@copilotkit/aimock 1.46.0 → 1.47.0

package/dist/sse-writer.cjs

@@ -1,13 +1,18 @@
+const require_frame_loop = require('./response/frame-loop.cjs');
 
 //#region src/sse-writer.ts
 function delay(ms, signal) {
 	if (ms <= 0 || signal?.aborted) return Promise.resolve();
 	return new Promise((resolve) => {
-		const timer = setTimeout(resolve, ms);
-		signal?.addEventListener("abort", () => {
+		const onAbort = () => {
 			clearTimeout(timer);
 			resolve();
-		}, { once: true });
+		};
+		const timer = setTimeout(() => {
+			signal?.removeEventListener("abort", onAbort);
+			resolve();
+		}, ms);
+		signal?.addEventListener("abort", onAbort, { once: true });
 	});
 }
 /** OpenRouter's SSE keepalive comment line (colon-prefixed, per SSE syntax). */
@@ -48,33 +53,45 @@ function calculateTotalDelay(contentLength, chunkSize, profile, recordedTimings,
 }
 async function writeSSEStream(res, chunks, optionsOrLatency) {
 	const opts = typeof optionsOrLatency === "number" ? { latency: optionsOrLatency } : optionsOrLatency ?? {};
-	const latency = opts.latency ?? 0;
-	const profile = opts.streamingProfile;
-	const { recordedTimings, replaySpeed } = opts;
-	const signal = opts.signal;
-	const onChunkSent = opts.onChunkSent;
-	if (res.writableEnded) return true;
-	res.setHeader("Content-Type", "text/event-stream");
-	res.setHeader("Cache-Control", "no-cache");
-	res.setHeader("Connection", "keep-alive");
-	if (opts.openRouterProcessing && !res.writableEnded) res.write(OPENROUTER_PROCESSING_COMMENT);
-	let chunkIndex = 0;
-	for (const chunk of chunks) {
-		const chunkDelay = calculateDelay(chunkIndex, profile, latency, recordedTimings, replaySpeed);
-		if (chunkDelay > 0) await delay(chunkDelay, signal);
-		if (signal?.aborted) return false;
-		if (res.writableEnded) return true;
-		res.write(`data: ${JSON.stringify(chunk)}\n\n`);
-		onChunkSent?.();
-		if (signal?.aborted) return false;
-		chunkIndex++;
-	}
-	if (!res.writableEnded) {
-		if (opts.usageChunk) res.write(`data: ${JSON.stringify(opts.usageChunk)}\n\n`);
-		res.write("data: [DONE]\n\n");
-		res.end();
-	}
-	return true;
+	return require_frame_loop.runFrameLoop(require_frame_loop.httpSink(res), chunks, sseCodec({
+		done: true,
+		usageChunk: opts.usageChunk,
+		openRouterProcessing: opts.openRouterProcessing
+	}), {
+		latency: opts.latency,
+		streamingProfile: opts.streamingProfile,
+		recordedTimings: opts.recordedTimings,
+		replaySpeed: opts.replaySpeed
+	}, {
+		signal: chunks.length > 0 ? opts.signal : void 0,
+		onChunkSent: opts.onChunkSent
+	});
+}
+const SSE_HEADERS = {
+	"Content-Type": "text/event-stream",
+	"Cache-Control": "no-cache",
+	Connection: "keep-alive"
+};
+function sseCodec(opts) {
+	return {
+		headers: SSE_HEADERS,
+		...opts.openRouterProcessing ? { prelude: () => OPENROUTER_PROCESSING_COMMENT } : {},
+		encode: (frame) => `data: ${JSON.stringify(frame)}\n\n`,
+		trailer: () => {
+			if (!opts.done) return [];
+			const parts = [];
+			if (opts.usageChunk) parts.push(`data: ${JSON.stringify(opts.usageChunk)}\n\n`);
+			parts.push("data: [DONE]\n\n");
+			return parts;
+		}
+	};
+}
+function typedSseCodec(opts) {
+	return {
+		headers: SSE_HEADERS,
+		encode: (frame) => `event: ${frame.type}\ndata: ${JSON.stringify(frame)}\n\n`,
+		trailer: () => opts?.terminalSentinel ? ["data: [DONE]\n\n"] : []
+	};
 }
 /**
 * Default rate-limit response headers matching OpenAI's format.
@@ -102,6 +119,8 @@ function writeErrorResponse(res, status, body, options) {
 exports.calculateDelay = calculateDelay;
 exports.calculateTotalDelay = calculateTotalDelay;
 exports.delay = delay;
+exports.sseCodec = sseCodec;
+exports.typedSseCodec = typedSseCodec;
 exports.writeErrorResponse = writeErrorResponse;
 exports.writeSSEStream = writeSSEStream;
 //# sourceMappingURL=sse-writer.cjs.map

