sandboxy 0.0.6__py3-none-any.whl → 0.0.7__py3-none-any.whl

sandboxy/api/app.py

@@ -36,6 +36,19 @@ def create_local_app(
         root_dir: Working directory for scenarios/tools/agents.
         local_ui_path: Path to local UI static files.
     """
+    # Load .env from root_dir to pick up API keys from the project directory
+    # This is important when root_dir differs from cwd
+    env_file = root_dir / ".env"
+    if env_file.exists():
+        load_dotenv(env_file, override=True)
+        logger.info(f"Loaded environment from {env_file}")
+
+        # Reset provider registry so it picks up the newly loaded API keys
+        from sandboxy.providers.registry import reset_registry
+
+        reset_registry()
+        logger.info("Provider registry reset after loading project .env")
+
     from sandboxy.local.context import LocalContext, set_local_context
 
     ctx = LocalContext(root_dir=root_dir)

sandboxy/api/routes/local.py

@@ -493,7 +493,10 @@ async def run_scenario(request: RunScenarioRequest) -> RunScenarioResponse:
                     experiment_name=mlflow_config.experiment,
                 )
 
-            with mlflow_run_context(mlflow_config, run_name=request.model) as run_id:
+            # Short run name - just the model (scenario is in experiment name)
+            run_name = request.model.split("/")[-1] if "/" in request.model else request.model
+
+            with mlflow_run_context(mlflow_config, run_name=run_name) as run_id:
                 result = await runner.run(
                     scenario=spec,
                     model=request.model,
@@ -633,12 +636,17 @@ async def compare_models(request: CompareModelsRequest) -> CompareModelsResponse
                         scenario_name=spec.name,
                     )
                     exporter = MLflowExporter(config)
+
+                    # Short run name - just the model
+                    run_name = result.model.split("/")[-1] if "/" in result.model else result.model
+
                     exporter.export(
                         result=result.to_dict(),
                         scenario_path=scenario_path,
                         scenario_name=spec.name,
                         scenario_id=spec.id,
                         agent_name=result.model,
+                        run_name=run_name,
                     )
             except ImportError:
                 logger.warning("MLflow not installed, skipping export")
@@ -1513,7 +1521,9 @@ async def run_with_dataset(request: RunDatasetRequest) -> RunDatasetResponse:
         if mlflow_config and mlflow_config.enabled:
             from sandboxy.mlflow import mlflow_run_context
 
-            run_name = f"{request.model}-{request.dataset_id}"
+            # Short run name - model + dataset
+            model_short = request.model.split("/")[-1] if "/" in request.model else request.model
+            run_name = f"{model_short} ({request.dataset_id})"
             with mlflow_run_context(mlflow_config, run_name=run_name) as run_id:
                 result = await run_dataset_benchmark()
 

sandboxy/cli/main.py

@@ -833,7 +833,9 @@ def scenario(
                 )
 
             # Start run, execute scenario, then log metrics - all connected
-            with mlflow_run_context(mlflow_config, run_name=model_id) as run_id:
+            # Short run name - just the model
+            run_name = model_id.split("/")[-1] if "/" in model_id else model_id
+            with mlflow_run_context(mlflow_config, run_name=run_name) as run_id:
                 runner = ScenarioRunner(scenario=spec, agent=agent)
                 result = runner.run(max_turns=max_turns)
 

sandboxy/providers/__init__.py

@@ -40,6 +40,7 @@ from sandboxy.providers.registry import (
     get_provider,
     get_registry,
     reload_local_providers,
+    reset_registry,
 )
 
 __all__ = [
@@ -53,6 +54,7 @@ __all__ = [
     "get_provider",
     "get_registry",
     "reload_local_providers",
+    "reset_registry",
     # Local provider
     "LocalProvider",
     "LocalProviderConnectionError",

sandboxy/providers/registry.py

@@ -154,23 +154,34 @@ class ProviderRegistry:
         if "/" in model_id:
             provider_name, model_name = model_id.split("/", 1)
 
-            # Check for local provider first (e.g., "ollama/llama3")
+            # Priority 1: Check for LOCAL provider with matching name
+            # Local providers take precedence over cloud providers
             if provider_name in self.providers:
                 provider = self.providers[provider_name]
-                # Verify it's a local provider or supports the model
-                if hasattr(provider, "config") or provider.supports_model(model_id):
+                # Only use if it's a local provider (has config attribute)
+                if hasattr(provider, "config"):
                     return provider
 
-            # OpenRouter format (e.g., "openai/gpt-4o")
+            # Priority 2: OpenRouter for provider/model format (e.g., "openai/gpt-4o")
             if "openrouter" in self.providers:
                 return self.providers["openrouter"]
 
-            # Fallback to direct provider if prefix matches
+            # Priority 3: Fallback to direct cloud provider if available
             if provider_name == "openai" and "openai" in self.providers:
                 return self.providers["openai"]
             if provider_name == "anthropic" and "anthropic" in self.providers:
                 return self.providers["anthropic"]
 
+            # No valid provider found for prefixed model - raise clear error
+            available = list(self.providers.keys())
+            raise ProviderError(
+                f"No provider available for model '{model_id}'. "
+                f"The '{provider_name}/' prefix requires OpenRouter (set OPENROUTER_API_KEY) "
+                f"or a local provider named '{provider_name}'. "
+                f"Available providers: {available}",
+                provider="registry",
+            )
+
         # No prefix - use direct providers
         model_lower = model_id.lower()
 
@@ -280,6 +291,16 @@ def get_registry() -> ProviderRegistry:
     return _registry
 
 
+def reset_registry() -> None:
+    """Reset the global provider registry.
+
+    Forces re-initialization on next get_registry() call.
+    Useful after loading new environment variables.
+    """
+    global _registry
+    _registry = None
+
+
 def get_provider(model_id: str) -> BaseProvider:
     """Get a provider for a model (convenience function).
 

