageInfo() == nil {
@@ -1361,7 +1423,7 @@ func (x *GroupPerformance) ProtoReflect() protoreflect.Message {
 
 // Deprecated: Use GroupPerformance.ProtoReflect.Descriptor instead.
 func (*GroupPerformance) Descriptor() ([]byte, []int) {
-	return file_agentevals_api_v1alpha1_evaluation_proto_rawDescGZIP(), []int{15}
+	return file_agentevals_api_v1alpha1_evaluation_proto_rawDescGZIP(), []int{16}
 }
 
 func (x *GroupPerformance) GetTraceId() string {
@@ -1400,7 +1462,7 @@ type EvaluateTracesResponse struct {
 
 func (x *EvaluateTracesResponse) Reset() {
 	*x = EvaluateTracesResponse{}
-	mi := &file_agentevals_api_v1alpha1_evaluation_proto_msgTypes[16]
+	mi := &file_agentevals_api_v1alpha1_evaluation_proto_msgTypes[17]
 	ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
 	ms.StoreMessageInfo(mi)
 }
@@ -1412,7 +1474,7 @@ func (x *EvaluateTracesResponse) String() string {
 func (*EvaluateTracesResponse) ProtoMessage() {}
 
 func (x *EvaluateTracesResponse) ProtoReflect() protoreflect.Message {
-	mi := &file_agentevals_api_v1alpha1_evaluation_proto_msgTypes[16]
+	mi := &file_agentevals_api_v1alpha1_evaluation_proto_msgTypes[17]
 	if x != nil {
 		ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
 		if ms.LoadMessageInfo() == nil {
@@ -1425,7 +1487,7 @@ func (x *EvaluateTracesResponse) ProtoReflect() protoreflect.Message {
 
 // Deprecated: Use EvaluateTracesResponse.ProtoReflect.Descriptor instead.
 func (*EvaluateTracesResponse) Descriptor() ([]byte, []int) {
-	return file_agentevals_api_v1alpha1_evaluation_proto_rawDescGZIP(), []int{16}
+	return file_agentevals_api_v1alpha1_evaluation_proto_rawDescGZIP(), []int{17}
 }
 
 func (x *EvaluateTracesResponse) GetEvent() isEvaluateTracesResponse_Event {
@@ -1517,7 +1579,7 @@ const file_agentevals_api_v1alpha1_evaluation_proto_rawDesc = "" +
 	"\x10requires_rubrics\x18\a \x01(\bR\x0frequiresRubrics\x12\x18\n" +
 	"\aworking\x18\b \x01(\bR\aworking\"P\n" +
 	"\x13ListMetricsResponse\x129\n" +
-	"\ametrics\x18\x01 \x03(\v2\x1f.agentevals.api.v1alpha1.MetricR\ametrics\"\xa7\x02\n" +
+	"\ametrics\x18\x01 \x03(\v2\x1f.agentevals.api.v1alpha1.MetricR\ametrics\"\xe2\x02\n" +
 	"\x10BuiltinEvaluator\x12\x12\n" +
 	"\x04name\x18\x01 \x01(\tR\x04name\x12!\n" +
 	"\tthreshold\x18\x02 \x01(\x01H\x00R\tthreshold\x88\x01\x01\x12\x1f\n" +
@@ -1525,9 +1587,13 @@ const file_agentevals_api_v1alpha1_evaluation_proto_rawDesc = "" +
 	"judgeModel\x12`\n" +
 	"\x15trajectory_match_type\x18\x04 \x01(\x0e2,.agentevals.api.v1alpha1.TrajectoryMatchTypeR\x13trajectoryMatchType\x12%\n" +
 	"\x0ecredential_ref\x18\x05 \x01(\tR\rcredentialRef\x12$\n" +
-	"\x0ejudge_base_url\x18\x06 \x01(\tR\fjudgeBaseUrlB\f\n" +
+	"\x0ejudge_base_url\x18\x06 \x01(\tR\fjudgeBaseUrl\x129\n" +
+	"\arubrics\x18\a \x03(\v2\x1f.agentevals.api.v1alpha1.RubricR\arubricsB\f\n" +
 	"\n" +
-	"_threshold\"\xe0\x01\n" +
+	"_threshold\",\n" +
+	"\x06Rubric\x12\x0e\n" +
+	"\x02id\x18\x01 \x01(\tR\x02id\x12\x12\n" +
+	"\x04text\x18\x02 \x01(\tR\x04text\"\xe0\x01\n" +
 	"\rCodeEvaluator\x12\x12\n" +
 	"\x04name\x18\x01 \x01(\tR\x04name\x12!\n" +
 	"\tthreshold\x18\x02 \x01(\x01H\x00R\tthreshold\x88\x01\x01\x12\x1d\n" +
@@ -1663,7 +1729,7 @@ func file_agentevals_api_v1alpha1_evaluation_proto_rawDescGZIP() []byte {
 }
 
 var file_agentevals_api_v1alpha1_evaluation_proto_enumTypes = make([]protoimpl.EnumInfo, 2)
-var file_agentevals_api_v1alpha1_evaluation_proto_msgTypes = make([]protoimpl.MessageInfo, 18)
+var file_agentevals_api_v1alpha1_evaluation_proto_msgTypes = make([]protoimpl.MessageInfo, 19)
 var file_agentevals_api_v1alpha1_evaluation_proto_goTypes = []any{
 	(TrajectoryMatchType)(0),       // 0: agentevals.api.v1alpha1.TrajectoryMatchType
 	(EvalStatus)(0),                // 1: agentevals.api.v1alpha1.EvalStatus
@@ -1671,63 +1737,65 @@ var file_agentevals_api_v1alpha1_evaluation_proto_goTypes = []any{
 	(*Metric)(nil),                 // 3: agentevals.api.v1alpha1.Metric
 	(*ListMetricsResponse)(nil),    // 4: agentevals.api.v1alpha1.ListMetricsResponse
 	(*BuiltinEvaluator)(nil),       // 5: agentevals.api.v1alpha1.BuiltinEvaluator
-	(*CodeEvaluator)(nil),          // 6: agentevals.api.v1alpha1.CodeEvaluator
-	(*RemoteEvaluator)(nil),        // 7: agentevals.api.v1alpha1.RemoteEvaluator
-	(*Evaluator)(nil),              // 8: agentevals.api.v1alpha1.Evaluator
-	(*EvalParams)(nil),             // 9: agentevals.api.v1alpha1.EvalParams
-	(*InvocationResult)(nil),       // 10: agentevals.api.v1alpha1.InvocationResult
-	(*MetricResult)(nil),           // 11: agentevals.api.v1alpha1.MetricResult
-	(*TurnRef)(nil),                // 12: agentevals.api.v1alpha1.TurnRef
-	(*TraceResult)(nil),            // 13: agentevals.api.v1alpha1.TraceResult
-	(*RunResult)(nil),              // 14: agentevals.api.v1alpha1.RunResult
-	(*EvaluateTracesRequest)(nil),  // 15: agentevals.api.v1alpha1.EvaluateTracesRequest
-	(*EvaluationProgress)(nil),     // 16: agentevals.api.v1alpha1.EvaluationProgress
-	(*GroupPerformance)(nil),       // 17: agentevals.api.v1alpha1.GroupPerformance
-	(*EvaluateTracesResponse)(nil), // 18: agentevals.api.v1alpha1.EvaluateTracesResponse
-	nil,                            // 19: agentevals.api.v1alpha1.EvaluateTracesRequest.CredentialRefsEntry
-	(*structpb.Struct)(nil),        // 20: google.protobuf.Struct
-	(GroupBy)(0),                   // 21: agentevals.api.v1alpha1.GroupBy
-	(*TraceInput)(nil),             // 22: agentevals.api.v1alpha1.TraceInput
+	(*Rubric)(nil),                 // 6: agentevals.api.v1alpha1.Rubric
+	(*CodeEvaluator)(nil),          // 7: agentevals.api.v1alpha1.CodeEvaluator
+	(*RemoteEvaluator)(nil),        // 8: agentevals.api.v1alpha1.RemoteEvaluator
+	(*Evaluator)(nil),              // 9: agentevals.api.v1alpha1.Evaluator
+	(*EvalParams)(nil),             // 10: agentevals.api.v1alpha1.EvalParams
+	(*InvocationResult)(nil),       // 11: agentevals.api.v1alpha1.InvocationResult
+	(*MetricResult)(nil),           // 12: agentevals.api.v1alpha1.MetricResult
+	(*TurnRef)(nil),                // 13: agentevals.api.v1alpha1.TurnRef
+	(*TraceResult)(nil),            // 14: agentevals.api.v1alpha1.TraceResult
+	(*RunResult)(nil),              // 15: agentevals.api.v1alpha1.RunResult
+	(*EvaluateTracesRequest)(nil),  // 16: agentevals.api.v1alpha1.EvaluateTracesRequest
+	(*EvaluationProgress)(nil),     // 17: agentevals.api.v1alpha1.EvaluationProgress
+	(*GroupPerformance)(nil),       // 18: agentevals.api.v1alpha1.GroupPerformance
+	(*EvaluateTracesResponse)(nil), // 19: agentevals.api.v1alpha1.EvaluateTracesResponse
+	nil,                            // 20: agentevals.api.v1alpha1.EvaluateTracesRequest.CredentialRefsEntry
+	(*structpb.Struct)(nil),        // 21: google.protobuf.Struct
+	(GroupBy)(0),                   // 22: agentevals.api.v1alpha1.GroupBy
+	(*TraceInput)(nil),             // 23: agentevals.api.v1alpha1.TraceInput
 }
 var file_agentevals_api_v1alpha1_evaluation_proto_depIdxs = []int32{
 	3,  // 0: agentevals.api.v1alpha1.ListMetricsResponse.metrics:type_name -> agentevals.api.v1alpha1.Metric
 	0,  // 1: agentevals.api.v1alpha1.BuiltinEvaluator.trajectory_match_type:type_name -> agentevals.api.v1alpha1.TrajectoryMatchType
-	20, // 2: agentevals.api.v1alpha1.CodeEvaluator.config:type_name -> google.protobuf.Struct
-	20, // 3: agentevals.api.v1alpha1.RemoteEvaluator.config:type_name -> google.protobuf.Struct
-	5,  // 4: agentevals.api.v1alpha1.Evaluator.builtin:type_name -> agentevals.api.v1alpha1.BuiltinEvaluator
-	6,  // 5: agentevals.api.v1alpha1.Evaluator.code:type_name -> agentevals.api.v1alpha1.CodeEvaluator
-	7,  // 6: agentevals.api.v1alpha1.Evaluator.remote:type_name -> agentevals.api.v1alpha1.RemoteEvaluator
-	8,  // 7: agentevals.api.v1alpha1.EvalParams.evaluators:type_name -> agentevals.api.v1alpha1.Evaluator
-	21, // 8: agentevals.api.v1alpha1.EvalParams.group_by:type_name -> agentevals.api.v1alpha1.GroupBy
-	1,  // 9: agentevals.api.v1alpha1.InvocationResult.status:type_name -> agentevals.api.v1alpha1.EvalStatus
-	1,  // 10: agentevals.api.v1alpha1.MetricResult.eval_status:type_name -> agentevals.api.v1alpha1.EvalStatus
-	10, // 11: agentevals.api.v1alpha1.MetricResult.per_invocation:type_name -> agentevals.api.v1alpha1.InvocationResult
-	20, // 12: agentevals.api.v1alpha1.MetricResult.details:type_name -> google.protobuf.Struct
-	11, // 13: agentevals.api.v1alpha1.TraceResult.metric_results:type_name -> agentevals.api.v1alpha1.MetricResult
-	20, // 14: agentevals.api.v1alpha1.TraceResult.performance_metrics:type_name -> google.protobuf.Struct
-	12, // 15: agentevals.api.v1alpha1.TraceResult.turn_refs:type_name -> agentevals.api.v1alpha1.TurnRef
-	13, // 16: agentevals.api.v1alpha1.RunResult.trace_results:type_name -> agentevals.api.v1alpha1.TraceResult
-	20, // 17: agentevals.api.v1alpha1.RunResult.performance_metrics:type_name -> google.protobuf.Struct
-	22, // 18: agentevals.api.v1alpha1.EvaluateTracesRequest.traces:type_name -> agentevals.api.v1alpha1.TraceInput
-	20, // 19: agentevals.api.v1alpha1.EvaluateTracesRequest.adk_eval_set:type_name -> google.protobuf.Struct
-	9,  // 20: agentevals.api.v1alpha1.EvaluateTracesRequest.params:type_name -> agentevals.api.v1alpha1.EvalParams
-	19, // 21: agentevals.api.v1alpha1.EvaluateTracesRequest.credential_refs:type_name -> agentevals.api.v1alpha1.EvaluateTracesRequest.CredentialRefsEntry
-	20, // 22: agentevals.api.v1alpha1.GroupPerformance.performance_metrics:type_name -> google.protobuf.Struct
-	20, // 23: agentevals.api.v1alpha1.GroupPerformance.trace_metadata:type_name -> google.protobuf.Struct
-	16, // 24: agentevals.api.v1alpha1.EvaluateTracesResponse.progress:type_name -> agentevals.api.v1alpha1.EvaluationProgress
-	17, // 25: agentevals.api.v1alpha1.EvaluateTracesResponse.performance:type_name -> agentevals.api.v1alpha1.GroupPerformance
-	13, // 26: agentevals.api.v1alpha1.EvaluateTracesResponse.trace_result:type_name -> agentevals.api.v1alpha1.TraceResult
-	14, // 27: agentevals.api.v1alpha1.EvaluateTracesResponse.done:type_name -> agentevals.api.v1alpha1.RunResult
-	20, // 28: agentevals.api.v1alpha1.EvaluateTracesRequest.CredentialRefsEntry.value:type_name -> google.protobuf.Struct
-	2,  // 29: agentevals.api.v1alpha1.EvaluationService.ListMetrics:input_type -> agentevals.api.v1alpha1.ListMetricsRequest
-	15, // 30: agentevals.api.v1alpha1.EvaluationService.EvaluateTraces:input_type -> agentevals.api.v1alpha1.EvaluateTracesRequest
-	4,  // 31: agentevals.api.v1alpha1.EvaluationService.ListMetrics:output_type -> agentevals.api.v1alpha1.ListMetricsResponse
-	18, // 32: agentevals.api.v1alpha1.EvaluationService.EvaluateTraces:output_type -> agentevals.api.v1alpha1.EvaluateTracesResponse
-	31, // [31:33] is the sub-list for method output_type
-	29, // [29:31] is the sub-list for method input_type
-	29, // [29:29] is the sub-list for extension type_name
-	29, // [29:29] is the sub-list for extension extendee
-	0,  // [0:29] is the sub-list for field type_name
+	6,  // 2: agentevals.api.v1alpha1.BuiltinEvaluator.rubrics:type_name -> agentevals.api.v1alpha1.Rubric
+	21, // 3: agentevals.api.v1alpha1.CodeEvaluator.config:type_name -> google.protobuf.Struct
+	21, // 4: agentevals.api.v1alpha1.RemoteEvaluator.config:type_name -> google.protobuf.Struct
+	5,  // 5: agentevals.api.v1alpha1.Evaluator.builtin:type_name -> agentevals.api.v1alpha1.BuiltinEvaluator
+	7,  // 6: agentevals.api.v1alpha1.Evaluator.code:type_name -> agentevals.api.v1alpha1.CodeEvaluator
+	8,  // 7: agentevals.api.v1alpha1.Evaluator.remote:type_name -> agentevals.api.v1alpha1.RemoteEvaluator
+	9,  // 8: agentevals.api.v1alpha1.EvalParams.evaluators:type_name -> agentevals.api.v1alpha1.Evaluator
+	22, // 9: agentevals.api.v1alpha1.EvalParams.group_by:type_name -> agentevals.api.v1alpha1.GroupBy
+	1,  // 10: agentevals.api.v1alpha1.InvocationResult.status:type_name -> agentevals.api.v1alpha1.EvalStatus
+	1,  // 11: agentevals.api.v1alpha1.MetricResult.eval_status:type_name -> agentevals.api.v1alpha1.EvalStatus
+	11, // 12: agentevals.api.v1alpha1.MetricResult.per_invocation:type_name -> agentevals.api.v1alpha1.InvocationResult
+	21, // 13: agentevals.api.v1alpha1.MetricResult.details:type_name -> google.protobuf.Struct
+	12, // 14: agentevals.api.v1alpha1.TraceResult.metric_results:type_name -> agentevals.api.v1alpha1.MetricResult
+	21, // 15: agentevals.api.v1alpha1.TraceResult.performance_metrics:type_name -> google.protobuf.Struct
+	13, // 16: agentevals.api.v1alpha1.TraceResult.turn_refs:type_name -> agentevals.api.v1alpha1.TurnRef
+	14, // 17: agentevals.api.v1alpha1.RunResult.trace_results:type_name -> agentevals.api.v1alpha1.TraceResult
+	21, // 18: agentevals.api.v1alpha1.RunResult.performance_metrics:type_name -> google.protobuf.Struct
+	23, // 19: agentevals.api.v1alpha1.EvaluateTracesRequest.traces:type_name -> agentevals.api.v1alpha1.TraceInput
+	21, // 20: agentevals.api.v1alpha1.EvaluateTracesRequest.adk_eval_set:type_name -> google.protobuf.Struct
+	10, // 21: agentevals.api.v1alpha1.EvaluateTracesRequest.params:type_name -> agentevals.api.v1alpha1.EvalParams
+	20, // 22: agentevals.api.v1alpha1.EvaluateTracesRequest.credential_refs:type_name -> agentevals.api.v1alpha1.EvaluateTracesRequest.CredentialRefsEntry
+	21, // 23: agentevals.api.v1alpha1.GroupPerformance.performance_metrics:type_name -> google.protobuf.Struct
+	21, // 24: agentevals.api.v1alpha1.GroupPerformance.trace_metadata:type_name -> google.protobuf.Struct
+	17, // 25: agentevals.api.v1alpha1.EvaluateTracesResponse.progress:type_name -> agentevals.api.v1alpha1.EvaluationProgress
+	18, // 26: agentevals.api.v1alpha1.EvaluateTracesResponse.performance:type_name -> agentevals.api.v1alpha1.GroupPerformance
+	14, // 27: agentevals.api.v1alpha1.EvaluateTracesResponse.trace_result:type_name -> agentevals.api.v1alpha1.TraceResult
+	15, // 28: agentevals.api.v1alpha1.EvaluateTracesResponse.done:type_name -> agentevals.api.v1alpha1.RunResult
+	21, // 29: agentevals.api.v1alpha1.EvaluateTracesRequest.CredentialRefsEntry.value:type_name -> google.protobuf.Struct
+	2,  // 30: agentevals.api.v1alpha1.EvaluationService.ListMetrics:input_type -> agentevals.api.v1alpha1.ListMetricsRequest
+	16, // 31: agentevals.api.v1alpha1.EvaluationService.EvaluateTraces:input_type -> agentevals.api.v1alpha1.EvaluateTracesRequest
+	4,  // 32: agentevals.api.v1alpha1.EvaluationService.ListMetrics:output_type -> agentevals.api.v1alpha1.ListMetricsResponse
+	19, // 33: agentevals.api.v1alpha1.EvaluationService.EvaluateTraces:output_type -> agentevals.api.v1alpha1.EvaluateTracesResponse
+	32, // [32:34] is the sub-list for method output_type
+	30, // [30:32] is the sub-list for method input_type
+	30, // [30:30] is the sub-list for extension type_name
+	30, // [30:30] is the sub-list for extension extendee
+	0,  // [0:30] is the sub-list for field type_name
 }
 