package/dist/sse-writer.cjs.map

@@ -1 +1 @@
-{"version":3,"file":"sse-writer.cjs","names":[],"sources":["../src/sse-writer.ts"],"sourcesContent":["import type * as http from \"node:http\";\nimport type { SSEChunk, StreamingProfile, RecordedTimings } from \"./types.js\";\n\nexport function delay(ms: number, signal?: AbortSignal): Promise<void> {\n  if (ms <= 0 || signal?.aborted) return Promise.resolve();\n  return new Promise((resolve) => {\n    const timer = setTimeout(resolve, ms);\n    signal?.addEventListener(\n      \"abort\",\n      () => {\n        clearTimeout(timer);\n        resolve();\n      },\n      { once: true },\n    );\n  });\n}\n\nexport interface StreamOptions {\n  latency?: number;\n  streamingProfile?: StreamingProfile;\n  recordedTimings?: RecordedTimings;\n  replaySpeed?: number;\n  signal?: AbortSignal;\n  onChunkSent?: () => void;\n  /** When set, emitted as the final chunk before [DONE] (OpenAI stream_options.include_usage). */\n  usageChunk?: SSEChunk;\n  /**\n   * When true, a single OpenRouter `: OPENROUTER PROCESSING` SSE comment\n   * (keepalive) line is emitted FIRST — before the first data chunk — matching\n   * real OpenRouter bytes (or-capture/chat-stream.sse). A faithful client skips\n   * `:`-prefixed lines; opt-in, default off (so strict OpenAI parsers aren't\n   * surprised). Used only by the OpenRouter chat surface.\n   */\n  openRouterProcessing?: boolean;\n}\n\n/** OpenRouter's SSE keepalive comment line (colon-prefixed, per SSE syntax). */\nconst OPENROUTER_PROCESSING_COMMENT = \": OPENROUTER PROCESSING\\n\\n\";\n\nexport function calculateDelay(\n  chunkIndex: number,\n  profile?: StreamingProfile,\n  fallbackLatency?: number,\n  recordedTimings?: RecordedTimings,\n  replaySpeed?: number,\n): number {\n  const speed = replaySpeed ?? 1.0;\n  let delayMs: number;\n\n  if (profile) {\n    // StreamingProfile has highest precedence\n    let fromProfile = true;\n    if (chunkIndex === 0 && profile.ttft !== undefined) {\n      delayMs = profile.ttft;\n    } else if (profile.tps !== undefined && profile.tps > 0) {\n      delayMs = 1000 / profile.tps;\n    } else {\n      delayMs = fallbackLatency ?? 0;\n      fromProfile = false;\n    }\n    // Jitter only applies when the delay came from ttft/tps, not fallback\n    if (fromProfile && profile.jitter && profile.jitter > 0) {\n      delayMs *= 1 + (Math.random() * 2 - 1) * profile.jitter;\n      if (delayMs < 0) delayMs = 0;\n    }\n  } else if (recordedTimings) {\n    // Recorded timings (second precedence)\n    if (chunkIndex === 0) {\n      delayMs = recordedTimings.ttftMs;\n    } else {\n      const idx = chunkIndex - 1;\n      if (idx < recordedTimings.interChunkDelaysMs.length) {\n        delayMs = recordedTimings.interChunkDelaysMs[idx];\n      } else {\n        // Excess chunks: derive average from recorded inter-chunk delays\n        const totalInterChunk = recordedTimings.interChunkDelaysMs.reduce((a, b) => a + b, 0);\n        delayMs =\n          recordedTimings.interChunkDelaysMs.length > 0\n            ? totalInterChunk / recordedTimings.interChunkDelaysMs.length\n            : 0;\n      }\n    }\n  } else {\n    delayMs = fallbackLatency ?? 0;\n  }\n\n  delayMs = Math.max(0, delayMs);\n  return speed > 0 ? delayMs / speed : delayMs;\n}\n\n// Internal (not part of the public API): the total wait before a non-streaming\n// (`stream: false`) JSON response is written. Sums calculateDelay over the\n// content chunk count that `contentLength` at `chunkSize` would produce, so a\n// fixture with an explicit `streamingProfile` or `recordedTimings` pays roughly\n// the same ttft/tps cost as its streamed content chunks. It is an\n// approximation, not an exact match: a streamed reply also pays delays for its\n// role and finish chunks. There is deliberately no fallback-latency parameter:\n// bare `latency` / `--latency` is an inter-SSE-chunk gap and never delays a\n// non-streaming response. With neither a profile nor recorded timings, this\n// returns 0.\n// Kept as a line comment, not JSDoc: the bundled .d.ts drops this non-public\n// declaration but keeps a JSDoc block, which would then attach to writeSSEStream.\nexport function calculateTotalDelay(\n  contentLength: number,\n  chunkSize: number,\n  profile?: StreamingProfile,\n  recordedTimings?: RecordedTimings,\n  replaySpeed?: number,\n): number {\n  const chunkCount = Math.max(1, Math.ceil(contentLength / chunkSize));\n  let total = 0;\n  for (let chunkIndex = 0; chunkIndex < chunkCount; chunkIndex++) {\n    total += calculateDelay(chunkIndex, profile, undefined, recordedTimings, replaySpeed);\n  }\n  return total;\n}\n\nexport async function writeSSEStream(\n  res: http.ServerResponse,\n  chunks: SSEChunk[],\n  optionsOrLatency?: number | StreamOptions,\n): Promise<boolean> {\n  const opts: StreamOptions =\n    typeof optionsOrLatency === \"number\" ? { latency: optionsOrLatency } : (optionsOrLatency ?? {});\n  const latency = opts.latency ?? 0;\n  const profile = opts.streamingProfile;\n  const { recordedTimings, replaySpeed } = opts;\n  const signal = opts.signal;\n  const onChunkSent = opts.onChunkSent;\n\n  if (res.writableEnded) return true;\n  res.setHeader(\"Content-Type\", \"text/event-stream\");\n  res.setHeader(\"Cache-Control\", \"no-cache\");\n  res.setHeader(\"Connection\", \"keep-alive\");\n\n  // OpenRouter emits exactly ONE `: OPENROUTER PROCESSING` keepalive comment\n  // first — before any data frame (opt-in; see or-capture/chat-stream.sse). A\n  // faithful client discards `:`-prefixed lines before JSON-parsing a `data:`.\n  if (opts.openRouterProcessing && !res.writableEnded) {\n    res.write(OPENROUTER_PROCESSING_COMMENT);\n  }\n\n  let chunkIndex = 0;\n  for (const chunk of chunks) {\n    const chunkDelay = calculateDelay(chunkIndex, profile, latency, recordedTimings, replaySpeed);\n    if (chunkDelay > 0) {\n      await delay(chunkDelay, signal);\n    }\n    if (signal?.aborted) return false;\n    if (res.writableEnded) return true;\n    res.write(`data: ${JSON.stringify(chunk)}\\n\\n`);\n    onChunkSent?.();\n    if (signal?.aborted) return false;\n    chunkIndex++;\n  }\n\n  if (!res.writableEnded) {\n    if (opts.usageChunk) {\n      res.write(`data: ${JSON.stringify(opts.usageChunk)}\\n\\n`);\n    }\n    res.write(\"data: [DONE]\\n\\n\");\n    res.end();\n  }\n  return true;\n}\n\n/**\n * Default rate-limit response headers matching OpenAI's format.\n * Values are static — aimock doesn't track actual request counts.\n */\nconst RATE_LIMIT_HEADERS: Record<string, string> = {\n  \"x-ratelimit-limit-requests\": \"60\",\n  \"x-ratelimit-limit-tokens\": \"150000\",\n  \"x-ratelimit-remaining-requests\": \"0\",\n  \"x-ratelimit-remaining-tokens\": \"0\",\n  \"x-ratelimit-reset-requests\": \"1s\",\n  \"x-ratelimit-reset-tokens\": \"6m0s\",\n};\n\nexport interface ErrorResponseOptions {\n  /** Override the Retry-After header value (seconds). Default: 1. Only applied on 429. */\n  retryAfter?: number;\n}\n\nexport function writeErrorResponse(\n  res: http.ServerResponse,\n  status: number,\n  body: string,\n  options?: ErrorResponseOptions,\n): void {\n  const headers: Record<string, string> = { \"Content-Type\": \"application/json\" };\n  if (status === 429) {\n    headers[\"Retry-After\"] = String(options?.retryAfter ?? 