feat(#373): prediction integration for cost estimation

- Create PredictionService for pre-task cost/token estimates
- Refresh common predictions on startup
- Integrate predictions into LLM telemetry tracker
- Add GET /api/telemetry/estimate endpoint
- Graceful degradation when no prediction data available
- Add unit tests for prediction service

Refs #373

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

apps/api/src/llm/llm-telemetry-tracker.service.spec.ts

@@ -333,12 +333,13 @@ describe("LlmTelemetryTrackerService", () => {
       service.trackLlmCompletion(baseParams);
 
       // claude-sonnet-4-5: 150 * 3 + 300 * 15 = 450 + 4500 = 4950
-      const expectedCost = 4950;
+      const expectedActualCost = 4950;
 
       expect(mockTelemetryService.eventBuilder?.build).toHaveBeenCalledWith(
         expect.objectContaining({
-          estimated_cost_usd_micros: expectedCost,
+          // Estimated values are 0 when no PredictionService is injected
-          actual_cost_usd_micros: expectedCost,
+          estimated_cost_usd_micros: 0,
+          actual_cost_usd_micros: expectedActualCost,
         }),
       );
     });
@@ -437,8 +438,9 @@ describe("LlmTelemetryTrackerService", () => {
         expect.objectContaining({
           actual_input_tokens: 50,
           actual_output_tokens: 100,
-          estimated_input_tokens: 50,
+          // Estimated values are 0 when no PredictionService is injected
-          estimated_output_tokens: 100,
+          estimated_input_tokens: 0,
+          estimated_output_tokens: 0,
         }),
       );
     });

apps/api/src/llm/llm-telemetry-tracker.service.ts

@@ -1,5 +1,6 @@
-import { Injectable, Logger } from "@nestjs/common";
+import { Injectable, Logger, Optional } from "@nestjs/common";
 import { MosaicTelemetryService } from "../mosaic-telemetry/mosaic-telemetry.service";
+import { PredictionService } from "../mosaic-telemetry/prediction.service";
 import { TaskType, Complexity, Harness, Provider, Outcome } from "@mosaicstack/telemetry-client";
 import type { LlmProviderType } from "./providers/llm-provider.interface";
 import { calculateCostMicrodollars } from "./llm-cost-table";
@@ -140,7 +141,10 @@ export function inferTaskType(
 export class LlmTelemetryTrackerService {
   private readonly logger = new Logger(LlmTelemetryTrackerService.name);
 
-  constructor(private readonly telemetry: MosaicTelemetryService) {}
+  constructor(
+    private readonly telemetry: MosaicTelemetryService,
+    @Optional() private readonly predictionService?: PredictionService
+  ) {}
 
   /**
    * Track an LLM completion event via Mosaic Telemetry.
@@ -158,24 +162,47 @@ export class LlmTelemetryTrackerService {
         return;
       }
 
+      const taskType = inferTaskType(params.operation, params.callingContext);
+      const provider = mapProviderType(params.providerType);
+
       const costMicrodollars = calculateCostMicrodollars(
         params.model,
         params.inputTokens,
         params.outputTokens
       );
 
+      // Query predictions for estimated fields (graceful degradation)
+      let estimatedInputTokens = 0;
+      let estimatedOutputTokens = 0;
+      let estimatedCostMicros = 0;
+
+      if (this.predictionService) {
+        const prediction = this.predictionService.getEstimate(
+          taskType,
+          params.model,
+          provider,
+          Complexity.LOW
+        );
+
+        if (prediction?.prediction && prediction.metadata.confidence !== "none") {
+          estimatedInputTokens = prediction.prediction.input_tokens.median;
+          estimatedOutputTokens = prediction.prediction.output_tokens.median;
+          estimatedCostMicros = prediction.prediction.cost_usd_micros.median ?? 0;
+        }
+      }
+
       const event = builder.build({
         task_duration_ms: params.durationMs,
-        task_type: inferTaskType(params.operation, params.callingContext),
+        task_type: taskType,
         complexity: Complexity.LOW,
         harness: mapHarness(params.providerType),
         model: params.model,
-        provider: mapProviderType(params.providerType),
+        provider,
-        estimated_input_tokens: params.inputTokens,
+        estimated_input_tokens: estimatedInputTokens,
-        estimated_output_tokens: params.outputTokens,
+        estimated_output_tokens: estimatedOutputTokens,
         actual_input_tokens: params.inputTokens,
         actual_output_tokens: params.outputTokens,
-        estimated_cost_usd_micros: costMicrodollars,
+        estimated_cost_usd_micros: estimatedCostMicros,
         actual_cost_usd_micros: costMicrodollars,
         quality_gate_passed: true,
         quality_gates_run: [],