 func init() { file_agentevals_api_v1alpha1_evaluation_proto_init() }
@@ -1737,20 +1805,20 @@ func file_agentevals_api_v1alpha1_evaluation_proto_init() {
 	}
 	file_agentevals_api_v1alpha1_common_proto_init()
 	file_agentevals_api_v1alpha1_evaluation_proto_msgTypes[3].OneofWrappers = []any{}
-	file_agentevals_api_v1alpha1_evaluation_proto_msgTypes[4].OneofWrappers = []any{}
 	file_agentevals_api_v1alpha1_evaluation_proto_msgTypes[5].OneofWrappers = []any{}
-	file_agentevals_api_v1alpha1_evaluation_proto_msgTypes[6].OneofWrappers = []any{
+	file_agentevals_api_v1alpha1_evaluation_proto_msgTypes[6].OneofWrappers = []any{}
+	file_agentevals_api_v1alpha1_evaluation_proto_msgTypes[7].OneofWrappers = []any{
 		(*Evaluator_Builtin)(nil),
 		(*Evaluator_Code)(nil),
 		(*Evaluator_Remote)(nil),
 	}
-	file_agentevals_api_v1alpha1_evaluation_proto_msgTypes[7].OneofWrappers = []any{}
 	file_agentevals_api_v1alpha1_evaluation_proto_msgTypes[8].OneofWrappers = []any{}
 	file_agentevals_api_v1alpha1_evaluation_proto_msgTypes[9].OneofWrappers = []any{}
-	file_agentevals_api_v1alpha1_evaluation_proto_msgTypes[13].OneofWrappers = []any{
+	file_agentevals_api_v1alpha1_evaluation_proto_msgTypes[10].OneofWrappers = []any{}
+	file_agentevals_api_v1alpha1_evaluation_proto_msgTypes[14].OneofWrappers = []any{
 		(*EvaluateTracesRequest_AdkEvalSet)(nil),
 	}
-	file_agentevals_api_v1alpha1_evaluation_proto_msgTypes[16].OneofWrappers = []any{
+	file_agentevals_api_v1alpha1_evaluation_proto_msgTypes[17].OneofWrappers = []any{
 		(*EvaluateTracesResponse_Progress)(nil),
 		(*EvaluateTracesResponse_Performance)(nil),
 		(*EvaluateTracesResponse_TraceResult)(nil),
@@ -1762,7 +1830,7 @@ func file_agentevals_api_v1alpha1_evaluation_proto_init() {
 			GoPackagePath: reflect.TypeOf(x{}).PkgPath(),
 			RawDescriptor: unsafe.Slice(unsafe.StringData(file_agentevals_api_v1alpha1_evaluation_proto_rawDesc), len(file_agentevals_api_v1alpha1_evaluation_proto_rawDesc)),
 			NumEnums:      2,
-			NumMessages:   18,
+			NumMessages:   19,
 			NumExtensions: 0,
 			NumServices:   1,
 		},
diff --git a/proto/agentevals/api/v1alpha1/evaluation.proto b/proto/agentevals/api/v1alpha1/evaluation.proto
index ebfb98c..8c59a65 100644
--- a/proto/agentevals/api/v1alpha1/evaluation.proto
+++ b/proto/agentevals/api/v1alpha1/evaluation.proto
@@ -27,7 +27,7 @@ message Metric {
   bool requires_llm = 5;
   // Needs Google Cloud credentials.
   bool requires_gcp = 6;
-  // Needs rubrics in the eval set.
+  // Needs rubrics, from the evaluator or the eval set.
   bool requires_rubrics = 7;
   // Usable with recorded traces.
   bool working = 8;
@@ -56,6 +56,14 @@ message BuiltinEvaluator {
   string credential_ref = 5;
   // Base URL of an OpenAI compatible judge endpoint.
   string judge_base_url = 6;
+  // Rubrics for the rubric_based_* metrics that take them. Ids are unique.
+  repeated Rubric rubrics = 7;
+}
+
+// One property a rubric metric's judge checks.
+message Rubric {
+  string id = 1;
+  string text = 2;
 }
 
 // An evaluator file on the server, run as a subprocess.
diff --git a/src/agentevals/adk_bridge.py b/src/agentevals/adk_bridge.py
index 1ef84e7..dc74d29 100644
--- a/src/agentevals/adk_bridge.py
+++ b/src/agentevals/adk_bridge.py
@@ -14,9 +14,11 @@
 from typing import Any
 
 from google.adk.evaluation.eval_case import EvalCase, IntermediateData, Invocation
+from google.adk.evaluation.eval_rubrics import Rubric, RubricContent
 from google.adk.evaluation.eval_set import EvalSet
 from google.genai import types as genai_types
 
+from .config import RubricDef
 from .genai.matching import ExpectedConversation, ExpectedTurn
 from .genai.messages import text_of
 from .genai.model import Turn
@@ -71,6 +73,16 @@ def expected_invocations(expected: ExpectedConversation | None) -> list[Invocati
     return list(expected.raw.conversation or [])
 
 
+def criterion_rubrics(definitions: list[RubricDef] | None) -> list[Rubric]:
+    return [Rubric(rubric_id=r.id, rubric_content=RubricContent(text_property=r.text)) for r in definitions or []]
+
+
+def expected_rubrics(expected: ExpectedConversation | None) -> list[Rubric]:
+    if expected is None or not isinstance(expected.raw, EvalCase):
+        return []
+    return list(expected.raw.rubrics or [])
+
+
 def multi_turn_cases(eval_set: EvalSet | None) -> bool:
     return any(len(case.conversation or []) > 1 for case in (eval_set.eval_cases if eval_set else []))
 
diff --git a/src/agentevals/api/rpc/convert.py b/src/agentevals/api/rpc/convert.py
index e62a110..777b6c9 100644
--- a/src/agentevals/api/rpc/convert.py
+++ b/src/agentevals/api/rpc/convert.py
@@ -179,6 +179,8 @@ def _evaluator_dict(evaluator: evaluation_pb2.Evaluator) -> dict[str, Any]:
                 out[field] = getattr(message, field)
         if message.trajectory_match_type in TRAJECTORY_MATCH_FROM_PB:
             out["trajectory_match_type"] = TRAJECTORY_MATCH_FROM_PB[message.trajectory_match_type]
+        if message.rubrics:
+            out["rubrics"] = [{"id": r.id, "text": r.text} for r in message.rubrics]
         return out
     if message.HasField("timeout"):
         out["timeout"] = message.timeout
@@ -223,6 +225,7 @@ def _evaluator_to_pb(evaluator: Any) -> evaluation_pb2.Evaluator:
             trajectory_match_type=TRAJECTORY_MATCH.get(evaluator.trajectory_match_type or "", 0),
             credential_ref=evaluator.credential_ref or "",
             judge_base_url=evaluator.judge_base_url or "",
+            rubrics=[evaluation_pb2.Rubric(id=r.id, text=r.text) for r in evaluator.rubrics or []],
         )
         if evaluator.threshold is not None:
             builtin.threshold = evaluator.threshold
diff --git a/src/agentevals/api/v1alpha1/evaluation_pb2.py b/src/agentevals/api/v1alpha1/evaluation_pb2.py
index 821d2cc..ad0e12a 100644
--- a/src/agentevals/api/v1alpha1/evaluation_pb2.py
+++ b/src/agentevals/api/v1alpha1/evaluation_pb2.py
@@ -26,7 +26,7 @@
 from google.protobuf import struct_pb2 as google_dot_protobuf_dot_struct__pb2
 