1);\n    Object.assign(headers, RATE_LIMIT_HEADERS);\n  }\n  res.writeHead(status, headers);\n  res.end(body);\n}\n"],"mappings":";;AAGA,SAAgB,MAAM,IAAY,QAAqC;AACrE,KAAI,MAAM,KAAK,QAAQ,QAAS,QAAO,QAAQ,SAAS;AACxD,QAAO,IAAI,SAAS,YAAY;EAC9B,MAAM,QAAQ,WAAW,SAAS,GAAG;AACrC,UAAQ,iBACN,eACM;AACJ,gBAAa,MAAM;AACnB,YAAS;KAEX,EAAE,MAAM,MAAM,CACf;GACD;;;AAuBJ,MAAM,gCAAgC;AAEtC,SAAgB,eACd,YACA,SACA,iBACA,iBACA,aACQ;CACR,MAAM,QAAQ,eAAe;CAC7B,IAAI;AAEJ,KAAI,SAAS;EAEX,IAAI,cAAc;AAClB,MAAI,eAAe,KAAK,QAAQ,SAAS,OACvC,WAAU,QAAQ;WACT,QAAQ,QAAQ,UAAa,QAAQ,MAAM,EACpD,WAAU,MAAO,QAAQ;OACpB;AACL,aAAU,mBAAmB;AAC7B,iBAAc;;AAGhB,MAAI,eAAe,QAAQ,UAAU,QAAQ,SAAS,GAAG;AACvD,cAAW,KAAK,KAAK,QAAQ,GAAG,IAAI,KAAK,QAAQ;AACjD,OAAI,UAAU,EAAG,WAAU;;YAEpB,gBAET,KAAI,eAAe,EACjB,WAAU,gBAAgB;MACrB;EACL,MAAM,MAAM,aAAa;AACzB,MAAI,MAAM,gBAAgB,mBAAmB,OAC3C,WAAU,gBAAgB,mBAAmB;OACxC;GAEL,MAAM,kBAAkB,gBAAgB,mBAAmB,QAAQ,GAAG,MAAM,IAAI,GAAG,EAAE;AACrF,aACE,gBAAgB,mBAAmB,SAAS,IACxC,kBAAkB,gBAAgB,mBAAmB,SACrD;;;KAIV,WAAU,mBAAmB;AAG/B,WAAU,KAAK,IAAI,GAAG,QAAQ;AAC9B,QAAO,QAAQ,IAAI,UAAU,QAAQ;;AAevC,SAAgB,oBACd,eACA,WACA,SACA,iBACA,aACQ;CACR,MAAM,aAAa,KAAK,IAAI,GAAG,KAAK,KAAK,gBAAgB,UAAU,CAAC;CACpE,IAAI,QAAQ;AACZ,MAAK,IAAI,aAAa,GAAG,aAAa,YAAY,aAChD,UAAS,eAAe,YAAY,SAAS,QAAW,iBAAiB,YAAY;AAEvF,QAAO;;AAGT,eAAsB,eACpB,KACA,QACA,kBACkB;CAClB,MAAM,OACJ,OAAO,qBAAqB,WAAW,EAAE,SAAS,kBAAkB,GAAI,oBAAoB,EAAE;CAChG,MAAM,UAAU,KAAK,WAAW;CAChC,MAAM,UAAU,KAAK;CACrB,MAAM,EAAE,iBAAiB,gBAAgB;CACzC,MAAM,SAAS,KAAK;CACpB,MAAM,cAAc,KAAK;AAEzB,KAAI,IAAI,cAAe,QAAO;AAC9B,KAAI,UAAU,gBAAgB,oBAAoB;AAClD,KAAI,UAAU,iBAAiB,WAAW;AAC1C,KAAI,UAAU,cAAc,aAAa;AAKzC,KAAI,KAAK,wBAAwB,CAAC,IAAI,cACpC,KAAI,MAAM,8BAA8B;CAG1C,IAAI,aAAa;AACjB,MAAK,MAAM,SAAS,QAAQ;EAC1B,MAAM,aAAa,eAAe,YAAY,SAAS,SAAS,iBAAiB,YAAY;AAC7F,MAAI,aAAa,EACf,OAAM,MAAM,YAAY,OAAO;AAEjC,MAAI,QAAQ,QAAS,QAAO;AAC5B,MAAI,IAAI,cAAe,QAAO;AAC9B,MAAI,MAAM,SAAS,KAAK,UAAU,MAAM,CAAC,MAAM;AAC/C,iBAAe;AACf,MAAI,QAAQ,QAAS,QAAO;AAC5B;;AAGF,KAAI,CAAC,IAAI,eAAe;AACtB,MAAI,KAAK,WACP,KAAI,MAAM,SAAS,KAAK,UAAU,KAAK,WAAW,CAAC,MAAM;AAE3D,MAAI,MAAM,mBAAmB;AAC7B,MAAI,KAAK;;AAEX,QAAO;;;;;;AAOT,MAAM,qBAA6C;CACjD,8BAA8B;CAC9B,4BAA4B;CAC5B,kCAAkC;CAClC,gCAAgC;CAChC,8BAA8B;CAC9B,4BAA4B;CAC7B;AAOD,SAAgB,mBACd,KACA,QACA,MACA,SACM;CACN,MAAM,UAAkC,EAAE,gBAAgB,oBAAoB;AAC9E,KAAI,WAAW,KAAK;AAClB,UAAQ,iBAAiB,OAAO,SAAS,cAAc,EAAE;AACzD,SAAO,OAAO,SAAS,mBAAmB;;AAE5C,KAAI,UAAU,QAAQ,QAAQ;AAC9B,KAAI,IAAI,KAAK"}
+{"version":3,"file":"sse-writer.cjs","names":["runFrameLoop","httpSink"],"sources":["../src/sse-writer.ts"],"sourcesContent":["import type * as http from \"node:http\";\nimport type { SSEChunk, StreamingProfile, RecordedTimings } from \"./types.js\";\nimport { httpSink, runFrameLoop, type FrameCodec } from \"./response/frame-loop.js\";\n\nexport function delay(ms: number, signal?: AbortSignal): Promise<void> {\n  if (ms <= 0 || signal?.aborted) return Promise.resolve();\n  return new Promise((resolve) => {\n    const onAbort = (): void => {\n      clearTimeout(timer);\n      resolve();\n    };\n    // Remove the abort listener when the timer fires, so a long stream sharing\n    // one lifecycle signal does not keep one listener per delayed frame.\n    const timer = setTimeout(() => {\n      signal?.removeEventListener(\"abort\", onAbort);\n      resolve();\n    }, ms);\n    signal?.addEventListener(\"abort\", onAbort, { once: true });\n  });\n}\n\nexport interface StreamOptions {\n  latency?: number;\n  streamingProfile?: StreamingProfile;\n  recordedTimings?: RecordedTimings;\n  replaySpeed?: number;\n  signal?: AbortSignal;\n  onChunkSent?: () => void;\n  /** When set, emitted as the final chunk before [DONE] (OpenAI stream_options.include_usage). */\n  usageChunk?: SSEChunk;\n  /**\n   * When true, a single OpenRouter `: OPENROUTER PROCESSING` SSE comment\n   * (keepalive) line is emitted FIRST — before the first data chunk — matching\n   * real OpenRouter bytes (or-capture/chat-stream.sse). A faithful client skips\n   * `:`-prefixed lines; opt-in, default off (so strict OpenAI parsers aren't\n   * surprised). Used only by the OpenRouter chat surface.\n   */\n  openRouterProcessing?: boolean;\n}\n\n/** OpenRouter's SSE keepalive comment line (colon-prefixed, per SSE syntax). */\nconst OPENROUTER_PROCESSING_COMMENT = \": OPENROUTER PROCESSING\\n\\n\";\n\nexport function calculateDelay(\n  chunkIndex: number,\n  profile?: StreamingProfile,\n  fallbackLatency?: number,\n  recordedTimings?: RecordedTimings,\n  replaySpeed?: number,\n): number {\n  const speed = replaySpeed ?? 1.0;\n  let delayMs: number;\n\n  if (profile) {\n    // StreamingProfile has highest precedence\n    let fromProfile = true;\n    if (chunkIndex === 0 && profile.ttft !== undefined) {\n      delayMs = profile.ttft;\n    } else if (profile.tps !== undefined && profile.tps > 0) {\n      delayMs = 1000 / profile.tps;\n    } else {\n      delayMs = fallbackLatency ?? 0;\n      fromProfile = false;\n    }\n    // Jitter only applies when the delay came from ttft/tps, not fallback\n    if (fromProfile && profile.jitter && profile.jitter > 0) {\n      delayMs *= 1 + (Math.random() * 2 - 1) * profile.jitter;\n      if (delayMs < 0) delayMs = 0;\n    }\n  } else if (recordedTimings) {\n    // Recorded timings (second precedence)\n    if (chunkIndex === 0) {\n      delayMs = recordedTimings.ttftMs;\n    } else {\n      const idx = chunkIndex - 1;\n      if (idx < recordedTimings.interChunkDelaysMs.length) {\n        delayMs = recordedTimings.interChunkDelaysMs[idx];\n      } else {\n        // Excess chunks: derive average from recorded inter-chunk delays\n        const totalInterChunk = recordedTimings.interChunkDelaysMs.reduce((a, b) => a + b, 0);\n        delayMs =\n          recordedTimings.interChunkDelaysMs.length > 0\n            ? totalInterChunk / recordedTimings.interChunkDelaysMs.length\n            : 0;\n      }\n    }\n  } else {\n    delayMs = fallbackLatency ?? 0;\n  }\n\n  delayMs = Math.max(0, delayMs);\n  return speed > 0 ? delayMs / speed : delayMs;\n}\n\n// Internal (not part of the public API): the total wait before a non-streaming\n// (`stream: false`) JSON response is written. Sums calculateDelay over the\n// content chunk count that `contentLength` at `chunkSize` would produce, so a\n// fixture with an explicit `streamingProfile` or `recordedTimings` pays roughly\n// the same ttft/tps cost as its streamed content chunks. It is an\n// approximation, not an exact match: a streamed reply also pays delays for its\n// role and finish chunks. There is deliberately no fallback-latency parameter:\n// bare `latency` / `--latency` is an inter-SSE-chunk gap and never delays a\n// non-streaming response. With neither a profile nor recorded timings, this\n// returns 0.