sandboxy/ui/dist/assets/{index-BZFjoK-_.js → index-DfKu4uXt.js}

@@ -283,7 +283,7 @@ Please change the parent <Route path="${h}"> to <Route path="${h==="/"?"*":`${h}
  *
  * This source code is licensed under the ISC license.
  * See the LICENSE file in the root directory of this source tree.
- */const bn=W("Zap",[["path",{d:"M4 14a1 1 0 0 1-.78-1.63l9.9-10.2a.5.5 0 0 1 .86.46l-1.92 6.02A1 1 0 0 0 13 10h7a1 1 0 0 1 .78 1.63l-9.9 10.2a.5.5 0 0 1-.86-.46l1.92-6.02A1 1 0 0 0 11 14z",key:"1xq2db"}]]),Dg=[{path:"/",label:"Dashboard",icon:Pg},{path:"/results",label:"Results",icon:Hr},{path:"/builder",label:"Builder",icon:wn},{path:"/datasets",label:"Datasets",icon:sn}];function Ig(){const e=Ft(),[t,n]=x.useState(()=>typeof window<"u"?localStorage.getItem("theme")==="light":!1);return x.useEffect(()=>{t?(document.documentElement.classList.add("light"),localStorage.setItem("theme","light")):(document.documentElement.classList.remove("light"),localStorage.setItem("theme","dark"))},[t]),l.jsxs("div",{className:"app-shell",children:[l.jsxs("aside",{className:"sidebar",children:[l.jsx("div",{className:"p-5 border-b border-slate-800/70",children:l.jsxs("div",{className:"flex items-center justify-between",children:[l.jsxs("div",{children:[l.jsx("h1",{className:"text-xl font-semibold",style:{color:"var(--app-text)"},children:"Sandboxy Local"}),l.jsx("p",{className:"text-sm",style:{color:"var(--app-muted)"},children:"Development Server"})]}),l.jsx("button",{onClick:()=>n(!t),className:"p-2 rounded-lg transition-colors hover:bg-slate-500/20",title:t?"Switch to dark mode":"Switch to light mode",children:t?l.jsx(Ag,{size:20,style:{color:"var(--app-muted)"}}):l.jsx(zg,{size:20,style:{color:"var(--app-muted)"}})})]})}),l.jsx("nav",{className:"p-4",children:l.jsx("ul",{className:"space-y-2",children:Dg.map(({path:s,label:r,icon:a})=>l.jsx("li",{children:l.jsxs(de,{to:s,"data-active":e.pathname===s,className:"nav-link",children:[l.jsx(a,{size:20}),r]})},s))})})]}),l.jsx("main",{className:"flex-1 overflow-auto",children:l.jsx(Uh,{})})]})}const $g="/api/v1";class Ug{async fetch(t,n){const s=await fetch(`${$g}${t}`,{...n,headers:{"Content-Type":"application/json",...n==null?void 0:n.headers}});if(!s.ok){const r=await s.json().catch(()=>({detail:"Unknown error"}));throw new Error(r.detail||`HTTP ${s.status}`)}return s.json()}async getStatus(){return this.fetch("/local/status")}async listScenarios(){return this.fetch("/local/scenarios")}async getScenario(t){return this.fetch(`/local/scenarios/${encodeURIComponent(t)}`)}async getScenarioGoals(t){return this.fetch(`/local/scenarios/${encodeURIComponent(t)}/goals`)}async getScenarioTools(t){return this.fetch(`/local/scenarios/${encodeURIComponent(t)}/tools`)}async listTools(){return this.fetch("/local/tools")}async getTool(t){return this.fetch(`/local/tools/${encodeURIComponent(t)}`)}async listAgents(){return this.fetch("/local/agents")}async getAgent(t){return this.fetch(`/local/agents/${encodeURIComponent(t)}`)}async listModels(){return this.fetch("/local/models")}async listRuns(){return this.fetch("/local/runs")}async getRun(t){return this.fetch(`/local/runs/${encodeURIComponent(t)}`)}async runScenario(t){return this.fetch("/local/run",{method:"POST",body:JSON.stringify(t)})}async compareModels(t){return this.fetch("/local/compare",{method:"POST",body:JSON.stringify(t)})}async listDatasets(){return this.fetch("/local/datasets")}async getDataset(t){return this.fetch(`/local/datasets/${encodeURIComponent(t)}`)}async saveDataset(t,n){return this.fetch("/local/datasets",{method:"POST",body:JSON.stringify({id:t,content:n})})}async updateDataset(t,n){return this.fetch(`/local/datasets/${encodeURIComponent(t)}`,{method:"PUT",body:JSON.stringify({id:t,content:n})})}async deleteDataset(t){return this.fetch(`/local/datasets/${encodeURIComponent(t)}`,{method:"DELETE"})}async runDataset(t){return this.fetch("/local/run-dataset",{method:"POST",body:JSON.stringify(t)})}}class Bg extends Ug{async listProviders(){return(await this.fetch("/providers")).providers}async addProvider(t){return this.fetch("/providers",{method:"POST",body:JSON.stringify(t)})}async getProvider(t){return this.fetch(`/providers/${encodeURIComponent(t)}`)}async deleteProvider(t){await this.fetch(`/providers/${encodeURIComponent(t)}`,{method:"DELETE"})}async testProvider(t){return this.fetch(`/providers/${encodeURIComponent(t)}/test`,{method:"POST"})}}const Z=new Bg;function Hg(){const[e,t]=x.useState(null),[n,s]=x.useState(!0),[r,a]=x.useState(null),[i,o]=x.useState([]),[c,u]=x.useState(!0),[m,f]=x.useState(!1),[g,y]=x.useState(null),[j,v]=x.useState({}),[N,p]=x.useState({name:"",type:"ollama",base_url:"http://localhost:11434/v1"}),[d,h]=x.useState(null),[k,w]=x.useState(!1),R=async()=>{try{const A=await Z.listProviders();o(A)}catch{o([])}finally{u(!1)}},P=async A=>{A.preventDefault(),h(null),w(!0);try{await Z.addProvider(N),await R(),f(!1),p({name:"",type:"ollama",base_url:"http://localhost:11434/v1"})}catch(L){h(L instanceof Error?L.message:"Failed to add provider")}finally{w(!1)}},M=async A=>{y(A);try{const L=await Z.testProvider(A);v(U=>({...U,[A]:{success:L.success,message:L.success?`Connected! Found ${L.models_found.length} models (${L.latency_ms}ms)`:L.error||"Connection failed"}}))}catch(L){v(U=>({...U,[A]:{success:!1,message:L instanceof Error?L.message:"Test failed"}}))}finally{y(null)}},O=async A=>{if(confirm(`Delete provider "${A}"?`))try{await Z.deleteProvider(A),await R(),v(L=>{const U={...L};return delete U[A],U})}catch(L){alert(L instanceof Error?L.message:"Failed to delete provider")}},z=A=>{switch(A){case"ollama":return"http://localhost:11434/v1";case"lmstudio":return"http://localhost:1234/v1";case"vllm":return"http://localhost:8000/v1";default:return"http://localhost:8000/v1"}};return x.useEffect(()=>{fetch("/api/v1/local/status").then(A=>{if(!A.ok)throw new Error("Failed to fetch status");return A.json()}).then(t).catch(A=>a(A.message)).finally(()=>s(!1)),R()},[]),n?l.jsx("div",{className:"p-8 page",children:l.jsx("div",{className:"text-slate-400",children:"Loading..."})}):r?l.jsx("div",{className:"p-8 page",children:l.jsxs("div",{className:"text-red-400",children:["Error: ",r]})}):e?l.jsxs("div",{className:"p-8 page",children:[l.jsxs("div",{className:"mb-8",children:[l.jsx("h1",{className:"text-2xl font-semibold text-slate-100 mb-2",children:"Dashboard"}),l.jsxs("p",{className:"text-slate-400",children:["Root: ",l.jsx("code",{className:"panel-subtle px-2 py-1 rounded",children:e.root_dir})]})]}),l.jsxs("section",{className:"mb-8",children:[l.jsxs("div",{className:"flex items-center justify-between mb-4",children:[l.jsxs("h2",{className:"text-lg font-semibold text-slate-100 flex items-center gap-2",children:[l.jsx(hp,{size:20}),"Scenarios (",e.scenarios.length,")"]}),l.jsxs(de,{to:"/builder",className:"flex items-center gap-1 px-3 py-1.5 bg-orange-400 hover:bg-orange-300 text-slate-900 rounded-lg text-sm font-semibold",children:[l.jsx(Ne,{size:16}),"Create Scenario"]})]}),e.scenarios.length===0?l.jsx("p",{className:"text-slate-400",children:"No scenarios found in scenarios/"}):l.jsx("div",{className:"grid gap-4 md:grid-cols-2 lg:grid-cols-3",children:e.scenarios.map(A=>l.jsxs("div",{className:"panel-card p-4 hover:border-slate-600 transition-colors",children:[l.jsxs(de,{to:`/scenario/${A.id}`,className:"block mb-3",children:[l.jsxs("div",{className:"flex items-start justify-between",children:[l.jsx("h3",{className:"font-medium text-slate-100 mb-1 hover:text-orange-400 transition-colors",children:A.name}),l.jsx(ge,{size:16,className:"text-slate-600 mt-1 flex-shrink-0"})]}),l.jsx("p",{className:"text-sm text-slate-400 line-clamp-2",children:A.description||"No description"})]}),l.jsxs("div",{className:"flex justify-between items-center pt-2 border-t border-slate-700/40",children:[l.jsx("span",{className:"text-xs text-slate-500",children:A.relative_path}),l.jsxs(de,{to:`/run/${A.id}`,className:"flex items-center gap-1 px-3 py-1 bg-orange-400 text-slate-900 rounded hover:bg-orange-300 transition-colors text-sm font-medium",children:[l.jsx(Xt,{size:14}),"Run"]})]})]},A.id))})]}),l.jsxs("section",{className:"mb-8",children:[l.jsxs("div",{className:"flex items-center justify-between mb-4",children:[l.jsxs("h2",{className:"text-lg font-semibold text-slate-100 flex items-center gap-2",children:[l.jsx(wn,{size:20}),"Tools (",e.tools.length,")"]}),l.jsxs(de,{to:"/tool-builder",className:"flex items-center gap-1 px-3 py-1.5 bg-emerald-400 hover:bg-emerald-300 text-slate-900 rounded-lg text-sm font-semibold",children:[l.jsx(Ne,{size:16}),"Create Tool"]})]}),e.tools.length===0?l.jsx("p",{className:"text-slate-400",children:"No tools found in tools/"}):l.jsx("div",{className:"flex flex-wrap gap-2",children:e.tools.map(A=>l.jsx("span",{className:"px-3 py-1 panel-subtle rounded-full text-sm text-slate-200 border border-slate-700/60",children:A.name},A.id))})]}),l.jsxs("section",{className:"mb-8",children:[l.jsxs("div",{className:"flex items-center justify-between mb-4",children:[l.jsxs("h2",{className:"text-lg font-semibold text-slate-100 flex items-center gap-2",children:[l.jsx(Ao,{size:20}),"Local Providers (",i.length,")"]}),l.jsxs("button",{onClick:()=>f(!m),className:"flex items-center gap-1 px-3 py-1.5 bg-purple-500 hover:bg-purple-400 text-white rounded-lg text-sm font-semibold",children:[l.jsx(Ne,{size:16}),"Add Provider"]})]}),m&&l.jsxs("div",{className:"panel-card p-4 mb-4",children:[l.jsx("h3",{className:"text-sm font-semibold text-slate-200 mb-3",children:"Add Local Provider"}),l.jsxs("form",{onSubmit:P,className:"space-y-3",children:[l.jsxs("div",{className:"grid grid-cols-2 gap-3",children:[l.jsxs("div",{children:[l.jsx("label",{className:"block text-xs text-slate-400 mb-1",children:"Name"}),l.jsx("input",{type:"text",value:N.name,onChange:A=>p({...N,name:A.target.value}),placeholder:"ollama-local",className:"w-full px-3 py-1.5 bg-slate-800 border border-slate-700 rounded text-sm text-slate-100 placeholder-slate-500",required:!0})]}),l.jsxs("div",{children:[l.jsx("label",{className:"block text-xs text-slate-400 mb-1",children:"Type"}),l.jsxs("select",{value:N.type,onChange:A=>{const L=A.target.value;p({...N,type:L,base_url:z(L)})},className:"w-full px-3 py-1.5 bg-slate-800 border border-slate-700 rounded text-sm text-slate-100",children:[l.jsx("option",{value:"ollama",children:"Ollama"}),l.jsx("option",{value:"lmstudio",children:"LM Studio"}),l.jsx("option",{value:"vllm",children:"vLLM"}),l.jsx("option",{value:"openai-compatible",children:"OpenAI Compatible"})]})]})]}),l.jsxs("div",{children:[l.jsx("label",{className:"block text-xs text-slate-400 mb-1",children:"Base URL"}),l.jsx("input",{type:"url",value:N.base_url,onChange:A=>p({...N,base_url:A.target.value}),placeholder:"http://localhost:11434/v1",className:"w-full px-3 py-1.5 bg-slate-800 border border-slate-700 rounded text-sm text-slate-100 placeholder-slate-500",required:!0})]}),d&&l.jsx("p",{className:"text-red-400 text-xs",children:d}),l.jsxs("div",{className:"flex gap-2 justify-end",children:[l.jsx("button",{type:"button",onClick:()=>f(!1),className:"px-3 py-1.5 text-slate-400 hover:text-slate-200 text-sm",children:"Cancel"}),l.jsx("button",{type:"submit",disabled:k,className:"px-3 py-1.5 bg-purple-500 hover:bg-purple-400 text-white rounded text-sm font-medium disabled:opacity-50",children:k?"Adding...":"Add Provider"})]})]})]}),c?l.jsx("p",{className:"text-slate-400",children:"Loading providers..."}):i.length===0?l.jsxs("div",{className:"text-slate-400",children:[l.jsx("p",{className:"mb-2",children:"No local providers configured."}),l.jsx("p",{className:"text-sm",children:"Add a provider to use local models like Ollama, LM Studio, or vLLM."})]}):l.jsx("div",{className:"grid gap-3",children:i.map(A=>{var L,U;return l.jsxs("div",{className:"panel-card p-4 flex items-center justify-between",children:[l.jsxs("div",{className:"flex items-center gap-3",children:[A.status==="connected"?l.jsx(fl,{size:18,className:"text-green-400"}):A.status==="error"?l.jsx(Eo,{size:18,className:"text-red-400"}):l.jsx(nn,{size:18,className:"text-slate-500"}),l.jsxs("div",{className:"flex-1 min-w-0",children:[l.jsxs("div",{className:"flex items-center gap-2",children:[l.jsx("span",{className:"font-medium text-slate-100",children:A.name}),l.jsx("span",{className:"text-xs px-1.5 py-0.5 bg-slate-700 rounded text-slate-300",children:A.type}),A.enabled&&l.jsx("span",{className:"text-xs px-1.5 py-0.5 bg-green-900/50 text-green-400 rounded",children:"enabled"})]}),l.jsx("div",{className:"text-xs text-slate-400 mt-0.5",children:A.base_url}),((L=A.models)==null?void 0:L.length)>0&&l.jsx("div",{className:"flex flex-wrap gap-1.5 mt-2",children:A.models.map(B=>l.jsx("span",{className:"text-xs px-2 py-0.5 bg-emerald-900/40 border border-emerald-700/50 rounded text-emerald-300",children:B},B))}),(((U=A.models)==null?void 0:U.length)===0||!A.models)&&A.status==="connected"&&l.jsx("div",{className:"text-xs text-slate-500 mt-1",children:"No models available"}),j[A.name]&&l.jsx("div",{className:`text-xs mt-1 ${j[A.name].success?"text-green-400":"text-red-400"}`,children:j[A.name].message})]})]}),l.jsxs("div",{className:"flex items-center gap-2",children:[l.jsx("button",{onClick:()=>M(A.name),disabled:g===A.name,className:"p-1.5 text-slate-400 hover:text-slate-200 hover:bg-slate-700 rounded disabled:opacity-50",title:"Test connection",children:l.jsx(Mg,{size:16,className:g===A.name?"animate-spin":""})}),l.jsx("button",{onClick:()=>O(A.name),className:"p-1.5 text-slate-400 hover:text-red-400 hover:bg-slate-700 rounded",title:"Delete provider",children:l.jsx(Ae,{size:16})})]})]},A.name)})})]})]}):null}function Vg(){const{scenarioId:e}=rp(),[t,n]=x.useState(null),[s,r]=x.useState([]),[a,i]=x.useState([]),[o,c]=x.useState(!0),[u,m]=x.useState(null),[f,g]=x.useState(new Set(["tools","evaluation","interaction"])),y=L=>{g(U=>{const B=new Set(U);return B.has(L)?B.delete(L):B.add(L),B})};if(x.useEffect(()=>{e&&(c(!0),Promise.all([Z.getScenario(e),Z.getScenarioGoals(e).catch(()=>[]),Z.getScenarioTools(e).catch(()=>[])]).then(([L,U,B])=>{n(L),r(U),i(B)}).catch(L=>m(L.message)).finally(()=>c(!1)))},[e]),o)return l.jsx("div",{className:"p-8 page",children:l.jsx("div",{className:"text-slate-400",children:"Loading scenario..."})});if(u||!t)return l.jsxs("div",{className:"p-8 page",children:[l.jsxs("div",{className:"text-red-400",children:["Error: ",u||"Scenario not found"]}),l.jsx(de,{to:"/",className:"text-orange-400 hover:underline mt-4 inline-block",children:"← Back to Dashboard"})]});const j=t.content||{},v=j.steps||[],N=j.mcp_servers||[],p=j.tools||{},d=j.tools_from||[],h=j.evaluation?j.evaluation.judge:null,k=j.evaluation,w=k==null?void 0:k.goals,R=k?k.max_score||((w==null?void 0:w.reduce((L,U)=>L+(U.points||0),0))??0):null,P=t.variables||[],M=v.length>0,O=j.category||null,z=j.tags||[],A=a.length>0?a.length:Object.keys(p).length;return l.jsxs("div",{className:"p-8 page max-w-5xl",children:[l.jsxs(de,{to:"/",className:"inline-flex items-center gap-1 text-slate-400 hover:text-slate-200 mb-6 text-sm",children:[l.jsx(Gs,{size:16}),"Back to Dashboard"]}),l.jsxs("div",{className:"panel-card p-6 mb-6",children:[l.jsxs("div",{className:"flex items-start justify-between",children:[l.jsxs("div",{className:"flex-1",children:[l.jsx("h1",{className:"text-2xl font-semibold text-slate-100 mb-2",children:t.name}),l.jsx("p",{className:"text-slate-400 mb-4",children:t.description}),l.jsxs("div",{className:"flex flex-wrap gap-2",children:[O&&l.jsx("span",{className:"px-2 py-1 bg-purple-900/40 border border-purple-700/50 rounded text-xs text-purple-300",children:O}),z.map(L=>l.jsx("span",{className:"px-2 py-1 bg-slate-700/50 border border-slate-600/50 rounded text-xs text-slate-300",children:L},L))]})]}),l.jsxs("div",{className:"flex gap-2 ml-4",children:[l.jsxs(de,{to:`/builder?scenario=${e}`,className:"flex items-center gap-1.5 px-3 py-2 bg-slate-700 hover:bg-slate-600 text-slate-200 rounded-lg text-sm font-medium transition-colors",children:[l.jsx(Lg,{size:16}),"Edit"]}),l.jsxs(de,{to:`/run/${e}`,className:"flex items-center gap-1.5 px-4 py-2 bg-orange-400 hover:bg-orange-300 text-slate-900 rounded-lg text-sm font-semibold transition-colors",children:[l.jsx(Xt,{size:16}),"Run"]})]})]}),l.jsxs("div",{className:"flex gap-6 mt-6 pt-4 border-t border-slate-700/60",children:[l.jsxs("div",{className:"flex items-center gap-2 text-sm",children:[l.jsx(vr,{size:16,className:"text-slate-500"}),l.jsx("span",{className:"text-slate-400",children:M?`Multi-turn (${v.length} steps)`:"Single-turn"})]}),l.jsxs("div",{className:"flex items-center gap-2 text-sm",children:[l.jsx(wn,{size:16,className:"text-slate-500"}),l.jsx("span",{className:"text-slate-400",children:N.length>0&&A===0?`${N.length} MCP server${N.length!==1?"s":""}`:`${A} tool${A!==1?"s":""}${N.length>0?` + ${N.length} MCP`:""}`})]}),l.jsxs("div",{className:"flex items-center gap-2 text-sm",children:[l.jsx(yi,{size:16,className:"text-slate-500"}),l.jsxs("span",{className:"text-slate-400",children:[s.length," goal",s.length!==1?"s":"",R&&` (${R} pt${R!==1?"s":""})`]})]}),P.length>0&&l.jsxs("div",{className:"flex items-center gap-2 text-sm",children:[l.jsx(ji,{size:16,className:"text-slate-500"}),l.jsxs("span",{className:"text-slate-400",children:[P.length," variables"]})]})]})]}),l.jsxs("div",{className:"panel-card mb-4",children:[l.jsxs("button",{onClick:()=>y("tools"),className:"w-full flex items-center justify-between p-4 text-left hover:bg-slate-800/30 transition-colors",children:[l.jsxs("div",{className:"flex items-center gap-2",children:[l.jsx(wn,{size:18,className:"text-emerald-400"}),l.jsxs("h2",{className:"font-semibold text-slate-100",children:["Tools",N.length>0&&A===0?l.jsxs("span",{className:"ml-2 text-sm font-normal text-slate-400",children:["(",N.length," MCP server",N.length>1?"s":"",")"]}):l.jsxs("span",{className:"ml-2 text-sm font-normal text-slate-400",children:["(",A,N.length>0?` + ${N.length} MCP`:"",")"]})]})]}),f.has("tools")?l.jsx(pe,{size:18,className:"text-slate-500"}):l.jsx(ge,{size:18,className:"text-slate-500"})]}),f.has("tools")&&l.jsxs("div",{className:"px-4 pb-4 space-y-4",children:[N.length>0&&l.jsxs("div",{children:[l.jsx("h3",{className:"text-xs font-semibold text-slate-500 uppercase tracking-wider mb-2",children:"MCP Servers"}),N.map(L=>{var U;return l.jsxs("div",{className:"p-3 bg-slate-800/50 border border-slate-700/60 rounded-lg mb-2",children:[l.jsxs("div",{className:"flex items-center gap-2 mb-2",children:[l.jsx(Ao,{size:14,className:"text-cyan-400"}),l.jsx("span",{className:"font-medium text-slate-200",children:L.name}),l.jsx("span",{className:"text-xs px-1.5 py-0.5 bg-cyan-900/40 text-cyan-300 rounded",children:"MCP"})]}),l.jsx("div",{className:"text-xs text-slate-500 mb-2",children:L.url?l.jsx("span",{className:"font-mono",children:L.url}):L.command?l.jsxs("span",{className:"font-mono",children:[L.command," ",(U=L.args)==null?void 0:U.join(" ")]}):null}),a.length>0&&l.jsx("div",{className:"flex flex-wrap gap-1.5",children:a.map(B=>l.jsx("span",{className:"text-xs px-2 py-1 bg-emerald-900/30 border border-emerald-700/40 rounded text-emerald-300",title:B.description,children:B.name},B.name))})]},L.name)})]}),d.length>0&&l.jsxs("div",{children:[l.jsx("h3",{className:"text-xs font-semibold text-slate-500 uppercase tracking-wider mb-2",children:"Tool Libraries"}),l.jsx("div",{className:"flex flex-wrap gap-2",children:d.map(L=>l.jsx("span",{className:"px-3 py-1.5 bg-slate-800/50 border border-slate-700/60 rounded text-sm text-slate-300",children:L},L))})]}),Object.keys(p).length>0&&l.jsxs("div",{children:[l.jsx("h3",{className:"text-xs font-semibold text-slate-500 uppercase tracking-wider mb-2",children:"Inline Tools (Mock)"}),l.jsx("div",{className:"space-y-2",children:Object.entries(p).map(([L,U])=>{const B=U,C=B.actions,E=C?Object.keys(C):[];return l.jsxs("div",{className:"p-3 bg-slate-800/50 border border-slate-700/60 rounded-lg",children:[l.jsxs("div",{className:"flex items-center gap-2 mb-1",children:[l.jsx(bn,{size:14,className:"text-yellow-400"}),l.jsx("span",{className:"font-medium text-slate-200",children:L}),l.jsx("span",{className:"text-xs px-1.5 py-0.5 bg-yellow-900/40 text-yellow-300 rounded",children:"Mock"})]}),typeof B.description=="string"&&B.description&&l.jsx("p",{className:"text-xs text-slate-500 mb-2",children:B.description}),E.length>0&&l.jsx("div",{className:"flex flex-wrap gap-1.5",children:E.map(_=>l.jsx("span",{className:"text-xs px-2 py-1 bg-slate-700/50 rounded text-slate-400",children:_},_))})]},L)})})]}),N.length===0&&d.length===0&&Object.keys(p).length===0&&l.jsx("p",{className:"text-slate-500 text-sm",children:"No tools configured"})]})]}),l.jsxs("div",{className:"panel-card mb-4",children:[l.jsxs("button",{onClick:()=>y("evaluation"),className:"w-full flex items-center justify-between p-4 text-left hover:bg-slate-800/30 transition-colors",children:[l.jsxs("div",{className:"flex items-center gap-2",children:[l.jsx(yi,{size:18,className:"text-orange-400"}),l.jsxs("h2",{className:"font-semibold text-slate-100",children:["Evaluation",R&&l.jsxs("span",{className:"ml-2 text-sm font-normal text-slate-400",children:["(",R," pts max)"]})]})]}),f.has("evaluation")?l.jsx(pe,{size:18,className:"text-slate-500"}):l.jsx(ge,{size:18,className:"text-slate-500"})]}),f.has("evaluation")&&l.jsxs("div",{className:"px-4 pb-4 space-y-4",children:[s.length>0&&l.jsxs("div",{children:[l.jsx("h3",{className:"text-xs font-semibold text-slate-500 uppercase tracking-wider mb-3",children:"Goals"}),l.jsx("div",{className:"space-y-2",children:s.map(L=>{const U=j.evaluation,B=U==null?void 0:U.goals,C=B==null?void 0:B.find(S=>S.id===L.id),E=C==null?void 0:C.detection,_=E==null?void 0:E.type,b=C==null?void 0:C.points;return l.jsxs("div",{className:"flex items-start gap-3 p-3 bg-slate-800/50 border border-slate-700/60 rounded-lg",children:[l.jsx(fl,{size:16,className:"text-slate-600 mt-0.5 flex-shrink-0"}),l.jsxs("div",{className:"flex-1 min-w-0",children:[l.jsxs("div",{className:"flex items-center gap-2 mb-1",children:[l.jsx("span",{className:"font-medium text-slate-200",children:L.name}),L.outcome&&l.jsx("span",{className:"text-xs px-1.5 py-0.5 bg-blue-900/40 text-blue-300 rounded",children:"outcome"})]}),L.description&&l.jsx("p",{className:"text-xs text-slate-500",children:L.description})]}),l.jsxs("div",{className:"flex items-center gap-2 flex-shrink-0",children:[_&&l.jsx("span",{className:"text-xs px-2 py-1 bg-slate-700/50 rounded text-slate-400 font-mono",children:_}),b!==void 0&&l.jsxs("span",{className:"text-sm font-semibold text-orange-400",children:[b," pts"]})]})]},L.id)})})]}),h&&l.jsxs("div",{children:[l.jsx("h3",{className:"text-xs font-semibold text-slate-500 uppercase tracking-wider mb-3",children:"Judge"}),l.jsxs("div",{className:"p-3 bg-slate-800/50 border border-slate-700/60 rounded-lg",children:[l.jsxs("div",{className:"flex items-center gap-2 mb-2",children:[l.jsx(Fg,{size:14,className:"text-purple-400"}),l.jsxs("span",{className:"font-medium text-slate-200 capitalize",children:[h.type," Judge"]}),h.model&&l.jsx("span",{className:"text-xs px-1.5 py-0.5 bg-purple-900/40 text-purple-300 rounded",children:h.model})]}),h.pass_threshold!==void 0&&l.jsxs("p",{className:"text-xs text-slate-500 mb-2",children:["Pass threshold: ",(h.pass_threshold*100).toFixed(0),"%"]}),h.rubric&&l.jsxs("details",{className:"text-xs",children:[l.jsx("summary",{className:"text-slate-400 cursor-pointer hover:text-slate-300",children:"View rubric"}),l.jsx("pre",{className:"mt-2 p-2 bg-slate-900 rounded text-slate-400 overflow-auto max-h-48 whitespace-pre-wrap",children:h.rubric})]})]})]}),s.length===0&&!h&&l.jsx("p",{className:"text-slate-500 text-sm",children:"No evaluation configured"})]})]}),l.jsxs("div",{className:"panel-card mb-4",children:[l.jsxs("button",{onClick:()=>y("interaction"),className:"w-full flex items-center justify-between p-4 text-left hover:bg-slate-800/30 transition-colors",children:[l.jsxs("div",{className:"flex items-center gap-2",children:[l.jsx(vr,{size:18,className:"text-blue-400"}),l.jsxs("h2",{className:"font-semibold text-slate-100",children:["Interaction",l.jsxs("span",{className:"ml-2 text-sm font-normal text-slate-400",children:["(",M?"Multi-turn":"Single-turn",")"]})]})]}),f.has("interaction")?l.jsx(pe,{size:18,className:"text-slate-500"}):l.jsx(ge,{size:18,className:"text-slate-500"})]}),f.has("interaction")&&l.jsxs("div",{className:"px-4 pb-4 space-y-4",children:[typeof j.system_prompt=="string"&&j.system_prompt&&l.jsxs("div",{children:[l.jsx("h3",{className:"text-xs font-semibold text-slate-500 uppercase tracking-wider mb-2",children:"System Prompt"}),l.jsx("pre",{className:"p-3 bg-slate-800/50 border border-slate-700/60 rounded-lg text-xs text-slate-300 overflow-auto max-h-48 whitespace-pre-wrap",children:j.system_prompt})]}),M&&l.jsxs("div",{children:[l.jsx("h3",{className:"text-xs font-semibold text-slate-500 uppercase tracking-wider mb-3",children:"Steps"}),l.jsx("div",{className:"space-y-2",children:v.map((L,U)=>{var B;return l.jsxs("div",{className:"flex items-start gap-3 p-3 bg-slate-800/50 border border-slate-700/60 rounded-lg",children:[l.jsx("div",{className:"w-6 h-6 rounded-full bg-blue-900/40 border border-blue-700/50 flex items-center justify-center text-xs text-blue-300 flex-shrink-0",children:U+1}),l.jsxs("div",{className:"flex-1",children:[l.jsxs("div",{className:"flex items-center gap-2 mb-1",children:[l.jsx("span",{className:"font-medium text-slate-200",children:L.id}),l.jsx("span",{className:"text-xs px-1.5 py-0.5 bg-slate-700/50 text-slate-400 rounded font-mono",children:L.action})]}),typeof((B=L.params)==null?void 0:B.content)=="string"&&L.params.content&&l.jsxs("p",{className:"text-xs text-slate-500 line-clamp-2",children:[L.params.content.substring(0,100),L.params.content.length>100&&"..."]})]})]},L.id)})})]}),!M&&typeof j.prompt=="string"&&j.prompt&&l.jsxs("div",{children:[l.jsx("h3",{className:"text-xs font-semibold text-slate-500 uppercase tracking-wider mb-2",children:"Prompt"}),l.jsx("pre",{className:"p-3 bg-slate-800/50 border border-slate-700/60 rounded-lg text-xs text-slate-300 overflow-auto max-h-48 whitespace-pre-wrap",children:j.prompt})]})]})]}),P.length>0&&l.jsxs("div",{className:"panel-card mb-4",children:[l.jsxs("button",{onClick:()=>y("variables"),className:"w-full flex items-center justify-between p-4 text-left hover:bg-slate-800/30 transition-colors",children:[l.jsxs("div",{className:"flex items-center gap-2",children:[l.jsx(ji,{size:18,className:"text-indigo-400"}),l.jsxs("h2",{className:"font-semibold text-slate-100",children:["Variables (",P.length,")"]})]}),f.has("variables")?l.jsx(pe,{size:18,className:"text-slate-500"}):l.jsx(ge,{size:18,className:"text-slate-500"})]}),f.has("variables")&&l.jsx("div",{className:"px-4 pb-4",children:l.jsx("div",{className:"grid gap-2",children:P.map(L=>l.jsxs("div",{className:"flex items-center justify-between p-3 bg-slate-800/50 border border-slate-700/60 rounded-lg",children:[l.jsxs("div",{className:"flex items-center gap-3",children:[l.jsx("code",{className:"text-sm font-mono text-indigo-300",children:`{{${L.name}}}`}),l.jsx("span",{className:"text-xs px-1.5 py-0.5 bg-slate-700/50 text-slate-400 rounded",children:L.type}),L.required&&l.jsx("span",{className:"text-xs px-1.5 py-0.5 bg-red-900/40 text-red-300 rounded",children:"required"})]}),L.default!==void 0&&l.jsxs("span",{className:"text-xs text-slate-500",children:["default: ",JSON.stringify(L.default)]})]},L.name))})})]}),l.jsxs("div",{className:"panel-card",children:[l.jsxs("button",{onClick:()=>y("file"),className:"w-full flex items-center justify-between p-4 text-left hover:bg-slate-800/30 transition-colors",children:[l.jsxs("div",{className:"flex items-center gap-2",children:[l.jsx(Hr,{size:18,className:"text-slate-500"}),l.jsx("h2",{className:"font-semibold text-slate-100",children:"File"})]}),f.has("file")?l.jsx(pe,{size:18,className:"text-slate-500"}):l.jsx(ge,{size:18,className:"text-slate-500"})]}),f.has("file")&&l.jsx("div",{className:"px-4 pb-4",children:l.jsx("p",{className:"text-xs text-slate-500 font-mono",children:t.path})})]})]})}function Wg(){const[e,t]=x.useState("idle"),[n,s]=x.useState(null),[r,a]=x.useState(null),[i,o]=x.useState(null),c=x.useCallback(()=>{t("idle"),s(null),a(null),o(null)},[]),u=x.useCallback(async(f,g,y,j)=>{c(),t("running");try{const v=await Z.runScenario({scenario_id:f,model:g,variables:y,mlflow_export:j==null?void 0:j.enabled,mlflow_tracking_uri:j==null?void 0:j.trackingUri,mlflow_experiment:j==null?void 0:j.experiment,mlflow_tracing:j==null?void 0:j.tracing});v.error?(t("error"),o(v.error)):(t("completed"),s(v))}catch(v){t("error"),o(v instanceof Error?v.message:"Unknown error")}},[c]),m=x.useCallback(async(f,g,y=1,j,v)=>{c(),t("running");try{const N=await Z.compareModels({scenario_id:f,models:g,runs_per_model:y,variables:j,mlflow_export:v==null?void 0:v.enabled,mlflow_tracking_uri:v==null?void 0:v.trackingUri,mlflow_experiment:v==null?void 0:v.experiment,mlflow_tracing:v==null?void 0:v.tracing});t("completed"),a(N)}catch(N){t("error"),o(N instanceof Error?N.message:"Unknown error")}},[c]);return{state:e,result:n,comparison:r,error:i,runScenario:u,compareModels:m,reset:c}}function jp(e){return e==null?"-":e<1e-4?"<$0.0001":e<.01?`$${e.toFixed(4)}`:`$${e.toFixed(3)}`}function Mo({result:e}){var s,r,a;const[t,n]=x.useState(!1);return l.jsxs("div",{className:"space-y-6",children:[l.jsxs("div",{className:"grid grid-cols-2 md:grid-cols-3 lg:grid-cols-6 gap-3",children:[e.evaluation&&l.jsxs("div",{className:"panel-card p-4 col-span-2 md:col-span-1",children:[l.jsxs("div",{className:"flex items-center gap-2 text-slate-400 mb-2",children:[l.jsx(Lo,{className:"w-4 h-4"}),"Score"]}),l.jsxs("div",{className:"text-2xl font-bold text-slate-100",children:[e.evaluation.percentage.toFixed(0),"%"]}),l.jsxs("div",{className:"text-xs text-slate-500",children:[e.evaluation.total_score.toFixed(0)," / ",e.evaluation.max_score.toFixed(0)," pts"]})]}),l.jsxs("div",{className:"panel-card p-4",children:[l.jsxs("div",{className:"flex items-center gap-2 text-slate-400 mb-2",children:[l.jsx(_o,{className:"w-4 h-4"}),"Latency"]}),l.jsxs("div",{className:"text-2xl font-bold text-slate-100",children:[((e.latency_ms||0)/1e3).toFixed(1),"s"]})]}),l.jsxs("div",{className:"panel-card p-4",children:[l.jsxs("div",{className:"flex items-center gap-2 text-slate-400 mb-2",children:[l.jsx(vr,{className:"w-4 h-4"}),"Messages"]}),l.jsx("div",{className:"text-2xl font-bold text-slate-100",children:((s=e.history)==null?void 0:s.length)||0})]}),l.jsxs("div",{className:"panel-card p-4",children:[l.jsxs("div",{className:"flex items-center gap-2 text-slate-400 mb-2",children:[l.jsx(bn,{className:"w-4 h-4"}),"Tool Calls"]}),l.jsx("div",{className:"text-2xl font-bold text-slate-100",children:((r=e.tool_calls)==null?void 0:r.length)||0})]}),l.jsxs("div",{className:"panel-card p-4",children:[l.jsxs("div",{className:"flex items-center gap-2 text-slate-400 mb-2",children:[l.jsx(Tg,{className:"w-4 h-4"}),"Tokens"]}),l.jsx("div",{className:"text-2xl font-bold text-slate-100",children:((e.input_tokens||0)+(e.output_tokens||0)).toLocaleString()}),l.jsxs("div",{className:"text-xs text-slate-500",children:[(e.input_tokens||0).toLocaleString()," in / ",(e.output_tokens||0).toLocaleString()," out"]})]}),l.jsxs("div",{className:"panel-card p-4",children:[l.jsxs("div",{className:"flex items-center gap-2 text-slate-400 mb-2",children:[l.jsx(To,{className:"w-4 h-4"}),"Cost"]}),l.jsx("div",{className:"text-2xl font-bold text-emerald-300",children:jp(e.cost_usd)})]})]}),e.evaluation&&e.evaluation.goals&&e.evaluation.goals.length>0&&l.jsxs("div",{className:"panel-card p-6",children:[l.jsx("h3",{className:"text-lg font-semibold text-slate-100 mb-4",children:"Goals"}),l.jsx("div",{className:"space-y-3",children:e.evaluation.goals.map(i=>l.jsxs("div",{className:`flex items-center justify-between p-3 rounded-lg border ${i.achieved?"bg-emerald-500/10 border-emerald-400/40":"bg-slate-950/40 border-slate-800/70"}`,children:[l.jsxs("div",{className:"flex items-center gap-3",children:[i.achieved?l.jsx(fl,{className:"w-5 h-5 text-emerald-400"}):l.jsx(Eo,{className:"w-5 h-5 text-slate-500"}),l.jsxs("div",{children:[l.jsx("div",{className:"font-medium text-slate-100",children:i.name||i.id}),i.reason&&l.jsx("div",{className:"text-sm text-slate-400",children:i.reason})]})]}),l.jsxs("div",{className:`font-bold ${i.achieved?"text-emerald-300":"text-slate-500"}`,children:[i.achieved?`+${i.points}`:"0"," pts"]})]},i.id))})]}),((a=e.evaluation)==null?void 0:a.judge)&&l.jsxs("div",{className:"panel-card p-6",children:[l.jsx("h3",{className:"text-lg font-semibold text-slate-100 mb-4",children:"Judge Evaluation"}),l.jsxs("div",{className:"flex items-center gap-4 mb-4",children:[l.jsxs("div",{className:`text-3xl font-bold ${e.evaluation.judge.passed?"text-emerald-300":"text-red-400"}`,children:[(e.evaluation.judge.score*100).toFixed(0),"%"]}),l.jsx("div",{className:`px-3 py-1 rounded-full text-sm ${e.evaluation.judge.passed?"bg-emerald-500/15 text-emerald-300":"bg-red-900/50 text-red-400"}`,children:e.evaluation.judge.passed?"Passed":"Failed"})]}),l.jsx("p",{className:"text-slate-400",children:e.evaluation.judge.reasoning})]}),e.response&&l.jsxs("div",{className:"panel-card p-6",children:[l.jsx("h3",{className:"text-lg font-semibold text-slate-100 mb-4",children:"Response"}),l.jsx("div",{className:"panel-subtle rounded-lg p-4 whitespace-pre-wrap text-slate-200",children:e.response||"(No response)"})]}),e.history&&e.history.length>0&&l.jsxs("div",{className:"panel-card",children:[l.jsxs("button",{onClick:()=>n(!t),className:"w-full p-4 text-left flex items-center justify-between hover:bg-slate-800/60 transition-colors rounded-lg",children:[l.jsxs("span",{className:"font-semibold text-slate-100",children:["Conversation History (",e.history.length," messages)"]}),l.jsx("span",{className:"text-slate-400",children:t?"Hide":"Show"})]}),t&&l.jsx("div",{className:"px-6 pb-6 space-y-4",children:e.history.map((i,o)=>l.jsxs("div",{className:`p-3 rounded-lg ${i.role==="user"?"bg-orange-500/10 border border-orange-400/40":i.role==="assistant"?"panel-subtle":i.role==="tool"?"bg-amber-500/10 border border-amber-400/40":"panel-subtle"}`,children:[l.jsx("div",{className:"text-xs font-medium text-slate-500 mb-1 uppercase",children:i.role}),l.jsx("div",{className:"text-slate-200 whitespace-pre-wrap text-sm",children:i.content||"(empty)"})]},o))})]}),e.tool_calls&&e.tool_calls.length>0&&l.jsxs("div",{className:"panel-card p-6",children:[l.jsxs("h3",{className:"text-lg font-semibold text-slate-100 mb-4",children:["Tool Calls (",e.tool_calls.length,")"]}),l.jsx("div",{className:"space-y-3",children:e.tool_calls.map((i,o)=>l.jsx(Yg,{call:i},o))})]}),e.error&&l.jsxs("div",{className:"panel-card p-4 border-l-4 border-l-red-500 bg-red-950/30",children:[l.jsxs("div",{className:"flex items-center gap-2 text-red-400 mb-1",children:[l.jsx(nn,{className:"w-4 h-4"}),l.jsx("span",{className:"font-medium",children:"Error"})]}),l.jsx("p",{className:"text-sm text-red-300",children:e.error})]})]})}function uu(e){if(typeof e=="string")try{const t=JSON.parse(e);return JSON.stringify(t,null,2)}catch{return e}return JSON.stringify(e,null,2)}function Yg({call:e}){const[t,n]=x.useState(!1),s=e.result!