 
-DESCRIPTOR = _descriptor_pool.Default().AddSerializedFile(b'\n(agentevals/api/v1alpha1/evaluation.proto\x12\x17\x61gentevals.api.v1alpha1\x1a$agentevals/api/v1alpha1/common.proto\x1a\x1cgoogle/protobuf/struct.proto\"\x14\n\x12ListMetricsRequest\"\x91\x02\n\x06Metric\x12\x12\n\x04name\x18\x01 \x01(\tR\x04name\x12\x1a\n\x08\x63\x61tegory\x18\x02 \x01(\tR\x08\x63\x61tegory\x12 \n\x0b\x64\x65scription\x18\x03 \x01(\tR\x0b\x64\x65scription\x12*\n\x11requires_eval_set\x18\x04 \x01(\x08R\x0frequiresEvalSet\x12!\n\x0crequires_llm\x18\x05 \x01(\x08R\x0brequiresLlm\x12!\n\x0crequires_gcp\x18\x06 \x01(\x08R\x0brequiresGcp\x12)\n\x10requires_rubrics\x18\x07 \x01(\x08R\x0frequiresRubrics\x12\x18\n\x07working\x18\x08 \x01(\x08R\x07working\"P\n\x13ListMetricsResponse\x12\x39\n\x07metrics\x18\x01 \x03(\x0b\x32\x1f.agentevals.api.v1alpha1.MetricR\x07metrics\"\xa7\x02\n\x10\x42uiltinEvaluator\x12\x12\n\x04name\x18\x01 \x01(\tR\x04name\x12!\n\tthreshold\x18\x02 \x01(\x01H\x00R\tthreshold\x88\x01\x01\x12\x1f\n\x0bjudge_model\x18\x03 \x01(\tR\njudgeModel\x12`\n\x15trajectory_match_type\x18\x04 \x01(\x0e\x32,.agentevals.api.v1alpha1.TrajectoryMatchTypeR\x13trajectoryMatchType\x12%\n\x0e\x63redential_ref\x18\x05 \x01(\tR\rcredentialRef\x12$\n\x0ejudge_base_url\x18\x06 \x01(\tR\x0cjudgeBaseUrlB\x0c\n\n_threshold\"\xe0\x01\n\rCodeEvaluator\x12\x12\n\x04name\x18\x01 \x01(\tR\x04name\x12!\n\tthreshold\x18\x02 \x01(\x01H\x00R\tthreshold\x88\x01\x01\x12\x1d\n\x07timeout\x18\x03 \x01(\x05H\x01R\x07timeout\x88\x01\x01\x12/\n\x06\x63onfig\x18\x04 \x01(\x0b\x32\x17.google.protobuf.StructR\x06\x63onfig\x12\x1a\n\x08\x65xecutor\x18\x05 \x01(\tR\x08\x65xecutor\x12\x12\n\x04path\x18\x06 \x01(\tR\x04pathB\x0c\n\n_thresholdB\n\n\x08_timeout\"\xf8\x01\n\x0fRemoteEvaluator\x12\x12\n\x04name\x18\x01 \x01(\tR\x04name\x12!\n\tthreshold\x18\x02 \x01(\x01H\x00R\tthreshold\x88\x01\x01\x12\x1d\n\x07timeout\x18\x03 \x01(\x05H\x01R\x07timeout\x88\x01\x01\x12/\n\x06\x63onfig\x18\x04 \x01(\x0b\x32\x17.google.protobuf.StructR\x06\x63onfig\x12\x1a\n\x08\x65xecutor\x18\x05 \x01(\tR\x08\x65xecutor\x12\x16\n\x06source\x18\x06 \x01(\tR\x06source\x12\x10\n\x03ref\x18\x07 \x01(\tR\x03refB\x0c\n\n_thresholdB\n\n\x08_timeout\"\xdc\x01\n\tEvaluator\x12\x45\n\x07\x62uiltin\x18\x01 \x01(\x0b\x32).agentevals.api.v1alpha1.BuiltinEvaluatorH\x00R\x07\x62uiltin\x12<\n\x04\x63ode\x18\x02 \x01(\x0b\x32&.agentevals.api.v1alpha1.CodeEvaluatorH\x00R\x04\x63ode\x12\x42\n\x06remote\x18\x03 \x01(\x0b\x32(.agentevals.api.v1alpha1.RemoteEvaluatorH\x00R\x06remoteB\x06\n\x04kind\"\xb0\x02\n\nEvalParams\x12\x42\n\nevaluators\x18\x01 \x03(\x0b\x32\".agentevals.api.v1alpha1.EvaluatorR\nevaluators\x12\x37\n\x15max_concurrent_traces\x18\x02 \x01(\x05H\x00R\x13maxConcurrentTraces\x88\x01\x01\x12\x35\n\x14max_concurrent_evals\x18\x03 \x01(\x05H\x01R\x12maxConcurrentEvals\x88\x01\x01\x12;\n\x08group_by\x18\x04 \x01(\x0e\x32 .agentevals.api.v1alpha1.GroupByR\x07groupByB\x18\n\x16_max_concurrent_tracesB\x17\n\x15_max_concurrent_evals\"t\n\x10InvocationResult\x12\x19\n\x05score\x18\x01 \x01(\x01H\x00R\x05score\x88\x01\x01\x12;\n\x06status\x18\x02 \x01(\x0e\x32#.agentevals.api.v1alpha1.EvalStatusR\x06statusB\x08\n\x06_score\"\xad\x03\n\x0cMetricResult\x12\x1f\n\x0bmetric_name\x18\x01 \x01(\tR\nmetricName\x12\x19\n\x05score\x18\x02 \x01(\x01H\x00R\x05score\x88\x01\x01\x12\x44\n\x0b\x65val_status\x18\x03 \x01(\x0e\x32#.agentevals.api.v1alpha1.EvalStatusR\nevalStatus\x12P\n\x0eper_invocation\x18\x04 \x03(\x0b\x32).agentevals.api.v1alpha1.InvocationResultR\rperInvocation\x12\x19\n\x05\x65rror\x18\x05 \x01(\tH\x01R\x05\x65rror\x88\x01\x01\x12\"\n\nerror_type\x18\x06 \x01(\tH\x02R\terrorType\x88\x01\x01\x12\x31\n\x07\x64\x65tails\x18\x07 \x01(\x0b\x32\x17.google.protobuf.StructR\x07\x64\x65tails\x12$\n\x0b\x64uration_ms\x18\x08 \x01(\x01H\x03R\ndurationMs\x88\x01\x01\x42\x08\n\x06_scoreB\x08\n\x06_errorB\r\n\x0b_error_typeB\x0e\n\x0c_duration_ms\"=\n\x07TurnRef\x12\x19\n\x08trace_id\x18\x01 \x01(\tR\x07traceId\x12\x17\n\x07span_id\x18\x02 \x01(\tR\x06spanId\"\xc8\x04\n\x0bTraceResult\x12\x19\n\x08trace_id\x18\x01 \x01(\tR\x07traceId\x12\'\n\x0fnum_invocations\x18\x02 \x01(\x05R\x0enumInvocations\x12L\n\x0emetric_results\x18\x03 \x03(\x0b\x32%.agentevals.api.v1alpha1.MetricResultR\rmetricResults\x12/\n\x13\x63onversion_warnings\x18\x04 \x03(\tR\x12\x63onversionWarnings\x12H\n\x13performance_metrics\x18\x05 \x01(\x0b\x32\x17.google.protobuf.StructR\x12performanceMetrics\x12!\n\x0cservice_name\x18\x06 \x01(\tR\x0bserviceName\x12\x1d\n\nagent_name\x18\x07 \x01(\tR\tagentName\x12\x1b\n\ttrace_ids\x18\x08 \x03(\tR\x08traceIds\x12\'\n\x0f\x63onversation_id\x18\t \x01(\tR\x0e\x63onversationId\x12\x1b\n\tgroup_key\x18\n \x01(\tR\x08groupKey\x12 \n\x0c\x65val_case_id\x18\x0b \x01(\tR\nevalCaseId\x12&\n\x0f\x65val_case_match\x18\x0c \x01(\tR\revalCaseMatch\x12=\n\tturn_refs\x18\r \x03(\x0b\x32 .agentevals.api.v1alpha1.TurnRefR\x08turnRefs\"\xcf\x01\n\tRunResult\x12I\n\rtrace_results\x18\x01 \x03(\x0b\x32$.agentevals.api.v1alpha1.TraceResultR\x0ctraceResults\x12\x16\n\x06\x65rrors\x18\x02 \x03(\tR\x06\x65rrors\x12H\n\x13performance_metrics\x18\x03 \x01(\x0b\x32\x17.google.protobuf.StructR\x12performanceMetrics\x12\x15\n\x06run_id\x18\x04 \x01(\tR\x05runId\"\xa3\x03\n\x15\x45valuateTracesRequest\x12;\n\x06traces\x18\x01 \x01(\x0b\x32#.agentevals.api.v1alpha1.TraceInputR\x06traces\x12;\n\x0c\x61\x64k_eval_set\x18\x02 \x01(\x0b\x32\x17.google.protobuf.StructH\x00R\nadkEvalSet\x12;\n\x06params\x18\x03 \x01(\x0b\x32#.agentevals.api.v1alpha1.EvalParamsR\x06params\x12k\n\x0f\x63redential_refs\x18\x04 \x03(\x0b\x32\x42.agentevals.api.v1alpha1.EvaluateTracesRequest.CredentialRefsEntryR\x0e\x63redentialRefs\x1aZ\n\x13\x43redentialRefsEntry\x12\x10\n\x03key\x18\x01 \x01(\tR\x03key\x12-\n\x05value\x18\x02 \x01(\x0b\x32\x17.google.protobuf.StructR\x05value:\x02\x38\x01\x42\n\n\x08\x65val_set\".\n\x12\x45valuationProgress\x12\x18\n\x07message\x18\x01 \x01(\tR\x07message\"\xb7\x01\n\x10GroupPerformance\x12\x19\n\x08trace_id\x18\x01 \x01(\tR\x07traceId\x12H\n\x13performance_metrics\x18\x02 \x01(\x0b\x32\x17.google.protobuf.StructR\x12performanceMetrics\x12>\n\x0etrace_metadata\x18\x03 \x01(\x0b\x32\x17.google.protobuf.StructR\rtraceMetadata\"\xc0\x02\n\x16\x45valuateTracesResponse\x12I\n\x08progress\x18\x01 \x01(\x0b\x32+.agentevals.api.v1alpha1.EvaluationProgressH\x00R\x08progress\x12M\n\x0bperformance\x18\x02 \x01(\x0b\x32).agentevals.api.v1alpha1.GroupPerformanceH\x00R\x0bperformance\x12I\n\x0ctrace_result\x18\x03 \x01(\x0b\x32$.agentevals.api.v1alpha1.TraceResultH\x00R\x0btraceResult\x12\x38\n\x04\x64one\x18\x04 \x01(\x0b\x32\".agentevals.api.v1alpha1.RunResultH\x00R\x04\x64oneB\x07\n\x05\x65vent*\xa6\x01\n\x13TrajectoryMatchType\x12%\n!TRAJECTORY_MATCH_TYPE_UNSPECIFIED\x10\x00\x12\x1f\n\x1bTRAJECTORY_MATCH_TYPE_EXACT\x10\x01\x12\"\n\x1eTRAJECTORY_MATCH_TYPE_IN_ORDER\x10\x02\x12#\n\x1fTRAJECTORY_MATCH_TYPE_ANY_ORDER\x10\x03*x\n\nEvalStatus\x12\x1b\n\x17\x45VAL_STATUS_UNSPECIFIED\x10\x00\x12\x16\n\x12\x45VAL_STATUS_PASSED\x10\x01\x12\x16\n\x12\x45VAL_STATUS_FAILED\x10\x02\x12\x1d\n\x19\x45VAL_STATUS_NOT_EVALUATED\x10\x03\x32\xf2\x01\n\x11\x45valuationService\x12h\n\x0bListMetrics\x12+.agentevals.api.v1alpha1.ListMetricsRequest\x1a,.agentevals.api.v1alpha1.ListMetricsResponse\x12s\n\x0e\x45valuateTraces\x12..agentevals.api.v1alpha1.EvaluateTracesRequest\x1a/.agentevals.api.v1alpha1.EvaluateTracesResponse0\x01\x42QZOgithub.com/agentevals-dev/agentevals/gen/go/agentevals/api/v1alpha1;apiv1alpha1b\x06proto3')