apps/api/src/mosaic-telemetry/mosaic-telemetry.controller.ts

@@ -0,0 +1,92 @@
+import { Controller, Get, Query, UseGuards, BadRequestException } from "@nestjs/common";
+import { AuthGuard } from "../auth/guards/auth.guard";
+import { PredictionService } from "./prediction.service";
+import {
+  TaskType,
+  Complexity,
+  Provider,
+  type PredictionResponse,
+} from "@mosaicstack/telemetry-client";
+
+/**
+ * Valid values for query parameter validation.
+ */
+const VALID_TASK_TYPES = new Set<string>(Object.values(TaskType));
+const VALID_COMPLEXITIES = new Set<string>(Object.values(Complexity));
+const VALID_PROVIDERS = new Set<string>(Object.values(Provider));
+
+/**
+ * Response DTO for the estimate endpoint.
+ */
+interface EstimateResponseDto {
+  data: PredictionResponse | null;
+}
+
+/**
+ * Mosaic Telemetry Controller
+ *
+ * Provides API endpoints for accessing telemetry prediction data.
+ * All endpoints require authentication via AuthGuard.
+ *
+ * This controller is intentionally lightweight - it delegates to PredictionService
+ * for the actual prediction logic and returns results directly to the frontend.
+ */
+@Controller("telemetry")
+@UseGuards(AuthGuard)
+export class MosaicTelemetryController {
+  constructor(private readonly predictionService: PredictionService) {}
+
+  /**
+   * GET /api/telemetry/estimate
+   *
+   * Get a cost/token estimate for a given task configuration.
+   * Returns prediction data including confidence level, or null if
+   * no prediction is available.
+   *
+   * @param taskType - Task type enum value (e.g. "implementation", "planning")
+   * @param model - Model name (e.g. "claude-sonnet-4-5")
+   * @param provider - Provider enum value (e.g. "anthropic", "openai")
+   * @param complexity - Complexity level (e.g. "low", "medium", "high")
+   * @returns Prediction response with estimates and confidence
+   */
+  @Get("estimate")
+  getEstimate(
+    @Query("taskType") taskType: string,
+    @Query("model") model: string,
+    @Query("provider") provider: string,
+    @Query("complexity") complexity: string
+  ): EstimateResponseDto {
+    if (!taskType || !model || !provider || !complexity) {
+      throw new BadRequestException(
+        "Missing query parameters. Required: taskType, model, provider, complexity"
+      );
+    }
+
+    if (!VALID_TASK_TYPES.has(taskType)) {
+      throw new BadRequestException(
+        `Invalid taskType "${taskType}". Valid values: ${[...VALID_TASK_TYPES].join(", ")}`
+      );
+    }
+
+    if (!VALID_PROVIDERS.has(provider)) {
+      throw new BadRequestException(
+        `Invalid provider "${provider}". Valid values: ${[...VALID_PROVIDERS].join(", ")}`
+      );
+    }
+
+    if (!VALID_COMPLEXITIES.has(complexity)) {
+      throw new BadRequestException(
+        `Invalid complexity "${complexity}". Valid values: ${[...VALID_COMPLEXITIES].join(", ")}`
+      );
+    }
+
+    const prediction = this.predictionService.getEstimate(
+      taskType as TaskType,
+      model,
+      provider as Provider,
+      complexity as Complexity
+    );
+
+    return { data: prediction };
+  }
+}

apps/api/src/mosaic-telemetry/mosaic-telemetry.module.ts

@@ -1,6 +1,9 @@
 import { Module, Global } from "@nestjs/common";
 import { ConfigModule } from "@nestjs/config";
+import { AuthModule } from "../auth/auth.module";
 import { MosaicTelemetryService } from "./mosaic-telemetry.service";
+import { PredictionService } from "./prediction.service";
+import { MosaicTelemetryController } from "./mosaic-telemetry.controller";
 