\n// Kept as a line comment, not JSDoc: the bundled .d.ts drops this non-public\n// declaration but keeps a JSDoc block, which would then attach to writeSSEStream.\nexport function calculateTotalDelay(\n  contentLength: number,\n  chunkSize: number,\n  profile?: StreamingProfile,\n  recordedTimings?: RecordedTimings,\n  replaySpeed?: number,\n): number {\n  const chunkCount = Math.max(1, Math.ceil(contentLength / chunkSize));\n  let total = 0;\n  for (let chunkIndex = 0; chunkIndex < chunkCount; chunkIndex++) {\n    total += calculateDelay(chunkIndex, profile, undefined, recordedTimings, replaySpeed);\n  }\n  return total;\n}\n\nexport async function writeSSEStream(\n  res: http.ServerResponse,\n  chunks: SSEChunk[],\n  optionsOrLatency?: number | StreamOptions,\n): Promise<boolean> {\n  const opts: StreamOptions =\n    typeof optionsOrLatency === \"number\" ? { latency: optionsOrLatency } : (optionsOrLatency ?? {});\n  // No lifecycle: an external caller gets close-awareness only through its own\n  // `signal` (a gone client still returns true).\n  // Base contract (public API): with no frames, an aborted signal does not\n  // stop the writer; it still finishes the stream (any trailer, then end())\n  // and returns true. The frame loop returns false there, so pass it no\n  // signal when there are no frames (it reads the signal only per frame and\n  // after the loop).\n  return runFrameLoop(\n    httpSink(res),\n    chunks,\n    sseCodec({\n      done: true,\n      usageChunk: opts.usageChunk,\n      openRouterProcessing: opts.openRouterProcessing,\n    }),\n    {\n      latency: opts.latency,\n      streamingProfile: opts.streamingProfile,\n      recordedTimings: opts.recordedTimings,\n      replaySpeed: opts.replaySpeed,\n    },\n    { signal: chunks.length > 0 ? opts.signal : undefined, onChunkSent: opts.onChunkSent },\n  );\n}\n\nconst SSE_HEADERS: Record<string, string> = {\n  \"Content-Type\": \"text/event-stream\",\n  \"Cache-Control\": \"no-cache\",\n  Connection: \"keep-alive\",\n};\n\n// Internal (not part of the public API): data-only SSE (`data: <json>\\n\\n`).\n// `done: true` writes the optional usage chunk then `data: [DONE]` after the\n// last frame (OpenAI chat); `done: false` writes no trailer.\n// `openRouterProcessing` writes the OpenRouter keepalive comment first. Line\n// comments, not JSDoc: see calculateTotalDelay.\nexport function sseCodec<F = SSEChunk>(opts: {\n  done: boolean;\n  usageChunk?: SSEChunk;\n  openRouterProcessing?: boolean;\n}): FrameCodec<F> {\n  return {\n    headers: SSE_HEADERS,\n    ...(opts.openRouterProcessing ? { prelude: () => OPENROUTER_PROCESSING_COMMENT } : {}),\n    encode: (frame) => `data: ${JSON.stringify(frame)}\\n\\n`,\n    trailer: () => {\n      if (!opts.done) return [];\n      const parts: string[] = [];\n      if (opts.usageChunk) parts.push(`data: ${JSON.stringify(opts.usageChunk)}\\n\\n`);\n      parts.push(\"data: [DONE]\\n\\n\");\n      return parts;\n    },\n  };\n}\n\n// Internal: typed SSE (`event: <type>\\ndata: <json>\\n\\n`), used by Anthropic.\n// `terminalSentinel` writes `data: [DONE]` after the last frame, only when the\n// stream was not aborted.\nexport function typedSseCodec<F extends { type: string }>(opts?: {\n  terminalSentinel?: boolean;\n}): FrameCodec<F> {\n  return {\n    headers: SSE_HEADERS,\n    encode: (frame) => `event: ${frame.type}\\ndata: ${JSON.stringify(frame)}\\n\\n`,\n    trailer: () => (opts?.terminalSentinel ? [\"data: [DONE]\\n\\n\"] : []),\n  };\n}\n\n/**\n * Default rate-limit response headers matching OpenAI's format.\n * Values are static — aimock doesn't track actual request counts.\n */\nconst RATE_LIMIT_HEADERS: Record<string, string> = {\n  \"x-ratelimit-limit-requests\": \"60\",\n  \"x-ratelimit-limit-tokens\": \"150000\",\n  \"x-ratelimit-remaining-requests\": \"0\",\n  \"x-ratelimit-remaining-tokens\": \"0\",\n  \"x-ratelimit-reset-requests\": \"1s\",\n  \"x-ratelimit-reset-tokens\": \"6m0s\",\n};\n\nexport interface ErrorResponseOptions {\n  /** Override the Retry-After header value (seconds). Default: 1. Only applied on 429. */\n  retryAfter?: number;\n}\n\nexport function writeErrorResponse(\n  res: http.ServerResponse,\n  status: number,\n  body: string,\n  options?: ErrorResponseOptions,\n): void {\n  const headers: Record<string, string> = { \"Content-Type\": \"application/json\" };\n  if (status === 429) {\n    headers[\"Retry-After\"] = String(options?.retryAfter ?? 1);\n    Object.assign(headers, RATE_LIMIT_HEADERS);\n  }\n  res.writeHead(status, headers);\n  res.end(body);\n}\n"],"mappings":";;;AAIA,SAAgB,MAAM,IAAY,QAAqC;AACrE,KAAI,MAAM,KAAK,QAAQ,QAAS,QAAO,QAAQ,SAAS;AACxD,QAAO,IAAI,SAAS,YAAY;EAC9B,MAAM,gBAAsB;AAC1B,gBAAa,MAAM;AACnB,YAAS;;EAIX,MAAM,QAAQ,iBAAiB;AAC7B,WAAQ,oBAAoB,SAAS,QAAQ;AAC7C,YAAS;KACR,GAAG;AACN,UAAQ,iBAAiB,SAAS,SAAS,EAAE,MAAM,MAAM,CAAC;GAC1D;;;AAuBJ,MAAM,gCAAgC;AAEtC,SAAgB,eACd,YACA,SACA,iBACA,iBACA,aACQ;CACR,MAAM,QAAQ,eAAe;CAC7B,IAAI;AAEJ,KAAI,SAAS;EAEX,IAAI,cAAc;AAClB,MAAI,eAAe,KAAK,QAAQ,SAAS,OACvC,WAAU,QAAQ;WACT,QAAQ,QAAQ,UAAa,QAAQ,MAAM,EACpD,WAAU,MAAO,QAAQ;OACpB;AACL,aAAU,mBAAmB;AAC7B,iBAAc;;AAGhB,MAAI,eAAe,QAAQ,UAAU,QAAQ,SAAS,GAAG;AACvD,cAAW,KAAK,KAAK,QAAQ,GAAG,IAAI,KAAK,QAAQ;AACjD,OAAI,UAAU,EAAG,WAAU;;YAEpB,gBAET,KAAI,eAAe,EACjB,WAAU,gBAAgB;MACrB;EACL,MAAM,MAAM,aAAa;AACzB,MAAI,MAAM,gBAAgB,mBAAmB,OAC3C,WAAU,gBAAgB,mBAAmB;OACxC;GAEL,MAAM,kBAAkB,gBAAgB,mBAAmB,QAAQ,GAAG,MAAM,IAAI,GAAG,EAAE;AACrF,aACE,gBAAgB,mBAAmB,SAAS,IACxC,kBAAkB,gBAAgB,mBAAmB,SACrD;;;KAIV,WAAU,mBAAmB;AAG/B,WAAU,KAAK,IAAI,GAAG,QAAQ;AAC9B,QAAO,QAAQ,IAAI,UAAU,QAAQ;;AAevC,SAAgB,oBACd,eACA,WACA,SACA,iBACA,aACQ;CACR,MAAM,aAAa,KAAK,IAAI,GAAG,KAAK,KAAK,gBAAgB,UAAU,CAAC;CACpE,IAAI,QAAQ;AACZ,MAAK,IAAI,aAAa,GAAG,aAAa,YAAY,aAChD,UAAS,eAAe,YAAY,SAAS,QAAW,iBAAiB,YAAY;AAEvF,QAAO;;AAGT,eAAsB,eACpB,KACA,QACA,kBACkB;CAClB,MAAM,OACJ,OAAO,qBAAqB,WAAW,EAAE,SAAS,kBAAkB,GAAI,oBAAoB,EAAE;AAQhG,QAAOA,gCACLC,4BAAS,IAAI,EACb,QACA,SAAS;EACP,MAAM;EACN,YAAY,KAAK;EACjB,sBAAsB,KAAK;EAC5B,CAAC,EACF;EACE,SAAS,KAAK;EACd,kBAAkB,KAAK;EACvB,iBAAiB,KAAK;EACtB,aAAa,KAAK;EACnB,EACD;EAAE,QAAQ,OAAO,SAAS,IAAI,KAAK,SAAS;EAAW,aAAa,KAAK;EAAa,CACvF;;AAGH,MAAM,cAAsC;CAC1C,gBAAgB;CAChB,iBAAiB;CACjB,YAAY;CACb;AAOD,SAAgB,SAAuB,MAIrB;AAChB,QAAO;EACL,SAAS;EACT,GAAI,KAAK,uBAAuB,EAAE,eAAe,+BAA+B,GAAG,EAAE;EACrF,SAAS,UAAU,SAAS,KAAK,UAAU,MAAM,CAAC;EAClD,eAAe;AACb,OAAI,CAAC,KAAK,KAAM,QAAO,EAAE;GACzB,MAAM,QAAkB,EAAE;AAC1B,OAAI,KAAK,WAAY,OAAM,KAAK,SAAS,KAAK,UAAU,KAAK,WAAW,CAAC,MAAM;AAC/E,SAAM,KAAK,mBAAmB;AAC9B,UAAO;;EAEV;;AAMH,SAAgB,cAA0C,MAExC;AAChB,QAAO;EACL,SAAS;EACT,SAAS,UAAU,UAAU,MAAM,KAAK,UAAU,KAAK,UAAU,MAAM,CAAC;EACxE,eAAgB,MAAM,mBAAmB,CAAC,mBAAmB,GAAG,EAAE;EACnE;;;;;;AAOH,MAAM,qBAA6C;CACjD,8BAA8B;CAC9B,4BAA4B;CAC5B,kCAAkC;CAClC,gCAAgC;CAChC,8BAA8B;CAC9B,4BAA4B;CAC7B;AAOD,SAAgB,mBACd,KACA,QACA,MACA,SACM;CACN,MAAM,UAAkC,EAAE,gBAAgB,oBAAoB;AAC9E,KAAI,WAAW,KAAK;AAClB,UAAQ,iBAAiB,OAAO,SAAS,cAAc,EAAE;AACzD,SAAO,OAAO,SAAS,mBAAmB;;AAE5C,KAAI,UAAU,QAAQ,QAAQ;AAC9B,KAAI,IAAI,KAAK"}