+DESCRIPTOR = _descriptor_pool.Default().AddSerializedFile(b'\n(agentevals/api/v1alpha1/evaluation.proto\x12\x17\x61gentevals.api.v1alpha1\x1a$agentevals/api/v1alpha1/common.proto\x1a\x1cgoogle/protobuf/struct.proto\"\x14\n\x12ListMetricsRequest\"\x91\x02\n\x06Metric\x12\x12\n\x04name\x18\x01 \x01(\tR\x04name\x12\x1a\n\x08\x63\x61tegory\x18\x02 \x01(\tR\x08\x63\x61tegory\x12 \n\x0b\x64\x65scription\x18\x03 \x01(\tR\x0b\x64\x65scription\x12*\n\x11requires_eval_set\x18\x04 \x01(\x08R\x0frequiresEvalSet\x12!\n\x0crequires_llm\x18\x05 \x01(\x08R\x0brequiresLlm\x12!\n\x0crequires_gcp\x18\x06 \x01(\x08R\x0brequiresGcp\x12)\n\x10requires_rubrics\x18\x07 \x01(\x08R\x0frequiresRubrics\x12\x18\n\x07working\x18\x08 \x01(\x08R\x07working\"P\n\x13ListMetricsResponse\x12\x39\n\x07metrics\x18\x01 \x03(\x0b\x32\x1f.agentevals.api.v1alpha1.MetricR\x07metrics\"\xe2\x02\n\x10\x42uiltinEvaluator\x12\x12\n\x04name\x18\x01 \x01(\tR\x04name\x12!\n\tthreshold\x18\x02 \x01(\x01H\x00R\tthreshold\x88\x01\x01\x12\x1f\n\x0bjudge_model\x18\x03 \x01(\tR\njudgeModel\x12`\n\x15trajectory_match_type\x18\x04 \x01(\x0e\x32,.agentevals.api.v1alpha1.TrajectoryMatchTypeR\x13trajectoryMatchType\x12%\n\x0e\x63redential_ref\x18\x05 \x01(\tR\rcredentialRef\x12$\n\x0ejudge_base_url\x18\x06 \x01(\tR\x0cjudgeBaseUrl\x12\x39\n\x07rubrics\x18\x07 \x03(\x0b\x32\x1f.agentevals.api.v1alpha1.RubricR\x07rubricsB\x0c\n\n_threshold\",\n\x06Rubric\x12\x0e\n\x02id\x18\x01 \x01(\tR\x02id\x12\x12\n\x04text\x18\x02 \x01(\tR\x04text\"\xe0\x01\n\rCodeEvaluator\x12\x12\n\x04name\x18\x01 \x01(\tR\x04name\x12!\n\tthreshold\x18\x02 \x01(\x01H\x00R\tthreshold\x88\x01\x01\x12\x1d\n\x07timeout\x18\x03 \x01(\x05H\x01R\x07timeout\x88\x01\x01\x12/\n\x06\x63onfig\x18\x04 \x01(\x0b\x32\x17.google.protobuf.StructR\x06\x63onfig\x12\x1a\n\x08\x65xecutor\x18\x05 \x01(\tR\x08\x65xecutor\x12\x12\n\x04path\x18\x06 \x01(\tR\x04pathB\x0c\n\n_thresholdB\n\n\x08_timeout\"\xf8\x01\n\x0fRemoteEvaluator\x12\x12\n\x04name\x18\x01 \x01(\tR\x04name\x12!\n\tthreshold\x18\x02 \x01(\x01H\x00R\tthreshold\x88\x01\x01\x12\x1d\n\x07timeout\x18\x03 \x01(\x05H\x01R\x07timeout\x88\x01\x01\x12/\n\x06\x63onfig\x18\x04 \x01(\x0b\x32\x17.google.protobuf.StructR\x06\x63onfig\x12\x1a\n\x08\x65xecutor\x18\x05 \x01(\tR\x08\x65xecutor\x12\x16\n\x06source\x18\x06 \x01(\tR\x06source\x12\x10\n\x03ref\x18\x07 \x01(\tR\x03refB\x0c\n\n_thresholdB\n\n\x08_timeout\"\xdc\x01\n\tEvaluator\x12\x45\n\x07\x62uiltin\x18\x01 \x01(\x0b\x32).agentevals.api.v1alpha1.BuiltinEvaluatorH\x00R\x07\x62uiltin\x12<\n\x04\x63ode\x18\x02 \x01(\x0b\x32&.agentevals.api.v1alpha1.CodeEvaluatorH\x00R\x04\x63ode\x12\x42\n\x06remote\x18\x03 \x01(\x0b\x32(.agentevals.api.v1alpha1.RemoteEvaluatorH\x00R\x06remoteB\x06\n\x04kind\"\xb0\x02\n\nEvalParams\x12\x42\n\nevaluators\x18\x01 \x03(\x0b\x32\".agentevals.api.v1alpha1.EvaluatorR\nevaluators\x12\x37\n\x15max_concurrent_traces\x18\x02 \x01(\x05H\x00R\x13maxConcurrentTraces\x88\x01\x01\x12\x35\n\x14max_concurrent_evals\x18\x03 \x01(\x05H\x01R\x12maxConcurrentEvals\x88\x01\x01\x12;\n\x08group_by\x18\x04 \x01(\x0e\x32 .agentevals.api.v1alpha1.GroupByR\x07groupByB\x18\n\x16_max_concurrent_tracesB\x17\n\x15_max_concurrent_evals\"t\n\x10InvocationResult\x12\x19\n\x05score\x18\x01 \x01(\x01H\x00R\x05score\x88\x01\x01\x12;\n\x06status\x18\x02 \x01(\x0e\x32#.agentevals.api.v1alpha1.EvalStatusR\x06statusB\x08\n\x06_score\"\xad\x03\n\x0cMetricResult\x12\x1f\n\x0bmetric_name\x18\x01 \x01(\tR\nmetricName\x12\x19\n\x05score\x18\x02 \x01(\x01H\x00R\x05score\x88\x01\x01\x12\x44\n\x0b\x65val_status\x18\x03 \x01(\x0e\x32#.agentevals.api.v1alpha1.EvalStatusR\nevalStatus\x12P\n\x0eper_invocation\x18\x04 \x03(\x0b\x32).agentevals.api.v1alpha1.InvocationResultR\rperInvocation\x12\x19\n\x05\x65rror\x18\x05 \x01(\tH\x01R\x05\x65rror\x88\x01\x01\x12\"\n\nerror_type\x18\x06 \x01(\tH\x02R\terrorType\x88\x01\x01\x12\x31\n\x07\x64\x65tails\x18\x07 \x01(\x0b\x32\x17.google.protobuf.StructR\x07\x64\x65tails\x12$\n\x0b\x64uration_ms\x18\x08 \x01(\x01H\x03R\ndurationMs\x88\x01\x01\x42\x08\n\x06_scoreB\x08\n\x06_errorB\r\n\x0b_error_typeB\x0e\n\x0c_duration_ms\"=\n\x07TurnRef\x12\x19\n\x08trace_id\x18\x01 \x01(\tR\x07traceId\x12\x17\n\x07span_id\x18\x02 \x01(\tR\x06spanId\"\xc8\x04\n\x0bTraceResult\x12\x19\n\x08trace_id\x18\x01 \x01(\tR\x07traceId\x12\'\n\x0fnum_invocations\x18\x02 \x01(\x05R\x0enumInvocations\x12L\n\x0emetric_results\x18\x03 \x03(\x0b\x32%.agentevals.api.v1alpha1.MetricResultR\rmetricResults\x12/\n\x13\x63onversion_warnings\x18\x04 \x03(\tR\x12\x63onversionWarnings\x12H\n\x13performance_metrics\x18\x05 \x01(\x0b\x32\x17.google.protobuf.StructR\x12performanceMetrics\x12!\n\x0cservice_name\x18\x06 \x01(\tR\x0bserviceName\x12\x1d\n\nagent_name\x18\x07 \x01(\tR\tagentName\x12\x1b\n\ttrace_ids\x18\x08 \x03(\tR\x08traceIds\x12\'\n\x0f\x63onversation_id\x18\t \x01(\tR\x0e\x63onversationId\x12\x1b\n\tgroup_key\x18\n \x01(\tR\x08groupKey\x12 \n\x0c\x65val_case_id\x18\x0b \x01(\tR\nevalCaseId\x12&\n\x0f\x65val_case_match\x18\x0c \x01(\tR\revalCaseMatch\x12=\n\tturn_refs\x18\r \x03(\x0b\x32 .agentevals.api.v1alpha1.TurnRefR\x08turnRefs\"\xcf\x01\n\tRunResult\x12I\n\rtrace_results\x18\x01 \x03(\x0b\x32$.agentevals.api.v1alpha1.TraceResultR\x0ctraceResults\x12\x16\n\x06\x65rrors\x18\x02 \x03(\tR\x06\x65rrors\x12H\n\x13performance_metrics\x18\x03 \x01(\x0b\x32\x17.google.protobuf.StructR\x12performanceMetrics\x12\x15\n\x06run_id\x18\x04 \x01(\tR\x05runId\"\xa3\x03\n\x15\x45valuateTracesRequest\x12;\n\x06traces\x18\x01 \x01(\x0b\x32#.agentevals.api.v1alpha1.TraceInputR\x06traces\x12;\n\x0c\x61\x64k_eval_set\x18\x02 \x01(\x0b\x32\x17.google.protobuf.StructH\x00R\nadkEvalSet\x12;\n\x06params\x18\x03 \x01(\x0b\x32#.agentevals.api.v1alpha1.EvalParamsR\x06params\x12k\n\x0f\x63redential_refs\x18\x04 \x03(\x0b\x32\x42.agentevals.api.v1alpha1.EvaluateTracesRequest.CredentialRefsEntryR\x0e\x63redentialRefs\x1aZ\n\x13\x43redentialRefsEntry\x12\x10\n\x03key\x18\x01 \x01(\tR\x03key\x12-\n\x05value\x18\x02 \x01(\x0b\x32\x17.google.protobuf.StructR\x05value:\x02\x38\x01\x42\n\n\x08\x65val_set\".\n\x12\x45valuationProgress\x12\x18\n\x07message\x18\x01 \x01(\tR\x07message\"\xb7\x01\n\x10GroupPerformance\x12\x19\n\x08trace_id\x18\x01 \x01(\tR\x07traceId\x12H\n\x13performance_metrics\x18\x02 \x01(\x0b\x32\x17.google.protobuf.StructR\x12performanceMetrics\x12>\n\x0etrace_metadata\x18\x03 \x01(\x0b\x32\x17.google.protobuf.StructR\rtraceMetadata\"\xc0\x02\n\x16\x45valuateTracesResponse\x12I\n\x08progress\x18\x01 \x01(\x0b\x32+.agentevals.api.v1alpha1.EvaluationProgressH\x00R\x08progress\x12M\n\x0bperformance\x18\x02 \x01(\x0b\x32).agentevals.api.v1alpha1.GroupPerformanceH\x00R\x0bperformance\x12I\n\x0ctrace_result\x18\x03 \x01(\x0b\x32$.agentevals.api.v1alpha1.TraceResultH\x00R\x0btraceResult\x12\x38\n\x04\x64one\x18\x04 \x01(\x0b\x32\".agentevals.api.v1alpha1.RunResultH\x00R\x04\x64oneB\x07\n\x05\x65vent*\xa6\x01\n\x13TrajectoryMatchType\x12%\n!TRAJECTORY_MATCH_TYPE_UNSPECIFIED\x10\x00\x12\x1f\n\x1bTRAJECTORY_MATCH_TYPE_EXACT\x10\x01\x12\"\n\x1eTRAJECTORY_MATCH_TYPE_IN_ORDER\x10\x02\x12#\n\x1fTRAJECTORY_MATCH_TYPE_ANY_ORDER\x10\x03*x\n\nEvalStatus\x12\x1b\n\x17\x45VAL_STATUS_UNSPECIFIED\x10\x00\x12\x16\n\x12\x45VAL_STATUS_PASSED\x10\x01\x12\x16\n\x12\x45VAL_STATUS_FAILED\x10\x02\x12\x1d\n\x19\x45VAL_STATUS_NOT_EVALUATED\x10\x03\x32\xf2\x01\n\x11\x45valuationService\x12h\n\x0bListMetrics\x12+.agentevals.api.v1alpha1.ListMetricsRequest\x1a,.agentevals.api.v1alpha1.ListMetricsResponse\x12s\n\x0e\x45valuateTraces\x12..agentevals.api.v1alpha1.EvaluateTracesRequest\x1a/.agentevals.api.v1alpha1.EvaluateTracesResponse0\x01\x42QZOgithub.com/agentevals-dev/agentevals/gen/go/agentevals/api/v1alpha1;apiv1alpha1b\x06proto3')
 