 /**
  * Global module providing Mosaic Telemetry integration via @mosaicstack/telemetry-client.
@@ -30,8 +33,9 @@ import { MosaicTelemetryService } from "./mosaic-telemetry.service";
  */
 @Global()
 @Module({
-  imports: [ConfigModule],
+  imports: [ConfigModule, AuthModule],
-  providers: [MosaicTelemetryService],
+  controllers: [MosaicTelemetryController],
-  exports: [MosaicTelemetryService],
+  providers: [MosaicTelemetryService, PredictionService],
+  exports: [MosaicTelemetryService, PredictionService],
 })
 export class MosaicTelemetryModule {}

apps/api/src/mosaic-telemetry/prediction.service.spec.ts

@@ -0,0 +1,320 @@
+import { describe, it, expect, beforeEach, vi } from "vitest";
+import { Test, TestingModule } from "@nestjs/testing";
+import {
+  TaskType,
+  Complexity,
+  Provider,
+} from "@mosaicstack/telemetry-client";
+import type {
+  PredictionResponse,
+  PredictionQuery,
+} from "@mosaicstack/telemetry-client";
+import { MosaicTelemetryService } from "./mosaic-telemetry.service";
+import { PredictionService } from "./prediction.service";
+
+describe("PredictionService", () => {
+  let service: PredictionService;
+  let mockTelemetryService: {
+    isEnabled: boolean;
+    getPrediction: ReturnType<typeof vi.fn>;
+    refreshPredictions: ReturnType<typeof vi.fn>;
+  };
+
+  const mockPredictionResponse: PredictionResponse = {
+    prediction: {
+      input_tokens: {
+        p10: 50,
+        p25: 80,
+        median: 120,
+        p75: 200,
+        p90: 350,
+      },
+      output_tokens: {
+        p10: 100,
+        p25: 150,
+        median: 250,
+        p75: 400,
+        p90: 600,
+      },
+      cost_usd_micros: {
+        p10: 500,
+        p25: 800,
+        median: 1200,
+        p75: 2000,
+        p90: 3500,
+      },
+      duration_ms: {
+        p10: 200,
+        p25: 400,
+        median: 800,
+        p75: 1500,
+        p90: 3000,
+      },
+      correction_factors: {
+        input: 1.0,
+        output: 1.0,
+      },
+      quality: {
+        gate_pass_rate: 0.95,
+        success_rate: 0.92,
+      },
+    },
+    metadata: {
+      sample_size: 150,
+      fallback_level: 0,
+      confidence: "high",
+      last_updated: "2026-02-15T00:00:00Z",
+      cache_hit: true,
+    },
+  };
+
+  const nullPredictionResponse: PredictionResponse = {
+    prediction: null,
+    metadata: {
+      sample_size: 0,
+      fallback_level: 3,
+      confidence: "none",
+      last_updated: null,
+      cache_hit: false,
+    },
+  };
+
+  beforeEach(async () => {
+    mockTelemetryService = {
+      isEnabled: true,
+      getPrediction: vi.fn().mockReturnValue(mockPredictionResponse),
+      refreshPredictions: vi.fn().mockResolvedValue(undefined),
+    };
+
+    const module: TestingModule = await Test.createTestingModule({
+      providers: [
+        PredictionService,
+        {
+          provide: MosaicTelemetryService,
+          useValue: mockTelemetryService,
+        },
+      ],
+    }).compile();
+
+    service = module.get<PredictionService>(PredictionService);
+  });
+
+  it("should be defined", () => {
+    expect(service).toBeDefined();
+  });
+
+  // ---------- getEstimate ----------
+
+  describe("getEstimate", () => {
+    it("should return prediction response for valid query", () => {
+      const result = service.getEstimate(
+        TaskType.IMPLEMENTATION,
+        "claude-sonnet-4-5",