package/dist/sse-writer.d.cts

@@ -29,7 +29,6 @@ interface ErrorResponseOptions {
 }
 declare function writeErrorResponse(res: http$1.ServerResponse, status: number, body: string, options?: ErrorResponseOptions): void;
 //# sourceMappingURL=sse-writer.d.ts.map
-
 //#endregion
 export { ErrorResponseOptions, StreamOptions, calculateDelay, delay, writeErrorResponse, writeSSEStream };
 //# sourceMappingURL=sse-writer.d.cts.map

package/dist/sse-writer.d.cts.map

@@ -1 +1 @@
-{"version":3,"file":"sse-writer.d.cts","names":[],"sources":["../src/sse-writer.ts"],"sourcesContent":[],"mappings":";;;;iBAGgB,KAAA,sBAA2B,cAAc;UAexC,aAAA;EAfD,OAAA,CAAK,EAAA,MAAA;EAAA,gBAAA,CAAA,EAiBA,gBAjBA;iBAAsB,CAAA,EAkBvB,eAlBuB;aAAc,CAAA,EAAA,MAAA;EAAO,MAAA,CAAA,EAoBrD,WApBqD;EAe/C,WAAA,CAAA,EAAA,GAAa,GAAA,IAAA;EAAA;YAET,CAAA,EAMN,QANM;;;;;AAoBrB;;;sBAIoB,CAAA,EAAA,OAAA;;AA0EE,iBA9EN,cAAA,CA8EoB,UAAA,EAAA,MAAA,EAAA,OAAA,CAAA,EA5ExB,gBA4EwB,EAAA,eAAA,CAAA,EAAA,MAAA,EAAA,eAAA,CAAA,EA1EhB,eA0EgB,EAAA,WAAA,CAAA,EAAA,MAAA,CAAA,EAAA,MAAA;AAC7B,iBADe,cAAA,CACV,GAAA,EAAL,MAAA,CAAK,cAAA,EAAA,MAAA,EACF,QADE,EAAA,EAAA,gBAAA,CAAA,EAAA,MAAA,GAEkB,aAFlB,CAAA,EAGT,OAHS,CAAA,OAAA,CAAA;AACF,UA4DO,oBAAA,CA5DP;;YAEP,CAAA,EAAA,MAAA;;AA0Dc,iBAKD,kBAAA,CALqB,GAAA,EAM9B,MAAA,CAAK,cANyB,EAAA,MAAA,EAAA,MAAA,EAAA,IAAA,EAAA,MAAA,EAAA,OAAA,CAAA,EASzB,oBATyB,CAAA,EAAA,IAAA;AAKrC"}
+{"version":3,"file":"sse-writer.d.cts","names":[],"sources":["../src/sse-writer.ts"],"sourcesContent":[],"mappings":";;;;iBAIgB,KAAA,sBAA2B,cAAc;AAAzC,UAiBC,aAAA,CAjBI;EAAA,OAAA,CAAA,EAAA,MAAA;kBAAsB,CAAA,EAmBtB,gBAnBsB;iBAAc,CAAA,EAoBrC,eApBqC;EAAO,WAAA,CAAA,EAAA,MAAA;EAiB/C,MAAA,CAAA,EAKN,WALmB;EAAA,WAAA,CAAA,EAAA,GAAA,GAAA,IAAA;;YAGV,CAAA,EAKL,QALK;;;;AAmBpB;;;;EAImC,oBAAA,CAAA,EAAA,OAAA;AA0EnC;AAAoC,iBA9EpB,cAAA,CA8EoB,UAAA,EAAA,MAAA,EAAA,OAAA,CAAA,EA5ExB,gBA4EwB,EAAA,eAAA,CAAA,EAAA,MAAA,EAAA,eAAA,CAAA,EA1EhB,eA0EgB,EAAA,WAAA,CAAA,EAAA,MAAA,CAAA,EAAA,MAAA;AAE1B,iBAFY,cAAA,CAEZ,GAAA,EADH,MAAA,CAAK,cACF,EAAA,MAAA,EAAA,QAAA,EAAA,EAAA,gBAAA,CAAA,EAAA,MAAA,GACoB,aADpB,CAAA,EAEP,OAFO,CAAA,OAAA,CAAA;UAsFO,oBAAA;;;;iBAKD,kBAAA,MACT,MAAA,CAAK,wDAGA"}