 _globals = globals()
 _builder.BuildMessageAndEnumDescriptors(DESCRIPTOR, _globals)
@@ -36,10 +36,10 @@
   _globals['DESCRIPTOR']._serialized_options = b'ZOgithub.com/agentevals-dev/agentevals/gen/go/agentevals/api/v1alpha1;apiv1alpha1'
   _globals['_EVALUATETRACESREQUEST_CREDENTIALREFSENTRY']._loaded_options = None
   _globals['_EVALUATETRACESREQUEST_CREDENTIALREFSENTRY']._serialized_options = b'8\001'
-  _globals['_TRAJECTORYMATCHTYPE']._serialized_start=4213
-  _globals['_TRAJECTORYMATCHTYPE']._serialized_end=4379
-  _globals['_EVALSTATUS']._serialized_start=4381
-  _globals['_EVALSTATUS']._serialized_end=4501
+  _globals['_TRAJECTORYMATCHTYPE']._serialized_start=4318
+  _globals['_TRAJECTORYMATCHTYPE']._serialized_end=4484
+  _globals['_EVALSTATUS']._serialized_start=4486
+  _globals['_EVALSTATUS']._serialized_end=4606
   _globals['_LISTMETRICSREQUEST']._serialized_start=137
   _globals['_LISTMETRICSREQUEST']._serialized_end=157
   _globals['_METRIC']._serialized_start=160
@@ -47,35 +47,37 @@
   _globals['_LISTMETRICSRESPONSE']._serialized_start=435
   _globals['_LISTMETRICSRESPONSE']._serialized_end=515
   _globals['_BUILTINEVALUATOR']._serialized_start=518
-  _globals['_BUILTINEVALUATOR']._serialized_end=813
-  _globals['_CODEEVALUATOR']._serialized_start=816
-  _globals['_CODEEVALUATOR']._serialized_end=1040
-  _globals['_REMOTEEVALUATOR']._serialized_start=1043
-  _globals['_REMOTEEVALUATOR']._serialized_end=1291
-  _globals['_EVALUATOR']._serialized_start=1294
-  _globals['_EVALUATOR']._serialized_end=1514
-  _globals['_EVALPARAMS']._serialized_start=1517
-  _globals['_EVALPARAMS']._serialized_end=1821
-  _globals['_INVOCATIONRESULT']._serialized_start=1823
-  _globals['_INVOCATIONRESULT']._serialized_end=1939
-  _globals['_METRICRESULT']._serialized_start=1942
-  _globals['_METRICRESULT']._serialized_end=2371
-  _globals['_TURNREF']._serialized_start=2373
-  _globals['_TURNREF']._serialized_end=2434
-  _globals['_TRACERESULT']._serialized_start=2437
-  _globals['_TRACERESULT']._serialized_end=3021
-  _globals['_RUNRESULT']._serialized_start=3024
-  _globals['_RUNRESULT']._serialized_end=3231
-  _globals['_EVALUATETRACESREQUEST']._serialized_start=3234
-  _globals['_EVALUATETRACESREQUEST']._serialized_end=3653
-  _globals['_EVALUATETRACESREQUEST_CREDENTIALREFSENTRY']._serialized_start=3551
-  _globals['_EVALUATETRACESREQUEST_CREDENTIALREFSENTRY']._serialized_end=3641
-  _globals['_EVALUATIONPROGRESS']._serialized_start=3655
-  _globals['_EVALUATIONPROGRESS']._serialized_end=3701
-  _globals['_GROUPPERFORMANCE']._serialized_start=3704
-  _globals['_GROUPPERFORMANCE']._serialized_end=3887
-  _globals['_EVALUATETRACESRESPONSE']._serialized_start=3890
-  _globals['_EVALUATETRACESRESPONSE']._serialized_end=4210
-  _globals['_EVALUATIONSERVICE']._serialized_start=4504
-  _globals['_EVALUATIONSERVICE']._serialized_end=4746
+  _globals['_BUILTINEVALUATOR']._serialized_end=872
+  _globals['_RUBRIC']._serialized_start=874
+  _globals['_RUBRIC']._serialized_end=918
+  _globals['_CODEEVALUATOR']._serialized_start=921
+  _globals['_CODEEVALUATOR']._serialized_end=1145
+  _globals['_REMOTEEVALUATOR']._serialized_start=1148
+  _globals['_REMOTEEVALUATOR']._serialized_end=1396
+  _globals['_EVALUATOR']._serialized_start=1399
+  _globals['_EVALUATOR']._serialized_end=1619
+  _globals['_EVALPARAMS']._serialized_start=1622
+  _globals['_EVALPARAMS']._serialized_end=1926
+  _globals['_INVOCATIONRESULT']._serialized_start=1928
+  _globals['_INVOCATIONRESULT']._serialized_end=2044
+  _globals['_METRICRESULT']._serialized_start=2047
+  _globals['_METRICRESULT']._serialized_end=2476
+  _globals['_TURNREF']._serialized_start=2478
+  _globals['_TURNREF']._serialized_end=2539
+  _globals['_TRACERESULT']._serialized_start=2542
+  _globals['_TRACERESULT']._serialized_end=3126
+  _globals['_RUNRESULT']._serialized_start=3129
+  _globals['_RUNRESULT']._serialized_end=3336
+  _globals['_EVALUATETRACESREQUEST']._serialized_start=3339
+  _globals['_EVALUATETRACESREQUEST']._serialized_end=3758
+  _globals['_EVALUATETRACESREQUEST_CREDENTIALREFSENTRY']._serialized_start=3656
+  _globals['_EVALUATETRACESREQUEST_CREDENTIALREFSENTRY']._serialized_end=3746
+  _globals['_EVALUATIONPROGRESS']._serialized_start=3760
+  _globals['_EVALUATIONPROGRESS']._serialized_end=3806
+  _globals['_GROUPPERFORMANCE']._serialized_start=3809
+  _globals['_GROUPPERFORMANCE']._serialized_end=3992
+  _globals['_EVALUATETRACESRESPONSE']._serialized_start=3995
+  _globals['_EVALUATETRACESRESPONSE']._serialized_end=4315
+  _globals['_EVALUATIONSERVICE']._serialized_start=4609
+  _globals['_EVALUATIONSERVICE']._serialized_end=4851
 # @@protoc_insertion_point(module_scope)
diff --git a/src/agentevals/api/v1alpha1/evaluation_pb2.pyi b/src/agentevals/api/v1alpha1/evaluation_pb2.pyi
index 03da133..c558529 100644
--- a/src/agentevals/api/v1alpha1/evaluation_pb2.pyi
+++ b/src/agentevals/api/v1alpha1/evaluation_pb2.pyi
@@ -62,20 +62,30 @@ class ListMetricsResponse(_message.Message):
     def __init__(self, metrics: _Optional[_Iterable[_Union[Metric, _Mapping]]] = ...) -> None: ...
 