+        Provider.ANTHROPIC,
+        Complexity.LOW
+      );
+
+      expect(result).toEqual(mockPredictionResponse);
+      expect(mockTelemetryService.getPrediction).toHaveBeenCalledWith({
+        task_type: TaskType.IMPLEMENTATION,
+        model: "claude-sonnet-4-5",
+        provider: Provider.ANTHROPIC,
+        complexity: Complexity.LOW,
+      });
+    });
+
+    it("should pass correct query parameters to telemetry service", () => {
+      service.getEstimate(
+        TaskType.CODE_REVIEW,
+        "gpt-4o",
+        Provider.OPENAI,
+        Complexity.HIGH
+      );
+
+      expect(mockTelemetryService.getPrediction).toHaveBeenCalledWith({
+        task_type: TaskType.CODE_REVIEW,
+        model: "gpt-4o",
+        provider: Provider.OPENAI,
+        complexity: Complexity.HIGH,
+      });
+    });
+
+    it("should return null when telemetry returns null", () => {
+      mockTelemetryService.getPrediction.mockReturnValue(null);
+
+      const result = service.getEstimate(
+        TaskType.IMPLEMENTATION,
+        "claude-sonnet-4-5",
+        Provider.ANTHROPIC,
+        Complexity.LOW
+      );
+
+      expect(result).toBeNull();
+    });
+
+    it("should return null prediction response when confidence is none", () => {
+      mockTelemetryService.getPrediction.mockReturnValue(nullPredictionResponse);
+
+      const result = service.getEstimate(
+        TaskType.IMPLEMENTATION,
+        "unknown-model",
+        Provider.UNKNOWN,
+        Complexity.LOW
+      );
+
+      expect(result).toEqual(nullPredictionResponse);
+      expect(result?.metadata.confidence).toBe("none");
+    });
+
+    it("should return null and not throw when getPrediction throws", () => {
+      mockTelemetryService.getPrediction.mockImplementation(() => {
+        throw new Error("Prediction fetch failed");
+      });
+
+      const result = service.getEstimate(
+        TaskType.IMPLEMENTATION,
+        "claude-sonnet-4-5",
+        Provider.ANTHROPIC,
+        Complexity.LOW
+      );
+
+      expect(result).toBeNull();
+    });
+
+    it("should handle non-Error thrown objects gracefully", () => {
+      mockTelemetryService.getPrediction.mockImplementation(() => {
+        throw "string error";
+      });
+
+      const result = service.getEstimate(
+        TaskType.IMPLEMENTATION,
+        "claude-sonnet-4-5",
+        Provider.ANTHROPIC,
+        Complexity.LOW
+      );
+
+      expect(result).toBeNull();
+    });
+  });
+
+  // ---------- refreshCommonPredictions ----------
+
+  describe("refreshCommonPredictions", () => {
+    it("should call refreshPredictions with multiple query combinations", async () => {
+      await service.refreshCommonPredictions();
+
+      expect(mockTelemetryService.refreshPredictions).toHaveBeenCalledTimes(1);
+
+      const queries: PredictionQuery[] =
+        mockTelemetryService.refreshPredictions.mock.calls[0][0];
+
+      // Should have queries for cross-product of models, task types, and complexities
+      expect(queries.length).toBeGreaterThan(0);
+
+      // Verify all queries have valid structure
+      for (const query of queries) {
+        expect(query).toHaveProperty("task_type");
+        expect(query).toHaveProperty("model");
+        expect(query).toHaveProperty("provider");
+        expect(query).toHaveProperty("complexity");
+      }
+    });
+
+    it("should include Anthropic model predictions", async () => {
