fix: clarify run detail result metrics

.env.example

@@ -26,8 +26,10 @@ LLMBENCHLAB_WORKER_MAX_ATTEMPTS=3
 LLMBENCHLAB_WORKER_SHUTDOWN_GRACE_SECONDS=30
 LLMBENCHLAB_WORKER_PROGRESS_FLUSH_SECONDS=5
 LLMBENCHLAB_WORKER_PROGRESS_STALE_SECONDS=60
+LLMBENCHLAB_DEV_WORKER_PROCESSES=1
 LLMBENCHLAB_WORKER_EXPECTED_PROCESSES=1
 LLMBENCHLAB_WORKER_RECOVERY_ALERT_SECONDS=60
+LLMBENCHLAB_COMPOSE_WORKER_PROCESSES=2
 LLMBENCHLAB_REDIS_PUBLISH_TIMEOUT_SECONDS=1
 LLMBENCHLAB_REDIS_OPERATION_TIMEOUT_SECONDS=2
 

Makefile

@@ -1,12 +1,13 @@
 SHELL := /bin/bash
 .DEFAULT_GOAL := help
 
-.PHONY: help setup dev backend worker frontend test lint format smoke migrate phase2-acceptance phase2-capacity phase2-slo docker-up docker-down
+.PHONY: help setup dev dev-multi backend worker frontend test lint format smoke migrate phase2-acceptance phase2-capacity phase2-slo docker-up docker-down
 
 help:
 	@echo "LLMBenchLab developer commands:"
 	@echo "  make setup        Install dependencies and initialize the local database"
-	@echo "  make dev          Start API, independent Worker, and frontend together"
+	@echo "  make dev          Start local API, Worker(s), and frontend (DEV_WORKERS=N needs PostgreSQL)"
+	@echo "  make dev-multi    Start PostgreSQL/Redis/API/frontend with two Workers (WORKERS=N)"
 	@echo "  make backend      Start only the FastAPI development server"
 	@echo "  make worker       Start only the independent task Worker"
 	@echo "  make frontend     Start only the Vite development server"
@@ -25,7 +26,13 @@ setup:
 	@./scripts/setup.sh
 
 dev:
-	@./scripts/dev.sh
+	@if [[ "$(origin DEV_WORKERS)" != "undefined" ]]; then \
+		LLMBENCHLAB_DEV_WORKER_PROCESSES="$(DEV_WORKERS)" ./scripts/dev.sh; \
+	else \
+		./scripts/dev.sh; \
+	fi
+
+dev-multi: docker-up
 
 backend:
 	@./scripts/bootstrap_credential_keyring.sh
@@ -82,7 +89,11 @@ phase2-slo:
 
 docker-up:
 	@./scripts/bootstrap_credential_keyring.sh
-	@docker compose up --build --wait --wait-timeout 180 --remove-orphans
+	@if [[ "$(origin WORKERS)" != "undefined" ]]; then \
+		LLMBENCHLAB_COMPOSE_WORKER_PROCESSES="$(WORKERS)" ./scripts/compose_up.sh; \
+	else \
+		./scripts/compose_up.sh; \
+	fi
 
 docker-down:
 	@docker compose down --remove-orphans

backend/alembic/versions/20260830_0008_provider_api_protocols.py

@@ -0,0 +1,98 @@
+"""Expand registered Model provider types to explicit API protocols.
+
+Revision ID: 20260830_0008
+Revises: 20260830_0007
+Create Date: 2026-08-30 12:00:00 UTC
+"""
+
+from collections.abc import Sequence
+
+import sqlalchemy as sa
+
+from alembic import op
+
+revision: str = "20260830_0008"
+down_revision: str | None = "20260830_0007"
+branch_labels: str | Sequence[str] | None = None
+depends_on: str | Sequence[str] | None = None
+
+_OLD_PROVIDER_VALUES = "provider_type IN ('mock', 'openai_compatible')"
+_NEW_PROVIDER_VALUES = (
+    "provider_type IN ('mock', 'openai_compatible', 'openai_responses', 'anthropic_messages')"
+)
+_OLD_REMOTE_CONFIGURATION = (
+    "provider_type != 'openai_compatible' OR "
+    "(base_url IS NOT NULL AND remote_model_name IS NOT NULL AND "
+    "((credential_source = 'environment' AND api_key_env IS NOT NULL) OR "
+    "(credential_source = 'stored' AND api_key_env IS NULL)))"
+)
+_NEW_REMOTE_CONFIGURATION = (
+    "provider_type NOT IN "
+    "('openai_compatible', 'openai_responses', 'anthropic_messages') OR "
+    "(base_url IS NOT NULL AND remote_model_name IS NOT NULL AND "
+    "((credential_source = 'environment' AND api_key_env IS NOT NULL) OR "
+    "(credential_source = 'stored' AND api_key_env IS NULL)))"
+)
+
+
+def _replace_constraints(
+    *,
+    provider_values: str,
+    remote_configuration: str,
+    existing_length: int,
+    target_length: int,
+) -> None:
+    with op.batch_alter_table("models") as batch_op:
+        batch_op.drop_constraint(op.f("ck_models_provider_type_values"), type_="check")
+        batch_op.drop_constraint(
+            op.f("ck_models_openai_configuration_required"),
+            type_="check",
+        )
+        batch_op.alter_column(
+            "provider_type",
+            existing_type=sa.String(length=existing_length),
+            type_=sa.String(length=target_length),
+            existing_nullable=False,
+        )
+        batch_op.create_check_constraint(
+            op.f("ck_models_provider_type_values"),
+            provider_values,
+        )
+        batch_op.create_check_constraint(
+            op.f("ck_models_openai_configuration_required"),
+            remote_configuration,
+        )
+
+
+def upgrade() -> None:
+    _replace_constraints(
+        provider_values=_NEW_PROVIDER_VALUES,
+        remote_configuration=_NEW_REMOTE_CONFIGURATION,
+        existing_length=17,
+        target_length=18,
+    )
+
+
+def downgrade() -> None:
+    connection = op.get_bind()
+    if connection.dialect.name == "sqlite":
+        connection.exec_driver_sql("BEGIN IMMEDIATE")
+    elif connection.dialect.name == "postgresql":
+        connection.exec_driver_sql("LOCK TABLE models IN ACCESS EXCLUSIVE MODE")
+    new_type_count = connection.scalar(
+        sa.text(
+            "SELECT COUNT(*) FROM models "
+            "WHERE provider_type IN ('openai_responses', 'anthropic_messages')"
+        )
+    )
+    if int(new_type_count or 0) != 0:
+        raise RuntimeError(
+            "Cannot downgrade while openai_responses or anthropic_messages Models exist; "
+            "delete or convert those Model configurations first"
+        )
+    _replace_constraints(
+        provider_values=_OLD_PROVIDER_VALUES,
+        remote_configuration=_OLD_REMOTE_CONFIGURATION,
+        existing_length=18,
+        target_length=17,
+    )

