@openhoo/hoopilot 1.1.0 → 1.3.0

package/dist/index.d.cts

@@ -15,6 +15,13 @@ declare class MetricsRegistry {
     startRequest(): void;
     /** Record a completed request and clear its in-flight slot. */
     observe(observation: RequestObservation): void;
+    /**
+     * Record whether one upstream completion reported token usage. `missing`
+     * counts responses that carried no usage object — most often streamed Chat
+     * Completions sent without `stream_options: {"include_usage": true}` — so a
+     * rising miss rate flags clients whose token usage is going unaccounted.
+     */
+    recordTokenExtraction(extracted: boolean): void;
     /** Accumulate token counts for a model from one upstream completion. */
     recordTokens(model: string, usage: TokenUsage): void;
     /** Record one upstream Copilot call and whether it succeeded. */
@@ -43,7 +50,7 @@ declare class MetricsRegistry {
  * branch; combined with the runtime cancelling the client branch, that releases
  * the shared upstream connection instead of draining it in the background.
  */
-declare function observeResponseUsage(response: Response, fallbackModel: string, onUsage: (model: string, usage: TokenUsage) => void, signal?: AbortSignal): Response;
+declare function observeResponseUsage(response: Response, fallbackModel: string, onUsage: (model: string, usage: TokenUsage) => void, signal?: AbortSignal, onOutcome?: (extracted: boolean) => void): Response;
 
 type FetchLike = (input: string | URL | Request, init?: RequestInit) => Promise<Response>;
 interface Logger {
@@ -186,10 +193,28 @@ interface GithubRateLimitSnapshot {
     retryAfterSeconds?: number;
     used?: number;
 }
+/** Request-latency summary for one route, in milliseconds. */
+interface RouteLatency {
+    avgMs: number;
+    count: number;
+}
+/**
+ * Aggregate request-latency summary derived from the duration histogram. `avgMs`
+ * is exact; the percentiles are estimated from the histogram buckets (Prometheus-
+ * style linear interpolation), so they are approximate.
+ */
+interface LatencySnapshot {
+    avgMs: number;
+    byRoute: Record<string, RouteLatency>;
+    count: number;
+    p50Ms: number;
+    p95Ms: number;
+}
 /** A point-in-time JSON view of the in-process metrics. */
 interface MetricsSnapshot {
     githubRateLimit: Record<string, GithubRateLimitSnapshot>;
     inFlight: number;
+    latency: LatencySnapshot;
     requests: {
         byRoute: Record<string, number>;
         byStatus: Record<string, number>;
@@ -200,6 +225,10 @@ interface MetricsSnapshot {
         byModel: Record<string, ModelTokenTotals>;
         cached: number;
         completion: number;
+        extraction: {
+            extracted: number;
+            missing: number;
+        };
         prompt: number;
         reasoning: number;
         total: number;

package/dist/index.d.ts

@@ -15,6 +15,13 @@ declare class MetricsRegistry {
     startRequest(): void;
     /** Record a completed request and clear its in-flight slot. */
     observe(observation: RequestObservation): void;
+    /**
+     * Record whether one upstream completion reported token usage. `missing`
+     * counts responses that carried no usage object — most often streamed Chat
+     * Completions sent without `stream_options: {"include_usage": true}` — so a
+     * rising miss rate flags clients whose token usage is going unaccounted.
+     */
+    recordTokenExtraction(extracted: boolean): void;
     /** Accumulate token counts for a model from one upstream completion. */
     recordTokens(model: string, usage: TokenUsage): void;
     /** Record one upstream Copilot call and whether it succeeded. */
@@ -43,7 +50,7 @@ declare class MetricsRegistry {
  * branch; combined with the runtime cancelling the client branch, that releases
  * the shared upstream connection instead of draining it in the background.
  */
-declare function observeResponseUsage(response: Response, fallbackModel: string, onUsage: (model: string, usage: TokenUsage) => void, signal?: AbortSignal): Response;
+declare function observeResponseUsage(response: Response, fallbackModel: string, onUsage: (model: string, usage: TokenUsage) => void, signal?: AbortSignal, onOutcome?: (extracted: boolean) => void): Response;
 
 type FetchLike = (input: string | URL | Request, init?: RequestInit) => Promise<Response>;
 interface Logger {
@@ -186,10 +193,28 @@ interface GithubRateLimitSnapshot {
     retryAfterSeconds?: number;
     used?: number;
 }
+/** Request-latency summary for one route, in milliseconds. */
+interface RouteLatency {
+    avgMs: number;
+    count: number;
+}
+/**
+ * Aggregate request-latency summary derived from the duration histogram. `avgMs`
+ * is exact; the percentiles are estimated from the histogram buckets (Prometheus-
+ * style linear interpolation), so they are approximate.
+ */
+interface LatencySnapshot {
+    avgMs: number;
+    byRoute: Record<string, RouteLatency>;
+    count: number;
+    p50Ms: number;
+    p95Ms: number;
+}
 /** A point-in-time JSON view of the in-process metrics. */
 interface MetricsSnapshot {
     githubRateLimit: Record<string, GithubRateLimitSnapshot>;
     inFlight: number;
+    latency: LatencySnapshot;
     requests: {
         byRoute: Record<string, number>;
         byStatus: Record<string, number>;
@@ -200,6 +225,10 @@ interface MetricsSnapshot {
         byModel: Record<string, ModelTokenTotals>;
         cached: number;
         completion: number;
+        extraction: {
+            extracted: number;
+            missing: number;
+        };
         prompt: number;
         reasoning: number;
         total: number;