feat: add runner failover rules and cache affinity

apps/api/internal/clients/simulation.go

@@ -107,6 +107,7 @@ func simulationProfile(request Request) string {
 }
 
 func simulatedResult(request Request) map[string]any {
+	usage := simulatedUsage(request)
 	switch request.Kind {
 	case "chat.completions":
 		return map[string]any{
@@ -122,7 +123,7 @@ func simulatedResult(request Request) map[string]any {
 					"content": fmt.Sprintf("simulation response from %s", request.Candidate.Provider),
 				},
 			}},
-			"usage": map[string]any{"prompt_tokens": 12, "completion_tokens": 8, "total_tokens": 20},
+			"usage": simulatedOpenAIUsageMap(usage),
 		}
 	case "responses":
 		return map[string]any{
@@ -131,7 +132,7 @@ func simulatedResult(request Request) map[string]any {
 			"created_at":  nowUnix(),
 			"model":       request.Model,
 			"output_text": fmt.Sprintf("simulation response from %s", request.Candidate.Provider),
-			"usage":       map[string]any{"input_tokens": 12, "output_tokens": 8, "total_tokens": 20},
+			"usage":       simulatedResponseUsageMap(usage),
 		}
 	case "embeddings":
 		return simulatedEmbeddingResult(request)
@@ -373,6 +374,9 @@ func simulatedAudioData(request Request, fallbackPrompt string) []any {
 }
 
 func simulatedUsage(request Request) Usage {
+	if usage, ok := simulationUsageOverride(request); ok {
+		return usage
+	}
 	if request.ModelType == "chat" || request.ModelType == "text_generate" || request.Kind == "responses" {
 		return Usage{InputTokens: 12, OutputTokens: 8, TotalTokens: 20}
 	}
@@ -382,6 +386,56 @@ func simulatedUsage(request Request) Usage {
 	return Usage{}
 }
 
+func simulationUsageOverride(request Request) (Usage, bool) {
+	raw, _ := request.Body["simulationUsage"].(map[string]any)
+	if len(raw) == 0 {
+		raw, _ = request.Body["testUsage"].(map[string]any)
+	}
+	if len(raw) == 0 {
+		return Usage{}, false
+	}
+	usage := Usage{
+		InputTokens:       intFromAny(firstPresent(raw["inputTokens"], raw["input_tokens"], raw["promptTokens"], raw["prompt_tokens"])),
+		OutputTokens:      intFromAny(firstPresent(raw["outputTokens"], raw["output_tokens"], raw["completionTokens"], raw["completion_tokens"])),
+		CachedInputTokens: intFromAny(firstPresent(raw["cachedInputTokens"], raw["cached_input_tokens"], raw["cachedPromptTokens"], raw["cached_tokens"])),
+		TotalTokens:       intFromAny(firstPresent(raw["totalTokens"], raw["total_tokens"])),
+	}
+	if _, ok := firstPresentValue(raw["cachedInputTokens"], raw["cached_input_tokens"], raw["cachedPromptTokens"], raw["cached_tokens"]); ok {
+		usage.CachedInputTokensKnown = true
+	}
+	if usage.TotalTokens == 0 {
+		usage.TotalTokens = usage.InputTokens + usage.OutputTokens
+	}
+	if usage.CachedInputTokens > usage.InputTokens && usage.InputTokens > 0 {
+		usage.CachedInputTokens = usage.InputTokens
+	}
+	return usage, true
+}
+
+func simulatedOpenAIUsageMap(usage Usage) map[string]any {
+	out := map[string]any{
+		"prompt_tokens":     usage.InputTokens,
+		"completion_tokens": usage.OutputTokens,
+		"total_tokens":      usage.TotalTokens,
+	}
+	if usage.CachedInputTokensKnown {
+		out["prompt_tokens_details"] = map[string]any{"cached_tokens": usage.CachedInputTokens}
+	}
+	return out
+}
+
+func simulatedResponseUsageMap(usage Usage) map[string]any {
+	out := map[string]any{
+		"input_tokens":  usage.InputTokens,
+		"output_tokens": usage.OutputTokens,
+		"total_tokens":  usage.TotalTokens,
+	}
+	if usage.CachedInputTokensKnown {
+		out["input_tokens_details"] = map[string]any{"cached_tokens": usage.CachedInputTokens}
+	}
+	return out
+}
+
 func simulatedProgress(request Request) []Progress {
 	provider := request.Candidate.Provider
 	if provider == "" {