==void 0&&e.result!==null,r=e.error!==void 0&&e.error!==null;return l.jsxs("div",{className:`panel-subtle rounded-lg overflow-hidden ${e.success?"":"border border-red-500/30"}`,children:[l.jsxs("button",{onClick:()=>n(!t),className:"w-full p-4 text-left flex items-center justify-between hover:bg-slate-800/40 transition-colors",children:[l.jsxs("div",{className:"flex items-center gap-3",children:[t?l.jsx(pe,{className:"w-4 h-4 text-slate-500"}):l.jsx(ge,{className:"w-4 h-4 text-slate-500"}),l.jsxs("code",{className:"text-orange-300",children:[e.tool,".",e.action,"()"]})]}),l.jsx("span",{className:`text-sm ${e.success?"text-emerald-300":"text-red-400"}`,children:e.success?"✓ Success":"✗ Failed"})]}),t&&l.jsxs("div",{className:"px-4 pb-4 space-y-3",children:[l.jsxs("div",{children:[l.jsx("div",{className:"text-xs font-medium text-slate-500 mb-1",children:"Arguments"}),l.jsx("pre",{className:"text-xs text-slate-400 overflow-auto bg-slate-950/50 rounded p-2",children:uu(e.args)})]}),s&&l.jsxs("div",{children:[l.jsx("div",{className:"text-xs font-medium text-emerald-400 mb-1",children:"Result"}),l.jsx("pre",{className:"text-xs text-slate-400 overflow-auto bg-emerald-950/30 rounded p-2 max-h-48",children:uu(e.result)})]}),r&&l.jsxs("div",{children:[l.jsx("div",{className:"text-xs font-medium text-red-400 mb-1",children:"Error"}),l.jsx("pre",{className:"text-xs text-red-300 overflow-auto bg-red-950/30 rounded p-2",children:e.error})]})]})]})}function Qg(e){var o;const t=e.ranking||[],n=[];for(const c of t){const u=(o=e.stats)==null?void 0:o[c];if(!u)continue;const m=u.avg_score||0,f=u.avg_cost_usd||0,g=u.avg_latency_ms||0,y=f>0?m/(f*100):m>0?1/0:0;n.push({modelId:c,score:m,cost:f,latency:g,efficiency:y,badges:[]})}if(n.length===0)return n;const s=n.reduce((c,u)=>c.score>u.score?c:u),r=n.reduce((c,u)=>c.efficiency>u.efficiency?c:u),a=n.filter(c=>c.score>=s.score*.7),i=a.length>0?a.reduce((c,u)=>c.latency<u.latency?c:u):null;for(const c of n)c.modelId===s.modelId&&c.badges.push("quality"),c.modelId===r.modelId&&r.cost>0&&c.badges.push("value"),i&&c.modelId===i.modelId&&c.badges.push("fast");return n}function qg({type:e}){switch(e){case"quality":return l.jsxs("span",{className:"inline-flex items-center gap-1 px-2 py-0.5 rounded-full text-xs font-medium bg-amber-500/20 text-amber-300 border border-amber-500/30",children:[l.jsx(Lo,{className:"w-3 h-3"})," Best Quality"]});case"value":return l.jsxs("span",{className:"inline-flex items-center gap-1 px-2 py-0.5 rounded-full text-xs font-medium bg-emerald-500/20 text-emerald-300 border border-emerald-500/30",children:[l.jsx(To,{className:"w-3 h-3"})," Best Value"]});case"fast":return l.jsxs("span",{className:"inline-flex items-center gap-1 px-2 py-0.5 rounded-full text-xs font-medium bg-blue-500/20 text-blue-300 border border-blue-500/30",children:[l.jsx(bn,{className:"w-3 h-3"})," Fastest"]});default:return null}}function Jg({run:e,onClose:t}){return l.jsx("div",{className:"fixed inset-0 z-50 flex items-center justify-center p-4 bg-black/70",onClick:t,children:l.jsxs("div",{className:"bg-slate-900 rounded-xl max-w-4xl w-full max-h-[90vh] overflow-hidden flex flex-col",onClick:n=>n.stopPropagation(),children:[l.jsxs("div",{className:"flex items-center justify-between p-4 border-b border-slate-800",children:[l.jsxs("div",{children:[l.jsx("h2",{className:"text-lg font-semibold text-slate-100",children:"Run Details"}),l.jsx("p",{className:"text-sm text-slate-400",children:e.model})]}),l.jsx("button",{onClick:t,className:"p-2 text-slate-400 hover:text-slate-100 hover:bg-slate-800 rounded-lg",children:l.jsx(Nt,{className:"w-5 h-5"})})]}),l.jsx("div",{className:"overflow-auto p-4",children:l.jsx(Mo,{result:e})})]})})}function Np({comparison:e}){var m,f,g,y,j,v,N,p;const[t,n]=x.useState(null),[s,r]=x.useState(null),a=e.ranking||[],i=Qg(e),o=new Map(i.map(d=>[d.modelId,d])),c=new Map;if(e.results)for(const d of e.results){const h=c.get(d.model)||[];h.push(d),c.set(d.model,h)}const u=e.results&&e.results.length>0;return l.jsxs(l.Fragment,{children:[s&&l.jsx(Jg,{run:s,onClose:()=>r(null)}),l.jsxs("div",{className:"space-y-6",children:[i.length>0&&l.jsxs("div",{className:"grid grid-cols-1 md:grid-cols-3 gap-4",children:[i.find(d=>d.badges.includes("quality"))&&l.jsxs("div",{className:"panel-card p-4 border-l-4 border-l-amber-400",children:[l.jsxs("div",{className:"flex items-center gap-2 text-amber-300 mb-1",children:[l.jsx(Lo,{className:"w-4 h-4"}),l.jsx("span",{className:"text-sm font-medium",children:"Best Quality"})]}),l.jsx("div",{className:"text-lg font-semibold text-slate-100",children:((m=i.find(d=>d.badges.includes("quality")))==null?void 0:m.modelId.split("/")[1])||((f=i.find(d=>d.badges.includes("quality")))==null?void 0:f.modelId)}),l.jsxs("div",{className:"text-xs text-slate-400 mt-1",children:["Highest score: ",(g=i.find(d=>d.badges.includes("quality")))==null?void 0:g.score.toFixed(1)," pts"]})]}),i.find(d=>d.badges.includes("value"))&&l.jsxs("div",{className:"panel-card p-4 border-l-4 border-l-emerald-400",children:[l.jsxs("div",{className:"flex items-center gap-2 text-emerald-300 mb-1",children:[l.jsx(To,{className:"w-4 h-4"}),l.jsx("span",{className:"text-sm font-medium",children:"Best Value"})]}),l.jsx("div",{className:"text-lg font-semibold text-slate-100",children:((y=i.find(d=>d.badges.includes("value")))==null?void 0:y.modelId.split("/")[1])||((j=i.find(d=>d.badges.includes("value")))==null?void 0:j.modelId)}),l.jsx("div",{className:"text-xs text-slate-400 mt-1",children:(()=>{const d=i.find(R=>R.badges.includes("value")),h=i.find(R=>R.badges.includes("quality"));if(!d||!h)return"";if(d.cost===0)return"Free!";if(d.modelId===h.modelId)return"Also the best quality!";const k=Math.round(d.score/h.score*100),w=h.cost>0?Math.round(d.cost/h.cost*100):0;return w===0?`${k}% quality, nearly free`:`${k}% quality at ${w}% cost`})()})]}),i.find(d=>d.badges.includes("fast"))&&l.jsxs("div",{className:"panel-card p-4 border-l-4 border-l-blue-400",children:[l.jsxs("div",{className:"flex items-center gap-2 text-blue-300 mb-1",children:[l.jsx(bn,{className:"w-4 h-4"}),l.jsx("span",{className:"text-sm font-medium",children:"Fastest"})]}),l.jsx("div",{className:"text-lg font-semibold text-slate-100",children:((v=i.find(d=>d.badges.includes("fast")))==null?void 0:v.modelId.split("/")[1])||((N=i.find(d=>d.badges.includes("fast")))==null?void 0:N.modelId)}),l.jsxs("div",{className:"text-xs text-slate-400 mt-1",children:[((((p=i.find(d=>d.badges.includes("fast")))==null?void 0:p.latency)||0)/1e3).toFixed(1),"s avg latency"]})]})]}),l.jsxs("div",{className:"panel-card overflow-hidden",children:[l.jsxs("div",{className:"p-4 border-b border-slate-800/70",children:[l.jsx("h3",{className:"text-lg font-semibold text-slate-100",children:"Model Comparison"}),l.jsxs("p",{className:"text-sm text-slate-400",children:[e.runs_per_model," run(s) per model"]})]}),l.jsx("div",{className:"overflow-x-auto",children:l.jsxs("table",{className:"w-full",children:[l.jsx("thead",{className:"bg-slate-950/70",children:l.jsxs("tr",{children:[l.jsx("th",{className:"px-4 py-3 text-left text-sm font-medium text-slate-400",children:"Rank"}),l.jsx("th",{className:"px-4 py-3 text-left text-sm font-medium text-slate-400",children:"Model"}),l.jsx("th",{className:"px-4 py-3 text-right text-sm font-medium text-slate-400",children:"Score"}),l.jsx("th",{className:"px-4 py-3 text-right text-sm font-medium text-slate-400",children:"Judge"}),l.jsx("th",{className:"px-4 py-3 text-right text-sm font-medium text-slate-400",children:"Latency"}),l.jsx("th",{className:"px-4 py-3 text-right text-sm font-medium text-slate-400",children:"Tokens"}),l.jsx("th",{className:"px-4 py-3 text-right text-sm font-medium text-slate-400",children:"Cost"})]})}),l.jsx("tbody",{children:a.map((d,h)=>{var M,O;const k=(M=e.stats)==null?void 0:M[d],w=o.get(d),R=c.get(d)||[],P=t===d;return k?l.jsxs(l.Fragment,{children:[l.jsxs("tr",{className:`border-t border-slate-800/70 hover:bg-slate-900/40 ${u?"cursor-pointer":""}`,onClick:()=>u&&n(P?null:d),children:[l.jsx("td",{className:"px-4 py-3",children:l.jsxs("div",{className:"flex items-center gap-2",children:[u&&(P?l.jsx(pe,{className:"w-4 h-4 text-slate-500"}):l.jsx(ge,{className:"w-4 h-4 text-slate-500"})),l.jsxs("span",{className:`font-bold ${h===0?"text-amber-300":"text-slate-400"}`,children:["#",h+1]})]})}),l.jsxs("td",{className:"px-4 py-3",children:[l.jsx("div",{className:"text-slate-100 font-medium",children:d}),w&&w.badges.length>0&&l.jsx("div",{className:"flex flex-wrap gap-1 mt-1",children:w.badges.map(z=>l.jsx(qg,{type:z},z))})]}),l.jsx("td",{className:"px-4 py-3 text-right text-slate-100 font-semibold",children:((O=k.avg_score)==null?void 0:O.toFixed(1))||"0"}),l.jsx("td",{className:"px-4 py-3 text-right text-slate-400",children:k.avg_judge_score!=null?`${(k.avg_judge_score*100).toFixed(0)}%`:"-"}),l.jsxs("td",{className:"px-4 py-3 text-right text-slate-400",children:[((k.avg_latency_ms||0)/1e3).toFixed(1),"s"]}),l.jsx("td",{className:"px-4 py-3 text-right text-slate-400",children:((k.total_input_tokens||0)+(k.total_output_tokens||0)).toLocaleString()}),l.jsx("td",{className:"px-4 py-3 text-right text-emerald-300 font-medium",children:jp(k.avg_cost_usd)})]},d),P&&R.length>0&&l.jsx("tr",{children:l.jsx("td",{colSpan:7,className:"px-4 py-2 bg-slate-950/50",children:l.jsxs("div",{className:"ml-8 space-y-2",children:[l.jsxs("div",{className:"text-sm font-medium text-slate-400 mb-2",children:["Individual Runs (",R.length,")"]}),R.map((z,A)=>{var L,U,B;return l.jsxs("div",{className:"flex items-center justify-between p-3 rounded-lg bg-slate-900/50 border border-slate-800/50 hover:bg-slate-800/50 cursor-pointer",onClick:C=>{C.stopPropagation(),r(z)},children:[l.jsxs("div",{className:"flex items-center gap-3",children:[l.jsxs("span",{className:"text-slate-500 text-sm",children:["Run ",A+1]}),l.jsxs("span",{className:"text-slate-100 font-medium",children:[((U=(L=z.evaluation)==null?void 0:L.percentage)==null?void 0:U.toFixed(0))||0,"%"]}),((B=z.evaluation)==null?void 0:B.judge)&&l.jsx("span",{className:`text-sm ${z.evaluation.judge.passed?"text-emerald-300":"text-red-400"}`,children:z.evaluation.judge.passed?"✓ Passed":"✗ Failed"}),z.error&&l.jsxs("span",{className:"text-sm text-red-400 flex items-center gap-1",children:[l.jsx(nn,{className:"w-3 h-3"})," Error"]})]}),l.jsxs("button",{className:"flex items-center gap-1 text-sm text-orange-300 hover:text-orange-200",onClick:C=>{C.stopPropagation(),r(z)},children:[l.jsx(xp,{className:"w-4 h-4"}),"View Details"]})]},A)})]})})},`${d}-runs`)]}):null})})]})})]}),e.stats&&Object.values(e.stats).some(d=>d.goal_rates&&Object.keys(d.goal_rates).length>0)&&l.jsxs("div",{className:"panel-card p-6",children:[l.jsx("h3",{className:"text-lg font-semibold text-slate-100 mb-4",children:"Goal Achievement Rates"}),(()=>{const d=new Set;return Object.values(e.stats).forEach(h=>{h.goal_rates&&Object.keys(h.goal_rates).forEach(k=>d.add(k))}),l.jsx("div",{className:"overflow-x-auto",children:l.jsxs("table",{className:"w-full",children:[l.jsx("thead",{className:"bg-slate-950/70",children:l.jsxs("tr",{children:[l.jsx("th",{className:"px-4 py-3 text-left text-sm font-medium text-slate-400",children:"Model"}),Array.from(d).map(h=>l.jsx("th",{className:"px-4 py-3 text-right text-sm font-medium text-slate-400",children:h},h))]})}),l.jsx("tbody",{children:a.map(h=>{var w;const k=(w=e.stats)==null?void 0:w[h];return k?l.jsxs("tr",{className:"border-t border-slate-800/70",children:[l.jsx("td",{className:"px-4 py-3 text-slate-100 font-medium",children:h}),Array.from(d).map(R=>{var M;const P=((M=k.goal_rates)==null?void 0:M[R])??0;return l.jsx("td",{className:"px-4 py-3 text-right",children:l.jsxs("span",{className:P>=100?"text-emerald-300":P>=50?"text-amber-300":"text-red-400",children:[P.toFixed(0),"%"]})},R)})]},h):null})})]})})})()]})]})]})}function As(){return l.jsxs("span",{className:"inline-flex items-center gap-1 px-1.5 py-0.5 bg-emerald-500/20 border border-emerald-500/40 rounded text-xs text-emerald-400",children:[l.jsx(gp,{size:10}),"Local"]})}function du({models:e,value:t,onChange:n,disabled:s,placeholder:r="Select a model..."}){const[a,i]=x.useState(!1),[o,c]=x.useState(""),u=x.useRef(null),m=x.useRef(null),f=e.find(p=>p.id===t);x.useEffect(()=>{const p=d=>{u.current&&!u.current.contains(d.target)&&i(!1)};return document.addEventListener("mousedown",p),()=>document.removeEventListener("mousedown",p)},[]),x.useEffect(()=>{a&&m.current&&m.current.focus()},[a]);const g=e.filter(p=>p.name.toLowerCase().includes(o.toLowerCase())||p.id.toLowerCase().includes(o.toLowerCase())),y=g.reduce((p,d)=>{const h=d.provider_name||d.id.split("/")[0]||"other";return p[h]||(p[h]=[]),p[h].push(d),p},{}),j=["openai","anthropic","google","x-ai","deepseek","meta-llama","mistralai","qwen","perplexity"],v=p=>{var d;return(d=y[p])==null?void 0:d.some(h=>h.is_local)},N=Object.keys(y).sort((p,d)=>{const h=v(p),k=v(d);if(h&&!k)return-1;if(!h&&k)return 1;const w=j.indexOf(p),R=j.indexOf(d);return w===-1&&R===-1?p.localeCompare(d):w===-1?1:R===-1?-1:w-R});return l.jsxs("div",{ref:u,className:"relative",children:[l.jsxs("button",{type:"button",onClick:()=>!s&&i(!a),disabled:s,className:`w-full flex items-center justify-between gap-2 px-4 py-2.5 panel-subtle text-left transition-colors ${s?"opacity-50 cursor-not-allowed":"hover:bg-slate-800 cursor-pointer"} ${a?"ring-2 ring-orange-400":""}`,children:[f?l.jsxs("div",{className:"flex items-center justify-between flex-1 min-w-0 gap-2",children:[l.jsx("span",{className:"text-slate-100 truncate",children:f.name}),f.is_local&&l.jsx(As,{}),l.jsx("span",{className:"text-xs text-slate-500 shrink-0",children:f.price})]}):l.jsx("span",{className:"text-slate-500",children:r}),l.jsx(pe,{size:18,className:`text-slate-400 shrink-0 transition-transform ${a?"rotate-180":""}`})]}),a&&l.jsxs("div",{className:"absolute z-50 w-full mt-1 panel-card border border-slate-700/80 dark:border-slate-700/80 shadow-xl max-h-[28rem] overflow-hidden flex flex-col",children:[l.jsx("div",{className:"p-2 border-b border-slate-800/70",children:l.jsxs("div",{className:"relative",children:[l.jsx(Po,{size:16,className:"absolute left-3 top-1/2 -translate-y-1/2 text-slate-500"}),l.jsx("input",{ref:m,type:"text",value:o,onChange:p=>c(p.target.value),placeholder:"Search models...",className:"w-full pl-9 pr-3 py-2 bg-slate-900 border border-slate-800/70 rounded text-sm text-slate-100 placeholder-slate-500 focus:outline-none focus:border-orange-400/60"})]})}),l.jsxs("div",{className:"overflow-y-auto flex-1",children:[N.map(p=>l.jsxs("div",{children:[l.jsxs("div",{className:"px-3 py-1.5 text-xs font-medium text-slate-500 uppercase bg-slate-900 sticky top-0 flex items-center gap-2",children:[p,v(p)&&l.jsx(As,{})]}),y[p].map(d=>l.jsxs("button",{type:"button",onClick:()=>{n(d.id),i(!1),c("")},className:`w-full flex items-center gap-3 px-3 py-2 text-left transition-colors ${d.id===t?"bg-orange-500/20 text-orange-300":"hover:bg-slate-800 text-slate-100"}`,children:[l.jsxs("div",{className:"flex-1 min-w-0 flex items-center gap-2",children:[l.jsx("div",{className:"truncate",children:d.name}),d.is_local&&!v(p)&&l.jsx(As,{})]}),l.jsx("span",{className:"text-xs text-slate-500 shrink-0",children:d.price}),d.id===t&&l.jsx(Xs,{size:16,className:"text-orange-400 shrink-0"})]},d.id))]},p)),g.length===0&&l.jsx("div",{className:"px-3 py-4 text-center text-slate-500 text-sm",children:"No models found"})]})]})]})}function Kg({models:e,selected:t,onChange:n,disabled:s}){const[r,a]=x.useState(!1),[i,o]=x.useState(""),c=x.useRef(null),u=x.useRef(null);x.useEffect(()=>{const p=d=>{c.current&&!c.current.contains(d.target)&&a(!1)};return document.addEventListener("mousedown",p),()=>document.removeEventListener("mousedown",p)},[]),x.useEffect(()=>{r&&u.current&&u.current.focus()},[r]);const m=p=>{t.includes(p)?n(t.filter(d=>d!==p)):n([...t,p])},f=(p,d)=>{d.stopPropagation(),n(t.filter(h=>h!==p))},g=e.filter(p=>p.name.toLowerCase().includes(i.toLowerCase())||p.id.toLowerCase().includes(i.toLowerCase())),y=g.reduce((p,d)=>{const h=d.provider_name||d.id.split("/")[0]||"other";return p[h]||(p[h]=[]),p[h].push(d),p},{}),j=["openai","anthropic","google","x-ai","deepseek","meta-llama","mistralai","qwen","perplexity"],v=p=>{var d;return(d=y[p])==null?void 0:d.some(h=>h.is_local)},N=Object.keys(y).sort((p,d)=>{const h=v(p),k=v(d);if(h&&!k)return-1;if(!h&&k)return 1;const w=j.indexOf(p),R=j.indexOf(d);return w===-1&&R===-1?p.localeCompare(d):w===-1?1:R===-1?-1:w-R});return l.jsxs("div",{ref:c,className:"relative",children:[t.length>0&&l.jsx("div",{className:"flex flex-wrap gap-2 mb-2",children:t.map(p=>{const d=e.find(h=>h.id===p);return l.jsxs("span",{className:`flex items-center gap-1.5 px-2.5 py-1 rounded-full text-sm text-slate-100 ${d!=null&&d.is_local?"bg-emerald-500/20 border border-emerald-400/40":"bg-orange-500/20 border border-orange-400/40"}`,children:[(d==null?void 0:d.is_local)&&l.jsx(gp,{size:12,className:"text-emerald-400"}),(d==null?void 0:d.name)||p,l.jsx("button",{type:"button",onClick:h=>f(p,h),className:"hover:text-red-400 transition-colors",disabled:s,children:l.jsx(Nt,{size:14})})]},p)})}),l.jsxs("button",{type:"button",onClick:()=>!s&&a(!r),disabled:s,className:`w-full flex items-center justify-between gap-2 px-4 py-2.5 panel-subtle text-left transition-colors ${s?"opacity-50 cursor-not-allowed":"hover:bg-slate-800 cursor-pointer"} ${r?"ring-2 ring-orange-400":""}`,children:[l.jsx("span",{className:"text-slate-500",children:t.length===0?"Click to add models...":"Add more models..."}),l.jsx(pe,{size:18,className:`text-slate-400 shrink-0 transition-transform ${r?"rotate-180":""}`})]}),r&&l.jsxs("div",{className:"absolute z-50 w-full mt-1 panel-card border border-slate-700/80 dark:border-slate-700/80 shadow-xl max-h-[28rem] overflow-hidden flex flex-col",children:[l.jsx("div",{className:"p-2 border-b border-slate-800/70",children:l.jsxs("div",{className:"relative",children:[l.jsx(Po,{size:16,className:"absolute left-3 top-1/2 -translate-y-1/2 text-slate-500"}),l.jsx("input",{ref:u,type:"text",value:i,onChange:p=>o(p.target.value),placeholder:"Search models...",className:"w-full pl-9 pr-3 py-2 bg-slate-900 border border-slate-800/70 rounded text-sm text-slate-100 placeholder-slate-500 focus:outline-none focus:border-orange-400/60"})]})}),l.jsxs("div",{className:"overflow-y-auto flex-1",children:[N.map(p=>l.jsxs("div",{children:[l.jsxs("div",{className:"px-3 py-1.5 text-xs font-medium text-slate-500 uppercase bg-slate-900 sticky top-0 flex items-center gap-2",children:[p,v(p)&&l.jsx(As,{})]}),y[p].map(d=>{const h=t.includes(d.id);return l.jsxs("button",{type:"button",onClick:()=>m(d.id),className:`w-full flex items-center gap-3 px-3 py-2 text-left transition-colors ${h?"bg-orange-500/20 text-orange-300":"hover:bg-slate-800 text-slate-100"}`,children:[l.jsx("div",{className:`w-4 h-4 rounded border flex items-center justify-center shrink-0 ${h?"bg-orange-400 border-orange-400":"border-slate-600"}`,children:h&&l.jsx(Xs,{size:12,className:"text-slate-900"})}),l.jsxs("div",{className:"flex-1 min-w-0 flex items-center gap-2",children:[l.jsx("div",{className:"truncate",children:d.name}),d.is_local&&!v(p)&&l.jsx(As,{})]}),l.jsx("span",{className:"text-xs text-slate-500 shrink-0",children:d.price})]},d.id)})]},p)),g.length===0&&l.jsx("div",{className:"px-3 py-4 text-center text-slate-500 text-sm",children:"No models found"})]})]})]})}function fu(){var Ho;const{scenarioId:e}=rp(),[t]=yg(),n=t.get("dataset"),s=parseInt(t.get("parallel")||"5"),[r,a]=x.useState([]),[i,o]=x.useState(null),[c,u]=x.useState(e||""),[m,f]=x.useState([]),[g,y]=x.useState(""),[j,v]=x.useState([]),[N,p]=x.useState(n?"dataset":"single"),[d,h]=x.useState(1),[k,w]=x.useState({}),[R,P]=x.useState(!0),[M,O]=x.useState(null),[z,A]=x.useState([]),[L,U]=x.useState(n||""),[B,C]=x.useState(null),[E,_]=x.useState(!1),[b,S]=x.useState(s),[T,D]=x.useState(!1),[H,te]=x.useState(!1),[X,lt]=x.useState(""),[Fe,rt]=x.useState(""),[Cn,pl]=x.useState(!0),{state:ae,result:Ot,comparison:Uo,error:Bo,runScenario:dm,compareModels:fm}=Wg();x.useEffect(()=>{(async()=>{try{const[Oe,He,xm]=await Promise.all([Z.listScenarios(),Z.listModels(),Z.listDatasets()]);if(a(Oe),f(He),A(xm),He.length>0&&y(He[0].id),e){u(e);const Vo=await Z.getScenario(e);o(Vo);const Wo={};for(const Yo of Vo.variables||[])Wo[Yo.name]=Yo.default??"";w(Wo)}}catch(Oe){O(Oe instanceof Error?Oe.message:"Failed to load data")}finally{P(!1)}})()},[e]),x.useEffect(()=>{c&&c!==e&&Z.getScenario(c).then(V=>{o(V);const Oe={};for(const He of V.variables||[])Oe[He.name]=He.default??"";w(Oe)}).catch(V=>O(V.message))},[c,e]);const pm=async()=>{const V=c||e;if(!V)return;const Oe=T?{enabled:!0,trackingUri:X||void 0,experiment:Fe||void 0,tracing:Cn}:void 0;if(N==="dataset"){if(!L||!g)return;_(!0),C(null);try{const He=await Z.runDataset({scenario_id:V,dataset_id:L,model:g,parallel:b,mlflow_enabled:T,mlflow_experiment:Fe||void 0});C(He)}catch(He){O(He instanceof Error?He.message:"Dataset run failed")}finally{_(!1)}}else if(N==="single"){if(!g)return;await dm(V,g,k,Oe)}else{if(j.length===0)return;await fm(V,j,d,k,Oe)}},mm=(V,Oe)=>{w(He=>({...He,[V]:Oe}))};return R?l.jsx("div",{className:"flex items-center justify-center h-full",children:l.jsx(cu,{className:"w-8 h-8 animate-spin text-orange-400"})}):M?l.jsx("div",{className:"p-8 page",children:l.jsx("div",{className:"panel-solid p-4 text-red-400 border border-red-700/60",children:M})}):l.jsxs("div",{className:"p-8 max-w-6xl mx-auto page",children:[l.jsxs(de,{to:"/",className:"flex items-center gap-2 text-slate-400 hover:text-slate-100 mb-6",children:[l.jsx(Gs,{size:20}),"Back to Dashboard"]}),l.jsxs("div",{className:"mb-8",children:[l.jsx("h1",{className:"text-2xl font-semibold text-slate-100 mb-2",children:(i==null?void 0:i.name)||e}),(i==null?void 0:i.description)&&l.jsx("p",{className:"text-slate-400",children:i.description})]}),l.jsxs("div",{className:"panel-card p-6 mb-6",children:[l.jsxs("div",{className:"flex items-center justify-between mb-4",children:[l.jsx("h2",{className:"text-lg font-semibold text-slate-100",children:"Run Configuration"}),l.jsxs(de,{to:"/builder",className:"flex items-center gap-2 text-slate-400 hover:text-slate-100 text-sm",children:[l.jsx(Og,{size:16}),"Edit Scenario"]})]}),!e&&l.jsxs("div",{className:"mb-6",children:[l.jsx("label",{className:"block text-sm font-medium text-slate-400 mb-2",children:"Scenario"}),l.jsxs("select",{value:c,onChange:V=>u(V.target.value),disabled:ae==="running"||E,className:"w-full panel-subtle px-4 py-2 text-slate-100 focus:outline-none focus:ring-2 focus:ring-orange-400",children:[l.jsx("option",{value:"",children:"Select a scenario..."}),r.map(V=>l.jsx("option",{value:V.id,children:V.name},V.id))]})]}),l.jsxs("div",{className:"mb-6",children:[l.jsx("label",{className:"block text-sm font-medium text-slate-400 mb-2",children:"Mode"}),l.jsxs("div",{className:"flex gap-2",children:[l.jsx("button",{onClick:()=>p("single"),disabled:ae==="running"||E,className:`px-4 py-2 rounded-lg transition-colors ${N==="single"?"bg-orange-400 text-slate-900":"bg-slate-800 text-slate-300 hover:bg-slate-700"}`,children:"Single Model"}),l.jsx("button",{onClick:()=>p("compare"),disabled:ae==="running"||E,className:`px-4 py-2 rounded-lg transition-colors ${N==="compare"?"bg-orange-400 text-slate-900":"bg-slate-800 text-slate-300 hover:bg-slate-700"}`,children:"Compare Models"}),l.jsxs("button",{onClick:()=>p("dataset"),disabled:ae==="running"||E||z.length===0,className:`flex items-center gap-2 px-4 py-2 rounded-lg transition-colors ${N==="dataset"?"bg-blue-500 text-white":"bg-slate-800 text-slate-300 hover:bg-slate-700"}`,children:[l.jsx(sn,{size:16}),"Dataset Benchmark"]})]})]}),N==="single"?l.jsxs("div",{className:"mb-6",children:[l.jsx("label",{className:"block text-sm font-medium text-slate-400 mb-2",children:"Model"}),l.jsx(du,{models:m,value:g,onChange:y,disabled:ae==="running"})]}):N==="compare"?l.jsxs("div",{className:"space-y-4 mb-6",children:[l.jsxs("div",{children:[l.jsx("label",{className:"block text-sm font-medium text-slate-400 mb-2",children:"Models to Compare"}),l.jsx(Kg,{models:m,selected:j,onChange:v,disabled:ae==="running"})]}),l.jsxs("div",{children:[l.jsx("label",{className:"block text-sm font-medium text-slate-400 mb-2",children:"Runs per Model"}),l.jsx("input",{type:"number",min:1,max:10,value:d,onChange:V=>h(Math.max(1,Math.min(10,parseInt(V.target.value)||1))),disabled:ae==="running",className:"w-32 panel-subtle px-4 py-2 text-slate-100 focus:outline-none focus:ring-2 focus:ring-orange-400"}),l.jsx("p",{className:"mt-1 text-xs text-slate-500",children:"More runs = more statistical significance"})]})]}):l.jsxs("div",{className:"space-y-4 mb-6",children:[l.jsxs("div",{children:[l.jsx("label",{className:"block text-sm font-medium text-slate-400 mb-2",children:"Dataset"}),l.jsxs("select",{value:L,onChange:V=>U(V.target.value),disabled:E,className:"w-full panel-subtle px-4 py-2 text-slate-100 focus:outline-none focus:ring-2 focus:ring-blue-400",children:[l.jsx("option",{value:"",children:"Select a dataset..."}),z.map(V=>l.jsxs("option",{value:V.id,children:[V.name," (",V.case_count," cases)"]},V.id))]}),z.length===0&&l.jsxs("p",{className:"mt-1 text-xs text-slate-500",children:["No datasets found. ",l.jsx(de,{to:"/datasets",className:"text-blue-400 hover:underline",children:"Create one"})]})]}),l.jsxs("div",{children:[l.jsx("label",{className:"block text-sm font-medium text-slate-400 mb-2",children:"Model"}),l.jsx(du,{models:m,value:g,onChange:y,disabled:E})]}),l.jsxs("div",{children:[l.jsx("label",{className:"block text-sm font-medium text-slate-400 mb-2",children:"Parallel Runs"}),l.jsxs("div",{className:"flex items-center gap-3",children:[l.jsx("input",{type:"number",min:1,max:20,value:b,onChange:V=>S(Math.max(1,Math.min(20,parseInt(V.target.value)||1))),disabled:E,className:"w-20 panel-subtle px-3 py-2 text-slate-100 focus:outline-none focus:ring-2 focus:ring-blue-400"}),l.jsx("span",{className:"text-sm text-slate-500",children:"concurrent cases"})]}),l.jsx("p",{className:"mt-1 text-xs text-slate-500",children:"Higher = faster, but may hit rate limits"})]})]}),i&&i.variables&&i.variables.length>0&&l.jsxs("div",{className:"mb-6 p-4 panel-subtle",children:[l.jsxs("div",{className:"flex items-center gap-2 mb-4",children:[l.jsx(yp,{size:18,className:"text-slate-400"}),l.jsx("h3",{className:"font-medium text-slate-100",children:"Variables"})]}),l.jsx("div",{className:"grid grid-cols-1 md:grid-cols-2 gap-4",children:i.variables.map(V=>l.jsx(Xg,{variable:V,value:k[V.name],onChange:Oe=>mm(V.name,Oe),disabled:ae==="running"},V.name))})]}),l.jsxs("div",{className:"mb-6 p-4 panel-subtle",children:[l.jsxs("button",{onClick:()=>te(!H),className:"flex items-center gap-2 w-full text-left",children:[H?l.jsx(pe,{size:18}):l.jsx(ge,{size:18}),l.jsx("span",{className:"font-medium text-slate-100",children:"MLflow Tracking"}),T&&l.jsx("span",{className:"ml-2 px-2 py-0.5 text-xs bg-green-500/20 text-green-400 rounded",children:"Enabled"})]}),H&&l.jsxs("div",{className:"mt-4 space-y-4",children:[l.jsxs("label",{className:"flex items-center gap-3 cursor-pointer",children:[l.jsx("input",{type:"checkbox",checked:T,onChange:V=>D(V.target.checked),disabled:ae==="running"||E,className:"w-4 h-4 rounded border-slate-600 text-orange-400 focus:ring-orange-400"}),l.jsx("span",{className:"text-slate-200",children:"Enable MLflow tracking"})]}),T&&l.jsxs(l.Fragment,{children:[N!=="dataset"&&l.jsxs("div",{children:[l.jsx("label",{className:"block text-sm font-medium text-slate-400 mb-1",children:"Tracking URI"}),l.jsx("input",{type:"text",value:X,onChange:V=>lt(V.target.value),disabled:ae==="running",placeholder:"http://127.0.0.1:5000 (uses MLFLOW_TRACKING_URI if empty)",className:"w-full panel-subtle px-3 py-2 text-slate-100 text-sm focus:outline-none focus:ring-2 focus:ring-orange-400"})]}),l.jsxs("div",{children:[l.jsx("label",{className:"block text-sm font-medium text-slate-400 mb-1",children:"Experiment Name"}),l.jsx("input",{type:"text",value:Fe,onChange:V=>rt(V.target.value),disabled:ae==="running"||E,placeholder:N==="dataset"?`${(i==null?void 0:i.name)||"scenario"}-dataset`:(i==null?void 0:i.name)||"Defaults to scenario name",className:"w-full panel-subtle px-3 py-2 text-slate-100 text-sm focus:outline-none focus:ring-2 focus:ring-orange-400"})]}),N!=="dataset"&&l.jsxs("label",{className:"flex items-center gap-3 cursor-pointer",children:[l.jsx("input",{type:"checkbox",checked:Cn,onChange:V=>pl(V.target.checked),disabled:ae==="running",className:"w-4 h-4 rounded border-slate-600 text-orange-400 focus:ring-orange-400"}),l.jsxs("div",{children:[l.jsx("span",{className:"text-slate-200",children:"Enable LLM Tracing"}),l.jsx("p",{className:"text-xs text-slate-500",children:"Capture detailed traces of each LLM call"})]})]}),l.jsxs("p",{className:"text-xs text-slate-500 flex items-center gap-1",children:[l.jsx(_g,{size:12}),"View results at your MLflow server after the run completes"]})]})]})]}),l.jsx("button",{onClick:pm,disabled:ae==="running"||E||!c||(N==="single"?!g:N==="compare"?j.length===0:!L||!g),className:`flex items-center gap-2 ${N==="dataset"?"bg-blue-500 hover:bg-blue-400":"bg-orange-400 hover:bg-orange-300"} disabled:bg-slate-700 disabled:text-slate-400 disabled:cursor-not-allowed text-slate-900 px-6 py-2 rounded-lg transition-colors font-semibold`,children:ae==="running"||E?l.jsxs(l.Fragment,{children:[l.jsx(cu,{className:"w-4 h-4 animate-spin"}),"Running..."]}):l.jsxs(l.Fragment,{children:[N==="dataset"?l.jsx(sn,{className:"w-4 h-4"}):l.jsx(Xt,{className:"w-4 h-4"}),N==="single"?"Run Scenario":N==="compare"?`Compare ${j.length} Models`:`Run ${((Ho=z.find(V=>V.id===L))==null?void 0:Ho.case_count)||0} Cases`]})})]}),Bo&&l.jsx("div",{className:"panel-solid border border-red-700/60 p-4 mb-6 text-red-400",children:l.jsxs("div",{className:"flex items-center gap-2",children:[l.jsx(Eo,{className:"w-5 h-5"}),l.jsx("span",{children:Bo})]})}),Ot&&l.jsx(Mo,{result:Ot}),Uo&&l.jsx(Np,{comparison:Uo}),B&&l.jsx(Gg,{result:B})]})}function Gg({result:e}){const[t,n]=x.useState(new Set),s=r=>{const a=new Set(t);a.has(r)?a.delete(r):a.add(r),n(a)};return l.jsxs("div",{className:"panel-card p-6",children:[l.jsxs("h2",{className:"text-xl font-semibold text-slate-100 mb-4 flex items-center gap-2",children:[l.jsx(sn,{size:20}),"Dataset Benchmark Results"]}),l.jsxs("div",{className:"grid grid-cols-2 md:grid-cols-4 gap-4 mb-6",children:[l.jsxs("div",{className:"panel-subtle p-4 rounded-lg",children:[l.jsxs("div",{className:"text-2xl font-bold text-slate-100",children:[e.passed_cases,"/",e.total_cases]}),l.jsx("div",{className:"text-sm text-slate-400",children:"Cases Passed"})]}),l.jsxs("div",{className:"panel-subtle p-4 rounded-lg",children:[l.jsxs("div",{className:`text-2xl font-bold ${e.pass_rate>=.9?"text-green-400":e.pass_rate>=.7?"text-yellow-400":"text-red-400"}`,children:[(e.pass_rate*100).toFixed(1),"%"]}),l.jsx("div",{className:"text-sm text-slate-400",children:"Pass Rate"})]}),l.jsxs("div",{className:"panel-subtle p-4 rounded-lg",children:[l.jsxs("div",{className:"text-2xl font-bold text-slate-100",children:[e.avg_percentage.toFixed(1),"%"]}),l.jsx("div",{className:"text-sm text-slate-400",children:"Avg Score"})]}),l.jsxs("div",{className:"panel-subtle p-4 rounded-lg",children:[l.jsxs("div",{className:"text-2xl font-bold text-slate-100",children:[(e.total_time_ms/1e3).toFixed(1),"s"]}),l.jsx("div",{className:"text-sm text-slate-400",children:"Total Time"})]})]}),Object.keys(e.by_expected).length>0&&l.jsxs("div",{className:"mb-6",children:[l.jsx("h3",{className:"font-medium text-slate-300 mb-3",children:"By Expected Outcome"}),l.jsx("div",{className:"grid grid-cols-2 md:grid-cols-3 gap-2",children:Object.entries(e.by_expected).map(([r,a])=>{const i=a.passed+a.failed,o=i>0?a.passed/i:0;return l.jsxs("div",{className:"panel-subtle p-3 rounded",children:[l.jsx("div",{className:"font-medium text-slate-200",children:r}),l.jsxs("div",{className:"text-sm text-slate-400",children:[a.passed,"/",i," (",(o*100).toFixed(0),"%)"]})]},r)})})]}),l.jsxs("div",{children:[l.jsxs("h3",{className:"font-medium text-slate-300 mb-3",children:["Case Results (",e.failed_cases," failed)"]}),l.jsx("div",{className:"space-y-2 max-h-[40vh] overflow-y-auto",children:e.case_results.map(r=>l.jsxs("div",{className:`panel-subtle rounded-lg overflow-hidden ${r.passed?"":"border border-red-500/30"}`,children:[l.jsxs("button",{onClick:()=>s(r.case_id),className:"w-full px-4 py-3 flex items-center gap-3 text-left hover:bg-slate-700/50 transition-colors",children:[r.passed?l.jsx(Xs,{size:16,className:"text-green-400"}):l.jsx(Nt,{size:16,className:"text-red-400"}),l.jsx("span",{className:"font-medium text-slate-200",children:r.case_id}),r.expected&&l.jsxs("span",{className:"text-sm text-slate-400",children:["expected: ",r.expected]}),l.jsxs("span",{className:"ml-auto text-sm text-slate-400",children:[r.percentage.toFixed(0),"% | ",r.latency_ms,"ms"]})]}),t.has(r.case_id)&&l.jsxs("div",{className:"px-4 pb-3 pt-1 border-t border-slate-700",children:[r.failure_reason&&l.jsx("div",{className:"text-red-400 text-sm mb-2",children:r.failure_reason}),l.jsxs("div",{className:"grid grid-cols-2 gap-4 text-sm",children:[l.jsxs("div",{children:[l.jsx("span",{className:"text-slate-400",children:"Expected:"})," ",l.jsx("span",{className:"text-slate-200",children:r.expected||"N/A"})]}),l.jsxs("div",{children:[l.jsx("span",{className:"text-slate-400",children:"Actual:"})," ",l.jsx("span",{className:"text-slate-200",children:r.actual_outcome||"None"})]}),l.jsxs("div",{children:[l.jsx("span",{className:"text-slate-400",children:"Score:"})," ",l.jsxs("span",{className:"text-slate-200",children:[r.goal_score,"/",r.max_score]})]}),l.jsxs("div",{children:[l.jsx("span",{className:"text-slate-400",children:"Latency:"})," ",l.jsxs("span",{className:"text-slate-200",children:[r.latency_ms,"ms"]})]})]})]})]},r.case_id))})]})]})}function Xg({variable:e,value:t,onChange:n,disabled:s}){const r="w-full panel-subtle px-3 py-2 text-slate-100 text-sm focus:outline-none focus:ring-2 focus:ring-orange-400 disabled:opacity-50";return l.jsxs("div",{children:[l.jsxs("label",{className:"block text-sm font-medium text-slate-400 mb-1",children:[e.label,e.required&&l.jsx("span",{className:"text-red-400 ml-1",children:"*"})]}),e.type==="select"&&e.options.length>0?l.jsxs("select",{value:String(t??""),onChange:a=>n(a.target.value),disabled:s,className:r,children:[l.jsx("option",{value:"",children:"Select..."