 class BuiltinEvaluator(_message.Message):
-    __slots__ = ("name", "threshold", "judge_model", "trajectory_match_type", "credential_ref", "judge_base_url")
+    __slots__ = ("name", "threshold", "judge_model", "trajectory_match_type", "credential_ref", "judge_base_url", "rubrics")
     NAME_FIELD_NUMBER: _ClassVar[int]
     THRESHOLD_FIELD_NUMBER: _ClassVar[int]
     JUDGE_MODEL_FIELD_NUMBER: _ClassVar[int]
     TRAJECTORY_MATCH_TYPE_FIELD_NUMBER: _ClassVar[int]
     CREDENTIAL_REF_FIELD_NUMBER: _ClassVar[int]
     JUDGE_BASE_URL_FIELD_NUMBER: _ClassVar[int]
+    RUBRICS_FIELD_NUMBER: _ClassVar[int]
     name: str
     threshold: float
     judge_model: str
     trajectory_match_type: TrajectoryMatchType
     credential_ref: str
     judge_base_url: str
-    def __init__(self, name: _Optional[str] = ..., threshold: _Optional[float] = ..., judge_model: _Optional[str] = ..., trajectory_match_type: _Optional[_Union[TrajectoryMatchType, str]] = ..., credential_ref: _Optional[str] = ..., judge_base_url: _Optional[str] = ...) -> None: ...
+    rubrics: _containers.RepeatedCompositeFieldContainer[Rubric]
+    def __init__(self, name: _Optional[str] = ..., threshold: _Optional[float] = ..., judge_model: _Optional[str] = ..., trajectory_match_type: _Optional[_Union[TrajectoryMatchType, str]] = ..., credential_ref: _Optional[str] = ..., judge_base_url: _Optional[str] = ..., rubrics: _Optional[_Iterable[_Union[Rubric, _Mapping]]] = ...) -> None: ...
+
+class Rubric(_message.Message):
+    __slots__ = ("id", "text")
+    ID_FIELD_NUMBER: _ClassVar[int]
+    TEXT_FIELD_NUMBER: _ClassVar[int]
+    id: str
+    text: str
+    def __init__(self, id: _Optional[str] = ..., text: _Optional[str] = ...) -> None: ...
 
 class CodeEvaluator(_message.Message):
     __slots__ = ("name", "threshold", "timeout", "config", "executor", "path")
diff --git a/src/agentevals/builtin_metrics.py b/src/agentevals/builtin_metrics.py
index cfe5383..9e9a6b3 100644
--- a/src/agentevals/builtin_metrics.py
+++ b/src/agentevals/builtin_metrics.py
@@ -28,6 +28,7 @@
 from google.adk.evaluation.eval_rubrics import Rubric, RubricContent
 from google.adk.evaluation.evaluator import EvaluationResult, Evaluator
 
+from .config import RUBRIC_METRIC_TYPES
 from .resolvers import get_resolved_credential
 
 logger = logging.getLogger(__name__)
@@ -106,7 +107,7 @@ def metric_catalog() -> list[MetricDescription]:
                 requires_llm=name in METRICS_NEEDING_LLM,
                 requires_gcp=name in METRICS_NEEDING_GCP,
                 requires_rubrics=rubrics,
-                working=not rubrics,
+                working=not rubrics or name in RUBRIC_METRIC_TYPES,
             )
         )
     return out
@@ -235,14 +236,91 @@ def rubric_strings_to_objects(rubric_texts: list[str]) -> list[Rubric]:
     ]
 
 
+def to_rubric_objects(rubrics: list[str | Rubric] | None) -> list[Rubric]:
+    """Plain strings become positional rubrics; ADK ``Rubric`` objects pass through."""
+    return [
+        Rubric(rubric_id=f"rubric_{index}", rubric_content=RubricContent(text_property=r)) if isinstance(r, str) else r
+        for index, r in enumerate(rubrics or [])
+    ]
+
+
+def attach_case_rubrics(
+    metric_name: str,
+    actual_invocations: list[Invocation],
+    expected_invocations: list[Invocation] | None,
+    case_rubrics: list[Rubric] | None,
+    criterion_rubrics: list[Rubric] | None = None,
+) -> tuple[list[Invocation], list[str]]:
+    """Bind case and invocation rubrics without changing ADK types or turn scope.
+
+    Check effective properties before the first judge call, and report dropped
+    properties by id and source instead of silently defaulting their type.
+    """
+    metric_type = RUBRIC_METRIC_TYPES.get(metric_name)
+    attached = []
+    warnings = []
+    for index, actual in enumerate(actual_invocations):
+        expected = expected_invocations[index] if expected_invocations and index < len(expected_invocations) else None
+        sources = [
+            (f"actual invocation {index}", actual.rubrics or []),
+            ("eval case", case_rubrics or []),
+            (f"expected invocation {index}", expected.rubrics or [] if expected else []),
+        ]
+        combined = [r for _, rubrics in sources for r in rubrics]
+        for source, rubrics in sources:
+            for rubric in rubrics:
+                if rubric.type is None:
+                    warning = (
+                        f"{source} rubric '{rubric.rubric_id}' ignored: type {rubric.type!r}; "
+                        f"metric requires {metric_type!r}"
+                    )
+                    if warning not in warnings:
+                        warnings.append(warning)
+        effective = [*(criterion_rubrics or []), *(r for r in combined if r.type == metric_type)]
+        ids = [r.rubric_id for r in effective]
+        if len(ids) != len(set(ids)):
+            raise ValueError(f"Rubric ids must be distinct across metric, case and invocation {index}: {ids}")
+        attached.append(actual.model_copy(update={"rubrics": combined}))
+    return attached, warnings
+
+
+def extract_rubric_details(eval_result: EvaluationResult) -> dict[str, Any]:
+    """Per-rubric scores, per invocation and overall, from a rubric metric."""
+
+    def _score(rubric_score: Any) -> dict[str, Any]:
+        return {
+            "rubric_id": rubric_score.rubric_id,
+            "score": rubric_score.score,
+            "rationale": rubric_score.rationale,
+        }
+
+    per_invocation = []
+    for per_inv_result in eval_result.per_invocation_results:
+        actual_inv = per_inv_result.actual_invocation
+        per_invocation.append(
+            {
+                "invocation_id": actual_inv.invocation_id if actual_inv else None,
+                "rubric_scores": [_score(r) for r in per_inv_result.rubric_scores or []],
+            }
+        )
+    return {
+        "per_invocation": per_invocation,
+        "overall_rubric_scores": [_score(r) for r in eval_result.overall_rubric_scores or []],
+    }
+
+
 def build_eval_metric(
     metric_name: str,
     judge_model: str | None,
     threshold: float | None,
-    rubrics: list[str] | None = None,
+    rubrics: list[str | Rubric] | None = None,
     match_type: str | None = None,
 ) -> EvalMetric:
-    """Construct an ADK ``EvalMetric`` with the appropriate criterion."""
+    """Construct an ADK ``EvalMetric`` with the appropriate criterion.
+
+    ``rubrics`` feed the ``rubric_based_*`` metrics' criterion: plain strings
+    get positional ids, ADK ``Rubric`` objects are used as given.
+    """
     effective_threshold = threshold if threshold is not None else 0.5
 
     criterion: BaseCriterion | None = None
@@ -268,14 +346,11 @@ def build_eval_metric(
             threshold=effective_threshold,
             judge_model_options=judge_opts,
         )
-    elif metric_name in (
-        "rubric_based_final_response_quality_v1",
-        "rubric_based_tool_use_quality_v1",
-    ):
+    elif metric_name in RUBRIC_METRIC_TYPES:
         judge_opts = JudgeModelOptions()
         if judge_model:
             judge_opts.judge_model = judge_model
-        rubric_objects = rubric_strings_to_objects(rubrics) if rubrics else []
+        rubric_objects = to_rubric_objects(rubrics)
         criterion = RubricsBasedCriterion(
             threshold=effective_threshold,
             judge_model_options=judge_opts,
@@ -435,11 +510,12 @@ async def evaluate_builtin_metric(
     credential_ref: str | None = None,
     judge_base_url: str | None = None,
     expected_reason: str | None = None,
+    rubrics: list[str | Rubric] | None = None,
+    case_rubrics: list[Rubric] | None = None,
 ):
-    """Evaluate a single built-in ADK metric and return a ``MetricResult``.
-
-    A metric that compares against a golden conversation is ``NOT_EVALUATED`` when none was
-    selected, with the reason in ``details``; it is never scored against an arbitrary case.
+    """Evaluate a built-in metric. Case rubrics apply to every invocation;
+    invocation rubrics apply by position, with their ADK types unchanged.
+    Unmatched golden comparisons remain NOT_EVALUATED.
     """
     from .runner import MetricResult
 
@@ -450,8 +526,34 @@ async def evaluate_builtin_metric(
             details={"reason": expected_reason or "no eval set provided"},
         )
 
+    rubric_warnings: list[str] = []
+    if metric_name in RUBRIC_METRIC_TYPES:
+        rubric_objects = to_rubric_objects(rubrics)
+        try:
+            actual_invocations, rubric_warnings = attach_case_rubrics(
+                metric_name, actual_invocations, expected_invocations, case_rubrics, rubric_objects
+            )
+        except ValueError as exc:
+            return MetricResult(metric_name=metric_name, error=str(exc), error_type="ValueError")
+        missing = [
+            index
+            for index, inv in enumerate(actual_invocations)
+            if not rubric_objects and not any(r.type == RUBRIC_METRIC_TYPES[metric_name] for r in inv.rubrics or [])
+        ]
+        if missing or not actual_invocations:
+            return MetricResult(
+                metric_name=metric_name,
+                error=(
+                    f"Metric '{metric_name}' requires rubrics applicable to invocations {missing}: "
+                    "set 'rubrics' on the evaluator in the eval config, or correctly typed 'rubrics' "
+                    "on the matched eval case or its invocations."
+                ),
+                error_type="ValueError",
+                details={"warnings": rubric_warnings} if rubric_warnings else None,
+            )
+
     try:
-        eval_metric = build_eval_metric(metric_name, judge_model, threshold, match_type=match_type)
+        eval_metric = build_eval_metric(metric_name, judge_model, threshold, rubrics=rubrics, match_type=match_type)
         evaluator: Evaluator = get_evaluator(eval_metric)
 
         if credential_ref:
@@ -488,6 +590,10 @@ async def evaluate_builtin_metric(
         details = None
         if metric_name == "tool_trajectory_avg_score":
             details = extract_trajectory_details(eval_result)
+        elif metric_name in RUBRIC_METRIC_TYPES:
+            details = extract_rubric_details(eval_result)
+            if rubric_warnings:
+                details["warnings"] = rubric_warnings
 
         return MetricResult(
             metric_name=metric_name,
diff --git a/src/agentevals/config.py b/src/agentevals/config.py
index 0c773a5..e77d67b 100644
--- a/src/agentevals/config.py
+++ b/src/agentevals/config.py
@@ -22,9 +22,32 @@ def _normalize_trajectory_match_type(v: str | None) -> str | None:
     "per_turn_user_simulator_quality_v1": "it scores a simulated user, and agentevals evaluates recorded traces",
 }
 
+# Rubric metrics that take rubrics, with the ADK rubric type each one keeps from eval case and
+# invocation rubrics. ADK never filters config rubrics by type.
+RUBRIC_METRIC_TYPES = {
+    "rubric_based_final_response_quality_v1": "FINAL_RESPONSE_QUALITY",
+    "rubric_based_tool_use_quality_v1": "TOOL_USE_QUALITY",
+}
+
 GroupBy = Literal["auto", "trace", "conversation"]
 
 
+class RubricDef(BaseModel):
+    """A named property for a rubric metric, without ADK-specific configuration."""
+
+    model_config = ConfigDict(alias_generator=to_camel, populate_by_name=True, extra="forbid")
+
+    id: str = Field(min_length=1)
+    text: str = Field(min_length=1)
+
+    @field_validator("id", "text")
+    @classmethod
+    def _not_blank(cls, value: str) -> str:
+        if not value.strip():
+            raise ValueError("Rubric id and text must not be blank")
+        return value
+
+
 class BuiltinMetricDef(BaseModel):
     """A built-in ADK metric, optionally with threshold/judge overrides."""
 
@@ -35,6 +58,9 @@ class BuiltinMetricDef(BaseModel):
     threshold: float | None = Field(default=None, ge=0, le=1)
     judge_model: str | None = None
     trajectory_match_type: str | None = None
+    rubrics: list[RubricDef] | None = Field(
+        default=None, description="Named {id, text} properties, or strings with positional ids, for rubric metrics."
+    )
     credential_ref: str | None = Field(
         default=None,
         description="Logical name of a RunSpec.credential_refs entry whose resolved value is the judge API key.",
@@ -56,6 +82,35 @@ def _validate_name(cls, v: str) -> str:
             raise ValueError(f"Built-in metric '{v}' is not supported: {UNSUPPORTED_BUILTIN_METRICS[v]}")
         return v
 
+    @field_validator("rubrics", mode="before")
+    @classmethod
+    def _normalize_rubrics(cls, value: Any) -> Any:
+        if value is None:
+            return None
+        if not isinstance(value, list):
+            raise ValueError("'rubrics' must be a list of strings or {id, text} mappings")
+        return [
+            {"id": f"rubric_{index}", "text": entry} if isinstance(entry, str) else entry
+            for index, entry in enumerate(value)
+        ]
+
+    @field_validator("rubrics")
+    @classmethod
+    def _unique_rubric_ids(cls, value: list[RubricDef] | None) -> list[RubricDef] | None:
+        duplicates = sorted(key for key, count in Counter(r.id for r in value or []).items() if count > 1)
+        if duplicates:
+            raise ValueError("Rubric ids must be unique within a metric. Duplicate ids: " + ", ".join(duplicates))
+        return value
+
+    @model_validator(mode="after")
+    def _rubrics_only_on_rubric_metrics(self) -> BuiltinMetricDef:
+        if self.rubrics and self.name not in RUBRIC_METRIC_TYPES:
+            raise ValueError(
+                f"Built-in metric '{self.name}' does not take rubrics. "
+                f"Metrics that do: {', '.join(sorted(RUBRIC_METRIC_TYPES))}"
+            )
+        return self
+
 
 class BaseEvaluatorDef(BaseModel):
     """Shared fields for all executable evaluator definitions."""
diff --git a/src/agentevals/custom_evaluators.py b/src/agentevals/custom_evaluators.py
index 058a11a..398f946 100644
--- a/src/agentevals/custom_evaluators.py
+++ b/src/agentevals/custom_evaluators.py
@@ -344,7 +344,7 @@ async def evaluate_custom_evaluator(
     from .runner import MetricResult
 
     if isinstance(evaluator_def, BuiltinMetricDef):
-        from .adk_bridge import expected_invocations, to_adk_invocations
+        from .adk_bridge import criterion_rubrics, expected_invocations, expected_rubrics, to_adk_invocations
         from .builtin_metrics import evaluate_builtin_metric
 
         actual_invocations = to_adk_invocations(actual.turns)
@@ -365,6 +365,8 @@ async def evaluate_custom_evaluator(
             credential_ref=evaluator_def.credential_ref,
             judge_base_url=evaluator_def.judge_base_url,
             expected_reason=expected_reason,
+            rubrics=criterion_rubrics(evaluator_def.rubrics),
+            case_rubrics=expected_rubrics(expected),
         )
         if alignment:
             result.details = {**(result.details or {}), **alignment}
diff --git a/src/agentevals/mcp_server.py b/src/agentevals/mcp_server.py
index 62ba67c..4477d32 100644
--- a/src/agentevals/mcp_server.py
+++ b/src/agentevals/mcp_server.py
@@ -465,7 +465,7 @@ async def evaluate_sessions(
                 "golden_session_id": golden_session_id,
                 "eval_set_id": eval_set_id or f"eval-{golden_session_id[:12]}",
                 "evaluators": [
-                    e.model_dump(mode="json", by_alias=True)
+                    e.model_dump(mode="json", by_alias=True, exclude={"rubrics"} if e.rubrics is None else set())
                     for e in make_builtin_evaluator_entries(
                         metrics,
                         judge_model=judge_model,
diff --git a/tests/test_api.py b/tests/test_api.py
index 7acb625..f2c3fed 100644
--- a/tests/test_api.py
+++ b/tests/test_api.py
@@ -452,6 +452,13 @@ def test_metrics_envelope(self):
         assert isinstance(body["data"], list)
         assert body["error"] is None
 