package/dist/sse-writer.d.ts

@@ -29,7 +29,6 @@ interface ErrorResponseOptions {
 }
 declare function writeErrorResponse(res: http$1.ServerResponse, status: number, body: string, options?: ErrorResponseOptions): void;
 //# sourceMappingURL=sse-writer.d.ts.map
-
 //#endregion
 export { ErrorResponseOptions, StreamOptions, calculateDelay, delay, writeErrorResponse, writeSSEStream };
 //# sourceMappingURL=sse-writer.d.ts.map

package/dist/sse-writer.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"sse-writer.d.ts","names":[],"sources":["../src/sse-writer.ts"],"sourcesContent":[],"mappings":";;;;iBAGgB,KAAA,sBAA2B,cAAc;UAexC,aAAA;EAfD,OAAA,CAAK,EAAA,MAAA;EAAA,gBAAA,CAAA,EAiBA,gBAjBA;iBAAsB,CAAA,EAkBvB,eAlBuB;aAAc,CAAA,EAAA,MAAA;EAAO,MAAA,CAAA,EAoBrD,WApBqD;EAe/C,WAAA,CAAA,EAAA,GAAa,GAAA,IAAA;EAAA;YAET,CAAA,EAMN,QANM;;;;;AAoBrB;;;sBAIoB,CAAA,EAAA,OAAA;;AA0EE,iBA9EN,cAAA,CA8EoB,UAAA,EAAA,MAAA,EAAA,OAAA,CAAA,EA5ExB,gBA4EwB,EAAA,eAAA,CAAA,EAAA,MAAA,EAAA,eAAA,CAAA,EA1EhB,eA0EgB,EAAA,WAAA,CAAA,EAAA,MAAA,CAAA,EAAA,MAAA;AAC7B,iBADe,cAAA,CACV,GAAA,EAAL,MAAA,CAAK,cAAA,EAAA,MAAA,EACF,QADE,EAAA,EAAA,gBAAA,CAAA,EAAA,MAAA,GAEkB,aAFlB,CAAA,EAGT,OAHS,CAAA,OAAA,CAAA;AACF,UA4DO,oBAAA,CA5DP;;YAEP,CAAA,EAAA,MAAA;;AA0Dc,iBAKD,kBAAA,CALqB,GAAA,EAM9B,MAAA,CAAK,cANyB,EAAA,MAAA,EAAA,MAAA,EAAA,IAAA,EAAA,MAAA,EAAA,OAAA,CAAA,EASzB,oBATyB,CAAA,EAAA,IAAA;AAKrC"}
+{"version":3,"file":"sse-writer.d.ts","names":[],"sources":["../src/sse-writer.ts"],"sourcesContent":[],"mappings":";;;;iBAIgB,KAAA,sBAA2B,cAAc;AAAzC,UAiBC,aAAA,CAjBI;EAAA,OAAA,CAAA,EAAA,MAAA;kBAAsB,CAAA,EAmBtB,gBAnBsB;iBAAc,CAAA,EAoBrC,eApBqC;EAAO,WAAA,CAAA,EAAA,MAAA;EAiB/C,MAAA,CAAA,EAKN,WALmB;EAAA,WAAA,CAAA,EAAA,GAAA,GAAA,IAAA;;YAGV,CAAA,EAKL,QALK;;;;AAmBpB;;;;EAImC,oBAAA,CAAA,EAAA,OAAA;AA0EnC;AAAoC,iBA9EpB,cAAA,CA8EoB,UAAA,EAAA,MAAA,EAAA,OAAA,CAAA,EA5ExB,gBA4EwB,EAAA,eAAA,CAAA,EAAA,MAAA,EAAA,eAAA,CAAA,EA1EhB,eA0EgB,EAAA,WAAA,CAAA,EAAA,MAAA,CAAA,EAAA,MAAA;AAE1B,iBAFY,cAAA,CAEZ,GAAA,EADH,MAAA,CAAK,cACF,EAAA,MAAA,EAAA,QAAA,EAAA,EAAA,gBAAA,CAAA,EAAA,MAAA,GACoB,aADpB,CAAA,EAEP,OAFO,CAAA,OAAA,CAAA;UAsFO,oBAAA;;;;iBAKD,kBAAA,MACT,MAAA,CAAK,wDAGA"}