}),e.options.map(a=>l.jsx("option",{value:a,children:a},a))]}):e.type==="boolean"?l.jsxs("select",{value:String(t??""),onChange:a=>n(a.target.value==="true"),disabled:s,className:r,children:[l.jsx("option",{value:"",children:"Select..."}),l.jsx("option",{value:"true",children:"True"}),l.jsx("option",{value:"false",children:"False"})]}):e.type==="number"?l.jsx("input",{type:"number",value:String(t??""),onChange:a=>n(a.target.value?Number(a.target.value):""),disabled:s,placeholder:e.default!==null?`Default: ${e.default}`:"",className:r}):l.jsx("input",{type:"text",value:String(t??""),onChange:a=>n(a.target.value),disabled:s,placeholder:e.default!==null?`Default: ${e.default}`:"",className:r})]})}const pu=e=>e.scenario_id.endsWith("_comparison")||e.filename.includes("_comparison_"),mu=e=>e.scenario_id.includes("_dataset_")||e.filename.includes("_dataset_"),Al=e=>e.replace(/_comparison$/,"").replace(/_dataset_.*$/,"");function Zg(){const[e,t]=x.useState([]),[n,s]=x.useState(!0),[r,a]=x.useState(null),[i,o]=x.useState(null),[c,u]=x.useState(null),[m,f]=x.useState(!1),[g,y]=x.useState(!1),[j,v]=x.useState(""),[N,p]=x.useState("all"),[d,h]=x.useState(new Set),k=C=>{h(E=>{const _=new Set(E);return _.has(C)?_.delete(C):_.add(C),_})},{filteredResults:w,groupedResults:R,scenarioOrder:P}=x.useMemo(()=>{let C=e.filter(b=>{if(j){const S=j.toLowerCase();if(!b.scenario_id.toLowerCase().includes(S))return!1}if(N!=="all"){const S=pu(b),T=mu(b);if(N==="comparison"&&!S||N==="dataset"&&!T||N==="single"&&(S||T))return!1}return!0});C=C.sort((b,S)=>new Date(S.timestamp).getTime()-new Date(b.timestamp).getTime());const E={},_=[];for(const b of C){const S=Al(b.scenario_id);E[S]||(E[S]=[],_.push(S)),E[S].push(b)}return{filteredResults:C,groupedResults:E,scenarioOrder:_}},[e,j,N]);x.useEffect(()=>{fetch("/api/v1/local/runs").then(C=>{if(!C.ok)throw new Error("Failed to fetch results");return C.json()}).then(t).catch(C=>a(C.message)).finally(()=>s(!1))},[]);const M=async C=>{try{const E=await fetch(`/api/v1/local/runs/${encodeURIComponent(C)}`);if(!E.ok)throw new Error("Failed to fetch result");const _=await E.json();o(_),u(C),f(!1)}catch(E){console.error("Error fetching result:",E)}},O=async()=>{if(i)try{await navigator.clipboard.writeText(JSON.stringify(i,null,2)),y(!0),setTimeout(()=>y(!1),2e3)}catch(C){console.error("Failed to copy:",C)}},z=C=>{var D,H,te,X,lt;if(!C)return"unknown";const E=C.case_results||((D=C.result)==null?void 0:D.case_results),_=C.dataset_id||((H=C.result)==null?void 0:H.dataset_id);if(E||_)return"dataset";if(C.ranking||((te=C.result)==null?void 0:te.ranking))return"comparison";const S=C.model||((X=C.result)==null?void 0:X.model),T=C.response||((lt=C.result)==null?void 0:lt.response);return S||T!==void 0?"single":"unknown"},A=C=>{const E=C.result||C;return{id:E.id||"",scenario_id:E.scenario_id||C.scenario_id||"",model:E.model||"",response:E.response||"",history:E.history||[],tool_calls:E.tool_calls||[],final_state:E.final_state||{},evaluation:E.evaluation||null,latency_ms:E.latency_ms||0,input_tokens:E.input_tokens||0,output_tokens:E.output_tokens||0,cost_usd:E.cost_usd||null,error:E.error||null}},L=C=>{const E=C.result||C;return{scenario_id:E.scenario_id||C.scenario_id||"",scenario_name:E.scenario_name||C.scenario_id||"",models:E.models||[],runs_per_model:E.runs_per_model||1,stats:E.stats||{},ranking:E.ranking||[],winner:E.winner||null,results:E.results||[]}},U=C=>{const E=C.result||C;return{scenario_id:E.scenario_id||C.scenario_id||"",model:E.model||"",dataset_id:E.dataset_id||"",total_cases:E.total_cases||0,passed_cases:E.passed_cases||0,failed_cases:E.failed_cases||0,pass_rate:E.pass_rate||0,avg_score:E.avg_score||0,avg_percentage:E.avg_percentage||0,by_expected:E.by_expected||{},total_time_ms:E.total_time_ms||0,case_results:E.case_results||[]}};if(n)return l.jsx("div",{className:"p-8 page",children:l.jsx("div",{className:"text-slate-400",children:"Loading..."})});if(r)return l.jsx("div",{className:"p-8 page",children:l.jsxs("div",{className:"text-red-400",children:["Error: ",r]})});const B=z(i);return l.jsxs("div",{className:"p-8 page",children:[l.jsx("h1",{className:"text-2xl font-semibold text-slate-100 mb-6",children:"Run Results"}),e.length===0?l.jsxs("div",{className:"panel-card p-6 text-center",children:[l.jsx(Hr,{size:48,className:"mx-auto text-slate-600 mb-4"}),l.jsx("p",{className:"text-slate-400 mb-2",children:"No run results found"}),l.jsx("p",{className:"text-slate-500 text-sm mb-4",children:"Run a scenario to see results here."}),l.jsxs(de,{to:"/",className:"inline-flex items-center gap-2 text-orange-300 hover:text-orange-200",children:[l.jsx(Xt,{size:16}),"Go to Dashboard"]})]}):l.jsxs("div",{className:"grid grid-cols-1 lg:grid-cols-3 gap-6",children:[l.jsxs("div",{className:"lg:col-span-1 space-y-4",children:[l.jsxs("div",{className:"space-y-3",children:[l.jsxs("div",{className:"relative",children:[l.jsx(Po,{size:16,className:"absolute left-3 top-1/2 -translate-y-1/2 text-slate-500"}),l.jsx("input",{type:"text",placeholder:"Search scenarios...",value:j,onChange:C=>v(C.target.value),className:"w-full bg-slate-900/60 border border-slate-700/50 rounded-lg pl-10 pr-4 py-2 text-sm text-slate-100 placeholder-slate-500 focus:outline-none focus:border-orange-400/50"})]}),l.jsxs("div",{className:"flex flex-wrap gap-1 p-1 bg-slate-900/60 rounded-lg border border-slate-700/50",children:[l.jsx("button",{onClick:()=>p("all"),className:`flex-1 min-w-0 px-2 py-1.5 text-xs font-medium rounded-md transition-colors ${N==="all"?"bg-slate-700/80 text-slate-100":"text-slate-400 hover:text-slate-200"}`,children:"All"}),l.jsxs("button",{onClick:()=>p("single"),className:`flex-1 min-w-0 px-2 py-1.5 text-xs font-medium rounded-md transition-colors flex items-center justify-center gap-1 ${N==="single"?"bg-slate-700/80 text-slate-100":"text-slate-400 hover:text-slate-200"}`,children:[l.jsx(bn,{size:12}),"Single"]}),l.jsxs("button",{onClick:()=>p("comparison"),className:`flex-1 min-w-0 px-2 py-1.5 text-xs font-medium rounded-md transition-colors flex items-center justify-center gap-1 ${N==="comparison"?"bg-slate-700/80 text-slate-100":"text-slate-400 hover:text-slate-200"}`,children:[l.jsx(ou,{size:12}),"Compare"]}),l.jsxs("button",{onClick:()=>p("dataset"),className:`flex-1 min-w-0 px-2 py-1.5 text-xs font-medium rounded-md transition-colors flex items-center justify-center gap-1 ${N==="dataset"?"bg-blue-600/80 text-slate-100":"text-slate-400 hover:text-slate-200"}`,children:[l.jsx(sn,{size:12}),"Dataset"]})]})]}),l.jsxs("div",{className:"flex items-center justify-between",children:[l.jsxs("h2",{className:"text-sm font-medium text-slate-400",children:[w.length," result",w.length!==1?"s":""," in ",P.length," scenario",P.length!==1?"s":""]}),j&&l.jsx("button",{onClick:()=>v(""),className:"text-xs text-slate-500 hover:text-slate-300",children:"Clear"})]}),l.jsx("div",{className:"space-y-2 max-h-[70vh] overflow-y-auto pr-2",children:P.length===0?l.jsxs("div",{className:"text-center py-8 text-slate-500",children:[l.jsx(Rg,{size:24,className:"mx-auto mb-2 opacity-50"}),l.jsx("p",{className:"text-sm",children:"No results match your filters"})]}):P.map(C=>{const E=R[C],_=d.has(C);return l.jsxs("div",{className:"panel-card overflow-hidden",children:[l.jsxs("button",{onClick:()=>k(C),className:"w-full flex items-center justify-between p-3 hover:bg-slate-800/40 transition-colors",children:[l.jsxs("div",{className:"flex items-center gap-2",children:[_?l.jsx(ge,{size:16,className:"text-slate-500"}):l.jsx(pe,{size:16,className:"text-slate-500"}),l.jsx("span",{className:"font-medium text-slate-200 text-sm",children:C})]}),l.jsx("span",{className:"text-xs text-slate-500 bg-slate-800/60 px-2 py-0.5 rounded",children:E.length})]}),!_&&l.jsx("div",{className:"border-t border-slate-800/50",children:E.map(b=>{const S=pu(b),T=mu(b);return l.jsxs("div",{className:`flex items-center justify-between p-3 cursor-pointer transition-colors border-l-2 ${c===b.filename?"bg-slate-800/60 border-l-orange-400":"hover:bg-slate-800/40 border-l-transparent"}`,onClick:()=>M(b.filename),children:[l.jsxs("div",{className:"flex items-center gap-2 min-w-0",children:[T?l.jsx(sn,{size:14,className:"text-blue-400 flex-shrink-0"}):S?l.jsx(ou,{size:14,className:"text-purple-400 flex-shrink-0"}):l.jsx(bn,{size:14,className:"text-orange-400 flex-shrink-0"}),l.jsx("div",{className:"min-w-0",children:l.jsxs("p",{className:"text-xs text-slate-400 flex items-center gap-1",children:[l.jsx(_o,{size:10}),new Date(b.timestamp).toLocaleString()]})})]}),l.jsx(de,{to:`/run/${Al(b.scenario_id)}`,onClick:D=>D.stopPropagation(),className:"p-1.5 text-slate-500 hover:text-orange-300 hover:bg-slate-700/60 rounded flex-shrink-0",title:"Run again",children:l.jsx(Xt,{size:12})})]},b.filename)})})]},C)})})]}),l.jsx("div",{className:"lg:col-span-2",children:i?l.jsxs("div",{className:"space-y-4",children:[l.jsx("div",{className:"panel-card p-4",children:l.jsxs("div",{className:"flex items-center justify-between gap-4",children:[l.jsxs("div",{className:"min-w-0 flex-1",children:[l.jsx("h2",{className:"font-semibold text-slate-100 truncate",children:Al(i.scenario_id)}),B==="single"&&A(i).model&&l.jsxs("p",{className:"text-sm text-slate-400 mt-1",children:["Model: ",A(i).model]}),B==="comparison"&&l.jsx("p",{className:"text-sm text-slate-400 mt-1",children:"Model Comparison"}),B==="dataset"&&l.jsxs("p",{className:"text-sm text-slate-400 mt-1",children:["Dataset: ",U(i).dataset_id," | Model: ",U(i).model]})]}),l.jsxs("div",{className:"flex items-center gap-2 flex-shrink-0",children:[l.jsxs("button",{onClick:O,className:`flex items-center gap-1 text-xs px-2 py-1 border rounded whitespace-nowrap transition-colors ${g?"text-emerald-400 border-emerald-500/50 bg-emerald-500/10":"text-slate-400 hover:text-slate-100 border-slate-700/70"}`,children:[l.jsx(mp,{size:12}),g?"Copied!":"Copy JSON"]}),l.jsx("button",{onClick:()=>f(!m),className:"text-xs text-slate-400 hover:text-slate-100 px-2 py-1 border border-slate-700/70 rounded whitespace-nowrap",children:m?"View Details":"View JSON"}),l.jsxs(de,{to:`/run/${Al(i.scenario_id)}`,className:"flex items-center gap-1 bg-orange-400 hover:bg-orange-300 text-slate-900 px-3 py-1 rounded text-sm font-semibold whitespace-nowrap",children:[l.jsx(Xt,{size:14}),"Run Again"]})]})]})}),m?l.jsx("div",{className:"panel-card p-4",children:l.jsx("pre",{className:"text-xs text-slate-300 bg-slate-950/70 p-3 rounded overflow-auto max-h-[70vh]",children:JSON.stringify(i,null,2)})}):l.jsxs(l.Fragment,{children:[B==="single"&&l.jsx(Mo,{result:A(i)}),B==="comparison"&&l.jsx(Np,{comparison:L(i)}),B==="dataset"&&l.jsx(ev,{result:U(i)}),B==="unknown"&&l.jsx("div",{className:"panel-card p-6 text-center",children:l.jsx("p",{className:"text-slate-500",children:'Could not parse result format. Click "View JSON" to see raw data.'})})]})]}):l.jsxs("div",{className:"panel-card p-8 text-center",children:[l.jsx(xp,{size:32,className:"mx-auto text-slate-600 mb-3"}),l.jsx("p",{className:"text-slate-400",children:"Select a result to view details"})]})})]})]})}function ev({result:e}){const[t,n]=x.useState(new Set),s=o=>{const c=new Set(t);c.has(o)?c.delete(o):c.add(o),n(c)},r=e.passed_cases,a=e.total_cases,i=e.pass_rate;return l.jsxs("div",{className:"space-y-4",children:[l.jsxs("div",{className:"grid grid-cols-2 md:grid-cols-4 gap-3",children:[l.jsxs("div",{className:"panel-card p-4",children:[l.jsxs("div",{className:"flex items-center gap-2 text-slate-400 mb-2",children:[l.jsx(sn,{className:"w-4 h-4"}),"Cases Passed"]}),l.jsxs("div",{className:`text-2xl font-bold ${i>=.9?"text-emerald-300":i>=.7?"text-amber-300":"text-red-400"}`,children:[r,"/",a]}),l.jsxs("div",{className:"text-xs text-slate-500",children:[(i*100).toFixed(1),"% pass rate"]})]}),l.jsxs("div",{className:"panel-card p-4",children:[l.jsxs("div",{className:"flex items-center gap-2 text-slate-400 mb-2",children:[l.jsx(Xs,{className:"w-4 h-4"}),"Avg Score"]}),l.jsxs("div",{className:"text-2xl font-bold text-slate-100",children:[e.avg_percentage.toFixed(1),"%"]}),l.jsxs("div",{className:"text-xs text-slate-500",children:[e.avg_score.toFixed(1)," points"]})]}),l.jsxs("div",{className:"panel-card p-4",children:[l.jsxs("div",{className:"flex items-center gap-2 text-slate-400 mb-2",children:[l.jsx(_o,{className:"w-4 h-4"}),"Total Time"]}),l.jsxs("div",{className:"text-2xl font-bold text-slate-100",children:[(e.total_time_ms/1e3).toFixed(1),"s"]}),l.jsxs("div",{className:"text-xs text-slate-500",children:[Math.round(e.total_time_ms/Math.max(a,1)),"ms/case"]})]}),l.jsxs("div",{className:"panel-card p-4",children:[l.jsxs("div",{className:"flex items-center gap-2 text-slate-400 mb-2",children:[l.jsx(Nt,{className:"w-4 h-4"}),"Failed"]}),l.jsx("div",{className:"text-2xl font-bold text-red-400",children:e.failed_cases}),l.jsxs("div",{className:"text-xs text-slate-500",children:[a-r," case",a-r!==1?"s":""," failed"]})]})]}),Object.keys(e.by_expected).length>0&&l.jsxs("div",{className:"panel-card p-6",children:[l.jsx("h3",{className:"font-semibold text-slate-100 mb-4",children:"By Expected Outcome"}),l.jsx("div",{className:"grid grid-cols-2 md:grid-cols-3 gap-3",children:Object.entries(e.by_expected).map(([o,c])=>{const u=c.passed+c.failed,m=u>0?c.passed/u:0;return l.jsxs("div",{className:"panel-subtle p-3 rounded-lg",children:[l.jsx("div",{className:"font-medium text-slate-200",children:o}),l.jsxs("div",{className:"flex items-center gap-2 mt-1",children:[l.jsxs("div",{className:`text-lg font-semibold ${m>=.9?"text-emerald-300":m>=.7?"text-amber-300":"text-red-400"}`,children:[c.passed,"/",u]}),l.jsxs("span",{className:"text-xs text-slate-500",children:["(",(m*100).toFixed(0),"%)"]})]})]},o)})})]}),l.jsxs("div",{className:"panel-card p-6",children:[l.jsxs("h3",{className:"font-semibold text-slate-100 mb-4",children:["Case Results (",e.failed_cases," failed, ",e.passed_cases," passed)"]}),l.jsx("div",{className:"space-y-2 max-h-[50vh] overflow-y-auto",children:e.case_results.map(o=>l.jsxs("div",{className:`panel-subtle rounded-lg overflow-hidden ${o.passed?"":"border border-red-500/30"}`,children:[l.jsxs("button",{onClick:()=>s(o.case_id),className:"w-full px-4 py-3 flex items-center gap-3 text-left hover:bg-slate-700/50 transition-colors",children:[o.passed?l.jsx(Xs,{size:16,className:"text-emerald-400 flex-shrink-0"}):l.jsx(Nt,{size:16,className:"text-red-400 flex-shrink-0"}),l.jsx("span",{className:"font-medium text-slate-200",children:o.case_id}),o.expected&&o.expected.length>0&&l.jsxs("span",{className:"text-sm text-slate-400",children:["expected: ",o.expected.join(" or ")]}),l.jsxs("span",{className:"ml-auto text-sm text-slate-400 flex-shrink-0",children:[o.percentage.toFixed(0),"% | ",o.latency_ms,"ms"]}),t.has(o.case_id)?l.jsx(pe,{size:14,className:"text-slate-500 flex-shrink-0"}):l.jsx(ge,{size:14,className:"text-slate-500 flex-shrink-0"})]}),t.has(o.case_id)&&l.jsxs("div",{className:"px-4 pb-3 pt-1 border-t border-slate-700",children:[o.failure_reason&&l.jsx("div",{className:"text-red-400 text-sm mb-2 p-2 bg-red-950/30 rounded",children:o.failure_reason}),l.jsxs("div",{className:"grid grid-cols-2 gap-4 text-sm",children:[l.jsxs("div",{children:[l.jsx("span",{className:"text-slate-500",children:"Expected:"})," ",l.jsx("span",{className:"text-slate-200",children:o.expected&&o.expected.length>0?o.expected.join(" or "):"N/A"})]}),l.jsxs("div",{children:[l.jsx("span",{className:"text-slate-500",children:"Actual:"})," ",l.jsx("span",{className:"text-slate-200",children:o.actual_outcome||"None"})]}),l.jsxs("div",{children:[l.jsx("span",{className:"text-slate-500",children:"Score:"})," ",l.jsxs("span",{className:"text-slate-200",children:[o.goal_score,"/",o.max_score]})]}),l.jsxs("div",{children:[l.jsx("span",{className:"text-slate-500",children:"Latency:"})," ",l.jsxs("span",{className:"text-slate-200",children:[o.latency_ms,"ms"]})]})]})]})]},o.case_id))})]})]})}/*! js-yaml 4.1.1 https://github.com/nodeca/js-yaml @license MIT */function wp(e){return typeof e>"u"||e===null}function tv(e){return typeof e=="object"&&e!==null}function nv(e){return Array.isArray(e)?e:wp(e)?[]:[e]}function sv(e,t){var n,s,r,a;if(t)for(a=Object.keys(t),n=0,s=a.length;n<s;n+=1)r=a[n],e[r]=t[r];return e}function lv(e,t){var n="",s;for(s=0;s<t;s+=1)n+=e;return n}function rv(e){return e===0&&Number.NEGATIVE_INFINITY===1/e}var av=wp,iv=tv,ov=nv,cv=lv,uv=rv,dv=sv,me={isNothing:av,isObject:iv,toArray:ov,repeat:cv,isNegativeZero:uv,extend:dv};function bp(e,t){var n="",s=e.reason||"(unknown reason)";return e.mark?(e.mark.name&&(n+='in "'+e.mark.name+'" '),n+="("+(e.mark.line+1)+":"+(e.mark.column+1)+")",!t&&e.mark.snippet&&(n+=`
+ */const bn=W("Zap",[["path",{d:"M4 14a1 1 0 0 1-.78-1.63l9.9-10.2a.5.5 0 0 1 .86.46l-1.92 6.02A1 1 0 0 0 13 10h7a1 1 0 0 1 .78 1.63l-9.9 10.2a.5.5 0 0 1-.86-.46l1.92-6.02A1 1 0 0 0 11 14z",key:"1xq2db"}]]),Dg=[{path:"/",label:"Dashboard",icon:Pg},{path:"/results",label:"Results",icon:Hr},{path:"/builder",label:"Builder",icon:wn},{path:"/datasets",label:"Datasets",icon:sn}];function Ig(){const e=Ft(),[t,n]=x.useState(()=>typeof window<"u"?localStorage.getItem("theme")==="light":!1);return x.useEffect(()=>{t?(document.documentElement.classList.add("light"),localStorage.setItem("theme","light")):(document.documentElement.classList.remove("light"),localStorage.setItem("theme","dark"))},[t]),l.jsxs("div",{className:"app-shell",children:[l.jsxs("aside",{className:"sidebar",children:[l.jsx("div",{className:"p-5 border-b border-slate-800/70",children:l.jsxs("div",{className:"flex items-center justify-between",children:[l.jsxs("div",{children:[l.jsx("h1",{className:"text-xl font-semibold",style:{color:"var(--app-text)"},children:"Sandboxy Local"}),l.jsx("p",{className:"text-sm",style:{color:"var(--app-muted)"},children:"Development Server"})]}),l.jsx("button",{onClick:()=>n(!t),className:"p-2 rounded-lg transition-colors hover:bg-slate-500/20",title:t?"Switch to dark mode":"Switch to light mode",children:t?l.jsx(Ag,{size:20,style:{color:"var(--app-muted)"}}):l.jsx(zg,{size:20,style:{color:"var(--app-muted)"}})})]})}),l.jsx("nav",{className:"p-4",children:l.jsx("ul",{className:"space-y-2",children:Dg.map(({path:s,label:r,icon:a})=>l.jsx("li",{children:l.jsxs(de,{to:s,"data-active":e.pathname===s,className:"nav-link",children:[l.jsx(a,{size:20}),r]})},s))})})]}),l.jsx("main",{className:"flex-1 overflow-auto",children:l.jsx(Uh,{})})]})}const $g="/api/v1";class Ug{async fetch(t,n){const s=await fetch(`${$g}${t}`,{...n,headers:{"Content-Type":"application/json",...n==null?void 0:n.headers}});if(!s.ok){const r=await s.json().catch(()=>({detail:"Unknown error"}));throw new Error(r.detail||`HTTP ${s.status}`)}return s.json()}async getStatus(){return this.fetch("/local/status")}async listScenarios(){return this.fetch("/local/scenarios")}async getScenario(t){return this.fetch(`/local/scenarios/${encodeURIComponent(t)}`)}async getScenarioGoals(t){return this.fetch(`/local/scenarios/${encodeURIComponent(t)}/goals`)}async getScenarioTools(t){return this.fetch(`/local/scenarios/${encodeURIComponent(t)}/tools`)}async listTools(){return this.fetch("/local/tools")}async getTool(t){return this.fetch(`/local/tools/${encodeURIComponent(t)}`)}async listAgents(){return this.fetch("/local/agents")}async getAgent(t){return this.fetch(`/local/agents/${encodeURIComponent(t)}`)}async listModels(){return this.fetch("/local/models")}async listRuns(){return this.fetch("/local/runs")}async getRun(t){return this.fetch(`/local/runs/${encodeURIComponent(t)}`)}async runScenario(t){return this.fetch("/local/run",{method:"POST",body:JSON.stringify(t)})}async compareModels(t){return this.fetch("/local/compare",{method:"POST",body:JSON.stringify(t)})}async listDatasets(){return this.fetch("/local/datasets")}async getDataset(t){return this.fetch(`/local/datasets/${encodeURIComponent(t)}`)}async saveDataset(t,n){return this.fetch("/local/datasets",{method:"POST",body:JSON.stringify({id:t,content:n})})}async updateDataset(t,n){return this.fetch(`/local/datasets/${encodeURIComponent(t)}`,{method:"PUT",body:JSON.stringify({id:t,content:n})})}async deleteDataset(t){return this.fetch(`/local/datasets/${encodeURIComponent(t)}`,{method:"DELETE"})}async runDataset(t){return this.fetch("/local/run-dataset",{method:"POST",body:JSON.stringify(t)})}}class Bg extends Ug{async listProviders(){return(await this.fetch("/providers")).providers}async addProvider(t){return this.fetch("/providers",{method:"POST",body:JSON.stringify(t)})}async getProvider(t){return this.fetch(`/providers/${encodeURIComponent(t)}`)}async deleteProvider(t){await this.fetch(`/providers/${encodeURIComponent(t)}`,{method:"DELETE"})}async testProvider(t){return this.fetch(`/providers/${encodeURIComponent(t)}/test`,{method:"POST"})}}const Z=new Bg;function Hg(){const[e,t]=x.useState(null),[n,s]=x.useState(!0),[r,a]=x.useState(null),[i,o]=x.useState([]),[c,u]=x.useState(!0),[m,f]=x.useState(!1),[g,y]=x.useState(null),[j,v]=x.useState({}),[N,p]=x.useState({name:"",type:"ollama",base_url:"http://localhost:11434/v1"}),[d,h]=x.useState(null),[k,w]=x.useState(!1),R=async()=>{try{const A=await Z.listProviders();o(A)}catch{o([])}finally{u(!1)}},P=async A=>{A.preventDefault(),h(null),w(!0);try{await Z.addProvider(N),await R(),f(!1),p({name:"",type:"ollama",base_url:"http://localhost:11434/v1"})}catch(L){h(L instanceof Error?L.message:"Failed to add provider")}finally{w(!1)}},M=async A=>{y(A);try{const L=await Z.testProvider(A);v(U=>({...U,[A]:{success:L.success,message:L.success?`Connected! Found ${L.models_found.length} models (${L.latency_ms}ms)`:L.error||"Connection failed"}}))}catch(L){v(U=>({...U,[A]:{success:!1,message:L instanceof Error?L.message:"Test failed"}}))}finally{y(null)}},O=async A=>{if(confirm(`Delete provider "${A}"?`))try{await Z.deleteProvider(A),await R(),v(L=>{const U={...L};return delete U[A],U})}catch(L){alert(L instanceof Error?L.message:"Failed to delete provider")}},z=A=>{switch(A){case"ollama":return"http://localhost:11434/v1";case"lmstudio":return"http://localhost:1234/v1";case"vllm":return"http://localhost:8000/v1";default:return"http://localhost:8000/v1"}};return x.useEffect(()=>{fetch("/api/v1/local/status").then(A=>{if(!A.ok)throw new Error("Failed to fetch status");return A.json()}).then(t).catch(A=>a(A.message)).finally(()=>s(!1)),R()},[]),n?l.jsx("div",{className:"p-8 page",children:l.jsx("div",{className:"text-slate-400",children:"Loading..."})}):r?l.jsx("div",{className:"p-8 page",children:l.jsxs("div",{className:"text-red-400",children:["Error: ",r]})}):e?l.jsxs("div",{className:"p-8 page",children:[l.jsxs("div",{className:"mb-8",children:[l.jsx("h1",{className:"text-2xl font-semibold text-slate-100 mb-2",children:"Dashboard"}),l.jsxs("p",{className:"text-slate-400",children:["Root: ",l.jsx("code",{className:"panel-subtle px-2 py-1 rounded",children:e.root_dir})]})]}),l.jsxs("section",{className:"mb-8",children:[l.jsxs("div",{className:"flex items-center justify-between mb-4",children:[l.jsxs("h2",{className:"text-lg font-semibold text-slate-100 flex items-center gap-2",children:[l.jsx(hp,{size:20}),"Scenarios (",e.scenarios.length,")"]}),l.jsxs(de,{to:"/builder",className:"flex items-center gap-1 px-3 py-1.5 bg-orange-400 hover:bg-orange-300 text-slate-900 rounded-lg text-sm font-semibold",children:[l.jsx(Ne,{size:16}),"Create Scenario"]})]}),e.scenarios.length===0?l.jsx("p",{className:"text-slate-400",children:"No scenarios found in scenarios/"}):l.jsx("div",{className:"grid gap-4 md:grid-cols-2 lg:grid-cols-3",children:e.scenarios.map(A=>l.jsxs("div",{className:"panel-card p-4 hover:border-slate-600 transition-colors",children:[l.jsxs(de,{to:`/scenario/${A.id}`,className:"block mb-3",children:[l.jsxs("div",{className:"flex items-start justify-between",children:[l.jsx("h3",{className:"font-medium text-slate-100 mb-1 hover:text-orange-400 transition-colors",children:A.name}),l.jsx(ge,{size:16,className:"text-slate-600 mt-1 flex-shrink-0"})]}),l.jsx("p",{className:"text-sm text-slate-400 line-clamp-2",children:A.description||"No description"})]}),l.jsxs("div",{className:"flex justify-between items-center pt-2 border-t border-slate-700/40",children:[l.jsx("span",{className:"text-xs text-slate-500",children:A.relative_path}),l.jsxs(de,{to:`/run/${A.id}`,className:"flex items-center gap-1 px-3 py-1 bg-orange-400 text-slate-900 rounded hover:bg-orange-300 transition-colors text-sm font-medium",children:[l.jsx(Xt,{size:14}),"Run"]})]})]},A.id))})]}),l.jsxs("section",{className:"mb-8",children:[l.jsxs("div",{className:"flex items-center justify-between mb-4",children:[l.jsxs("h2",{className:"text-lg font-semibold text-slate-100 flex items-center gap-2",children:[l.jsx(wn,{size:20}),"Tools (",e.tools.length,")"]}),l.jsxs(de,{to:"/tool-builder",className:"flex items-center gap-1 px-3 py-1.5 bg-emerald-400 hover:bg-emerald-300 text-slate-900 rounded-lg text-sm font-semibold",children:[l.jsx(Ne,{size:16}),"Create Tool"]})]}),e.tools.length===0?l.jsx("p",{className:"text-slate-400",children:"No tools found in tools/"}):l.jsx("div",{className:"flex flex-wrap gap-2",children:e.tools.map(A=>l.jsx("span",{className:"px-3 py-1 panel-subtle rounded-full text-sm text-slate-200 border border-slate-700/60",children:A.name},A.id))})]}),l.jsxs("section",{className:"mb-8",children:[l.jsxs("div",{className:"flex items-center justify-between mb-4",children:[l.jsxs("h2",{className:"text-lg font-semibold text-slate-100 flex items-center gap-2",children:[l.jsx(Ao,{size:20}),"Local Providers (",i.length,")"]}),l.jsxs("button",{onClick:()=>f(!m),className:"flex items-center gap-1 px-3 py-1.5 bg-purple-500 hover:bg-purple-400 text-white rounded-lg text-sm font-semibold",children:[l.jsx(Ne,{size:16}),"Add Provider"]})]}),m&&l.jsxs("div",{className:"panel-card p-4 mb-4",children:[l.jsx("h3",{className:"text-sm font-semibold text-slate-200 mb-3",children:"Add Local Provider"}),l.jsxs("form",{onSubmit:P,className:"space-y-3",children:[l.jsxs("div",{className:"grid grid-cols-2 gap-3",children:[l.jsxs("div",{children:[l.jsx("label",{className:"block text-xs text-slate-400 mb-1",children:"Name"}),l.jsx("input",{type:"text",value:N.name,onChange:A=>p({...N,name:A.target.value}),placeholder:"ollama-local",className:"w-full px-3 py-1.5 bg-slate-800 border border-slate-700 rounded text-sm text-slate-100 placeholder-slate-500",required:!0})]}),l.jsxs("div",{children:[l.jsx("label",{className:"block text-xs text-slate-400 mb-1",children:"Type"}),l.jsxs("select",{value:N.type,onChange:A=>{const L=A.target.value;p({...N,type:L,base_url:z(L)})},className:"w-full px-3 py-1.5 bg-slate-800 border border-slate-700 rounded text-sm text-slate-100",children:[l.jsx("option",{value:"ollama",children:"Ollama"}),l.jsx("option",{value:"lmstudio",children:"LM Studio"}),l.jsx("option",{value:"vllm",children:"vLLM"}),l.jsx("option",{value:"openai-compatible",children:"OpenAI Compatible"})]})]})]}),l.jsxs("div",{children:[l.jsx("label",{className:"block text-xs text-slate-400 mb-1",children:"Base URL"}),l.jsx("input",{type:"url",value:N.base_url,onChange:A=>p({...N,base_url:A.target.value}),placeholder:"http://localhost:11434/v1",className:"w-full px-3 py-1.5 bg-slate-800 border border-slate-700 rounded text-sm text-slate-100 placeholder-slate-500",required:!0})]}),d&&l.jsx("p",{className:"text-red-400 text-xs",children:d}),l.jsxs("div",{className:"flex gap-2 justify-end",children:[l.jsx("button",{type:"button",onClick:()=>f(!1),className:"px-3 py-1.5 text-slate-400 hover:text-slate-200 text-sm",children:"Cancel"}),l.jsx("button",{type:"submit",disabled:k,className:"px-3 py-1.5 bg-purple-500 hover:bg-purple-400 text-white rounded text-sm font-medium disabled:opacity-50",children:k?"Adding...":"Add Provider"})]})]})]}),c?l.jsx("p",{className:"text-slate-400",children:"Loading providers..."}):i.length===0?l.jsxs("div",{className:"text-slate-400",children:[l.jsx("p",{className:"mb-2",children:"No local providers configured."}),l.jsx("p",{className:"text-sm",children:"Add a provider to use local models like Ollama, LM Studio, or vLLM."})]}):l.jsx("div",{className:"grid gap-3",children:i.map(A=>{var L,U;return l.jsxs("div",{className:"panel-card p-4 flex items-center justify-between",children:[l.jsxs("div",{className:"flex items-center gap-3",children:[A.status==="connected"?l.jsx(fl,{size:18,className:"text-green-400"}):A.status==="error"?l.jsx(Eo,{size:18,className:"text-red-400"}):l.jsx(nn,{size:18,className:"text-slate-500"}),l.jsxs("div",{className:"flex-1 min-w-0",children:[l.jsxs("div",{className:"flex items-center gap-2",children:[l.jsx("span",{className:"font-medium text-slate-100",children:A.name}),l.jsx("span",{className:"text-xs px-1.5 py-0.5 bg-slate-700 rounded text-slate-300",children:A.type}),A.enabled&&l.jsx("span",{className:"text-xs px-1.5 py-0.5 bg-green-900/50 text-green-400 rounded",children:"enabled"})]}),l.jsx("div",{className:"text-xs text-slate-400 mt-0.5",children:A.base_url}),((L=A.models)==null?void 0:L.length)>0&&l.jsx("div",{className:"flex flex-wrap gap-1.5 mt-2",children:A.models.map(B=>l.jsx("span",{className:"text-xs px-2 py-0.5 bg-emerald-900/40 border border-emerald-700/50 rounded text-emerald-300",children:B},B))}),(((U=A.models)==null?void 0:U.length)===0||!A.models)&&A.status==="connected"&&l.jsx("div",{className:"text-xs text-slate-500 mt-1",children:"No models available"}),j[A.name]&&l.jsx("div",{className:`text-xs mt-1 ${j[A.name].success?"text-green-400":"text-red-400"}`,children:j[A.name].message})]})]}),l.jsxs("div",{className:"flex items-center gap-2",children:[l.jsx("button",{onClick:()=>M(A.name),disabled:g===A.name,className:"p-1.5 text-slate-400 hover:text-slate-200 hover:bg-slate-700 rounded disabled:opacity-50",title:"Test connection",children:l.jsx(Mg,{size:16,className:g===A.name?"animate-spin":""})}),l.jsx("button",{onClick:()=>O(A.name),className:"p-1.5 text-slate-400 hover:text-red-400 hover:bg-slate-700 rounded",title:"Delete provider",children:l.jsx(Ae,{size:16})})]})]},A.name)})})]})]}):null}function Vg(){const{scenarioId:e}=rp(),[t,n]=x.useState(null),[s,r]=x.useState([]),[a,i]=x.useState([]),[o,c]=x.useState(!0),[u,m]=x.useState(null),[f,g]=x.useState(new Set(["tools","evaluation","interaction"])),y=L=>{g(U=>{const B=new Set(U);return B.has(L)?B.delete(L):B.add(L),B})};if(x.useEffect(()=>{e&&(c(!0),Promise.all([Z.getScenario(e),Z.getScenarioGoals(e).catch(()=>[]),Z.getScenarioTools(e).catch(()=>[])]).then(([L,U,B])=>{n(L),r(U),i(B)}).catch(L=>m(L.message)).finally(()=>c(!1)))},[e]),o)return l.jsx("div",{className:"p-8 page",children:l.jsx("div",{className:"text-slate-400",children:"Loading scenario..."})});if(u||!t)return l.jsxs("div",{className:"p-8 page",children:[l.jsxs("div",{className:"text-red-400",children:["Error: ",u||"Scenario not found"]}),l.jsx(de,{to:"/",className:"text-orange-400 hover:underline mt-4 inline-block",children:"← Back to Dashboard"})]});const j=t.content||{},v=j.steps||[],N=j.mcp_servers||[],p=j.tools||{},d=j.tools_from||[],h=j.evaluation?j.evaluation.judge:null,k=j.evaluation,w=k==null?void 0:k.goals,R=k?k.max_score||((w==null?void 0:w.reduce((L,U)=>L+(U.points||0),0))??0):null,P=t.variables||[],M=v.length>0,O=j.category||null,z=j.tags||[],A=a.length>0?a.length:Object.keys(p).length;return l.jsxs("div",{className:"p-8 page max-w-5xl",children:[l.jsxs(de,{to:"/",className:"inline-flex items-center gap-1 text-slate-400 hover:text-slate-200 mb-6 text-sm",children:[l.jsx(Gs,{size:16}),"Back to Dashboard"]}),l.jsxs("div",{className:"panel-card p-6 mb-6",children:[l.jsxs("div",{className:"flex items-start justify-between",children:[l.jsxs("div",{className:"flex-1",children:[l.jsx("h1",{className:"text-2xl font-semibold text-slate-100 mb-2",children:t.name}),l.jsx("p",{className:"text-slate-400 mb-4",children:t.description}),l.jsxs("div",{className:"flex flex-wrap gap-2",children:[O&&l.jsx("span",{className:"px-2 py-1 bg-purple-900/40 border border-purple-700/50 rounded text-xs text-purple-300",children:O}),z.map(L=>l.jsx("span",{className:"px-2 py-1 bg-slate-700/50 border border-slate-600/50 rounded text-xs text-slate-300",children:L},L))]})]}),l.jsxs("div",{className:"flex gap-2 ml-4",children:[l.jsxs(de,{to:`/builder?scenario=${e}`,className:"flex items-center gap-1.5 px-3 py-2 bg-slate-700 hover:bg-slate-600 text-slate-200 rounded-lg text-sm font-medium transition-colors",children:[l.jsx(Lg,{size:16}),"Edit"]}),l.jsxs(de,{to:`/run/${e}`,className:"flex items-center gap-1.5 px-4 py-2 bg-orange-400 hover:bg-orange-300 text-slate-900 rounded-lg text-sm font-semibold transition-colors",children:[l.jsx(Xt,{size:16}),"Run"]})]})]}),l.jsxs("div",{className:"flex gap-6 mt-6 pt-4 border-t border-slate-700/60",children:[l.jsxs("div",{className:"flex items-center gap-2 text-sm",children:[l.jsx(vr,{size:16,className:"text-slate-500"}),l.jsx("span",{className:"text-slate-400",children:M?`Multi-turn (${v.length} steps)`:"Single-turn"})]}),l.jsxs("div",{className:"flex items-center gap-2 text-sm",children:[l.jsx(wn,{size:16,className:"text-slate-500"}),l.jsx("span",{className:"text-slate-400",children:N.length>0&&A===0?`${N.length} MCP server${N.length!==1?"s":""}`:`${A} tool${A!==1?"s":""}${N.length>0?` + ${N.length} MCP`:""}`})]}),l.jsxs("div",{className:"flex items-center gap-2 text-sm",children:[l.jsx(yi,{size:16,className:"text-slate-500"}),l.jsxs("span",{className:"text-slate-400",children:[s.length," goal",s.length!==1?"s":"",R&&` (${R} pt${R!