+    def test_only_supported_rubric_metrics_are_available(self):
+        body = _assert_envelope(self.client.get("/api/metrics"))
+        metrics = {metric["name"]: metric for metric in body["data"]}
+        assert metrics["rubric_based_final_response_quality_v1"]["working"] is True
+        assert metrics["rubric_based_tool_use_quality_v1"]["working"] is True
+        assert metrics["rubric_based_multi_turn_trajectory_quality_v1"]["working"] is False
+
     def test_metrics_all_camel(self):
         body = self.client.get("/api/metrics").json()
         _assert_all_keys_camel(body)
diff --git a/tests/test_builtin_metrics.py b/tests/test_builtin_metrics.py
index df37c27..b274912 100644
--- a/tests/test_builtin_metrics.py
+++ b/tests/test_builtin_metrics.py
@@ -29,6 +29,7 @@
 from agentevals import builtin_metrics
 from agentevals.adk_bridge import to_adk_invocations
 from agentevals.builtin_metrics import _to_invocation_events, evaluate_builtin_metric
+from agentevals.config import RUBRIC_METRIC_TYPES
 from agentevals.genai.extract import extract_conversation
 from agentevals.loader import load_traces
 
@@ -287,7 +288,14 @@ async def test_every_prebuilt_metric_receives_invocation_events(monkeypatch, met
     actual = _invocation([("a", "1")], [("a", "1", "A")])
     expected = _invocation([("a", "1")], [])
 
-    result = await evaluate_builtin_metric(metric_name, [actual], [expected], None, 0.5)
+    result = await evaluate_builtin_metric(
+        metric_name,
+        [actual],
+        [expected],
+        None,
+        0.5,
+        rubrics=["Uses captured evidence."] if metric_name in RUBRIC_METRIC_TYPES else None,
+    )
 
     assert result.error is None
     for inv in [*recorder.actual, *recorder.expected]:
diff --git a/tests/test_rubrics.py b/tests/test_rubrics.py
new file mode 100644
index 0000000..c1a5e77
--- /dev/null
+++ b/tests/test_rubrics.py
@@ -0,0 +1,441 @@
+"""The public path for rubric_based_* metrics: rubrics in the eval config,
+on the matched eval case and on its invocations, resolved the way ADK's own
+eval service resolves them, with per-rubric scores in the result."""
+
+from __future__ import annotations
+
+import asyncio
+import json
+import textwrap
+
+import pytest
+from google.adk.evaluation.eval_case import EvalCase, Invocation
+from google.adk.evaluation.eval_rubrics import Rubric, RubricContent, RubricScore
+from google.adk.evaluation.eval_set import EvalSet
+from google.adk.evaluation.evaluator import EvalStatus, EvaluationResult, PerInvocationResult
+from google.genai import types as genai_types
+
+from agentevals import builtin_metrics
+from agentevals.adk_bridge import criterion_rubrics, expected_from_case
+from agentevals.builtin_metrics import (
+    attach_case_rubrics,
+    build_eval_metric,
+    evaluate_builtin_metric,
+    extract_rubric_details,
+    to_rubric_objects,
+)
+from agentevals.config import BuiltinMetricDef, RubricDef, apply_builtin_overrides
+from agentevals.custom_evaluators import evaluate_custom_evaluator
+from agentevals.eval_config_loader import load_eval_config
+from agentevals.genai.extract import extract_conversation
+from agentevals.runner import EvalRunConfig, run_evaluation
+from otel.builders import request, span
+from test_runner import _make_eval_set_json, _make_tool_trace
+
+METRIC = "rubric_based_final_response_quality_v1"
+
+
+def _rubric(rubric_id: str, text: str = "a statement", rubric_type: str | None = None) -> Rubric:
+    return Rubric(rubric_id=rubric_id, rubric_content=RubricContent(text_property=text), type=rubric_type)
+
+
+def _invocation(text: str = "hi", rubrics: list[Rubric] | None = None) -> Invocation:
+    return Invocation(
+        invocation_id=text,
+        user_content=genai_types.Content(role="user", parts=[genai_types.Part(text=text)]),
+        final_response=genai_types.Content(role="model", parts=[genai_types.Part(text="ok")]),
+        rubrics=rubrics,
+    )
+
+
+class TestConfig:
+    def test_rubrics_accept_mappings_and_strings(self):
+        metric = BuiltinMetricDef.model_validate(
+            {
+                "name": METRIC,
+                "type": "builtin",
+                "rubrics": [{"id": "grounded", "text": "No invented cause."}, "Uses the latest correction."],
+            }
+        )
+        assert [(r.id, r.text) for r in metric.rubrics] == [
+            ("grounded", "No invented cause."),
+            ("rubric_1", "Uses the latest correction."),
+        ]
+        assert criterion_rubrics(metric.rubrics)[0].type is None
+
+    @pytest.mark.parametrize(
+        ("rubrics", "message"),
+        [
+            ([{"id": "a", "text": "one"}, {"id": "a", "text": "two"}], "unique"),
+            (["  "], "blank"),
+            ("not a list", "must be a list"),
+            ([{"id": "a"}], "text"),
+            ([{"id": "a", "text": "one", "weight": 2}], "weight"),
+            ([{"id": "a", "text": "one", "type": "FINAL_RESPONSE_QUALITY"}], "type"),
+            ([{"id": "  ", "text": "one"}], "blank"),
+        ],
+    )
+    def test_malformed_rubrics_are_refused(self, rubrics, message):
+        with pytest.raises(ValueError, match=message):
+            BuiltinMetricDef.model_validate({"name": METRIC, "type": "builtin", "rubrics": rubrics})
+
+    def test_rubrics_survive_run_level_overrides(self):
+        metric = BuiltinMetricDef(name=METRIC, rubrics=[RubricDef(id="a", text="t")])
+        (updated,) = apply_builtin_overrides([metric], judge_model="gemini-2.5-flash", threshold=0.9)
+        assert updated.judge_model == "gemini-2.5-flash"
+        assert [r.id for r in updated.rubrics] == ["a"]
+
+    def test_loader_reads_rubrics_from_yaml(self, tmp_path):
+        config_file = tmp_path / "eval_config.yaml"
+        config_file.write_text(
+            textwrap.dedent(
+                f"""
+                evaluators:
+                  - name: {METRIC}
+                    type: builtin
+                    judge_model: gemini-2.5-flash
+                    rubrics:
+                      - id: corrected_value_wins
+                        text: The response uses the operator's latest correction.
+                      - The response states no root cause the conversation did not establish.
+                """
+            ),
+            encoding="utf-8",
+        )
+        (metric,) = load_eval_config(config_file).evaluators
+        assert isinstance(metric, BuiltinMetricDef)
+        assert [r.id for r in metric.rubrics] == ["corrected_value_wins", "rubric_1"]
+
+
+class TestCriterion:
+    def test_strings_and_objects_both_reach_the_criterion(self):
+        metric = build_eval_metric(METRIC, "gemini-2.5-flash", 0.5, rubrics=["plain", _rubric("named", "typed")])
+        rubrics = metric.criterion.rubrics
+        assert [(r.rubric_id, r.rubric_content.text_property) for r in rubrics] == [
+            ("rubric_0", "plain"),
+            ("named", "typed"),
+        ]
+        assert to_rubric_objects(None) == []
+
+
+class TestCaseRubrics:
+    def test_case_rubrics_reach_every_invocation_without_defaulting_type(self):
+        actual = [_invocation("a"), _invocation("b")]
+        attached, warnings = attach_case_rubrics(METRIC, actual, None, [_rubric("case")])
+        for inv in attached:
+            (rubric,) = inv.rubrics
+            assert (rubric.rubric_id, rubric.type) == ("case", None)
+        assert warnings == ["eval case rubric 'case' ignored: type None; metric requires 'FINAL_RESPONSE_QUALITY'"]
+        # The originals are untouched.
+        assert all(inv.rubrics is None for inv in actual)
+
+    def test_invocation_rubrics_apply_at_their_position_and_keep_their_type(self):
+        actual = [_invocation("a"), _invocation("b")]
+        expected = [_invocation("a", [_rubric("first", rubric_type="CUSTOM")]), _invocation("b")]
+        attached, warnings = attach_case_rubrics(METRIC, actual, expected, None)
+        assert [(r.rubric_id, r.type) for r in attached[0].rubrics] == [("first", "CUSTOM")]
+        assert attached[1].rubrics == []
+        assert warnings == []
+
+    def test_a_duplicate_id_between_case_and_invocation_is_an_error(self):
+        actual = [_invocation("a")]
+        expected = [_invocation("a", [_rubric("dup", rubric_type="FINAL_RESPONSE_QUALITY")])]
+        with pytest.raises(ValueError, match="distinct"):
+            attach_case_rubrics(METRIC, actual, expected, [_rubric("dup", rubric_type="FINAL_RESPONSE_QUALITY")])
+
+    @pytest.mark.parametrize("rubric_type", [None, "TOOL_USE_QUALITY"])
+    def test_dropped_rubrics_are_reported_before_judge_construction(self, monkeypatch, rubric_type):
+        def unexpected(*args):
+            pytest.fail("Judge must not be constructed without an applicable rubric")
+
+        monkeypatch.setattr(builtin_metrics, "get_evaluator", unexpected)
+        result = asyncio.run(
+            evaluate_builtin_metric(
+                METRIC,
+                [_invocation("a")],
+                None,
+                "gemini-2.5-flash",
+                0.5,
+                case_rubrics=[_rubric("dropped", rubric_type=rubric_type)],
+            )
+        )
+        assert result.score is None and result.eval_status == "NOT_EVALUATED"
+        assert "requires rubrics" in result.error
+        if rubric_type is None:
+            assert "eval case rubric 'dropped' ignored" in result.details["warnings"][0]
+        else:
+            assert result.details is None
+
+    def test_missing_later_turn_rubrics_prevents_partial_judging(self, monkeypatch):
+        monkeypatch.setattr(builtin_metrics, "get_evaluator", lambda *args: pytest.fail("Partial judge call"))
+        result = asyncio.run(
+            evaluate_builtin_metric(
+                METRIC,
+                [_invocation("a"), _invocation("b")],
+                [_invocation("a", [_rubric("first", rubric_type="FINAL_RESPONSE_QUALITY")]), _invocation("b")],
+                "gemini-2.5-flash",
+                0.5,
+            )
+        )
+        assert result.score is None and "invocations [1]" in result.error
+
+    def test_duplicate_between_criterion_and_case_is_rejected(self):
+        with pytest.raises(ValueError, match="distinct"):
+            attach_case_rubrics(
+                METRIC,
+                [_invocation()],
+                None,
+                [_rubric("same", rubric_type="FINAL_RESPONSE_QUALITY")],
+                [_rubric("same")],
+            )
+
+    def test_a_rubric_metric_without_any_rubric_is_an_error_before_any_judge_call(self):
+        result = asyncio.run(
+            evaluate_builtin_metric(
+                METRIC,
+                [_invocation("a")],
+                None,
+                judge_model="gemini-2.5-flash",
+                threshold=0.5,
+            )
+        )
+        assert result.score is None
+        assert "requires rubrics" in result.error
+
+    def test_the_dispatch_passes_config_and_case_rubrics(self, monkeypatch):
+        seen = {}
+
+        async def fake(**kwargs):
+            seen.update(kwargs)
+            from agentevals.runner import MetricResult
+
+            return MetricResult(metric_name=kwargs["metric_name"], score=1.0, eval_status="PASSED")
+
+        monkeypatch.setattr(builtin_metrics, "evaluate_builtin_metric", fake)
+        case = EvalCase(eval_id="c", conversation=[_invocation("a")], rubrics=[_rubric("case")])
+        metric = BuiltinMetricDef(name=METRIC, rubrics=[RubricDef(id="cfg", text="t")])
+        actual = extract_conversation([_make_tool_trace([])])
+        asyncio.run(evaluate_custom_evaluator(metric, actual, expected_from_case(case)))
+        assert [r.rubric_id for r in seen["rubrics"]] == ["cfg"]
+        assert seen["rubrics"][0].type is None
+        assert [r.rubric_id for r in seen["case_rubrics"]] == ["case"]
+
+
+class TestPublicPath:
+    @pytest.mark.parametrize("ends_at_clarification", [False, True])
+    def test_clarification_and_completion_properties_keep_their_authored_turn(
+        self, tmp_path, monkeypatch, ends_at_clarification
+    ):
+        turns = [
+            ("Draft a booking.", "Which destination?", "asks-destination"),
+            ("Paris.", "What date?", "asks-date"),
+            ("Friday; actually Lyon.", "Draft booking for Lyon on Friday.", "uses-latest-destination"),
+        ]
+        if ends_at_clarification:
+            turns = turns[:1]
+        spans = [
+            span(
+                span_id=f"{index + 1:016x}",
+                start=(index + 1) * 1000,
+                end=(index + 1) * 1000 + 100,
+                attrs={
+                    "gen_ai.operation.name": "invoke_agent",
+                    "gen_ai.agent.name": "booking-assistant",
+                    "gen_ai.conversation.id": "booking-conversation",
+                    "gen_ai.input.messages": json.dumps(
+                        [{"role": "user", "parts": [{"type": "text", "content": question}]}]
+                    ),
+                    "gen_ai.output.messages": json.dumps(
+                        [{"role": "assistant", "parts": [{"type": "text", "content": answer}]}]
+                    ),
+                },
+            )
+            for index, (question, answer, _) in enumerate(turns)
+        ]
+        trace_path = tmp_path / "trace.json"
+        trace_path.write_text(json.dumps(request(spans)))
+        golden = {
+            "eval_set_id": "booking",
+            "eval_cases": [
+                {
+                    "eval_id": "booking-case",
+                    "rubrics": [
+                        _rubric(
+                            "every-turn",
+                            "Keeps a clear distinction between a question and a completed draft.",
+                            "FINAL_RESPONSE_QUALITY",
+                        ).model_dump()
+                    ],
+                    "conversation": [
+                        {
+                            "invocation_id": f"expected-{index}",
+                            "user_content": {"role": "user", "parts": [{"text": question}]},
+                            "final_response": {"role": "model", "parts": [{"text": answer}]},
+                            "rubrics": [_rubric(key, answer, "FINAL_RESPONSE_QUALITY").model_dump()],
+                        }
+                        for index, (question, answer, key) in enumerate(turns)
+                    ],
+                }
+            ],
+        }
+        golden_path = tmp_path / "golden.json"
+        golden_path.write_text(json.dumps(golden))
+
+        class Evaluator:
+            async def evaluate_invocations(self, actual_invocations, expected_invocations):
+                assert len(actual_invocations) == len(turns)
+                results = []
+                for index, (actual, expected) in enumerate(zip(actual_invocations, expected_invocations, strict=True)):
+                    key = turns[index][2]
+                    assert [r.rubric_id for r in actual.rubrics] == ["every-turn", key]
+                    assert expected.final_response.parts[0].text == turns[index][1]
+                    results.append(
+                        PerInvocationResult(
+                            actual_invocation=actual,
+                            score=1.0,
+                            eval_status=EvalStatus.PASSED,
+                            rubric_scores=[
+                                RubricScore(rubric_id=property_id, score=1.0, rationale="Synthetic observed property.")
+                                for property_id in ("every-turn", key)
+                            ],
+                        )
+                    )
+                return EvaluationResult(
+                    overall_score=1.0, overall_eval_status=EvalStatus.PASSED, per_invocation_results=results
+                )
+
+        monkeypatch.setattr(builtin_metrics, "get_evaluator", lambda metric: Evaluator())
+        result = asyncio.run(
+            run_evaluation(
+                EvalRunConfig(
+                    trace_files=[str(trace_path)],
+                    eval_set_file=str(golden_path),
+                    group_by="conversation",
+                    evaluators=[BuiltinMetricDef(name=METRIC)],
+                )
+            )
+        )
+        payload = json.loads(result.model_dump_json(by_alias=True))
+        (trace,) = payload["traceResults"]
+        assert trace["numInvocations"] == len(turns)
+        assert trace["evalCaseId"] == "booking-case"
+        details = trace["metricResults"][0]["details"]["per_invocation"]
+        assert [item["invocation_id"] for item in details] == [ref["spanId"] for ref in trace["turnRefs"]]
+        assert [[prop["rubric_id"] for prop in item["rubric_scores"]] for item in details] == [
+            ["every-turn", key] for _, _, key in turns
+        ]
+
+    @pytest.mark.parametrize(
+        ("metric_name", "rubric_type"),
+        [(METRIC, "FINAL_RESPONSE_QUALITY"), ("rubric_based_tool_use_quality_v1", "TOOL_USE_QUALITY")],
+    )
+    @pytest.mark.parametrize("score", [0.0, 1.0, None])
+    def test_authored_properties_survive_matching_and_serialization(
+        self, tmp_path, monkeypatch, metric_name, rubric_type, score
+    ):
+        class Evaluator:
+            async def evaluate_invocations(self, actual_invocations, expected_invocations):
+                actual = actual_invocations[0]
+                assert actual.app_details.agent_details
+                assert [(r.rubric_id, r.type) for r in actual.rubrics] == [
+                    ("case", rubric_type),
+                    ("ignored", None),
+                    ("turn", rubric_type),
+                ]
+                status = (
+                    EvalStatus.NOT_EVALUATED if score is None else EvalStatus.PASSED if score else EvalStatus.FAILED
+                )
+                scores = [
+                    RubricScore(rubric_id=key, score=score, rationale="observed synthetic evidence")
+                    for key in ("cfg", "case", "turn")
+                ]
+                return EvaluationResult(
+                    overall_score=score,
+                    overall_eval_status=status,
+                    per_invocation_results=[
+                        PerInvocationResult(
+                            actual_invocation=actual, score=score, eval_status=status, rubric_scores=scores
+                        )
+                    ],
+                    overall_rubric_scores=scores,
+                )
+
+        def factory(metric):
+            assert [r.rubric_id for r in metric.criterion.rubrics] == ["cfg"]
+            return Evaluator()
+
+        monkeypatch.setattr(builtin_metrics, "get_evaluator", factory)
+        from agentevals.otel.encode import encode_traces
+
+        trace_path = tmp_path / "trace.json"
+        trace_path.write_text(json.dumps(encode_traces([_make_tool_trace(["lookup"])])))
+        golden = _make_eval_set_json(["lookup"])
+        case = golden["eval_cases"][0]
+        case["eval_id"] = "matched-case"
+        case["rubrics"] = [_rubric("case", rubric_type=rubric_type).model_dump(), _rubric("ignored").model_dump()]
+        case["conversation"][0]["rubrics"] = [_rubric("turn", rubric_type=rubric_type).model_dump()]
+        unrelated = json.loads(json.dumps(case))
+        unrelated["eval_id"] = "unrelated"
+        unrelated["conversation"][0]["user_content"]["parts"][0]["text"] = "other question"
+        golden["eval_cases"].insert(0, unrelated)
+        golden_path = tmp_path / "golden.json"
+        golden_path.write_text(json.dumps(golden))
+        config_path = tmp_path / "config.yaml"
+        config_path.write_text(
+            f"evaluators:\n  - name: {metric_name}\n    type: builtin\n    rubrics:\n      - id: cfg\n        text: Uses captured evidence.\n"
+        )
+        config = load_eval_config(config_path)
+        result = asyncio.run(
+            run_evaluation(
+                EvalRunConfig(
+                    trace_files=[str(trace_path)], eval_set_file=str(golden_path), evaluators=config.evaluators
+                )
+            )
+        )
+        payload = json.loads(result.model_dump_json(by_alias=True))
+        trace = payload["traceResults"][0]
+        metric = trace["metricResults"][0]
+        assert trace["evalCaseId"] == "matched-case"
+        assert metric["score"] == score
+        assert metric["evalStatus"] == ("NOT_EVALUATED" if score is None else "PASSED" if score else "FAILED")
+        details = metric["details"]
+        assert details["per_invocation"][0]["invocation_id"] == trace["turnRefs"][0]["spanId"]
+        assert [
+            (r["rubric_id"], r["score"], r["rationale"]) for r in details["per_invocation"][0]["rubric_scores"]
+        ] == [(key, score, "observed synthetic evidence") for key in ("cfg", "case", "turn")]
+        assert "eval case rubric 'ignored' ignored" in details["warnings"][0]
+
+
+class TestDetails:
+    def test_per_rubric_scores_are_reported_per_invocation_and_overall(self):
+        actual = _invocation("a")
+        result = EvaluationResult(
+            overall_score=0.5,
+            overall_eval_status=EvalStatus.FAILED,
+            per_invocation_results=[
+                PerInvocationResult(
+                    actual_invocation=actual,
+                    score=0.5,
+                    eval_status=EvalStatus.FAILED,
+                    rubric_scores=[
+                        RubricScore(rubric_id="grounded", score=1.0, rationale="no cause was invented"),
+                        RubricScore(rubric_id="corrected", score=0.0, rationale="the stale value is current"),
+                    ],
+                )
+            ],
+            overall_rubric_scores=[
+                RubricScore(rubric_id="grounded", score=1.0),
+                RubricScore(rubric_id="corrected", score=0.0),
+            ],
+        )
+        details = extract_rubric_details(result)
+        assert details["per_invocation"] == [
+            {
+                "invocation_id": "a",
+                "rubric_scores": [
+                    {"rubric_id": "grounded", "score": 1.0, "rationale": "no cause was invented"},
+                    {"rubric_id": "corrected", "score": 0.0, "rationale": "the stale value is current"},
+                ],
+            }
+        ]
+        assert [r["rubric_id"] for r in details["overall_rubric_scores"]] == ["grounded", "corrected"]
diff --git a/ui/src/generated/agentevals/api/v1alpha1/evaluation_pb.d.ts b/ui/src/generated/agentevals/api/v1alpha1/evaluation_pb.d.ts
index b344014..57e9e84 100644
--- a/ui/src/generated/agentevals/api/v1alpha1/evaluation_pb.d.ts
+++ b/ui/src/generated/agentevals/api/v1alpha1/evaluation_pb.d.ts
@@ -64,7 +64,7 @@ export declare type Metric = Message<"agentevals.api.v1alpha1.Metric"> & {
   requiresGcp: boolean;
 