backend/app/adapters/__init__.py

@@ -20,6 +20,49 @@
 )
 from .mock import MockModelAdapter
 from .openai_compatible import OpenAICompatibleAdapter, sanitize_error_message
+from .provider_protocols import AnthropicMessagesAdapter, OpenAIResponsesAdapter
+
+
+def _remote_adapter_options(kwargs: dict[str, Any]) -> dict[str, Any]:
+    """Normalize the shared constructor surface for every remote adapter."""
+
+    options = dict(kwargs)
+    if "remote_model_name" not in options and "model_name" in options:
+        options["remote_model_name"] = options.pop("model_name")
+    aliases = {
+        "connect_timeout": "connect_timeout_seconds",
+        "read_timeout": "read_timeout_seconds",
+        "write_timeout": "write_timeout_seconds",
+        "pool_timeout": "pool_timeout_seconds",
+        "retry_count": "max_retries",
+        "backoff_base_seconds": "retry_backoff_base_seconds",
+        "backoff_cap_seconds": "retry_backoff_cap_seconds",
+    }
+    for old_name, new_name in aliases.items():
+        if old_name in options and new_name not in options:
+            options[new_name] = options.pop(old_name)
+    allowed = {
+        "base_url",
+        "remote_model_name",
+        "api_key_env",
+        "api_key",
+        "connect_timeout_seconds",
+        "read_timeout_seconds",
+        "write_timeout_seconds",
+        "pool_timeout_seconds",
+        "max_retries",
+        "retry_backoff_base_seconds",
+        "retry_backoff_cap_seconds",
+        "client",
+        "sleep",
+        "attempt_controller",
+    }
+    required = {"base_url", "remote_model_name"}
+    return {
+        key: value
+        for key, value in options.items()
+        if key in allowed and (value is not None or key in required)
+    }
 
 
 def build_adapter(provider_type: str, **kwargs: Any) -> ModelAdapter:
@@ -34,44 +77,11 @@ def build_adapter(provider_type: str, **kwargs: Any) -> ModelAdapter:
         }
         return MockModelAdapter(**mock_options)
     if normalized in {"openai", "openai_compatible"}:
-        options = dict(kwargs)
-        if "remote_model_name" not in options and "model_name" in options:
-            options["remote_model_name"] = options.pop("model_name")
-        aliases = {
-            "connect_timeout": "connect_timeout_seconds",
-            "read_timeout": "read_timeout_seconds",
-            "write_timeout": "write_timeout_seconds",
-            "pool_timeout": "pool_timeout_seconds",
-            "retry_count": "max_retries",
-            "backoff_base_seconds": "retry_backoff_base_seconds",
-            "backoff_cap_seconds": "retry_backoff_cap_seconds",
-        }
-        for old_name, new_name in aliases.items():
-            if old_name in options and new_name not in options:
-                options[new_name] = options.pop(old_name)
-        allowed = {
-            "base_url",
-            "remote_model_name",
-            "api_key_env",
-            "api_key",
-            "connect_timeout_seconds",
-            "read_timeout_seconds",
-            "write_timeout_seconds",
-            "pool_timeout_seconds",
-            "max_retries",
-            "retry_backoff_base_seconds",
-            "retry_backoff_cap_seconds",
-            "client",
-            "sleep",
-            "attempt_controller",
-        }
-        required = {"base_url", "remote_model_name"}
-        adapter_options = {
-            key: value
-            for key, value in options.items()
-            if key in allowed and (value is not None or key in required)
-        }
-        return OpenAICompatibleAdapter(**adapter_options)
+        return OpenAICompatibleAdapter(**_remote_adapter_options(kwargs))
+    if normalized == "openai_responses":
+        return OpenAIResponsesAdapter(**_remote_adapter_options(kwargs))
+    if normalized == "anthropic_messages":
+        return AnthropicMessagesAdapter(**_remote_adapter_options(kwargs))
     raise ValueError(f"Unsupported provider_type: {provider_type!r}")
 
 
@@ -80,13 +90,15 @@ def build_adapter(provider_type: str, **kwargs: Any) -> ModelAdapter:
 
 __all__ = [
     "AdapterError",
+    "AnthropicMessagesAdapter",
     "GenerationConfig",
     "Message",
     "MockModelAdapter",
     "ModelAdapter",
     "ModelGenerationResult",
     "OpenAICompatibleAdapter",
     "OpenAICompatibleModelAdapter",
+    "OpenAIResponsesAdapter",
     "ProviderAttemptContext",
     "ProviderAttemptController",
     "ProviderAttemptDisposition",