==1?"s":""})`]})]}),P.length>0&&l.jsxs("div",{className:"flex items-center gap-2 text-sm",children:[l.jsx(ji,{size:16,className:"text-slate-500"}),l.jsxs("span",{className:"text-slate-400",children:[P.length," variables"]})]})]})]}),l.jsxs("div",{className:"panel-card mb-4",children:[l.jsxs("button",{onClick:()=>y("tools"),className:"w-full flex items-center justify-between p-4 text-left hover:bg-slate-800/30 transition-colors",children:[l.jsxs("div",{className:"flex items-center gap-2",children:[l.jsx(wn,{size:18,className:"text-emerald-400"}),l.jsxs("h2",{className:"font-semibold text-slate-100",children:["Tools",N.length>0&&A===0?l.jsxs("span",{className:"ml-2 text-sm font-normal text-slate-400",children:["(",N.length," MCP server",N.length>1?"s":"",")"]}):l.jsxs("span",{className:"ml-2 text-sm font-normal text-slate-400",children:["(",A,N.length>0?` + ${N.length} MCP`:"",")"]})]})]}),f.has("tools")?l.jsx(pe,{size:18,className:"text-slate-500"}):l.jsx(ge,{size:18,className:"text-slate-500"})]}),f.has("tools")&&l.jsxs("div",{className:"px-4 pb-4 space-y-4",children:[N.length>0&&l.jsxs("div",{children:[l.jsx("h3",{className:"text-xs font-semibold text-slate-500 uppercase tracking-wider mb-2",children:"MCP Servers"}),N.map(L=>{var U;return l.jsxs("div",{className:"p-3 bg-slate-800/50 border border-slate-700/60 rounded-lg mb-2",children:[l.jsxs("div",{className:"flex items-center gap-2 mb-2",children:[l.jsx(Ao,{size:14,className:"text-cyan-400"}),l.jsx("span",{className:"font-medium text-slate-200",children:L.name}),l.jsx("span",{className:"text-xs px-1.5 py-0.5 bg-cyan-900/40 text-cyan-300 rounded",children:"MCP"})]}),l.jsx("div",{className:"text-xs text-slate-500 mb-2",children:L.url?l.jsx("span",{className:"font-mono",children:L.url}):L.command?l.jsxs("span",{className:"font-mono",children:[L.command," ",(U=L.args)==null?void 0:U.join(" ")]}):null}),a.length>0&&l.jsx("div",{className:"flex flex-wrap gap-1.5",children:a.map(B=>l.jsx("span",{className:"text-xs px-2 py-1 bg-emerald-900/30 border border-emerald-700/40 rounded text-emerald-300",title:B.description,children:B.name},B.name))})]},L.name)})]}),d.length>0&&l.jsxs("div",{children:[l.jsx("h3",{className:"text-xs font-semibold text-slate-500 uppercase tracking-wider mb-2",children:"Tool Libraries"}),l.jsx("div",{className:"flex flex-wrap gap-2",children:d.map(L=>l.jsx("span",{className:"px-3 py-1.5 bg-slate-800/50 border border-slate-700/60 rounded text-sm text-slate-300",children:L},L))})]}),Object.keys(p).length>0&&l.jsxs("div",{children:[l.jsx("h3",{className:"text-xs font-semibold text-slate-500 uppercase tracking-wider mb-2",children:"Inline Tools (Mock)"}),l.jsx("div",{className:"space-y-2",children:Object.entries(p).map(([L,U])=>{const B=U,C=B.actions,E=C?Object.keys(C):[];return l.jsxs("div",{className:"p-3 bg-slate-800/50 border border-slate-700/60 rounded-lg",children:[l.jsxs("div",{className:"flex items-center gap-2 mb-1",children:[l.jsx(bn,{size:14,className:"text-yellow-400"}),l.jsx("span",{className:"font-medium text-slate-200",children:L}),l.jsx("span",{className:"text-xs px-1.5 py-0.5 bg-yellow-900/40 text-yellow-300 rounded",children:"Mock"})]}),typeof B.description=="string"&&B.description&&l.jsx("p",{className:"text-xs text-slate-500 mb-2",children:B.description}),E.length>0&&l.jsx("div",{className:"flex flex-wrap gap-1.5",children:E.map(_=>l.jsx("span",{className:"text-xs px-2 py-1 bg-slate-700/50 rounded text-slate-400",children:_},_))})]},L)})})]}),N.length===0&&d.length===0&&Object.keys(p).length===0&&l.jsx("p",{className:"text-slate-500 text-sm",children:"No tools configured"})]})]}),l.jsxs("div",{className:"panel-card mb-4",children:[l.jsxs("button",{onClick:()=>y("evaluation"),className:"w-full flex items-center justify-between p-4 text-left hover:bg-slate-800/30 transition-colors",children:[l.jsxs("div",{className:"flex items-center gap-2",children:[l.jsx(yi,{size:18,className:"text-orange-400"}),l.jsxs("h2",{className:"font-semibold text-slate-100",children:["Evaluation",R&&l.jsxs("span",{className:"ml-2 text-sm font-normal text-slate-400",children:["(",R," pts max)"]})]})]}),f.has("evaluation")?l.jsx(pe,{size:18,className:"text-slate-500"}):l.jsx(ge,{size:18,className:"text-slate-500"})]}),f.has("evaluation")&&l.jsxs("div",{className:"px-4 pb-4 space-y-4",children:[s.length>0&&l.jsxs("div",{children:[l.jsx("h3",{className:"text-xs font-semibold text-slate-500 uppercase tracking-wider mb-3",children:"Goals"}),l.jsx("div",{className:"space-y-2",children:s.map(L=>{const U=j.evaluation,B=U==null?void 0:U.goals,C=B==null?void 0:B.find(S=>S.id===L.id),E=C==null?void 0:C.detection,_=E==null?void 0:E.type,b=C==null?void 0:C.points;return l.jsxs("div",{className:"flex items-start gap-3 p-3 bg-slate-800/50 border border-slate-700/60 rounded-lg",children:[l.jsx(fl,{size:16,className:"text-slate-600 mt-0.5 flex-shrink-0"}),l.jsxs("div",{className:"flex-1 min-w-0",children:[l.jsxs("div",{className:"flex items-center gap-2 mb-1",children:[l.jsx("span",{className:"font-medium text-slate-200",children:L.name}),L.outcome&&l.jsx("span",{className:"text-xs px-1.5 py-0.5 bg-blue-900/40 text-blue-300 rounded",children:"outcome"})]}),L.description&&l.jsx("p",{className:"text-xs text-slate-500",children:L.description})]}),l.jsxs("div",{className:"flex items-center gap-2 flex-shrink-0",children:[_&&l.jsx("span",{className:"text-xs px-2 py-1 bg-slate-700/50 rounded text-slate-400 font-mono",children:_}),b!==void 0&&l.jsxs("span",{className:"text-sm font-semibold text-orange-400",children:[b," pts"]})]})]},L.id)})})]}),h&&l.jsxs("div",{children:[l.jsx("h3",{className:"text-xs font-semibold text-slate-500 uppercase tracking-wider mb-3",children:"Judge"}),l.jsxs("div",{className:"p-3 bg-slate-800/50 border border-slate-700/60 rounded-lg",children:[l.jsxs("div",{className:"flex items-center gap-2 mb-2",children:[l.jsx(Fg,{size:14,className:"text-purple-400"}),l.jsxs("span",{className:"font-medium text-slate-200 capitalize",children:[h.type," Judge"]}),h.model&&l.jsx("span",{className:"text-xs px-1.5 py-0.5 bg-purple-900/40 text-purple-300 rounded",children:h.model})]}),h.pass_threshold!==void 0&&l.jsxs("p",{className:"text-xs text-slate-500 mb-2",children:["Pass threshold: ",(h.pass_threshold*100).toFixed(0),"%"]}),h.rubric&&l.jsxs("details",{className:"text-xs",children:[l.jsx("summary",{className:"text-slate-400 cursor-pointer hover:text-slate-300",children:"View rubric"}),l.jsx("pre",{className:"mt-2 p-2 bg-slate-900 rounded text-slate-400 overflow-auto max-h-48 whitespace-pre-wrap",children:h.rubric})]})]})]}),s.length===0&&!h&&l.jsx("p",{className:"text-slate-500 text-sm",children:"No evaluation configured"})]})]}),l.jsxs("div",{className:"panel-card mb-4",children:[l.jsxs("button",{onClick:()=>y("interaction"),className:"w-full flex items-center justify-between p-4 text-left hover:bg-slate-800/30 transition-colors",children:[l.jsxs("div",{className:"flex items-center gap-2",children:[l.jsx(vr,{size:18,className:"text-blue-400"}),l.jsxs("h2",{className:"font-semibold text-slate-100",children:["Interaction",l.jsxs("span",{className:"ml-2 text-sm font-normal text-slate-400",children:["(",M?"Multi-turn":"Single-turn",")"]})]})]}),f.has("interaction")?l.jsx(pe,{size:18,className:"text-slate-500"}):l.jsx(ge,{size:18,className:"text-slate-500"})]}),f.has("interaction")&&l.jsxs("div",{className:"px-4 pb-4 space-y-4",children:[typeof j.system_prompt=="string"&&j.system_prompt&&l.jsxs("div",{children:[l.jsx("h3",{className:"text-xs font-semibold text-slate-500 uppercase tracking-wider mb-2",children:"System Prompt"}),l.jsx("pre",{className:"p-3 bg-slate-800/50 border border-slate-700/60 rounded-lg text-xs text-slate-300 overflow-auto max-h-48 whitespace-pre-wrap",children:j.system_prompt})]}),M&&l.jsxs("div",{children:[l.jsx("h3",{className:"text-xs font-semibold text-slate-500 uppercase tracking-wider mb-3",children:"Steps"}),l.jsx("div",{className:"space-y-2",children:v.map((L,U)=>{var B;return l.jsxs("div",{className:"flex items-start gap-3 p-3 bg-slate-800/50 border border-slate-700/60 rounded-lg",children:[l.jsx("div",{className:"w-6 h-6 rounded-full bg-blue-900/40 border border-blue-700/50 flex items-center justify-center text-xs text-blue-300 flex-shrink-0",children:U+1}),l.jsxs("div",{className:"flex-1",children:[l.jsxs("div",{className:"flex items-center gap-2 mb-1",children:[l.jsx("span",{className:"font-medium text-slate-200",children:L.id}),l.jsx("span",{className:"text-xs px-1.5 py-0.5 bg-slate-700/50 text-slate-400 rounded font-mono",children:L.action})]}),typeof((B=L.params)==null?void 0:B.content)=="string"&&L.params.content&&l.jsxs("p",{className:"text-xs text-slate-500 line-clamp-2",children:[L.params.content.substring(0,100),L.params.content.length>100&&"..."]})]})]},L.id)})})]}),!M&&typeof j.prompt=="string"&&j.prompt&&l.jsxs("div",{children:[l.jsx("h3",{className:"text-xs font-semibold text-slate-500 uppercase tracking-wider mb-2",children:"Prompt"}),l.jsx("pre",{className:"p-3 bg-slate-800/50 border border-slate-700/60 rounded-lg text-xs text-slate-300 overflow-auto max-h-48 whitespace-pre-wrap",children:j.prompt})]})]})]}),P.length>0&&l.jsxs("div",{className:"panel-card mb-4",children:[l.jsxs("button",{onClick:()=>y("variables"),className:"w-full flex items-center justify-between p-4 text-left hover:bg-slate-800/30 transition-colors",children:[l.jsxs("div",{className:"flex items-center gap-2",children:[l.jsx(ji,{size:18,className:"text-indigo-400"}),l.jsxs("h2",{className:"font-semibold text-slate-100",children:["Variables (",P.length,")"]})]}),f.has("variables")?l.jsx(pe,{size:18,className:"text-slate-500"}):l.jsx(ge,{size:18,className:"text-slate-500"})]}),f.has("variables")&&l.jsx("div",{className:"px-4 pb-4",children:l.jsx("div",{className:"grid gap-2",children:P.map(L=>l.jsxs("div",{className:"flex items-center justify-between p-3 bg-slate-800/50 border border-slate-700/60 rounded-lg",children:[l.jsxs("div",{className:"flex items-center gap-3",children:[l.jsx("code",{className:"text-sm font-mono text-indigo-300",children:`{{${L.name}}}`}),l.jsx("span",{className:"text-xs px-1.5 py-0.5 bg-slate-700/50 text-slate-400 rounded",children:L.type}),L.required&&l.jsx("span",{className:"text-xs px-1.5 py-0.5 bg-red-900/40 text-red-300 rounded",children:"required"})]}),L.default!==void 0&&l.jsxs("span",{className:"text-xs text-slate-500",children:["default: ",JSON.stringify(L.default)]})]},L.name))})})]}),l.jsxs("div",{className:"panel-card",children:[l.jsxs("button",{onClick:()=>y("file"),className:"w-full flex items-center justify-between p-4 text-left hover:bg-slate-800/30 transition-colors",children:[l.jsxs("div",{className:"flex items-center gap-2",children:[l.jsx(Hr,{size:18,className:"text-slate-500"}),l.jsx("h2",{className:"font-semibold text-slate-100",children:"File"})]}),f.has("file")?l.jsx(pe,{size:18,className:"text-slate-500"}):l.jsx(ge,{size:18,className:"text-slate-500"})]}),f.has("file")&&l.jsx("div",{className:"px-4 pb-4",children:l.jsx("p",{className:"text-xs text-slate-500 font-mono",children:t.path})})]})]})}function Wg(){const[e,t]=x.useState("idle"),[n,s]=x.useState(null),[r,a]=x.useState(null),[i,o]=x.useState(null),c=x.useCallback(()=>{t("idle"),s(null),a(null),o(null)},[]),u=x.useCallback(async(f,g,y,j)=>{c(),t("running");try{const v=await Z.runScenario({scenario_id:f,model:g,variables:y,mlflow_export:j==null?void 0:j.enabled,mlflow_tracking_uri:j==null?void 0:j.trackingUri,mlflow_experiment:j==null?void 0:j.experiment,mlflow_tracing:j==null?void 0:j.tracing});v.error?(t("error"),o(v.error)):(t("completed"),s(v))}catch(v){t("error"),o(v instanceof Error?v.message:"Unknown error")}},[c]),m=x.useCallback(async(f,g,y=1,j,v)=>{c(),t("running");try{const N=await Z.compareModels({scenario_id:f,models:g,runs_per_model:y,variables:j,mlflow_export:v==null?void 0:v.enabled,mlflow_tracking_uri:v==null?void 0:v.trackingUri,mlflow_experiment:v==null?void 0:v.experiment,mlflow_tracing:v==null?void 0:v.tracing});t("completed"),a(N)}catch(N){t("error"),o(N instanceof Error?N.message:"Unknown error")}},[c]);return{state:e,result:n,comparison:r,error:i,runScenario:u,compareModels:m,reset:c}}function jp(e,t){return e==null?"-":t===!0||e===0?"Local":e<1e-4?"<$0.0001":e<.01?`$${e.toFixed(4)}`:`$${e.toFixed(3)}`}function Mo({result:e}){var s,r,a;const[t,n]=x.useState(!1);return l.jsxs("div",{className:"space-y-6",children:[l.jsxs("div",{className:"grid grid-cols-2 md:grid-cols-3 lg:grid-cols-6 gap-3",children:[e.evaluation&&l.jsxs("div",{className:"panel-card p-4 col-span-2 md:col-span-1",children:[l.jsxs("div",{className:"flex items-center gap-2 text-slate-400 mb-2",children:[l.jsx(Lo,{className:"w-4 h-4"}),"Score"]}),l.jsxs("div",{className:"text-2xl font-bold text-slate-100",children:[e.evaluation.percentage.toFixed(0),"%"]}),l.jsxs("div",{className:"text-xs text-slate-500",children:[e.evaluation.total_score.toFixed(0)," / ",e.evaluation.max_score.toFixed(0)," pts"]})]}),l.jsxs("div",{className:"panel-card p-4",children:[l.jsxs("div",{className:"flex items-center gap-2 text-slate-400 mb-2",children:[l.jsx(_o,{className:"w-4 h-4"}),"Latency"]}),l.jsxs("div",{className:"text-2xl font-bold text-slate-100",children:[((e.latency_ms||0)/1e3).toFixed(1),"s"]})]}),l.jsxs("div",{className:"panel-card p-4",children:[l.jsxs("div",{className:"flex items-center gap-2 text-slate-400 mb-2",children:[l.jsx(vr,{className:"w-4 h-4"}),"Messages"]}),l.jsx("div",{className:"text-2xl font-bold text-slate-100",children:((s=e.history)==null?void 0:s.length)||0})]}),l.jsxs("div",{className:"panel-card p-4",children:[l.jsxs("div",{className:"flex items-center gap-2 text-slate-400 mb-2",children:[l.jsx(bn,{className:"w-4 h-4"}),"Tool Calls"]}),l.jsx("div",{className:"text-2xl font-bold text-slate-100",children:((r=e.tool_calls)==null?void 0:r.length)||0})]}),l.jsxs("div",{className:"panel-card p-4",children:[l.jsxs("div",{className:"flex items-center gap-2 text-slate-400 mb-2",children:[l.jsx(Tg,{className:"w-4 h-4"}),"Tokens"]}),l.jsx("div",{className:"text-2xl font-bold text-slate-100",children:((e.input_tokens||0)+(e.output_tokens||0)).toLocaleString()}),l.jsxs("div",{className:"text-xs text-slate-500",children:[(e.input_tokens||0).toLocaleString()," in / ",(e.output_tokens||0).toLocaleString()," out"]})]}),l.jsxs("div",{className:"panel-card p-4",children:[l.jsxs("div",{className:"flex items-center gap-2 text-slate-400 mb-2",children:[l.jsx(To,{className:"w-4 h-4"}),"Cost"]}),l.jsx("div",{className:"text-2xl font-bold text-emerald-300",children:jp(e.cost_usd,e.is_local)})]})]}),e.evaluation&&e.evaluation.goals&&e.evaluation.goals.length>0&&l.jsxs("div",{className:"panel-card p-6",children:[l.jsx("h3",{className:"text-lg font-semibold text-slate-100 mb-4",children:"Goals"}),l.jsx("div",{className:"space-y-3",children:e.evaluation.goals.map(i=>l.jsxs("div",{className:`flex items-center justify-between p-3 rounded-lg border ${i.achieved?"bg-emerald-500/10 border-emerald-400/40":"bg-slate-950/40 border-slate-800/70"}`,children:[l.jsxs("div",{className:"flex items-center gap-3",children:[i.achieved?l.jsx(fl,{className:"w-5 h-5 text-emerald-400"}):l.jsx(Eo,{className:"w-5 h-5 text-slate-500"}),l.jsxs("div",{children:[l.jsx("div",{className:"font-medium text-slate-100",children:i.name||i.id}),i.reason&&l.jsx("div",{className:"text-sm text-slate-400",children:i.reason})]})]}),l.jsxs("div",{className:`font-bold ${i.achieved?"text-emerald-300":"text-slate-500"}`,children:[i.achieved?`+${i.points}`:"0"," pts"]})]},i.id))})]}),((a=e.evaluation)==null?void 0:a.judge)&&l.jsxs("div",{className:"panel-card p-6",children:[l.jsx("h3",{className:"text-lg font-semibold text-slate-100 mb-4",children:"Judge Evaluation"}),l.jsxs("div",{className:"flex items-center gap-4 mb-4",children:[l.jsxs("div",{className:`text-3xl font-bold ${e.evaluation.judge.passed?"text-emerald-300":"text-red-400"}`,children:[(e.evaluation.judge.score*100).toFixed(0),"%"]}),l.jsx("div",{className:`px-3 py-1 rounded-full text-sm ${e.evaluation.judge.passed?"bg-emerald-500/15 text-emerald-300":"bg-red-900/50 text-red-400"}`,children:e.evaluation.judge.passed?"Passed":"Failed"})]}),l.jsx("p",{className:"text-slate-400",children:e.evaluation.judge.reasoning})]}),e.response&&l.jsxs("div",{className:"panel-card p-6",children:[l.jsx("h3",{className:"text-lg font-semibold text-slate-100 mb-4",children:"Response"}),l.jsx("div",{className:"panel-subtle rounded-lg p-4 whitespace-pre-wrap text-slate-200",children:e.response||"(No response)"})]}),e.history&&e.history.length>0&&l.jsxs("div",{className:"panel-card",children:[l.jsxs("button",{onClick:()=>n(!t),className:"w-full p-4 text-left flex items-center justify-between hover:bg-slate-800/60 transition-colors rounded-lg",children:[l.jsxs("span",{className:"font-semibold text-slate-100",children:["Conversation History (",e.history.length," messages)"]}),l.jsx("span",{className:"text-slate-400",children:t?"Hide":"Show"})]}),t&&l.jsx("div",{className:"px-6 pb-6 space-y-4",children:e.history.map((i,o)=>l.jsxs("div",{className:`p-3 rounded-lg ${i.role==="user"?"bg-orange-500/10 border border-orange-400/40":i.role==="assistant"?"panel-subtle":i.role==="tool"?"bg-amber-500/10 border border-amber-400/40":"panel-subtle"}`,children:[l.jsx("div",{className:"text-xs font-medium text-slate-500 mb-1 uppercase",children:i.role}),l.jsx("div",{className:"text-slate-200 whitespace-pre-wrap text-sm",children:i.content||"(empty)"})]},o))})]}),e.tool_calls&&e.tool_calls.length>0&&l.jsxs("div",{className:"panel-card p-6",children:[l.jsxs("h3",{className:"text-lg font-semibold text-slate-100 mb-4",children:["Tool Calls (",e.tool_calls.length,")"]}),l.jsx("div",{className:"space-y-3",children:e.tool_calls.map((i,o)=>l.jsx(Yg,{call:i},o))})]}),e.error&&l.jsxs("div",{className:"panel-card p-4 border-l-4 border-l-red-500 bg-red-950/30",children:[l.jsxs("div",{className:"flex items-center gap-2 text-red-400 mb-1",children:[l.jsx(nn,{className:"w-4 h-4"}),l.jsx("span",{className:"font-medium",children:"Error"})]}),l.jsx("p",{className:"text-sm text-red-300",children:e.error})]})]})}function uu(e){if(typeof e=="string")try{const t=JSON.parse(e);return JSON.stringify(t,null,2)}catch{return e}return JSON.stringify(e,null,2)}function Yg({call:e}){const[t,n]=x.useState(!1),s=e.result!==void 0&&e.result!==null,r=e.error!==void 0&&e.error!==null;return l.jsxs("div",{className:`panel-subtle rounded-lg overflow-hidden ${e.success?"":"border border-red-500/30"}`,children:[l.jsxs("button",{onClick:()=>n(!t),className:"w-full p-4 text-left flex items-center justify-between hover:bg-slate-800/40 transition-colors",children:[l.jsxs("div",{className:"flex items-center gap-3",children:[t?l.jsx(pe,{className:"w-4 h-4 text-slate-500"}):l.jsx(ge,{className:"w-4 h-4 text-slate-500"}),l.jsxs("code",{className:"text-orange-300",children:[e.tool,".",e.action,"()"]})]}),l.jsx("span",{className:`text-sm ${e.success?"text-emerald-300":"text-red-400"}`,children:e.success?"✓ Success":"✗ Failed"})]}),t&&l.jsxs("div",{className:"px-4 pb-4 space-y-3",children:[l.jsxs("div",{children:[l.jsx("div",{className:"text-xs font-medium text-slate-500 mb-1",children:"Arguments"}),l.jsx("pre",{className:"text-xs text-slate-400 overflow-auto bg-slate-950/50 rounded p-2",children:uu(e.args)})]}),s&&l.jsxs("div",{children:[l.jsx("div",{className:"text-xs font-medium text-emerald-400 mb-1",children:"Result"}),l.jsx("pre",{className:"text-xs text-slate-400 overflow-auto bg-emerald-950/30 rounded p-2 max-h-48",children:uu(e.result)})]}),r&&l.jsxs("div",{children:[l.jsx("div",{className:"text-xs font-medium text-red-400 mb-1",children:"Error"}),l.jsx("pre",{className:"text-xs text-red-300 overflow-auto bg-red-950/30 rounded p-2",children:e.error})]})]})]})}function Qg(e){var o;const t=e.ranking||[],n=[];for(const c of t){const u=(o=e.stats)==null?void 0:o[c];if(!u)continue;const m=u.avg_score||0,f=u.avg_cost_usd||0,g=u.avg_latency_ms||0,y=f>0?m/(f*100):m>0?1/0:0;n.push({modelId:c,score:m,cost:f,latency:g,efficiency:y,badges:[]})}if(n.length===0)return n;const s=n.reduce((c,u)=>c.score>u.score?c:u),r=n.reduce((c,u)=>c.efficiency>u.efficiency?c:u),a=n.filter(c=>c.score>=s.score*.7),i=a.length>0?a.reduce((c,u)=>c.latency<u.latency?c:u):null;for(const c of n)c.modelId===s.modelId&&c.badges.push("quality"),c.modelId===r.modelId&&r.cost>0&&c.badges.push("value"),i&&c.modelId===i.modelId&&c.badges.push("fast");return n}function qg({type:e}){switch(e){case"quality":return l.jsxs("span",{className:"inline-flex items-center gap-1 px-2 py-0.5 rounded-full text-xs font-medium bg-amber-500/20 text-amber-300 border border-amber-500/30",children:[l.jsx(Lo,{className:"w-3 h-3"})," Best Quality"]});case"value":return l.jsxs("span",{className:"inline-flex items-center gap-1 px-2 py-0.5 rounded-full text-xs font-medium bg-emerald-500/20 text-emerald-300 border border-emerald-500/30",children:[l.jsx(To,{className:"w-3 h-3"})," Best Value"]});case"fast":return l.jsxs("span",{className:"inline-flex items-center gap-1 px-2 py-0.5 rounded-full text-xs font-medium bg-blue-500/20 text-blue-300 border border-blue-500/30",children:[l.jsx(bn,{className:"w-3 h-3"})," Fastest"]});default:return null}}function Jg({run:e,onClose:t}){return l.jsx("div",{className:"fixed inset-0 z-50 flex items-center justify-center p-4 bg-black/70",onClick:t,children:l.jsxs("div",{className:"bg-slate-900 rounded-xl max-w-4xl w-full max-h-[90vh] overflow-hidden flex flex-col",onClick:n=>n.stopPropagation(),children:[l.jsxs("div",{className:"flex items-center justify-between p-4 border-b border-slate-800",children:[l.jsxs("div",{children:[l.jsx("h2",{className:"text-lg font-semibold text-slate-100",children:"Run Details"}),l.jsx("p",{className:"text-sm text-slate-400",children:e.model})]}),l.jsx("button",{onClick:t,className:"p-2 text-slate-400 hover:text-slate-100 hover:bg-slate-800 rounded-lg",children:l.jsx(Nt,{className:"w-5 h-5"})})]}),l.jsx("div",{className:"overflow-auto p-4",children:l.jsx(Mo,{result:e})})]})})}function Np({comparison:e}){var m,f,g,y,j,v,N,p;const[t,n]=x.useState(null),[s,r]=x.useState(null),a=e.ranking||[],i=Qg(e),o=new Map(i.map(d=>[d.modelId,d])),c=new Map;if(e.results)for(const d of e.results){const h=c.get(d.model)||[];h.push(d),c.set(d.model,h)}const u=e.results&&e.results.length>0;return l.jsxs(l.Fragment,{children:[s&&l.jsx(Jg,{run:s,onClose:()=>r(null)}),l.jsxs("div",{className:"space-y-6",children:[i.length>0&&l.jsxs("div",{className:"grid grid-cols-1 md:grid-cols-3 gap-4",children:[i.find(d=>d.badges.includes("quality"))&&l.jsxs("div",{className:"panel-card p-4 border-l-4 border-l-amber-400",children:[l.jsxs("div",{className:"flex items-center gap-2 text-amber-300 mb-1",children:[l.jsx(Lo,{className:"w-4 h-4"}),l.jsx("span",{className:"text-sm font-medium",children:"Best Quality"})]}),l.jsx("div",{className:"text-lg font-semibold text-slate-100",children:((m=i.find(d=>d.badges.includes("quality")))==null?void 0:m.modelId.split("/")[1])||((f=i.find(d=>d.badges.includes("quality")))==null?void 0:f.modelId)}),l.jsxs("div",{className:"text-xs text-slate-400 mt-1",children:["Highest score: ",(g=i.find(d=>d.badges.includes("quality")))==null?void 0:g.score.toFixed(1)," pts"]})]}),i.find(d=>d.badges.includes("value"))&&l.jsxs("div",{className:"panel-card p-4 border-l-4 border-l-emerald-400",children:[l.jsxs("div",{className:"flex items-center gap-2 text-emerald-300 mb-1",children:[l.jsx(To,{className:"w-4 h-4"}),l.jsx("span",{className:"text-sm font-medium",children:"Best Value"})]}),l.jsx("div",{className:"text-lg font-semibold text-slate-100",children:((y=i.find(d=>d.badges.includes("value")))==null?void 0:y.modelId.split("/")[1])||((j=i.find(d=>d.badges.includes("value")))==null?void 0:j.modelId)}),l.jsx("div",{className:"text-xs text-slate-400 mt-1",children:(()=>{const d=i.find(R=>R.badges.includes("value")),h=i.find(R=>R.badges.includes("quality"));if(!d||!h)return"";if(d.cost===0)return"Free!";if(d.modelId===h.modelId)return"Also the best quality!";const k=Math.round(d.score/h.score*100),w=h.cost>0?Math.round(d.cost/h.cost*100):0;return w===0?`${k}% quality, nearly free`:`${k}% quality at ${w}% cost`})()})]}),i.find(d=>d.badges.includes("fast"))&&l.jsxs("div",{className:"panel-card p-4 border-l-4 border-l-blue-400",children:[l.jsxs("div",{className:"flex items-center gap-2 text-blue-300 mb-1",children:[l.jsx(bn,{className:"w-4 h-4"}),l.jsx("span",{className:"text-sm font-medium",children:"Fastest"})]}),l.jsx("div",{className:"text-lg font-semibold text-slate-100",children:((v=i.find(d=>d.badges.includes("fast")))==null?void 0:v.modelId.split("/")[1])||((N=i.find(d=>d.badges.includes("fast")))==null?void 0:N.modelId)}),l.jsxs("div",{className:"text-xs text-slate-400 mt-1",children:[((((p=i.find(d=>d.badges.includes("fast")))==null?void 0:p.latency)||0)/1e3).toFixed(1),"s avg latency"]})]})]}),l.jsxs("div",{className:"panel-card overflow-hidden",children:[l.jsxs("div",{className:"p-4 border-b border-slate-800/70",children:[l.jsx("h3",{className:"text-lg font-semibold text-slate-100",children:"Model Comparison"}),l.jsxs("p",{className:"text-sm text-slate-400",children:[e.runs_per_model," run(s) per model"]})]}),l.jsx("div",{className:"overflow-x-auto",children:l.jsxs("table",{className:"w-full",children:[l.jsx("thead",{className:"bg-slate-950/70",children:l.jsxs("tr",{children:[l.jsx("th",{className:"px-4 py-3 text-left text-sm font-medium text-slate-400",children:"Rank"}),l.jsx("th",{className:"px-4 py-3 text-left text-sm font-medium text-slate-400",children:"Model"}),l.jsx("th",{className:"px-4 py-3 text-right text-sm font-medium text-slate-400",children:"Score"}),l.jsx("th",{className:"px-4 py-3 text-right text-sm font-medium text-slate-400",children:"Judge"}),l.jsx("th",{className:"px-4 py-3 text-right text-sm font-medium text-slate-400",children:"Latency"}),l.jsx("th",{className:"px-4 py-3 text-right text-sm font-medium text-slate-400",children:"Tokens"}),l.jsx("th",{className:"px-4 py-3 text-right text-sm font-medium text-slate-400",children:"Cost"})]})}),l.jsx("tbody",{children:a.map((d,h)=>{var M,O;const k=(M=e.stats)==null?void 0:M[d],w=o.get(d),R=c.get(d)||[],P=t===d;return k?l.jsxs(l.Fragment,{children:[l.jsxs("tr",{className:`border-t border-slate-800/70 hover:bg-slate-900/40 ${u?"cursor-pointer":""}`,onClick:()=>u&&n(P?null:d),children:[l.jsx("td",{className:"px-4 py-3",children:l.jsxs("div",{className:"flex items-center gap-2",children:[u&&(P?l.jsx(pe,{className:"w-4 h-4 text-slate-500"}):l.jsx(ge,{className:"w-4 h-4 text-slate-500"})),l.jsxs("span",{className:`font-bold ${h===0?"text-amber-300":"text-slate-400"}`,children:["#",h+1]})]})}),l.jsxs("td",{className:"px-4 py-3",children:[l.jsx("div",{className:"text-slate-100 font-medium",children:d}),w&&w.badges.length>0&&l.jsx("div",{className:"flex flex-wrap gap-1 mt-1",children:w.badges.map(z=>l.jsx(qg,{type:z},z))})]}),l.jsx("td",{className:"px-4 py-3 text-right text-slate-100 font-semibold",children:((O=k.avg_score)==null?void 0:O.toFixed(1))||"0"}),l.jsx("td",{className:"px-4 py-3 text-right text-slate-400",children:k.avg_judge_score!=null?`${(k.avg_judge_score*100).toFixed(0)}%`:"-"}),l.jsxs("td",{className:"px-4 py-3 text-right text-slate-400",children:[((k.avg_latency_ms||0)/1e3).toFixed(1),"s"]}),l.jsx("td",{className:"px-4 py-3 text-right text-slate-400",children:((k.total_input_tokens||0)+(k.total_output_tokens||0)).toLocaleString()}),l.jsx("td",{className:"px-4 py-3 text-right text-emerald-300 font-medium",children:jp(k.avg_cost_usd)})]},d),P&&R.length>0&&l.jsx("tr",{children:l.jsx("td",{colSpan:7,className:"px-4 py-2 bg-slate-950/50",children:l.jsxs("div",{className:"ml-8 space-y-2",children:[l.jsxs("div",{className:"text-sm font-medium text-slate-400 mb-2",children:["Individual Runs (",R.length,")"]}),R.map((z,A)=>{var L,U,B;return l.jsxs("div",{className:"flex items-center justify-between p-3 rounded-lg bg-slate-900/50 border border-slate-800/50 hover:bg-slate-800/50 cursor-pointer",onClick:C=>{C.stopPropagation(),r(z)},children:[l.jsxs("div",{className:"flex items-center gap-3",children:[l.jsxs("span",{className:"text-slate-500 text-sm",children:["Run ",A+1]}),l.jsxs("span",{className:"text-slate-100 font-medium",children:[((U=(L=z.evaluation)==null?void 0:L.percentage)==null?void 0:U.toFixed(0))||0,"%"]}),((B=z.evaluation)==null?void 0:B.judge)&&l.jsx("span",{className:`text-sm ${z.evaluation.judge.passed?"text-emerald-300":"text-red-400"}`,children:z.evaluation.judge.passed?"✓ Passed":"✗ Failed"}),z.error&&l.jsxs("span",{className:"text-sm text-red-400 flex items-center gap-1",children:[l.jsx(nn,{className:"w-3 h-3"})," Error"]})]}),l.jsxs("button",{className:"flex items-center gap-1 text-sm text-orange-300 hover:text-orange-200",onClick:C=>{C.stopPropagation(),r(z)},children:[l.jsx(xp,{className:"w-4 h-4"}),"View Details"]})]},A)})]})})},`${d}-runs`)]}):null})})]})})]}),e.stats&&Object.values(e.stats).some(d=>d.goal_rates&&Object.keys(d.goal_rates).length>0)&&l.jsxs("div",{className:"panel-card p-6",children:[l.jsx("h3",{className:"text-lg font-semibold text-slate-100 mb-4",children:"Goal Achievement Rates"}),(()=>{const d=new Set;return Object.values(e.stats).forEach(h=>{h.goal_rates&&Object.keys(h.goal_rates).forEach(k=>d.add(k))}),l.jsx("div",{className:"overflow-x-auto",children:l.jsxs("table",{className:"w-full",children:[l.jsx("thead",{className:"bg-slate-950/70",children:l.jsxs("tr",{children:[l.jsx("th",{className:"px-4 py-3 text-left text-sm font-medium text-slate-400",children:"Model"}),Array.from(d).map(h=>l.jsx("th",{className:"px-4 py-3 text-right text-sm font-medium text-slate-400",children:h},h))]})}),l.jsx("tbody",{children:a.map(h=>{var w;const k=(w=e.stats)==null?void 0:w[h];return k?l.jsxs("tr",{className:"border-t border-slate-800/70",children:[l.jsx("td",{className:"px-4 py-3 text-slate-100 font-medium",children:h}),Array.from(d).map(R=>{var M;const P=((M=k.goal_rates)==null?void 0:M[R])??0;return l.jsx("td",{className:"px-4 py-3 text-right",children:l.jsxs("span",{className:P>=100?"text-emerald-300":P>=50?"text-amber-300":"text-red-400",children:[P.toFixed(0),"%"]})},R)})]},h):null})})]})})})()]})]})]})}function As(){return l.jsxs("span",{className:"inline-flex items-center gap-1 px-1.5 py-0.5 bg-emerald-500/20 border border-emerald-500/40 rounded text-xs text-emerald-400",children:[l.jsx(gp,{size:10}),"Local"]})}function du({models:e,value:t,onChange:n,disabled:s,placeholder:r="Select a model..."}){const[a,i]=x.useState(!1),[o,c]=x.useState(""),u=x.useRef(null),m=x.useRef(null),f=e.find(p=>p.id===t);x.useEffect(()=>{const p=d=>{u.current&&!u.current.contains(d.target)&&i(!1)};return document.addEventListener("mousedown",p),()=>document.removeEventListener("mousedown",p)},[]),x.useEffect(()=>{a&&m.current&&m.current.focus()},[a]);const g=e.filter(p=>p.name.toLowerCase().includes(o.toLowerCase())||p.id.toLowerCase().includes(o.toLowerCase())),y=g.reduce((p,d)=>{const h=d.provider_name||d.id.split("/")[0]||"other";return p[h]||(p[h]=[]),p[h].push(d),p},{}),j=["openai","anthropic","google","x-ai","deepseek","meta-llama","mistralai","qwen","perplexity"],v=p=>{var d;return(d=y[p])==null?void 0:d.some(h=>h.is_local)},N=Object.keys(y).sort((p,d)=>{const h=v(p),k=v(d);if(h&&!k)return-1;if(!h&&k)return 1;const w=j.indexOf(p),R=j.indexOf(d);return w===-1&&R===-1?p.localeCompare(d):w===-1?1:R===-1?-1:w-R});return l.jsxs("div",{ref:u,className:"relative",children:[l.jsxs("button",{type:"button",onClick:()=>!s&&i(!a),disabled:s,className:`w-full flex items-center justify-between gap-2 px-4 py-2.5 panel-subtle text-left transition-colors ${s?"opacity-50 cursor-not-allowed":"hover:bg-slate-800 cursor-pointer"} ${a?"ring-2 ring-orange-400":""}`,children:[f?l.jsxs("div",{className:"flex items-center justify-between flex-1 min-w-0 gap-2",children:[l.jsx("span",{className:"text-slate-100 truncate",children:f.name}),f.is_local&&l.jsx(As,{}),l.jsx("span",{className:"text-xs text-slate-500 shrink-0",children:f.price})]}):l.jsx("span",{className:"text-slate-500",children:r}),l.jsx(pe,{size:18,className:`text-slate-400 shrink-0 transition-transform ${a?"rotate-180":""}`})]}),a&&l.jsxs("div",{className:"absolute z-50 w-full mt-1 panel-card border border-slate-700/80 dark:border-slate-700/80 shadow-xl max-h-[28rem] overflow-hidden flex flex-col",children:[l.jsx("div",{className:"p-2 border-b border-slate-800/70",children:l.jsxs("div",{className:"relative",children:[l.jsx(Po,{size:16,className:"absolute left-3 top-1/2 -translate-y-1/2 text-slate-500"}),l.jsx("input",{ref:m,type:"text",value:o,onChange:p=>c(p.target.value),placeholder:"Search models...",className:"w-full pl-9 pr-3 py-2 bg-slate-900 border border-slate-800/70 rounded text-sm text-slate-100 placeholder-slate-500 focus:outline-none focus:border-orange-400/60"})]})}),l.jsxs("div",{className:"overflow-y-auto flex-1",children:[N.map(p=>l.jsxs("div",{children:[l.jsxs("div",{className:"px-3 py-1.5 text-xs font-medium text-slate-500 uppercase bg-slate-900 sticky top-0 flex items-center gap-2",children:[p,v(p)&&l.jsx(As,{})]}),y[p].map(d=>l.jsxs("button",{type:"button",onClick:()=>{n(d.id),i(!1),c("")},className:`w-full flex items-center gap-3 px-3 py-2 text-left transition-colors ${d.id===t?"bg-orange-500/20 text-orange-300":"hover:bg-slate-800 text-slate-100"}`,children:[l.jsxs("div",{className:"flex-1 min-w-0 flex items-center gap-2",children:[l.jsx("div",{className:"truncate",children:d.name}),d.is_local&&!v(p)&&l.jsx(As,{})]}),l.jsx("span",{className:"text-xs text-slate-500 shrink-0",children:d.price}),d.id===t&&l.jsx(Xs,{size:16,className:"text-orange-400 shrink-0"})]},d.id))]},p)),g.length===0&&l.jsx("div",{className:"px-3 py-4 text-center text-slate-500 text-sm",children:"No models found"})]})]})]})}function Kg({models:e,selected:t,onChange:n,disabled:s}){const[r,a]=x.useState(!1),[i,o]=x.useState(""),c=x.useRef(null),u=x.useRef(null);x.useEffect(()=>{const p=d=>{c.current&&!c.current.contains(d.target)&&a(!1)};return document.addEventListener("mousedown",p),()=>document.removeEventListener("mousedown",p)},[]),x.useEffect(()=>{r&&u.current&&u.current.focus()},[r]);const m=p=>{t.includes(p)?n(t.filter(d=>d!==p)):n([...t,p])},f=(p,d)=>{d.stopPropagation(),n(t.filter(h=>h!==p))},g=e.filter(p=>p.name.toLowerCase().includes(i.toLowerCase())||p.id.toLowerCase().includes(i.toLowerCase())),y=g.reduce((p,d)=>{const h=d.provider_name||d.id.split("/")[0]||"other";return p[h]||(p[h]=[]),p[h].push(d),p},{}),j=["openai","anthropic","google","x-ai","deepseek","meta-llama","mistralai","qwen","perplexity"],v=p=>{var d;return(d=y[p])==null?void 0:d.some(h=>h.is_local)},N=Object.keys(y).sort((p,d)=>{const h=v(p),k=v(d);if(h&&!k)return-1;if(!h&&k)return 1;const w=j.indexOf(p),R=j.indexOf(d);return w===-1&&R===-1?p.localeCompare(d):w===-1?1:R===-1?