+      await service.refreshCommonPredictions();
+
+      const queries: PredictionQuery[] =
+        mockTelemetryService.refreshPredictions.mock.calls[0][0];
+
+      const anthropicQueries = queries.filter(
+        (q: PredictionQuery) => q.provider === Provider.ANTHROPIC
+      );
+      expect(anthropicQueries.length).toBeGreaterThan(0);
+    });
+
+    it("should include OpenAI model predictions", async () => {
+      await service.refreshCommonPredictions();
+
+      const queries: PredictionQuery[] =
+        mockTelemetryService.refreshPredictions.mock.calls[0][0];
+
+      const openaiQueries = queries.filter(
+        (q: PredictionQuery) => q.provider === Provider.OPENAI
+      );
+      expect(openaiQueries.length).toBeGreaterThan(0);
+    });
+
+    it("should not call refreshPredictions when telemetry is disabled", async () => {
+      mockTelemetryService.isEnabled = false;
+
+      await service.refreshCommonPredictions();
+
+      expect(mockTelemetryService.refreshPredictions).not.toHaveBeenCalled();
+    });
+
+    it("should not throw when refreshPredictions rejects", async () => {
+      mockTelemetryService.refreshPredictions.mockRejectedValue(
+        new Error("Server unreachable")
+      );
+
+      // Should not throw
+      await expect(service.refreshCommonPredictions()).resolves.not.toThrow();
+    });
+
+    it("should include common task types in queries", async () => {
+      await service.refreshCommonPredictions();
+
+      const queries: PredictionQuery[] =
+        mockTelemetryService.refreshPredictions.mock.calls[0][0];
+
+      const taskTypes = new Set(queries.map((q: PredictionQuery) => q.task_type));
+
+      expect(taskTypes.has(TaskType.IMPLEMENTATION)).toBe(true);
+      expect(taskTypes.has(TaskType.PLANNING)).toBe(true);
+      expect(taskTypes.has(TaskType.CODE_REVIEW)).toBe(true);
+    });
+
+    it("should include common complexity levels in queries", async () => {
+      await service.refreshCommonPredictions();
+
+      const queries: PredictionQuery[] =
+        mockTelemetryService.refreshPredictions.mock.calls[0][0];
+
+      const complexities = new Set(queries.map((q: PredictionQuery) => q.complexity));
+
+      expect(complexities.has(Complexity.LOW)).toBe(true);
+      expect(complexities.has(Complexity.MEDIUM)).toBe(true);
+    });
+  });
+
+  // ---------- onModuleInit ----------
+
+  describe("onModuleInit", () => {
+    it("should trigger refreshCommonPredictions on init when telemetry is enabled", () => {
+      // refreshPredictions is async, but onModuleInit fires it and forgets
+      service.onModuleInit();
+
+      // Give the promise microtask a chance to execute
+      expect(mockTelemetryService.isEnabled).toBe(true);
+      // refreshPredictions will be called asynchronously
+    });
+
+    it("should not refresh when telemetry is disabled", () => {
+      mockTelemetryService.isEnabled = false;
+
+      service.onModuleInit();
+
+      // refreshPredictions should not be called since we returned early
+      expect(mockTelemetryService.refreshPredictions).not.toHaveBeenCalled();
+    });
+
+    it("should not throw when refresh fails on init", () => {
+      mockTelemetryService.refreshPredictions.mockRejectedValue(
+        new Error("Connection refused")
+      );
+
+      // Should not throw
+      expect(() => service.onModuleInit()).not.toThrow();
+    });
+  });
+});