   /**
-   * Needs rubrics in the eval set.
+   * Needs rubrics, from the evaluator or the eval set.
    *
    * @generated from field: bool requires_rubrics = 7;
    */
@@ -141,6 +141,13 @@ export declare type BuiltinEvaluator = Message<"agentevals.api.v1alpha1.BuiltinE
    * @generated from field: string judge_base_url = 6;
    */
   judgeBaseUrl: string;
+
+  /**
+   * Rubrics for the rubric_based_* metrics that take them. Ids are unique.
+   *
+   * @generated from field: repeated agentevals.api.v1alpha1.Rubric rubrics = 7;
+   */
+  rubrics: Rubric[];
 };
 
 /**
@@ -149,6 +156,29 @@ export declare type BuiltinEvaluator = Message<"agentevals.api.v1alpha1.BuiltinE
  */
 export declare const BuiltinEvaluatorSchema: GenMessage<BuiltinEvaluator>;
 
+/**
+ * One property a rubric metric's judge checks.
+ *
+ * @generated from message agentevals.api.v1alpha1.Rubric
+ */
+export declare type Rubric = Message<"agentevals.api.v1alpha1.Rubric"> & {
+  /**
+   * @generated from field: string id = 1;
+   */
+  id: string;
+
+  /**
+   * @generated from field: string text = 2;
+   */
+  text: string;
+};
+
+/**
+ * Describes the message agentevals.api.v1alpha1.Rubric.
+ * Use `create(RubricSchema)` to create a new message.
+ */
+export declare const RubricSchema: GenMessage<Rubric>;
+
 /**
  * An evaluator file on the server, run as a subprocess.
  *
diff --git a/ui/src/generated/agentevals/api/v1alpha1/evaluation_pb.js b/ui/src/generated/agentevals/api/v1alpha1/evaluation_pb.js
index f3f758b..abe5dfd 100644
--- a/ui/src/generated/agentevals/api/v1alpha1/evaluation_pb.js
+++ b/ui/src/generated/agentevals/api/v1alpha1/evaluation_pb.js
@@ -10,7 +10,7 @@ import { file_google_protobuf_struct } from "@bufbuild/protobuf/wkt";
  * Describes the file agentevals/api/v1alpha1/evaluation.proto.
  */
 export const file_agentevals_api_v1alpha1_evaluation = /*@__PURE__*/
-  fileDesc("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", [file_agentevals_api_v1alpha1_common, file_google_protobuf_struct]);
+  fileDesc("CihhZ2VudGV2YWxzL2FwaS92MWFscGhhMS9ldmFsdWF0aW9uLnByb3RvEhdhZ2VudGV2YWxzLmFwaS52MWFscGhhMSIUChJMaXN0TWV0cmljc1JlcXVlc3QirwEKBk1ldHJpYxIMCgRuYW1lGAEgASgJEhAKCGNhdGVnb3J5GAIgASgJEhMKC2Rlc2NyaXB0aW9uGAMgASgJEhkKEXJlcXVpcmVzX2V2YWxfc2V0GAQgASgIEhQKDHJlcXVpcmVzX2xsbRgFIAEoCBIUCgxyZXF1aXJlc19nY3AYBiABKAgSGAoQcmVxdWlyZXNfcnVicmljcxgHIAEoCBIPCgd3b3JraW5nGAggASgIIkcKE0xpc3RNZXRyaWNzUmVzcG9uc2USMAoHbWV0cmljcxgBIAMoCzIfLmFnZW50ZXZhbHMuYXBpLnYxYWxwaGExLk1ldHJpYyKKAgoQQnVpbHRpbkV2YWx1YXRvchIMCgRuYW1lGAEgASgJEhYKCXRocmVzaG9sZBgCIAEoAUgAiAEBEhMKC2p1ZGdlX21vZGVsGAMgASgJEksKFXRyYWplY3RvcnlfbWF0Y2hfdHlwZRgEIAEoDjIsLmFnZW50ZXZhbHMuYXBpLnYxYWxwaGExLlRyYWplY3RvcnlNYXRjaFR5cGUSFgoOY3JlZGVudGlhbF9yZWYYBSABKAkSFgoOanVkZ2VfYmFzZV91cmwYBiABKAkSMAoHcnVicmljcxgHIAMoCzIfLmFnZW50ZXZhbHMuYXBpLnYxYWxwaGExLlJ1YnJpY0IMCgpfdGhyZXNob2xkIiIKBlJ1YnJpYxIKCgJpZBgBIAEoCRIMCgR0ZXh0GAIgASgJIq4BCg1Db2RlRXZhbHVhdG9yEgwKBG5hbWUYASABKAkSFgoJdGhyZXNob2xkGAIgASgBSACIAQESFAoHdGltZW91dBgDIAEoBUgBiAEBEicKBmNvbmZpZxgEIAEoCzIXLmdvb2dsZS5wcm90b2J1Zi5TdHJ1Y3QSEAoIZXhlY3V0b3IYBSABKAkSDAoEcGF0aBgGIAEoCUIMCgpfdGhyZXNob2xkQgoKCF90aW1lb3V0Ir8BCg9SZW1vdGVFdmFsdWF0b3ISDAoEbmFtZRgBIAEoCRIWCgl0aHJlc2hvbGQYAiABKAFIAIgBARIUCgd0aW1lb3V0GAMgASgFSAGIAQESJwoGY29uZmlnGAQgASgLMhcuZ29vZ2xlLnByb3RvYnVmLlN0cnVjdBIQCghleGVjdXRvchgFIAEoCRIOCgZzb3VyY2UYBiABKAkSCwoDcmVmGAcgASgJQgwKCl90aHJlc2hvbGRCCgoIX3RpbWVvdXQixQEKCUV2YWx1YXRvchI8CgdidWlsdGluGAEgASgLMikuYWdlbnRldmFscy5hcGkudjFhbHBoYTEuQnVpbHRpbkV2YWx1YXRvckgAEjYKBGNvZGUYAiABKAsyJi5hZ2VudGV2YWxzLmFwaS52MWFscGhhMS5Db2RlRXZhbHVhdG9ySAASOgoGcmVtb3RlGAMgASgLMiguYWdlbnRldmFscy5hcGkudjFhbHBoYTEuUmVtb3RlRXZhbHVhdG9ySABCBgoEa2luZCLyAQoKRXZhbFBhcmFtcxI2CgpldmFsdWF0b3JzGAEgAygLMiIuYWdlbnRldmFscy5hcGkudjFhbHBoYTEuRXZhbHVhdG9yEiIKFW1heF9jb25jdXJyZW50X3RyYWNlcxgCIAEoBUgAiAEBEiEKFG1heF9jb25jdXJyZW50X2V2YWxzGAMgASgFSAGIAQESMgoIZ3JvdXBfYnkYBCABKA4yIC5hZ2VudGV2YWxzLmFwaS52MWFscGhhMS5Hcm91cEJ5QhgKFl9tYXhfY29uY3VycmVudF90cmFjZXNCFwoVX21heF9jb25jdXJyZW50X2V2YWxzImUKEEludm9jYXRpb25SZXN1bHQSEgoFc2NvcmUYASABKAFIAIgBARIzCgZzdGF0dXMYAiABKA4yIy5hZ2VudGV2YWxzLmFwaS52MWFscGhhMS5FdmFsU3RhdHVzQggKBl9zY29yZSLYAgoMTWV0cmljUmVzdWx0EhMKC21ldHJpY19uYW1lGAEgASgJEhIKBXNjb3JlGAIgASgBSACIAQESOAoLZXZhbF9zdGF0dXMYAyABKA4yIy5hZ2VudGV2YWxzLmFwaS52MWFscGhhMS5FdmFsU3RhdHVzEkEKDnBlcl9pbnZvY2F0aW9uGAQgAygLMikuYWdlbnRldmFscy5hcGkudjFhbHBoYTEuSW52b2NhdGlvblJlc3VsdBISCgVlcnJvchgFIAEoCUgBiAEBEhcKCmVycm9yX3R5cGUYBiABKAlIAogBARIoCgdkZXRhaWxzGAcgASgLMhcuZ29vZ2xlLnByb3RvYnVmLlN0cnVjdBIYCgtkdXJhdGlvbl9tcxgIIAEoAUgDiAEBQggKBl9zY29yZUIICgZfZXJyb3JCDQoLX2Vycm9yX3R5cGVCDgoMX2R1cmF0aW9uX21zIiwKB1R1cm5SZWYSEAoIdHJhY2VfaWQYASABKAkSDwoHc3Bhbl9pZBgCIAEoCSKXAwoLVHJhY2VSZXN1bHQSEAoIdHJhY2VfaWQYASABKAkSFwoPbnVtX2ludm9jYXRpb25zGAIgASgFEj0KDm1ldHJpY19yZXN1bHRzGAMgAygLMiUuYWdlbnRldmFscy5hcGkudjFhbHBoYTEuTWV0cmljUmVzdWx0EhsKE2NvbnZlcnNpb25fd2FybmluZ3MYBCADKAkSNAoTcGVyZm9ybWFuY2VfbWV0cmljcxgFIAEoCzIXLmdvb2dsZS5wcm90b2J1Zi5TdHJ1Y3QSFAoMc2VydmljZV9uYW1lGAYgASgJEhIKCmFnZW50X25hbWUYByABKAkSEQoJdHJhY2VfaWRzGAggAygJEhcKD2NvbnZlcnNhdGlvbl9pZBgJIAEoCRIRCglncm91cF9rZXkYCiABKAkSFAoMZXZhbF9jYXNlX2lkGAsgASgJEhcKD2V2YWxfY2FzZV9tYXRjaBgMIAEoCRIzCgl0dXJuX3JlZnMYDSADKAsyIC5hZ2VudGV2YWxzLmFwaS52MWFscGhhMS5UdXJuUmVmIp4BCglSdW5SZXN1bHQSOwoNdHJhY2VfcmVzdWx0cxgBIAMoCzIkLmFnZW50ZXZhbHMuYXBpLnYxYWxwaGExLlRyYWNlUmVzdWx0Eg4KBmVycm9ycxgCIAMoCRI0ChNwZXJmb3JtYW5jZV9tZXRyaWNzGAMgASgLMhcuZ29vZ2xlLnByb3RvYnVmLlN0cnVjdBIOCgZydW5faWQYBCABKAki6wIKFUV2YWx1YXRlVHJhY2VzUmVxdWVzdBIzCgZ0cmFjZXMYASABKAsyIy5hZ2VudGV2YWxzLmFwaS52MWFscGhhMS5UcmFjZUlucHV0Ei8KDGFka19ldmFsX3NldBgCIAEoCzIXLmdvb2dsZS5wcm90b2J1Zi5TdHJ1Y3RIABIzCgZwYXJhbXMYAyABKAsyIy5hZ2VudGV2YWxzLmFwaS52MWFscGhhMS5FdmFsUGFyYW1zElsKD2NyZWRlbnRpYWxfcmVmcxgEIAMoCzJCLmFnZW50ZXZhbHMuYXBpLnYxYWxwaGExLkV2YWx1YXRlVHJhY2VzUmVxdWVzdC5DcmVkZW50aWFsUmVmc0VudHJ5Gk4KE0NyZWRlbnRpYWxSZWZzRW50cnkSCwoDa2V5GAEgASgJEiYKBXZhbHVlGAIgASgLMhcuZ29vZ2xlLnByb3RvYnVmLlN0cnVjdDoCOAFCCgoIZXZhbF9zZXQiJQoSRXZhbHVhdGlvblByb2dyZXNzEg8KB21lc3NhZ2UYASABKAkiiwEKEEdyb3VwUGVyZm9ybWFuY2USEAoIdHJhY2VfaWQYASABKAkSNAoTcGVyZm9ybWFuY2VfbWV0cmljcxgCIAEoCzIXLmdvb2dsZS5wcm90b2J1Zi5TdHJ1Y3QSLwoOdHJhY2VfbWV0YWRhdGEYAyABKAsyFy5nb29nbGUucHJvdG9idWYuU3RydWN0IpYCChZFdmFsdWF0ZVRyYWNlc1Jlc3BvbnNlEj8KCHByb2dyZXNzGAEgASgLMisuYWdlbnRldmFscy5hcGkudjFhbHBoYTEuRXZhbHVhdGlvblByb2dyZXNzSAASQAoLcGVyZm9ybWFuY2UYAiABKAsyKS5hZ2VudGV2YWxzLmFwaS52MWFscGhhMS5Hcm91cFBlcmZvcm1hbmNlSAASPAoMdHJhY2VfcmVzdWx0GAMgASgLMiQuYWdlbnRldmFscy5hcGkudjFhbHBoYTEuVHJhY2VSZXN1bHRIABIyCgRkb25lGAQgASgLMiIuYWdlbnRldmFscy5hcGkudjFhbHBoYTEuUnVuUmVzdWx0SABCBwoFZXZlbnQqpgEKE1RyYWplY3RvcnlNYXRjaFR5cGUSJQohVFJBSkVDVE9SWV9NQVRDSF9UWVBFX1VOU1BFQ0lGSUVEEAASHwobVFJBSkVDVE9SWV9NQVRDSF9UWVBFX0VYQUNUEAESIgoeVFJBSkVDVE9SWV9NQVRDSF9UWVBFX0lOX09SREVSEAISIwofVFJBSkVDVE9SWV9NQVRDSF9UWVBFX0FOWV9PUkRFUhADKngKCkV2YWxTdGF0dXMSGwoXRVZBTF9TVEFUVVNfVU5TUEVDSUZJRUQQABIWChJFVkFMX1NUQVRVU19QQVNTRUQQARIWChJFVkFMX1NUQVRVU19GQUlMRUQQAhIdChlFVkFMX1NUQVRVU19OT1RfRVZBTFVBVEVEEAMy8gEKEUV2YWx1YXRpb25TZXJ2aWNlEmgKC0xpc3RNZXRyaWNzEisuYWdlbnRldmFscy5hcGkudjFhbHBoYTEuTGlzdE1ldHJpY3NSZXF1ZXN0GiwuYWdlbnRldmFscy5hcGkudjFhbHBoYTEuTGlzdE1ldHJpY3NSZXNwb25zZRJzCg5FdmFsdWF0ZVRyYWNlcxIuLmFnZW50ZXZhbHMuYXBpLnYxYWxwaGExLkV2YWx1YXRlVHJhY2VzUmVxdWVzdBovLmFnZW50ZXZhbHMuYXBpLnYxYWxwaGExLkV2YWx1YXRlVHJhY2VzUmVzcG9uc2UwAUJRWk9naXRodWIuY29tL2FnZW50ZXZhbHMtZGV2L2FnZW50ZXZhbHMvZ2VuL2dvL2FnZW50ZXZhbHMvYXBpL3YxYWxwaGExO2FwaXYxYWxwaGExYgZwcm90bzM", [file_agentevals_api_v1alpha1_common, file_google_protobuf_struct]);
 
 /**
  * Describes the message agentevals.api.v1alpha1.ListMetricsRequest.
@@ -40,96 +40,103 @@ export const ListMetricsResponseSchema = /*@__PURE__*/
 export const BuiltinEvaluatorSchema = /*@__PURE__*/
   messageDesc(file_agentevals_api_v1alpha1_evaluation, 3);
 
+/**
+ * Describes the message agentevals.api.v1alpha1.Rubric.
+ * Use `create(RubricSchema)` to create a new message.
+ */
+export const RubricSchema = /*@__PURE__*/
+  messageDesc(file_agentevals_api_v1alpha1_evaluation, 4);
+
 /**
  * Describes the message agentevals.api.v1alpha1.CodeEvaluator.
  * Use `create(CodeEvaluatorSchema)` to create a new message.
  */
 export const CodeEvaluatorSchema = /*@__PURE__*/
-  messageDesc(file_agentevals_api_v1alpha1_evaluation, 4);
+  messageDesc(file_agentevals_api_v1alpha1_evaluation, 5);
 
 /**
  * Describes the message agentevals.api.v1alpha1.RemoteEvaluator.
  * Use `create(RemoteEvaluatorSchema)` to create a new message.
  */
 export const RemoteEvaluatorSchema = /*@__PURE__*/
-  messageDesc(file_agentevals_api_v1alpha1_evaluation, 5);
+  messageDesc(file_agentevals_api_v1alpha1_evaluation, 6);
 
 /**
  * Describes the message agentevals.api.v1alpha1.Evaluator.
  * Use `create(EvaluatorSchema)` to create a new message.
  */
 export const EvaluatorSchema = /*@__PURE__*/
-  messageDesc(file_agentevals_api_v1alpha1_evaluation, 6);
+  messageDesc(file_agentevals_api_v1alpha1_evaluation, 7);
 
 /**
  * Describes the message agentevals.api.v1alpha1.EvalParams.
  * Use `create(EvalParamsSchema)` to create a new message.
  */
 export const EvalParamsSchema = /*@__PURE__*/
-  messageDesc(file_agentevals_api_v1alpha1_evaluation, 7);
+  messageDesc(file_agentevals_api_v1alpha1_evaluation, 8);
 
 /**
  * Describes the message agentevals.api.v1alpha1.InvocationResult.
  * Use `create(InvocationResultSchema)` to create a new message.
  */
 export const InvocationResultSchema = /*@__PURE__*/
-  messageDesc(file_agentevals_api_v1alpha1_evaluation, 8);
+  messageDesc(file_agentevals_api_v1alpha1_evaluation, 9);
 
 /**
  * Describes the message agentevals.api.v1alpha1.MetricResult.
  * Use `create(MetricResultSchema)` to create a new message.
  */
 export const MetricResultSchema = /*@__PURE__*/
-  messageDesc(file_agentevals_api_v1alpha1_evaluation, 9);
+  messageDesc(file_agentevals_api_v1alpha1_evaluation, 10);
 
 /**
  * Describes the message agentevals.api.v1alpha1.TurnRef.
  * Use `create(TurnRefSchema)` to create a new message.
  */
 export const TurnRefSchema = /*@__PURE__*/
-  messageDesc(file_agentevals_api_v1alpha1_evaluation, 10);
+  messageDesc(file_agentevals_api_v1alpha1_evaluation, 11);
 
 /**
  * Describes the message agentevals.api.v1alpha1.TraceResult.
  * Use `create(TraceResultSchema)` to create a new message.
  */
 export const TraceResultSchema = /*@__PURE__*/
-  messageDesc(file_agentevals_api_v1alpha1_evaluation, 11);
+  messageDesc(file_agentevals_api_v1alpha1_evaluation, 12);
 
 /**
  * Describes the message agentevals.api.v1alpha1.RunResult.
  * Use `create(RunResultSchema)` to create a new message.
  */
 export const RunResultSchema = /*@__PURE__*/
-  messageDesc(file_agentevals_api_v1alpha1_evaluation, 12);
+  messageDesc(file_agentevals_api_v1alpha1_evaluation, 13);
 
 /**
  * Describes the message agentevals.api.v1alpha1.EvaluateTracesRequest.
  * Use `create(EvaluateTracesRequestSchema)` to create a new message.
  */
 export const EvaluateTracesRequestSchema = /*@__PURE__*/
-  messageDesc(file_agentevals_api_v1alpha1_evaluation, 13);
+  messageDesc(file_agentevals_api_v1alpha1_evaluation, 14);
 
 /**
  * Describes the message agentevals.api.v1alpha1.EvaluationProgress.
  * Use `create(EvaluationProgressSchema)` to create a new message.
  */
 export const EvaluationProgressSchema = /*@__PURE__*/
-  messageDesc(file_agentevals_api_v1alpha1_evaluation, 14);
+  messageDesc(file_agentevals_api_v1alpha1_evaluation, 15);
 
 /**
  * Describes the message agentevals.api.v1alpha1.GroupPerformance.
  * Use `create(GroupPerformanceSchema)` to create a new message.
  */
 export const GroupPerformanceSchema = /*@__PURE__*/
-  messageDesc(file_agentevals_api_v1alpha1_evaluation, 15);
+  messageDesc(file_agentevals_api_v1alpha1_evaluation, 16);
 
 /**
  * Describes the message agentevals.api.v1alpha1.EvaluateTracesResponse.
  * Use `create(EvaluateTracesResponseSchema)` to create a new message.
  */
 export const EvaluateTracesResponseSchema = /*@__PURE__*/
-  messageDesc(file_agentevals_api_v1alpha1_evaluation, 16);
+  messageDesc(file_agentevals_api_v1alpha1_evaluation, 17);
 
 /**
  * Describes the enum agentevals.api.v1alpha1.TrajectoryMatchType.
diff --git a/ui/src/lib/types.ts b/ui/src/lib/types.ts
index a058403..9d68858 100644
--- a/ui/src/lib/types.ts
+++ b/ui/src/lib/types.ts
@@ -412,7 +412,7 @@ export const AVAILABLE_METRICS: MetricMetadata[] = [
     requiresLLM: true,
     requiresGCP: false,
     requiresRubrics: true,
-    working: false,
+    working: true,
     description: 'Rubric-based quality assessment of responses (requires rubrics config)'
   },
   {
@@ -422,7 +422,7 @@ export const AVAILABLE_METRICS: MetricMetadata[] = [
     requiresLLM: true,
     requiresGCP: false,
     requiresRubrics: true,
-    working: false,
+    working: true,
     description: 'Rubric-based assessment of tool usage quality (requires rubrics config)'
   },
   // Safety metrics