-1:w-R});return l.jsxs("div",{ref:c,className:"relative",children:[t.length>0&&l.jsx("div",{className:"flex flex-wrap gap-2 mb-2",children:t.map(p=>{const d=e.find(h=>h.id===p);return l.jsxs("span",{className:`flex items-center gap-1.5 px-2.5 py-1 rounded-full text-sm text-slate-100 ${d!=null&&d.is_local?"bg-emerald-500/20 border border-emerald-400/40":"bg-orange-500/20 border border-orange-400/40"}`,children:[(d==null?void 0:d.is_local)&&l.jsx(gp,{size:12,className:"text-emerald-400"}),(d==null?void 0:d.name)||p,l.jsx("button",{type:"button",onClick:h=>f(p,h),className:"hover:text-red-400 transition-colors",disabled:s,children:l.jsx(Nt,{size:14})})]},p)})}),l.jsxs("button",{type:"button",onClick:()=>!s&&a(!r),disabled:s,className:`w-full flex items-center justify-between gap-2 px-4 py-2.5 panel-subtle text-left transition-colors ${s?"opacity-50 cursor-not-allowed":"hover:bg-slate-800 cursor-pointer"} ${r?"ring-2 ring-orange-400":""}`,children:[l.jsx("span",{className:"text-slate-500",children:t.length===0?"Click to add models...":"Add more models..."}),l.jsx(pe,{size:18,className:`text-slate-400 shrink-0 transition-transform ${r?"rotate-180":""}`})]}),r&&l.jsxs("div",{className:"absolute z-50 w-full mt-1 panel-card border border-slate-700/80 dark:border-slate-700/80 shadow-xl max-h-[28rem] overflow-hidden flex flex-col",children:[l.jsx("div",{className:"p-2 border-b border-slate-800/70",children:l.jsxs("div",{className:"relative",children:[l.jsx(Po,{size:16,className:"absolute left-3 top-1/2 -translate-y-1/2 text-slate-500"}),l.jsx("input",{ref:u,type:"text",value:i,onChange:p=>o(p.target.value),placeholder:"Search models...",className:"w-full pl-9 pr-3 py-2 bg-slate-900 border border-slate-800/70 rounded text-sm text-slate-100 placeholder-slate-500 focus:outline-none focus:border-orange-400/60"})]})}),l.jsxs("div",{className:"overflow-y-auto flex-1",children:[N.map(p=>l.jsxs("div",{children:[l.jsxs("div",{className:"px-3 py-1.5 text-xs font-medium text-slate-500 uppercase bg-slate-900 sticky top-0 flex items-center gap-2",children:[p,v(p)&&l.jsx(As,{})]}),y[p].map(d=>{const h=t.includes(d.id);return l.jsxs("button",{type:"button",onClick:()=>m(d.id),className:`w-full flex items-center gap-3 px-3 py-2 text-left transition-colors ${h?"bg-orange-500/20 text-orange-300":"hover:bg-slate-800 text-slate-100"}`,children:[l.jsx("div",{className:`w-4 h-4 rounded border flex items-center justify-center shrink-0 ${h?"bg-orange-400 border-orange-400":"border-slate-600"}`,children:h&&l.jsx(Xs,{size:12,className:"text-slate-900"})}),l.jsxs("div",{className:"flex-1 min-w-0 flex items-center gap-2",children:[l.jsx("div",{className:"truncate",children:d.name}),d.is_local&&!v(p)&&l.jsx(As,{})]}),l.jsx("span",{className:"text-xs text-slate-500 shrink-0",children:d.price})]},d.id)})]},p)),g.length===0&&l.jsx("div",{className:"px-3 py-4 text-center text-slate-500 text-sm",children:"No models found"})]})]})]})}function fu(){var Ho;const{scenarioId:e}=rp(),[t]=yg(),n=t.get("dataset"),s=parseInt(t.get("parallel")||"5"),[r,a]=x.useState([]),[i,o]=x.useState(null),[c,u]=x.useState(e||""),[m,f]=x.useState([]),[g,y]=x.useState(""),[j,v]=x.useState([]),[N,p]=x.useState(n?"dataset":"single"),[d,h]=x.useState(1),[k,w]=x.useState({}),[R,P]=x.useState(!0),[M,O]=x.useState(null),[z,A]=x.useState([]),[L,U]=x.useState(n||""),[B,C]=x.useState(null),[E,_]=x.useState(!1),[b,S]=x.useState(s),[T,D]=x.useState(!1),[H,te]=x.useState(!1),[X,lt]=x.useState(""),[Fe,rt]=x.useState(""),[Cn,pl]=x.useState(!0),{state:ae,result:Ot,comparison:Uo,error:Bo,runScenario:dm,compareModels:fm}=Wg();x.useEffect(()=>{(async()=>{try{const[Oe,He,xm]=await Promise.all([Z.listScenarios(),Z.listModels(),Z.listDatasets()]);if(a(Oe),f(He),A(xm),He.length>0&&y(He[0].id),e){u(e);const Vo=await Z.getScenario(e);o(Vo);const Wo={};for(const Yo of Vo.variables||[])Wo[Yo.name]=Yo.default??"";w(Wo)}}catch(Oe){O(Oe instanceof Error?Oe.message:"Failed to load data")}finally{P(!1)}})()},[e]),x.useEffect(()=>{c&&c!==e&&Z.getScenario(c).then(V=>{o(V);const Oe={};for(const He of V.variables||[])Oe[He.name]=He.default??"";w(Oe)}).catch(V=>O(V.message))},[c,e]);const pm=async()=>{const V=c||e;if(!V)return;const Oe=T?{enabled:!0,trackingUri:X||void 0,experiment:Fe||void 0,tracing:Cn}:void 0;if(N==="dataset"){if(!L||!g)return;_(!0),C(null);try{const He=await Z.runDataset({scenario_id:V,dataset_id:L,model:g,parallel:b,mlflow_enabled:T,mlflow_experiment:Fe||void 0});C(He)}catch(He){O(He instanceof Error?He.message:"Dataset run failed")}finally{_(!1)}}else if(N==="single"){if(!g)return;await dm(V,g,k,Oe)}else{if(j.length===0)return;await fm(V,j,d,k,Oe)}},mm=(V,Oe)=>{w(He=>({...He,[V]:Oe}))};return R?l.jsx("div",{className:"flex items-center justify-center h-full",children:l.jsx(cu,{className:"w-8 h-8 animate-spin text-orange-400"})}):M?l.jsx("div",{className:"p-8 page",children:l.jsx("div",{className:"panel-solid p-4 text-red-400 border border-red-700/60",children:M})}):l.jsxs("div",{className:"p-8 max-w-6xl mx-auto page",children:[l.jsxs(de,{to:"/",className:"flex items-center gap-2 text-slate-400 hover:text-slate-100 mb-6",children:[l.jsx(Gs,{size:20}),"Back to Dashboard"]}),l.jsxs("div",{className:"mb-8",children:[l.jsx("h1",{className:"text-2xl font-semibold text-slate-100 mb-2",children:(i==null?void 0:i.name)||e}),(i==null?void 0:i.description)&&l.jsx("p",{className:"text-slate-400",children:i.description})]}),l.jsxs("div",{className:"panel-card p-6 mb-6",children:[l.jsxs("div",{className:"flex items-center justify-between mb-4",children:[l.jsx("h2",{className:"text-lg font-semibold text-slate-100",children:"Run Configuration"}),l.jsxs(de,{to:"/builder",className:"flex items-center gap-2 text-slate-400 hover:text-slate-100 text-sm",children:[l.jsx(Og,{size:16}),"Edit Scenario"]})]}),!e&&l.jsxs("div",{className:"mb-6",children:[l.jsx("label",{className:"block text-sm font-medium text-slate-400 mb-2",children:"Scenario"}),l.jsxs("select",{value:c,onChange:V=>u(V.target.value),disabled:ae==="running"||E,className:"w-full panel-subtle px-4 py-2 text-slate-100 focus:outline-none focus:ring-2 focus:ring-orange-400",children:[l.jsx("option",{value:"",children:"Select a scenario..."}),r.map(V=>l.jsx("option",{value:V.id,children:V.name},V.id))]})]}),l.jsxs("div",{className:"mb-6",children:[l.jsx("label",{className:"block text-sm font-medium text-slate-400 mb-2",children:"Mode"}),l.jsxs("div",{className:"flex gap-2",children:[l.jsx("button",{onClick:()=>p("single"),disabled:ae==="running"||E,className:`px-4 py-2 rounded-lg transition-colors ${N==="single"?"bg-orange-400 text-slate-900":"bg-slate-800 text-slate-300 hover:bg-slate-700"}`,children:"Single Model"}),l.jsx("button",{onClick:()=>p("compare"),disabled:ae==="running"||E,className:`px-4 py-2 rounded-lg transition-colors ${N==="compare"?"bg-orange-400 text-slate-900":"bg-slate-800 text-slate-300 hover:bg-slate-700"}`,children:"Compare Models"}),l.jsxs("button",{onClick:()=>p("dataset"),disabled:ae==="running"||E||z.length===0,className:`flex items-center gap-2 px-4 py-2 rounded-lg transition-colors ${N==="dataset"?"bg-blue-500 text-white":"bg-slate-800 text-slate-300 hover:bg-slate-700"}`,children:[l.jsx(sn,{size:16}),"Dataset Benchmark"]})]})]}),N==="single"?l.jsxs("div",{className:"mb-6",children:[l.jsx("label",{className:"block text-sm font-medium text-slate-400 mb-2",children:"Model"}),l.jsx(du,{models:m,value:g,onChange:y,disabled:ae==="running"})]}):N==="compare"?l.jsxs("div",{className:"space-y-4 mb-6",children:[l.jsxs("div",{children:[l.jsx("label",{className:"block text-sm font-medium text-slate-400 mb-2",children:"Models to Compare"}),l.jsx(Kg,{models:m,selected:j,onChange:v,disabled:ae==="running"})]}),l.jsxs("div",{children:[l.jsx("label",{className:"block text-sm font-medium text-slate-400 mb-2",children:"Runs per Model"}),l.jsx("input",{type:"number",min:1,max:10,value:d,onChange:V=>h(Math.max(1,Math.min(10,parseInt(V.target.value)||1))),disabled:ae==="running",className:"w-32 panel-subtle px-4 py-2 text-slate-100 focus:outline-none focus:ring-2 focus:ring-orange-400"}),l.jsx("p",{className:"mt-1 text-xs text-slate-500",children:"More runs = more statistical significance"})]})]}):l.jsxs("div",{className:"space-y-4 mb-6",children:[l.jsxs("div",{children:[l.jsx("label",{className:"block text-sm font-medium text-slate-400 mb-2",children:"Dataset"}),l.jsxs("select",{value:L,onChange:V=>U(V.target.value),disabled:E,className:"w-full panel-subtle px-4 py-2 text-slate-100 focus:outline-none focus:ring-2 focus:ring-blue-400",children:[l.jsx("option",{value:"",children:"Select a dataset..."}),z.map(V=>l.jsxs("option",{value:V.id,children:[V.name," (",V.case_count," cases)"]},V.id))]}),z.length===0&&l.jsxs("p",{className:"mt-1 text-xs text-slate-500",children:["No datasets found. ",l.jsx(de,{to:"/datasets",className:"text-blue-400 hover:underline",children:"Create one"})]})]}),l.jsxs("div",{children:[l.jsx("label",{className:"block text-sm font-medium text-slate-400 mb-2",children:"Model"}),l.jsx(du,{models:m,value:g,onChange:y,disabled:E})]}),l.jsxs("div",{children:[l.jsx("label",{className:"block text-sm font-medium text-slate-400 mb-2",children:"Parallel Runs"}),l.jsxs("div",{className:"flex items-center gap-3",children:[l.jsx("input",{type:"number",min:1,max:20,value:b,onChange:V=>S(Math.max(1,Math.min(20,parseInt(V.target.value)||1))),disabled:E,className:"w-20 panel-subtle px-3 py-2 text-slate-100 focus:outline-none focus:ring-2 focus:ring-blue-400"}),l.jsx("span",{className:"text-sm text-slate-500",children:"concurrent cases"})]}),l.jsx("p",{className:"mt-1 text-xs text-slate-500",children:"Higher = faster, but may hit rate limits"})]})]}),i&&i.variables&&i.variables.length>0&&l.jsxs("div",{className:"mb-6 p-4 panel-subtle",children:[l.jsxs("div",{className:"flex items-center gap-2 mb-4",children:[l.jsx(yp,{size:18,className:"text-slate-400"}),l.jsx("h3",{className:"font-medium text-slate-100",children:"Variables"})]}),l.jsx("div",{className:"grid grid-cols-1 md:grid-cols-2 gap-4",children:i.variables.map(V=>l.jsx(Xg,{variable:V,value:k[V.name],onChange:Oe=>mm(V.name,Oe),disabled:ae==="running"},V.name))})]}),l.jsxs("div",{className:"mb-6 p-4 panel-subtle",children:[l.jsxs("button",{onClick:()=>te(!H),className:"flex items-center gap-2 w-full text-left",children:[H?l.jsx(pe,{size:18}):l.jsx(ge,{size:18}),l.jsx("span",{className:"font-medium text-slate-100",children:"MLflow Tracking"}),T&&l.jsx("span",{className:"ml-2 px-2 py-0.5 text-xs bg-green-500/20 text-green-400 rounded",children:"Enabled"})]}),H&&l.jsxs("div",{className:"mt-4 space-y-4",children:[l.jsxs("label",{className:"flex items-center gap-3 cursor-pointer",children:[l.jsx("input",{type:"checkbox",checked:T,onChange:V=>D(V.target.checked),disabled:ae==="running"||E,className:"w-4 h-4 rounded border-slate-600 text-orange-400 focus:ring-orange-400"}),l.jsx("span",{className:"text-slate-200",children:"Enable MLflow tracking"})]}),T&&l.jsxs(l.Fragment,{children:[N!=="dataset"&&l.jsxs("div",{children:[l.jsx("label",{className:"block text-sm font-medium text-slate-400 mb-1",children:"Tracking URI"}),l.jsx("input",{type:"text",value:X,onChange:V=>lt(V.target.value),disabled:ae==="running",placeholder:"http://127.0.0.1:5000 (uses MLFLOW_TRACKING_URI if empty)",className:"w-full panel-subtle px-3 py-2 text-slate-100 text-sm focus:outline-none focus:ring-2 focus:ring-orange-400"})]}),l.jsxs("div",{children:[l.jsx("label",{className:"block text-sm font-medium text-slate-400 mb-1",children:"Experiment Name"}),l.jsx("input",{type:"text",value:Fe,onChange:V=>rt(V.target.value),disabled:ae==="running"||E,placeholder:N==="dataset"?`${(i==null?void 0:i.name)||"scenario"}-dataset`:(i==null?void 0:i.name)||"Defaults to scenario name",className:"w-full panel-subtle px-3 py-2 text-slate-100 text-sm focus:outline-none focus:ring-2 focus:ring-orange-400"})]}),N!=="dataset"&&l.jsxs("label",{className:"flex items-center gap-3 cursor-pointer",children:[l.jsx("input",{type:"checkbox",checked:Cn,onChange:V=>pl(V.target.checked),disabled:ae==="running",className:"w-4 h-4 rounded border-slate-600 text-orange-400 focus:ring-orange-400"}),l.jsxs("div",{children:[l.jsx("span",{className:"text-slate-200",children:"Enable LLM Tracing"}),l.jsx("p",{className:"text-xs text-slate-500",children:"Capture detailed traces of each LLM call"})]})]}),l.jsxs("p",{className:"text-xs text-slate-500 flex items-center gap-1",children:[l.jsx(_g,{size:12}),"View results at your MLflow server after the run completes"]})]})]})]}),l.jsx("button",{onClick:pm,disabled:ae==="running"||E||!c||(N==="single"?!g:N==="compare"?j.length===0:!L||!g),className:`flex items-center gap-2 ${N==="dataset"?"bg-blue-500 hover:bg-blue-400":"bg-orange-400 hover:bg-orange-300"} disabled:bg-slate-700 disabled:text-slate-400 disabled:cursor-not-allowed text-slate-900 px-6 py-2 rounded-lg transition-colors font-semibold`,children:ae==="running"||E?l.jsxs(l.Fragment,{children:[l.jsx(cu,{className:"w-4 h-4 animate-spin"}),"Running..."]}):l.jsxs(l.Fragment,{children:[N==="dataset"?l.jsx(sn,{className:"w-4 h-4"}):l.jsx(Xt,{className:"w-4 h-4"}),N==="single"?"Run Scenario":N==="compare"?`Compare ${j.length} Models`:`Run ${((Ho=z.find(V=>V.id===L))==null?void 0:Ho.case_count)||0} Cases`]})})]}),Bo&&l.jsx("div",{className:"panel-solid border border-red-700/60 p-4 mb-6 text-red-400",children:l.jsxs("div",{className:"flex items-center gap-2",children:[l.jsx(Eo,{className:"w-5 h-5"}),l.jsx("span",{children:Bo})]})}),Ot&&l.jsx(Mo,{result:Ot}),Uo&&l.jsx(Np,{comparison:Uo}),B&&l.jsx(Gg,{result:B})]})}function Gg({result:e}){const[t,n]=x.useState(new Set),s=r=>{const a=new Set(t);a.has(r)?a.delete(r):a.add(r),n(a)};return l.jsxs("div",{className:"panel-card p-6",children:[l.jsxs("h2",{className:"text-xl font-semibold text-slate-100 mb-4 flex items-center gap-2",children:[l.jsx(sn,{size:20}),"Dataset Benchmark Results"]}),l.jsxs("div",{className:"grid grid-cols-2 md:grid-cols-4 gap-4 mb-6",children:[l.jsxs("div",{className:"panel-subtle p-4 rounded-lg",children:[l.jsxs("div",{className:"text-2xl font-bold text-slate-100",children:[e.passed_cases,"/",e.total_cases]}),l.jsx("div",{className:"text-sm text-slate-400",children:"Cases Passed"})]}),l.jsxs("div",{className:"panel-subtle p-4 rounded-lg",children:[l.jsxs("div",{className:`text-2xl font-bold ${e.pass_rate>=.9?"text-green-400":e.pass_rate>=.7?"text-yellow-400":"text-red-400"}`,children:[(e.pass_rate*100).toFixed(1),"%"]}),l.jsx("div",{className:"text-sm text-slate-400",children:"Pass Rate"})]}),l.jsxs("div",{className:"panel-subtle p-4 rounded-lg",children:[l.jsxs("div",{className:"text-2xl font-bold text-slate-100",children:[e.avg_percentage.toFixed(1),"%"]}),l.jsx("div",{className:"text-sm text-slate-400",children:"Avg Score"})]}),l.jsxs("div",{className:"panel-subtle p-4 rounded-lg",children:[l.jsxs("div",{className:"text-2xl font-bold text-slate-100",children:[(e.total_time_ms/1e3).toFixed(1),"s"]}),l.jsx("div",{className:"text-sm text-slate-400",children:"Total Time"})]})]}),Object.keys(e.by_expected).length>0&&l.jsxs("div",{className:"mb-6",children:[l.jsx("h3",{className:"font-medium text-slate-300 mb-3",children:"By Expected Outcome"}),l.jsx("div",{className:"grid grid-cols-2 md:grid-cols-3 gap-2",children:Object.entries(e.by_expected).map(([r,a])=>{const i=a.passed+a.failed,o=i>0?a.passed/i:0;return l.jsxs("div",{className:"panel-subtle p-3 rounded",children:[l.jsx("div",{className:"font-medium text-slate-200",children:r}),l.jsxs("div",{className:"text-sm text-slate-400",children:[a.passed,"/",i," (",(o*100).toFixed(0),"%)"]})]},r)})})]}),l.jsxs("div",{children:[l.jsxs("h3",{className:"font-medium text-slate-300 mb-3",children:["Case Results (",e.failed_cases," failed)"]}),l.jsx("div",{className:"space-y-2 max-h-[40vh] overflow-y-auto",children:e.case_results.map(r=>l.jsxs("div",{className:`panel-subtle rounded-lg overflow-hidden ${r.passed?"":"border border-red-500/30"}`,children:[l.jsxs("button",{onClick:()=>s(r.case_id),className:"w-full px-4 py-3 flex items-center gap-3 text-left hover:bg-slate-700/50 transition-colors",children:[r.passed?l.jsx(Xs,{size:16,className:"text-green-400"}):l.jsx(Nt,{size:16,className:"text-red-400"}),l.jsx("span",{className:"font-medium text-slate-200",children:r.case_id}),r.expected&&l.jsxs("span",{className:"text-sm text-slate-400",children:["expected: ",r.expected]}),l.jsxs("span",{className:"ml-auto text-sm text-slate-400",children:[r.percentage.toFixed(0),"% | ",r.latency_ms,"ms"]})]}),t.has(r.case_id)&&l.jsxs("div",{className:"px-4 pb-3 pt-1 border-t border-slate-700",children:[r.failure_reason&&l.jsx("div",{className:"text-red-400 text-sm mb-2",children:r.failure_reason}),l.jsxs("div",{className:"grid grid-cols-2 gap-4 text-sm",children:[l.jsxs("div",{children:[l.jsx("span",{className:"text-slate-400",children:"Expected:"})," ",l.jsx("span",{className:"text-slate-200",children:r.expected||"N/A"})]}),l.jsxs("div",{children:[l.jsx("span",{className:"text-slate-400",children:"Actual:"})," ",l.jsx("span",{className:"text-slate-200",children:r.actual_outcome||"None"})]}),l.jsxs("div",{children:[l.jsx("span",{className:"text-slate-400",children:"Score:"})," ",l.jsxs("span",{className:"text-slate-200",children:[r.goal_score,"/",r.max_score]})]}),l.jsxs("div",{children:[l.jsx("span",{className:"text-slate-400",children:"Latency:"})," ",l.jsxs("span",{className:"text-slate-200",children:[r.latency_ms,"ms"]})]})]})]})]},r.case_id))})]})]})}function Xg({variable:e,value:t,onChange:n,disabled:s}){const r="w-full panel-subtle px-3 py-2 text-slate-100 text-sm focus:outline-none focus:ring-2 focus:ring-orange-400 disabled:opacity-50";return l.jsxs("div",{children:[l.jsxs("label",{className:"block text-sm font-medium text-slate-400 mb-1",children:[e.label,e.required&&l.jsx("span",{className:"text-red-400 ml-1",children:"*"})]}),e.type==="select"&&e.options.length>0?l.jsxs("select",{value:String(t??""),onChange:a=>n(a.target.value),disabled:s,className:r,children:[l.jsx("option",{value:"",children:"Select..."}),e.options.map(a=>l.jsx("option",{value:a,children:a},a))]}):e.type==="boolean"?l.jsxs("select",{value:String(t??""),onChange:a=>n(a.target.value==="true"),disabled:s,className:r,children:[l.jsx("option",{value:"",children:"Select..."}),l.jsx("option",{value:"true",children:"True"}),l.jsx("option",{value:"false",children:"False"})]}):e.type==="number"?l.jsx("input",{type:"number",value:String(t??""),onChange:a=>n(a.target.value?Number(a.target.value):""),disabled:s,placeholder:e.default!==null?`Default: ${e.default}`:"",className:r}):l.jsx("input",{type:"text",value:String(t??""),onChange:a=>n(a.target.value),disabled:s,placeholder:e.default!==null?`Default: ${e.default}`:"",className:r})]})}const pu=e=>e.scenario_id.endsWith("_comparison")||e.filename.includes("_comparison_"),mu=e=>e.scenario_id.includes("_dataset_")||e.filename.includes("_dataset_"),Al=e=>e.replace(/_comparison$/,"").replace(/_dataset_.*$/,"");function Zg(){const[e,t]=x.useState([]),[n,s]=x.useState(!0),[r,a]=x.useState(null),[i,o]=x.useState(null),[c,u]=x.useState(null),[m,f]=x.useState(!1),[g,y]=x.useState(!1),[j,v]=x.useState(""),[N,p]=x.useState("all"),[d,h]=x.useState(new Set),k=C=>{h(E=>{const _=new Set(E);return _.has(C)?_.delete(C):_.add(C),_})},{filteredResults:w,groupedResults:R,scenarioOrder:P}=x.useMemo(()=>{let C=e.filter(b=>{if(j){const S=j.toLowerCase();if(!b.scenario_id.toLowerCase().includes(S))return!1}if(N!=="all"){const S=pu(b),T=mu(b);if(N==="comparison"&&!S||N==="dataset"&&!T||N==="single"&&(S||T))return!1}return!0});C=C.sort((b,S)=>new Date(S.timestamp).getTime()-new Date(b.timestamp).getTime());const E={},_=[];for(const b of C){const S=Al(b.scenario_id);E[S]||(E[S]=[],_.push(S)),E[S].push(b)}return{filteredResults:C,groupedResults:E,scenarioOrder:_}},[e,j,N]);x.useEffect(()=>{fetch("/api/v1/local/runs").then(C=>{if(!C.ok)throw new Error("Failed to fetch results");return C.json()}).then(t).catch(C=>a(C.message)).finally(()=>s(!1))},[]);const M=async C=>{try{const E=await fetch(`/api/v1/local/runs/${encodeURIComponent(C)}`);if(!E.ok)throw new Error("Failed to fetch result");const _=await E.json();o(_),u(C),f(!1)}catch(E){console.error("Error fetching result:",E)}},O=async()=>{if(i)try{await navigator.clipboard.writeText(JSON.stringify(i,null,2)),y(!0),setTimeout(()=>y(!1),2e3)}catch(C){console.error("Failed to copy:",C)}},z=C=>{var D,H,te,X,lt;if(!C)return"unknown";const E=C.case_results||((D=C.result)==null?void 0:D.case_results),_=C.dataset_id||((H=C.result)==null?void 0:H.dataset_id);if(E||_)return"dataset";if(C.ranking||((te=C.result)==null?void 0:te.ranking))return"comparison";const S=C.model||((X=C.result)==null?void 0:X.model),T=C.response||((lt=C.result)==null?void 0:lt.response);return S||T!==void 0?"single":"unknown"},A=C=>{const E=C.result||C;return{id:E.id||"",scenario_id:E.scenario_id||C.scenario_id||"",model:E.model||"",response:E.response||"",history:E.history||[],tool_calls:E.tool_calls||[],final_state:E.final_state||{},evaluation:E.evaluation||null,latency_ms:E.latency_ms||0,input_tokens:E.input_tokens||0,output_tokens:E.output_tokens||0,cost_usd:E.cost_usd||null,error:E.error||null}},L=C=>{const E=C.result||C;return{scenario_id:E.scenario_id||C.scenario_id||"",scenario_name:E.scenario_name||C.scenario_id||"",models:E.models||[],runs_per_model:E.runs_per_model||1,stats:E.stats||{},ranking:E.ranking||[],winner:E.winner||null,results:E.results||[]}},U=C=>{const E=C.result||C;return{scenario_id:E.scenario_id||C.scenario_id||"",model:E.model||"",dataset_id:E.dataset_id||"",total_cases:E.total_cases||0,passed_cases:E.passed_cases||0,failed_cases:E.failed_cases||0,pass_rate:E.pass_rate||0,avg_score:E.avg_score||0,avg_percentage:E.avg_percentage||0,by_expected:E.by_expected||{},total_time_ms:E.total_time_ms||0,case_results:E.case_results||[]}};if(n)return l.jsx("div",{className:"p-8 page",children:l.jsx("div",{className:"text-slate-400",children:"Loading..."})});if(r)return l.jsx("div",{className:"p-8 page",children:l.jsxs("div",{className:"text-red-400",children:["Error: ",r]})});const B=z(i);return l.jsxs("div",{className:"p-8 page",children:[l.jsx("h1",{className:"text-2xl font-semibold text-slate-100 mb-6",children:"Run Results"}),e.length===0?l.jsxs("div",{className:"panel-card p-6 text-center",children:[l.jsx(Hr,{size:48,className:"mx-auto text-slate-600 mb-4"}),l.jsx("p",{className:"text-slate-400 mb-2",children:"No run results found"}),l.jsx("p",{className:"text-slate-500 text-sm mb-4",children:"Run a scenario to see results here."}),l.jsxs(de,{to:"/",className:"inline-flex items-center gap-2 text-orange-300 hover:text-orange-200",children:[l.jsx(Xt,{size:16}),"Go to Dashboard"]})]}):l.jsxs("div",{className:"grid grid-cols-1 lg:grid-cols-3 gap-6",children:[l.jsxs("div",{className:"lg:col-span-1 space-y-4",children:[l.jsxs("div",{className:"space-y-3",children:[l.jsxs("div",{className:"relative",children:[l.jsx(Po,{size:16,className:"absolute left-3 top-1/2 -translate-y-1/2 text-slate-500"}),l.jsx("input",{type:"text",placeholder:"Search scenarios...",value:j,onChange:C=>v(C.target.value),className:"w-full bg-slate-900/60 border border-slate-700/50 rounded-lg pl-10 pr-4 py-2 text-sm text-slate-100 placeholder-slate-500 focus:outline-none focus:border-orange-400/50"})]}),l.jsxs("div",{className:"flex flex-wrap gap-1 p-1 bg-slate-900/60 rounded-lg border border-slate-700/50",children:[l.jsx("button",{onClick:()=>p("all"),className:`flex-1 min-w-0 px-2 py-1.5 text-xs font-medium rounded-md transition-colors ${N==="all"?"bg-slate-700/80 text-slate-100":"text-slate-400 hover:text-slate-200"}`,children:"All"}),l.jsxs("button",{onClick:()=>p("single"),className:`flex-1 min-w-0 px-2 py-1.5 text-xs font-medium rounded-md transition-colors flex items-center justify-center gap-1 ${N==="single"?"bg-slate-700/80 text-slate-100":"text-slate-400 hover:text-slate-200"}`,children:[l.jsx(bn,{size:12}),"Single"]}),l.jsxs("button",{onClick:()=>p("comparison"),className:`flex-1 min-w-0 px-2 py-1.5 text-xs font-medium rounded-md transition-colors flex items-center justify-center gap-1 ${N==="comparison"?"bg-slate-700/80 text-slate-100":"text-slate-400 hover:text-slate-200"}`,children:[l.jsx(ou,{size:12}),"Compare"]}),l.jsxs("button",{onClick:()=>p("dataset"),className:`flex-1 min-w-0 px-2 py-1.5 text-xs font-medium rounded-md transition-colors flex items-center justify-center gap-1 ${N==="dataset"?"bg-blue-600/80 text-slate-100":"text-slate-400 hover:text-slate-200"}`,children:[l.jsx(sn,{size:12}),"Dataset"]})]})]}),l.jsxs("div",{className:"flex items-center justify-between",children:[l.jsxs("h2",{className:"text-sm font-medium text-slate-400",children:[w.length," result",w.length!==1?"s":""," in ",P.length," scenario",P.length!==1?"s":""]}),j&&l.jsx("button",{onClick:()=>v(""),className:"text-xs text-slate-500 hover:text-slate-300",children:"Clear"})]}),l.jsx("div",{className:"space-y-2 max-h-[70vh] overflow-y-auto pr-2",children:P.length===0?l.jsxs("div",{className:"text-center py-8 text-slate-500",children:[l.jsx(Rg,{size:24,className:"mx-auto mb-2 opacity-50"}),l.jsx("p",{className:"text-sm",children:"No results match your filters"})]}):P.map(C=>{const E=R[C],_=d.has(C);return l.jsxs("div",{className:"panel-card overflow-hidden",children:[l.jsxs("button",{onClick:()=>k(C),className:"w-full flex items-center justify-between p-3 hover:bg-slate-800/40 transition-colors",children:[l.jsxs("div",{className:"flex items-center gap-2",children:[_?l.jsx(ge,{size:16,className:"text-slate-500"}):l.jsx(pe,{size:16,className:"text-slate-500"}),l.jsx("span",{className:"font-medium text-slate-200 text-sm",children:C})]}),l.jsx("span",{className:"text-xs text-slate-500 bg-slate-800/60 px-2 py-0.5 rounded",children:E.length})]}),!_&&l.jsx("div",{className:"border-t border-slate-800/50",children:E.map(b=>{const S=pu(b),T=mu(b);return l.jsxs("div",{className:`flex items-center justify-between p-3 cursor-pointer transition-colors border-l-2 ${c===b.filename?"bg-slate-800/60 border-l-orange-400":"hover:bg-slate-800/40 border-l-transparent"}`,onClick:()=>M(b.filename),children:[l.jsxs("div",{className:"flex items-center gap-2 min-w-0",children:[T?l.jsx(sn,{size:14,className:"text-blue-400 flex-shrink-0"}):S?l.jsx(ou,{size:14,className:"text-purple-400 flex-shrink-0"}):l.jsx(bn,{size:14,className:"text-orange-400 flex-shrink-0"}),l.jsx("div",{className:"min-w-0",children:l.jsxs("p",{className:"text-xs text-slate-400 flex items-center gap-1",children:[l.jsx(_o,{size:10}),new Date(b.timestamp).toLocaleString()]})})]}),l.jsx(de,{to:`/run/${Al(b.scenario_id)}`,onClick:D=>D.stopPropagation(),className:"p-1.5 text-slate-500 hover:text-orange-300 hover:bg-slate-700/60 rounded flex-shrink-0",title:"Run again",children:l.jsx(Xt,{size:12})})]},b.filename)})})]},C)})})]}),l.jsx("div",{className:"lg:col-span-2",children:i?l.jsxs("div",{className:"space-y-4",children:[l.jsx("div",{className:"panel-card p-4",children:l.jsxs("div",{className:"flex items-center justify-between gap-4",children:[l.jsxs("div",{className:"min-w-0 flex-1",children:[l.jsx("h2",{className:"font-semibold text-slate-100 truncate",children:Al(i.scenario_id)}),B==="single"&&A(i).model&&l.jsxs("p",{className:"text-sm text-slate-400 mt-1",children:["Model: ",A(i).model]}),B==="comparison"&&l.jsx("p",{className:"text-sm text-slate-400 mt-1",children:"Model Comparison"}),B==="dataset"&&l.jsxs("p",{className:"text-sm text-slate-400 mt-1",children:["Dataset: ",U(i).dataset_id," | Model: ",U(i).model]})]}),l.jsxs("div",{className:"flex items-center gap-2 flex-shrink-0",children:[l.jsxs("button",{onClick:O,className:`flex items-center gap-1 text-xs px-2 py-1 border rounded whitespace-nowrap transition-colors ${g?"text-emerald-400 border-emerald-500/50 bg-emerald-500/10":"text-slate-400 hover:text-slate-100 border-slate-700/70"}`,children:[l.jsx(mp,{size:12}),g?"Copied!":"Copy JSON"]}),l.jsx("button",{onClick:()=>f(!m),className:"text-xs text-slate-400 hover:text-slate-100 px-2 py-1 border border-slate-700/70 rounded whitespace-nowrap",children:m?"View Details":"View JSON"}),l.jsxs(de,{to:`/run/${Al(i.scenario_id)}`,className:"flex items-center gap-1 bg-orange-400 hover:bg-orange-300 text-slate-900 px-3 py-1 rounded text-sm font-semibold whitespace-nowrap",children:[l.jsx(Xt,{size:14}),"Run Again"]})]})]})}),m?l.jsx("div",{className:"panel-card p-4",children:l.jsx("pre",{className:"text-xs text-slate-300 bg-slate-950/70 p-3 rounded overflow-auto max-h-[70vh]",children:JSON.stringify(i,null,2)})}):l.jsxs(l.Fragment,{children:[B==="single"&&l.jsx(Mo,{result:A(i)}),B==="comparison"&&l.jsx(Np,{comparison:L(i)}),B==="dataset"&&l.jsx(ev,{result:U(i)}),B==="unknown"&&l.jsx("div",{className:"panel-card p-6 text-center",children:l.jsx("p",{className:"text-slate-500",children:'Could not parse result format. Click "View JSON" to see raw data.'})})]})]}):l.jsxs("div",{className:"panel-card p-8 text-center",children:[l.jsx(xp,{size:32,className:"mx-auto text-slate-600 mb-3"}),l.jsx("p",{className:"text-slate-400",children:"Select a result to view details"})]})})]})]})}function ev({result:e}){const[t,n]=x.useState(new Set),s=o=>{const c=new Set(t);c.has(o)?c.delete(o):c.add(o),n(c)},r=e.passed_cases,a=e.total_cases,i=e.pass_rate;return l.jsxs("div",{className:"space-y-4",children:[l.jsxs("div",{className:"grid grid-cols-2 md:grid-cols-4 gap-3",children:[l.jsxs("div",{className:"panel-card p-4",children:[l.jsxs("div",{className:"flex items-center gap-2 text-slate-400 mb-2",children:[l.jsx(sn,{className:"w-4 h-4"}),"Cases Passed"]}),l.jsxs("div",{className:`text-2xl font-bold ${i>=.9?"text-emerald-300":i>=.7?"text-amber-300":"text-red-400"}`,children:[r,"/",a]}),l.jsxs("div",{className:"text-xs text-slate-500",children:[(i*100).toFixed(1),"% pass rate"]})]}),l.jsxs("div",{className:"panel-card p-4",children:[l.jsxs("div",{className:"flex items-center gap-2 text-slate-400 mb-2",children:[l.jsx(Xs,{className:"w-4 h-4"}),"Avg Score"]}),l.jsxs("div",{className:"text-2xl font-bold text-slate-100",children:[e.avg_percentage.toFixed(1),"%"]}),l.jsxs("div",{className:"text-xs text-slate-500",children:[e.avg_score.toFixed(1)," points"]})]}),l.jsxs("div",{className:"panel-card p-4",children:[l.jsxs("div",{className:"flex items-center gap-2 text-slate-400 mb-2",children:[l.jsx(_o,{className:"w-4 h-4"}),"Total Time"]}),l.jsxs("div",{className:"text-2xl font-bold text-slate-100",children:[(e.total_time_ms/1e3).toFixed(1),"s"]}),l.jsxs("div",{className:"text-xs text-slate-500",children:[Math.round(e.total_time_ms/Math.max(a,1)),"ms/case"]})]}),l.jsxs("div",{className:"panel-card p-4",children:[l.jsxs("div",{className:"flex items-center gap-2 text-slate-400 mb-2",children:[l.jsx(Nt,{className:"w-4 h-4"}),"Failed"]}),l.jsx("div",{className:"text-2xl font-bold text-red-400",children:e.failed_cases}),l.jsxs("div",{className:"text-xs text-slate-500",children:[a-r," case",a-r!==1?"s":""," failed"]})]})]}),Object.keys(e.by_expected).length>0&&l.jsxs("div",{className:"panel-card p-6",children:[l.jsx("h3",{className:"font-semibold text-slate-100 mb-4",children:"By Expected Outcome"}),l.jsx("div",{className:"grid grid-cols-2 md:grid-cols-3 gap-3",children:Object.entries(e.by_expected).map(([o,c])=>{const u=c.passed+c.failed,m=u>0?c.passed/u:0;return l.jsxs("div",{className:"panel-subtle p-3 rounded-lg",children:[l.jsx("div",{className:"font-medium text-slate-200",children:o}),l.jsxs("div",{className:"flex items-center gap-2 mt-1",children:[l.jsxs("div",{className:`text-lg font-semibold ${m>=.9?"text-emerald-300":m>=.7?"text-amber-300":"text-red-400"}`,children:[c.passed,"/",u]}),l.jsxs("span",{className:"text-xs text-slate-500",children:["(",(m*100).toFixed(0),"%)"]})]})]},o)})})]}),l.jsxs("div",{className:"panel-card p-6",children:[l.jsxs("h3",{className:"font-semibold text-slate-100 mb-4",children:["Case Results (",e.failed_cases," failed, ",e.passed_cases," passed)"]}),l.jsx("div",{className:"space-y-2 max-h-[50vh] overflow-y-auto",children:e.case_results.map(o=>l.jsxs("div",{className:`panel-subtle rounded-lg overflow-hidden ${o.passed?"":"border border-red-500/30"}`,children:[l.jsxs("button",{onClick:()=>s(o.case_id),className:"w-full px-4 py-3 flex items-center gap-3 text-left hover:bg-slate-700/50 transition-colors",children:[o.passed?l.jsx(Xs,{size:16,className:"text-emerald-400 flex-shrink-0"}):l.jsx(Nt,{size:16,className:"text-red-400 flex-shrink-0"}),l.jsx("span",{className:"font-medium text-slate-200",children:o.case_id}),o.expected&&o.expected.length>0&&l.jsxs("span",{className:"text-sm text-slate-400",children:["expected: ",o.expected.join(" or ")]}),l.jsxs("span",{className:"ml-auto text-sm text-slate-400 flex-shrink-0",children:[o.percentage.toFixed(0),"% | ",o.latency_ms,"ms"]}),t.has(o.case_id)?l.jsx(pe,{size:14,className:"text-slate-500 flex-shrink-0"}):l.jsx(ge,{size:14,className:"text-slate-500 flex-shrink-0"})]}),t.has(o.case_id)&&l.jsxs("div",{className:"px-4 pb-3 pt-1 border-t border-slate-700",children:[o.failure_reason&&l.jsx("div",{className:"text-red-400 text-sm mb-2 p-2 bg-red-950/30 rounded",children:o.failure_reason}),l.jsxs("div",{className:"grid grid-cols-2 gap-4 text-sm",children:[l.jsxs("div",{children:[l.jsx("span",{className:"text-slate-500",children:"Expected:"})," ",l.jsx("span",{className:"text-slate-200",children:o.expected&&o.expected.length>0?o.expected.join(" or "):"N/A"})]}),l.jsxs("div",{children:[l.jsx("span",{className:"text-slate-500",children:"Actual:"})," ",l.jsx("span",{className:"text-slate-200",children:o.actual_outcome||"None"})]}),l.jsxs("div",{children:[l.jsx("span",{className:"text-slate-500",children:"Score:"})," ",l.jsxs("span",{className:"text-slate-200",children:[o.goal_score,"/",o.max_score]})]}),l.jsxs("div",{children:[l.jsx("span",{className:"text-slate-500",children:"Latency:"})," ",l.jsxs("span",{className:"text-slate-200",children:[o.latency_ms,"ms"]})]})]})]})]},o.case_id))})]})]})}/*! js-yaml 4.1.1 https://github.com/nodeca/js-yaml @license MIT */function wp(e){return typeof e>"u"||e===null}function tv(e){return typeof e=="object"&&e!==null}function nv(e){return Array.isArray(e)?e:wp(e)?[]:[e]}function sv(e,t){var n,s,r,a;if(t)for(a=Object.keys(t),n=0,s=a.length;n<s;n+=1)r=a[n],e[r]=t[r];return e}function lv(e,t){var n="",s;for(s=0;s<t;s+=1)n+=e;return n}function rv(e){return e===0&&Number.NEGATIVE_INFINITY===1/e}var av=wp,iv=tv,ov=nv,cv=lv,uv=rv,dv=sv,me={isNothing:av,isObject:iv,toArray:ov,repeat:cv,isNegativeZero:uv,extend:dv};function bp(e,t){var n="",s=e.reason||"(unknown reason)";return e.mark?(e.mark.name&&(n+='in "'+e.mark.name+'" '),n+="("+(e.mark.line+1)+":"+(e.mark.column+1)+")",!t&&e.mark.snippet&&(n+=`
 