package/dist/sse-writer.js

@@ -1,12 +1,18 @@
+import { httpSink, runFrameLoop } from "./response/frame-loop.js";
+
 //#region src/sse-writer.ts
 function delay(ms, signal) {
 	if (ms <= 0 || signal?.aborted) return Promise.resolve();
 	return new Promise((resolve) => {
-		const timer = setTimeout(resolve, ms);
-		signal?.addEventListener("abort", () => {
+		const onAbort = () => {
 			clearTimeout(timer);
 			resolve();
-		}, { once: true });
+		};
+		const timer = setTimeout(() => {
+			signal?.removeEventListener("abort", onAbort);
+			resolve();
+		}, ms);
+		signal?.addEventListener("abort", onAbort, { once: true });
 	});
 }
 /** OpenRouter's SSE keepalive comment line (colon-prefixed, per SSE syntax). */
@@ -47,33 +53,45 @@ function calculateTotalDelay(contentLength, chunkSize, profile, recordedTimings,
 }
 async function writeSSEStream(res, chunks, optionsOrLatency) {
 	const opts = typeof optionsOrLatency === "number" ? { latency: optionsOrLatency } : optionsOrLatency ?? {};
-	const latency = opts.latency ?? 0;
-	const profile = opts.streamingProfile;
-	const { recordedTimings, replaySpeed } = opts;
-	const signal = opts.signal;
-	const onChunkSent = opts.onChunkSent;
-	if (res.writableEnded) return true;
-	res.setHeader("Content-Type", "text/event-stream");
-	res.setHeader("Cache-Control", "no-cache");
-	res.setHeader("Connection", "keep-alive");
-	if (opts.openRouterProcessing && !res.writableEnded) res.write(OPENROUTER_PROCESSING_COMMENT);
-	let chunkIndex = 0;
-	for (const chunk of chunks) {
-		const chunkDelay = calculateDelay(chunkIndex, profile, latency, recordedTimings, replaySpeed);
-		if (chunkDelay > 0) await delay(chunkDelay, signal);
-		if (signal?.aborted) return false;
-		if (res.writableEnded) return true;
-		res.write(`data: ${JSON.stringify(chunk)}\n\n`);
-		onChunkSent?.();
-		if (signal?.aborted) return false;
-		chunkIndex++;
-	}
-	if (!res.writableEnded) {
-		if (opts.usageChunk) res.write(`data: ${JSON.stringify(opts.usageChunk)}\n\n`);
-		res.write("data: [DONE]\n\n");
-		res.end();
-	}
-	return true;
+	return runFrameLoop(httpSink(res), chunks, sseCodec({
+		done: true,
+		usageChunk: opts.usageChunk,
+		openRouterProcessing: opts.openRouterProcessing
+	}), {
+		latency: opts.latency,
+		streamingProfile: opts.streamingProfile,
+		recordedTimings: opts.recordedTimings,
+		replaySpeed: opts.replaySpeed
+	}, {
+		signal: chunks.length > 0 ? opts.signal : void 0,
+		onChunkSent: opts.onChunkSent
+	});
+}
+const SSE_HEADERS = {
+	"Content-Type": "text/event-stream",
+	"Cache-Control": "no-cache",
+	Connection: "keep-alive"
+};
+function sseCodec(opts) {
+	return {
+		headers: SSE_HEADERS,
+		...opts.openRouterProcessing ? { prelude: () => OPENROUTER_PROCESSING_COMMENT } : {},
+		encode: (frame) => `data: ${JSON.stringify(frame)}\n\n`,
+		trailer: () => {
+			if (!opts.done) return [];
+			const parts = [];
+			if (opts.usageChunk) parts.push(`data: ${JSON.stringify(opts.usageChunk)}\n\n`);
+			parts.push("data: [DONE]\n\n");
+			return parts;
+		}
+	};
+}
+function typedSseCodec(opts) {
+	return {
+		headers: SSE_HEADERS,
+		encode: (frame) => `event: ${frame.type}\ndata: ${JSON.stringify(frame)}\n\n`,
+		trailer: () => opts?.terminalSentinel ? ["data: [DONE]\n\n"] : []
+	};
 }
 /**
 * Default rate-limit response headers matching OpenAI's format.
@@ -98,5 +116,5 @@ function writeErrorResponse(res, status, body, options) {
 }
 
 //#endregion
-export { calculateDelay, calculateTotalDelay, delay, writeErrorResponse, writeSSEStream };
+export { calculateDelay, calculateTotalDelay, delay, sseCodec, typedSseCodec, writeErrorResponse, writeSSEStream };
 //# sourceMappingURL=sse-writer.js.map