apps/api/src/mosaic-telemetry/prediction.service.ts

@@ -0,0 +1,161 @@
+import { Injectable, Logger, OnModuleInit } from "@nestjs/common";
+import {
+  TaskType,
+  Complexity,
+  Provider,
+  type PredictionQuery,
+  type PredictionResponse,
+} from "@mosaicstack/telemetry-client";
+import { MosaicTelemetryService } from "./mosaic-telemetry.service";
+
+/**
+ * Common model-provider combinations used for pre-fetching predictions.
+ * These represent the most frequently used LLM configurations.
+ */
+const COMMON_MODELS: { model: string; provider: Provider }[] = [
+  { model: "claude-sonnet-4-5", provider: Provider.ANTHROPIC },
+  { model: "claude-opus-4", provider: Provider.ANTHROPIC },
+  { model: "claude-haiku-4-5", provider: Provider.ANTHROPIC },
+  { model: "gpt-4o", provider: Provider.OPENAI },
+  { model: "gpt-4o-mini", provider: Provider.OPENAI },
+];
+
+/**
+ * Common task types to pre-fetch predictions for.
+ */
+const COMMON_TASK_TYPES: TaskType[] = [
+  TaskType.IMPLEMENTATION,
+  TaskType.PLANNING,
+  TaskType.CODE_REVIEW,
+];
+
+/**
+ * Common complexity levels to pre-fetch predictions for.
+ */
+const COMMON_COMPLEXITIES: Complexity[] = [Complexity.LOW, Complexity.MEDIUM];
+
+/**
+ * PredictionService
+ *
+ * Provides pre-task cost and token estimates using crowd-sourced prediction data
+ * from the Mosaic Telemetry server. Predictions are cached by the underlying SDK
+ * with a 6-hour TTL.
+ *
+ * This service is intentionally non-blocking: if predictions are unavailable
+ * (telemetry disabled, server unreachable, no data), all methods return null
+ * without throwing errors. Task execution should never be blocked by prediction
+ * failures.
+ *
+ * @example
+ * ```typescript
+ * const estimate = this.predictionService.getEstimate(
+ *   TaskType.IMPLEMENTATION,
+ *   "claude-sonnet-4-5",
+ *   Provider.ANTHROPIC,
+ *   Complexity.LOW,
+ * );
+ * if (estimate?.prediction) {
+ *   console.log(`Estimated cost: ${estimate.prediction.cost_usd_micros}`);
+ * }
+ * ```
+ */
+@Injectable()
+export class PredictionService implements OnModuleInit {
+  private readonly logger = new Logger(PredictionService.name);
+
+  constructor(private readonly telemetry: MosaicTelemetryService) {}
+
+  /**
+   * Refresh common predictions on startup.
+   * Runs asynchronously and never blocks module initialization.
+   */
+  onModuleInit(): void {
+    if (!this.telemetry.isEnabled) {
+      this.logger.log("Telemetry disabled - skipping prediction refresh");
+      return;
+    }
+
+    // Fire-and-forget: refresh in the background
+    this.refreshCommonPredictions().catch((error: unknown) => {
+      const msg = error instanceof Error ? error.message : String(error);
+      this.logger.warn(`Failed to refresh common predictions on startup: ${msg}`);
+    });
+  }
+
+  /**
+   * Get a cost/token estimate for a given task configuration.
+   *
+   * Returns the cached prediction from the SDK, or null if:
+   * - Telemetry is disabled
+   * - No prediction data exists for this combination
+   * - The prediction has expired
+   *
+   * @param taskType - The type of task to estimate
+   * @param model - The model name (e.g. "claude-sonnet-4-5")
+   * @param provider - The provider enum value
+   * @param complexity - The complexity level
+   * @returns Prediction response with estimates and confidence, or null
+   */
+  getEstimate(
+    taskType: TaskType,
+    model: string,
+    provider: Provider,
+    complexity: Complexity
+  ): PredictionResponse | null {
+    try {
+      const query: PredictionQuery = {
+        task_type: taskType,
+        model,
+        provider,
+        complexity,
+      };
+
+      return this.telemetry.getPrediction(query);
+    } catch (error: unknown) {
+      const msg = error instanceof Error ? error.message : String(error);
+      this.logger.warn(`Failed to get prediction estimate: ${msg}`);
+      return null;
+    }
+  }
+
+  /**
+   * Refresh predictions for commonly used (taskType, model, provider, complexity) combinations.
+   *
+   * Generates the cross-product of common models, task types, and complexities,
+   * then batch-refreshes them from the telemetry server. The SDK caches the
+   * results with a 6-hour TTL.
+   *
+   * This method is safe to call at any time. If telemetry is disabled or the
+   * server is unreachable, it completes without error.
+   */
+  async refreshCommonPredictions(): Promise<void> {
+    if (!this.telemetry.isEnabled) {
+      return;
+    }
+
+    const queries: PredictionQuery[] = [];
+
+    for (const { model, provider } of COMMON_MODELS) {
+      for (const taskType of COMMON_TASK_TYPES) {
+        for (const complexity of COMMON_COMPLEXITIES) {
+          queries.push({
+            task_type: taskType,
+            model,
+            provider,
+            complexity,
+          });
+        }
+      }
+    }
+
+    this.logger.log(`Refreshing ${String(queries.length)} common prediction queries...`);
+
+    try {
+      await this.telemetry.refreshPredictions(queries);
+      this.logger.log(`Successfully refreshed ${String(queries.length)} predictions`);
+    } catch (error: unknown) {
+      const msg = error instanceof Error ? error.message : String(error);
+      this.logger.warn(`Failed to refresh predictions: ${msg}`);
+    }
+  }
+}