 `+e.mark.snippet),s+" "+n):s}function Zs(e,t){Error.call(this),this.name="YAMLException",this.reason=e,this.mark=t,this.message=bp(this,!1),Error.captureStackTrace?Error.captureStackTrace(this,this.constructor):this.stack=new Error().stack||""}Zs.prototype=Object.create(Error.prototype);Zs.prototype.constructor=Zs;Zs.prototype.toString=function(t){return this.name+": "+bp(this,t)};var Te=Zs;function ja(e,t,n,s,r){var a="",i="",o=Math.floor(r/2)-1;return s-t>o&&(a=" ... ",t=s-o+a.length),n-s>o&&(i=" ...",n=s+o-i.length),{str:a+e.slice(t,n).replace(/\t/g,"→")+i,pos:s-t+a.length}}function Na(e,t){return me.repeat(" ",t-e.length)+e}function fv(e,t){if(t=Object.create(t||null),!e.buffer)return null;t.maxLength||(t.maxLength=79),typeof t.indent!="number"&&(t.indent=1),typeof t.linesBefore!="number"&&(t.linesBefore=3),typeof t.linesAfter!="number"&&(t.linesAfter=2);for(var n=/\r?\n|\r|\0/g,s=[0],r=[],a,i=-1;a=n.exec(e.buffer);)r.push(a.index),s.push(a.index+a[0].length),e.position<=a.index&&i<0&&(i=s.length-2);i<0&&(i=s.length-1);var o="",c,u,m=Math.min(e.line+t.linesAfter,r.length).toString().length,f=t.maxLength-(t.indent+m+3);for(c=1;c<=t.linesBefore&&!(i-c<0);c++)u=ja(e.buffer,s[i-c],r[i-c],e.position-(s[i]-s[i-c]),f),o=me.repeat(" ",t.indent)+Na((e.line-c+1).toString(),m)+" | "+u.str+`
 `+o;for(u=ja(e.buffer,s[i],r[i],e.position,f),o+=me.repeat(" ",t.indent)+Na((e.line+1).toString(),m)+" | "+u.str+`

sandboxy/ui/dist/index.html

@@ -5,7 +5,7 @@
     <link rel="icon" type="image/svg+xml" href="/vite.svg" />
     <meta name="viewport" content="width=device-width, initial-scale=1.0" />
     <title>Sandboxy Local</title>
-    <script type="module" crossorigin src="/assets/index-BZFjoK-_.js"></script>
+    <script type="module" crossorigin src="/assets/index-DfKu4uXt.js"></script>
     <link rel="stylesheet" crossorigin href="/assets/index-Qf7gGJk_.css">
   </head>
   <body>

{sandboxy-0.0.6.dist-info → sandboxy-0.0.7.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: sandboxy
-Version: 0.0.6
+Version: 0.0.7
 Summary: Open-source agent simulation and benchmarking platform
 Project-URL: Homepage, https://github.com/sandboxy-ai/sandboxy
 Project-URL: Repository, https://github.com/sandboxy-ai/sandboxy

{sandboxy-0.0.6.dist-info → sandboxy-0.0.7.dist-info}/RECORD

@@ -7,14 +7,14 @@ sandboxy/agents/base.py,sha256=LQQCb0nSoh9ubrDhj5_wsDuwZxA2lz7w3x4Vfk7EPiw,1828
 sandboxy/agents/llm_prompt.py,sha256=yIPIztjxl8yqMkHVo3V0PRSCoCsye1K6sF7cUUk_Ik8,13174
 sandboxy/agents/loader.py,sha256=iZJpsrI1o-2lmpWcRnq1gjCYmvCn19mJkhx8U3Y4B3c,6839
 sandboxy/api/__init__.py,sha256=3zw70CGbz3j8rK-8XcuGDx8KNibLfuFuvVEG02OIhek,119
-sandboxy/api/app.py,sha256=Xzipe24Ce6YcGbq2rCgK0W_Du8tQyW4iRfr1IsMRo7M,2181
+sandboxy/api/app.py,sha256=lbEFKRJUdrgCtkMbshJvDFMXk1zwSG6pFueUbuVGsXE,2716
 sandboxy/api/routes/__init__.py,sha256=CZS3yxK1CtW80ER3mbPqXbGN19HahbLC0Lu4A-8dw8o,31
 sandboxy/api/routes/agents.py,sha256=6LWJT_IAZrRRhBAzReSNsAI9eAp2oCKP7-5LvkZyeGw,2511
-sandboxy/api/routes/local.py,sha256=9DOX1i3zMLbHupziWvvAgDsL0z55KDh7B267dFIJNS0,50245
+sandboxy/api/routes/local.py,sha256=lbvlIO-Tp4n9LFc_0iP9x3qqyvq8PCRWnfq23DfynQ8,50759
 sandboxy/api/routes/providers.py,sha256=PZ5dWcog63iR-BoPQ7IME5VnIy5DqKZpQjLCtlPltrk,10894
 sandboxy/api/routes/tools.py,sha256=TPhHBIyo7NNSHWDaaanDfAF1bzqYHT_k1NJ96w8Dw0c,3308
 sandboxy/cli/__init__.py,sha256=WMN-u5rqs1uFuWArGP3crwz2_BXnZg_n6ShWZSSt68g,56
-sandboxy/cli/main.py,sha256=rBmddwJJZqecbew54uADfUhItlXJYxn_zRWNxFLUH2c,63400
+sandboxy/cli/main.py,sha256=549QVPGr3JxHdRVCU3QHuk7it6NbFCdnA8oXFEhy97E,63526
 sandboxy/cli/type_detector.py,sha256=WH_xxtmeqeqoYH6oy3voa0_pVL4kjbuRPb3oamaIG-k,1352
 sandboxy/core/__init__.py,sha256=OiBPYPUHF1Jw7qzYHHWqVubTijrQ9D_kazmdYnclHG0,62
 sandboxy/core/async_runner.py,sha256=YsAg3WHnMDwyhenpEl5udLxpVcvkAvYEYH1PAMo70JI,28878
@@ -37,7 +37,7 @@ sandboxy/mlflow/exporter.py,sha256=u5QvEsVI23ENWlSVWHG9iTgS23kZgYkihACCviBOxMk,1
 sandboxy/mlflow/metrics.py,sha256=JHrXQ8dHVGrmdNigTwUFDxf3C-3C5KorDowBuJn2Sw0,3263
 sandboxy/mlflow/tags.py,sha256=KIABLNoKekLv90AeUkVNDTfNVsnGPawKo2YFCiY7Of8,3779
 sandboxy/mlflow/tracing.py,sha256=aYr1ONFK2YNT4C0-qT9IwhsOZAJcDfFSCUJkT4eNOyw,3407
-sandboxy/providers/__init__.py,sha256=l5CbFKx8-yjMCmCsc_tDMWDk5-9xS1a-MzUZpHW5gF0,1713
+sandboxy/providers/__init__.py,sha256=VEaRpxZutP_3cBubJpDLQVOLCfnpoFpzR0_aG6--IZU,1755
 sandboxy/providers/anthropic_provider.py,sha256=vwUBd_wRZGlu-R1w4iXkpphpwwS53h0KxoqRdMk_tgk,8779
 sandboxy/providers/base.py,sha256=a3umHX2Jz_pYlvrgaTdWJzd_wk1LS4bu0XXf9g4omrM,3215
 sandboxy/providers/config.py,sha256=XckJjkG_AO016__ySGi2w7n9fpHv0U6bXUV4_BU6xvY,7013
@@ -45,7 +45,7 @@ sandboxy/providers/http_client.py,sha256=9iFfXG-XeqDm6iKNSzxvmjWkTcffnpkTnct--Tk
 sandboxy/providers/local.py,sha256=ytv7WIyS-SseQucef_p9eLxvKCY-XfNsy7u_4VMYRt4,16482
 sandboxy/providers/openai_provider.py,sha256=G2HKW_GV5jXCMXFb7m1zCiW4JHbcstlN1e5P-trFEg4,8676
 sandboxy/providers/openrouter.py,sha256=cQ9ddfrCNYlXMgiggVQN2BsMTUPtK_ulqoMBJQYvhu4,31932
-sandboxy/providers/registry.py,sha256=fnvtvznqk26LzB4VUYoC1hR1FidBY9oR4Vu8K45meCs,10120
+sandboxy/providers/registry.py,sha256=kpnzRkB6qRrmGCaZpA4ptkHBxRuYZ_Ma4l9gyBbnM64,10930
 sandboxy/scenarios/__init__.py,sha256=B8gOx-bCsl8TEFEcI1r9PC490NepprWE7_MauwDs5IQ,308
 sandboxy/scenarios/comparison.py,sha256=DaRmnACdltRdvSEb-VRKT-Wgj-f_bDotiBomUaC7wvw,16426
 sandboxy/scenarios/loader.py,sha256=YFqxV-87ngxqSog7kxxViqMqrMqPi4e49HCIH8dfMG0,9193
@@ -58,13 +58,13 @@ sandboxy/tools/base.py,sha256=LD7lygRk42MwOEHGf0pCOwFYsfuxOhkKhy3omdjF8fU,3728
 sandboxy/tools/loader.py,sha256=9QSqsexIYtRYcHWp6iyq551mYdODtYhKgR7Ld7vc_vM,8228
 sandboxy/tools/yaml_tools.py,sha256=CPJcVclGfUIERLqXO6KgJhEQudmr4UJIGbwQzAkaB0s,24184
 sandboxy/ui/__init__.py,sha256=iSb6TFb3CjM7chzVl--4cmPaIEOLtXghXChxaR0FJ_Y,608
-sandboxy/ui/dist/index.html,sha256=8e1lYQgl4ciFAN1ug7CSQtHc9DVk5nzSVKbT-jNPDZ8,461
-sandboxy/ui/dist/assets/index-BZFjoK-_.js,sha256=_mMAnyQ4uW7MTr86ViLAwXeyabEYtwljxDizOnYBY1k,415223
+sandboxy/ui/dist/index.html,sha256=yrXTkaNJWUVDKTrL7dMuud_cPsDgNB-zeomybPbtILc,461
+sandboxy/ui/dist/assets/index-DfKu4uXt.js,sha256=4fuf7nK9K6yfT-n19z4zOudC8ITskNud1OxK9Wgc3Qk,415258
 sandboxy/ui/dist/assets/index-Qf7gGJk_.css,sha256=QDuvLJqf7qe5HySqaBcixPBx6EOfWpC9CzhU8U29Wzg,31191
 sandboxy/utils/__init__.py,sha256=lT3hC8XALceDRwUHPInw7bRGodueKQ8JHMgQebPCIxw,61
 sandboxy/utils/time.py,sha256=9w5JEBA_t4mTdThUFBnY4gxzwSYIXcy0QWDIy8j-Xy4,511
-sandboxy-0.0.6.dist-info/METADATA,sha256=jI33tk-QHZg8spboPow_aTytPVznwwmUJLbuZ0Svmh0,8646
-sandboxy-0.0.6.dist-info/WHEEL,sha256=WLgqFyCfm_KASv4WHyYy0P3pM_m7J5L9k2skdKLirC8,87
-sandboxy-0.0.6.dist-info/entry_points.txt,sha256=RkY2kImLhO74KTbwWB58DO_7pisdAx3GZBoY1BJ1VdE,98
-sandboxy-0.0.6.dist-info/licenses/LICENSE,sha256=xx0jnfkXJvxRnG63LTGOxlggYnIysveWIZ6H3PNdCrQ,11357
-sandboxy-0.0.6.dist-info/RECORD,,
+sandboxy-0.0.7.dist-info/METADATA,sha256=7kX_dk4r7SknMxM8633hSHOX1B182hRoWqCyHcTif0Y,8646
+sandboxy-0.0.7.dist-info/WHEEL,sha256=WLgqFyCfm_KASv4WHyYy0P3pM_m7J5L9k2skdKLirC8,87
+sandboxy-0.0.7.dist-info/entry_points.txt,sha256=RkY2kImLhO74KTbwWB58DO_7pisdAx3GZBoY1BJ1VdE,98
+sandboxy-0.0.7.dist-info/licenses/LICENSE,sha256=xx0jnfkXJvxRnG63LTGOxlggYnIysveWIZ6H3PNdCrQ,11357
+sandboxy-0.0.7.dist-info/RECORD,,