package/dist/sse-writer.js.map

@@ -1 +1 @@
-{"version":3,"file":"sse-writer.js","names":[],"sources":["../src/sse-writer.ts"],"sourcesContent":["import type * as http from \"node:http\";\nimport type { SSEChunk, StreamingProfile, RecordedTimings } from \"./types.js\";\n\nexport function delay(ms: number, signal?: AbortSignal): Promise<void> {\n  if (ms <= 0 || signal?.aborted) return Promise.resolve();\n  return new Promise((resolve) => {\n    const timer = setTimeout(resolve, ms);\n    signal?.addEventListener(\n      \"abort\",\n      () => {\n        clearTimeout(timer);\n        resolve();\n      },\n      { once: true },\n    );\n  });\n}\n\nexport interface StreamOptions {\n  latency?: number;\n  streamingProfile?: StreamingProfile;\n  recordedTimings?: RecordedTimings;\n  replaySpeed?: number;\n  signal?: AbortSignal;\n  onChunkSent?: () => void;\n  /** When set, emitted as the final chunk before [DONE] (OpenAI stream_options.include_usage). */\n  usageChunk?: SSEChunk;\n  /**\n   * When true, a single OpenRouter `: OPENROUTER PROCESSING` SSE comment\n   * (keepalive) line is emitted FIRST — before the first data chunk — matching\n   * real OpenRouter bytes (or-capture/chat-stream.sse). A faithful client skips\n   * `:`-prefixed lines; opt-in, default off (so strict OpenAI parsers aren't\n   * surprised). Used only by the OpenRouter chat surface.\n   */\n  openRouterProcessing?: boolean;\n}\n\n/** OpenRouter's SSE keepalive comment line (colon-prefixed, per SSE syntax). */\nconst OPENROUTER_PROCESSING_COMMENT = \": OPENROUTER PROCESSING\\n\\n\";\n\nexport function calculateDelay(\n  chunkIndex: number,\n  profile?: StreamingProfile,\n  fallbackLatency?: number,\n  recordedTimings?: RecordedTimings,\n  replaySpeed?: number,\n): number {\n  const speed = replaySpeed ?? 1.0;\n  let delayMs: number;\n\n  if (profile) {\n    // StreamingProfile has highest precedence\n    let fromProfile = true;\n    if (chunkIndex === 0 && profile.ttft !== undefined) {\n      delayMs = profile.ttft;\n    } else if (profile.tps !== undefined && profile.tps > 0) {\n      delayMs = 1000 / profile.tps;\n    } else {\n      delayMs = fallbackLatency ?? 0;\n      fromProfile = false;\n    }\n    // Jitter only applies when the delay came from ttft/tps, not fallback\n    if (fromProfile && profile.jitter && profile.jitter > 0) {\n      delayMs *= 1 + (Math.random() * 2 - 1) * profile.jitter;\n      if (delayMs < 0) delayMs = 0;\n    }\n  } else if (recordedTimings) {\n    // Recorded timings (second precedence)\n    if (chunkIndex === 0) {\n      delayMs = recordedTimings.ttftMs;\n    } else {\n      const idx = chunkIndex - 1;\n      if (idx < recordedTimings.interChunkDelaysMs.length) {\n        delayMs = recordedTimings.interChunkDelaysMs[idx];\n      } else {\n        // Excess chunks: derive average from recorded inter-chunk delays\n        const totalInterChunk = recordedTimings.interChunkDelaysMs.reduce((a, b) => a + b, 0);\n        delayMs =\n          recordedTimings.interChunkDelaysMs.length > 0\n            ? totalInterChunk / recordedTimings.interChunkDelaysMs.length\n            : 0;\n      }\n    }\n  } else {\n    delayMs = fallbackLatency ?? 0;\n  }\n\n  delayMs = Math.max(0, delayMs);\n  return speed > 0 ? delayMs / speed : delayMs;\n}\n\n// Internal (not part of the public API): the total wait before a non-streaming\n// (`stream: false`) JSON response is written. Sums calculateDelay over the\n// content chunk count that `contentLength` at `chunkSize` would produce, so a\n// fixture with an explicit `streamingProfile` or `recordedTimings` pays roughly\n// the same ttft/tps cost as its streamed content chunks. It is an\n// approximation, not an exact match: a streamed reply also pays delays for its\n// role and finish chunks. There is deliberately no fallback-latency parameter:\n// bare `latency` / `--latency` is an inter-SSE-chunk gap and never delays a\n// non-streaming response. With neither a profile nor recorded timings, this\n// returns 0.\n// Kept as a line comment, not JSDoc: the bundled .d.ts drops this non-public\n// declaration but keeps a JSDoc block, which would then attach to writeSSEStream.\nexport function calculateTotalDelay(\n  contentLength: number,\n  chunkSize: number,\n  profile?: StreamingProfile,\n  recordedTimings?: RecordedTimings,\n  replaySpeed?: number,\n): number {\n  const chunkCount = Math.max(1, Math.ceil(contentLength / chunkSize));\n  let total = 0;\n  for (let chunkIndex = 0; chunkIndex < chunkCount; chunkIndex++) {\n    total += calculateDelay(chunkIndex, profile, undefined, recordedTimings, replaySpeed);\n  }\n  return total;\n}\n\nexport async function writeSSEStream(\n  res: http.ServerResponse,\n  chunks: SSEChunk[],\n  optionsOrLatency?: number | StreamOptions,\n): Promise<boolean> {\n  const opts: StreamOptions =\n    typeof optionsOrLatency === \"number\" ? { latency: optionsOrLatency } : (optionsOrLatency ?? {});\n  const latency = opts.latency ?? 0;\n  const profile = opts.streamingProfile;\n  const { recordedTimings, replaySpeed } = opts;\n  const signal = opts.signal;\n  const onChunkSent = opts.onChunkSent;\n\n  if (res.writableEnded) return true;\n  res.setHeader(\"Content-Type\", \"text/event-stream\");\n  res.setHeader(\"Cache-Control\", \"no-cache\");\n  res.setHeader(\"Connection\", \"keep-alive\");\n\n  // OpenRouter emits exactly ONE `: OPENROUTER PROCESSING` keepalive comment\n  // first — before any data frame (opt-in; see or-capture/chat-stream.sse). A\n  // faithful client discards `:`-prefixed lines before JSON-parsing a `data:`.\n  if (opts.openRouterProcessing && !res.writableEnded) {\n    res.write(OPENROUTER_PROCESSING_COMMENT);\n  }\n\n  let chunkIndex = 0;\n  for (const chunk of chunks) {\n    const chunkDelay = calculateDelay(chunkIndex, profile, latency, recordedTimings, replaySpeed);\n    if (chunkDelay > 0) {\n      await delay(chunkDelay, signal);\n    }\n    if (signal?.aborted) return false;\n    if (res.writableEnded) return true;\n    res.write(`data: ${JSON.stringify(chunk)}\\n\\n`);\n    onChunkSent?.();\n    if (signal?.aborted) return false;\n    chunkIndex++;\n  }\n\n  if (!res.writableEnded) {\n    if (opts.usageChunk) {\n      res.write(`data: ${JSON.stringify(opts.usageChunk)}\\n\\n`);\n    }\n    res.write(\"data: [DONE]\\n\\n\");\n    res.end();\n  }\n  return true;\n}\n\n/**\n * Default rate-limit response headers matching OpenAI's format.\n * Values are static — aimock doesn't track actual request counts.\n */\nconst RATE_LIMIT_HEADERS: Record<string, string> = {\n  \"x-ratelimit-limit-requests\": \"60\",\n  \"x-ratelimit-limit-tokens\": \"150000\",\n  \"x-ratelimit-remaining-requests\": \"0\",\n  \"x-ratelimit-remaining-tokens\": \"0\",\n  \"x-ratelimit-reset-requests\": \"1s\",\n  \"x-ratelimit-reset-tokens\": \"6m0s\",\n};\n\nexport interface ErrorResponseOptions {\n  /** Override the Retry-After header value (seconds). Default: 1. Only applied on 429. */\n  retryAfter?: number;\n}\n\nexport function writeErrorResponse(\n  res: http.ServerResponse,\n  status: number,\n  body: string,\n  options?: ErrorResponseOptions,\n): void {\n  const headers: Record<string, string> = { \"Content-Type\": \"application/json\" };\n  if (status === 429) {\n    headers[\"Retry-After\"] = String(options?.retryAfter ?? 1);\n    Object.assign(headers, RATE_LIMIT_HEADERS);\n  }\n  res.writeHead(status, headers);\n  res.end(body);\n}\n"],"mappings":";AAGA,SAAgB,MAAM,IAAY,QAAqC;AACrE,KAAI,MAAM,KAAK,QAAQ,QAAS,QAAO,QAAQ,SAAS;AACxD,QAAO,IAAI,SAAS,YAAY;EAC9B,MAAM,QAAQ,WAAW,SAAS,GAAG;AACrC,UAAQ,iBACN,eACM;AACJ,gBAAa,MAAM;AACnB,YAAS;KAEX,EAAE,MAAM,MAAM,CACf;GACD;;;AAuBJ,MAAM,gCAAgC;AAEtC,SAAgB,eACd,YACA,SACA,iBACA,iBACA,aACQ;CACR,MAAM,QAAQ,eAAe;CAC7B,IAAI;AAEJ,KAAI,SAAS;EAEX,IAAI,cAAc;AAClB,MAAI,eAAe,KAAK,QAAQ,SAAS,OACvC,WAAU,QAAQ;WACT,QAAQ,QAAQ,UAAa,QAAQ,MAAM,EACpD,WAAU,MAAO,QAAQ;OACpB;AACL,aAAU,mBAAmB;AAC7B,iBAAc;;AAGhB,MAAI,eAAe,QAAQ,UAAU,QAAQ,SAAS,GAAG;AACvD,cAAW,KAAK,KAAK,QAAQ,GAAG,IAAI,KAAK,QAAQ;AACjD,OAAI,UAAU,EAAG,WAAU;;YAEpB,gBAET,KAAI,eAAe,EACjB,WAAU,gBAAgB;MACrB;EACL,MAAM,MAAM,aAAa;AACzB,MAAI,MAAM,gBAAgB,mBAAmB,OAC3C,WAAU,gBAAgB,mBAAmB;OACxC;GAEL,MAAM,kBAAkB,gBAAgB,mBAAmB,QAAQ,GAAG,MAAM,IAAI,GAAG,EAAE;AACrF,aACE,gBAAgB,mBAAmB,SAAS,IACxC,kBAAkB,gBAAgB,mBAAmB,SACrD;;;KAIV,WAAU,mBAAmB;AAG/B,WAAU,KAAK,IAAI,GAAG,QAAQ;AAC9B,QAAO,QAAQ,IAAI,UAAU,QAAQ;;AAevC,SAAgB,oBACd,eACA,WACA,SACA,iBACA,aACQ;CACR,MAAM,aAAa,KAAK,IAAI,GAAG,KAAK,KAAK,gBAAgB,UAAU,CAAC;CACpE,IAAI,QAAQ;AACZ,MAAK,IAAI,aAAa,GAAG,aAAa,YAAY,aAChD,UAAS,eAAe,YAAY,SAAS,QAAW,iBAAiB,YAAY;AAEvF,QAAO;;AAGT,eAAsB,eACpB,KACA,QACA,kBACkB;CAClB,MAAM,OACJ,OAAO,qBAAqB,WAAW,EAAE,SAAS,kBAAkB,GAAI,oBAAoB,EAAE;CAChG,MAAM,UAAU,KAAK,WAAW;CAChC,MAAM,UAAU,KAAK;CACrB,MAAM,EAAE,iBAAiB,gBAAgB;CACzC,MAAM,SAAS,KAAK;CACpB,MAAM,cAAc,KAAK;AAEzB,KAAI,IAAI,cAAe,QAAO;AAC9B,KAAI,UAAU,gBAAgB,oBAAoB;AAClD,KAAI,UAAU,iBAAiB,WAAW;AAC1C,KAAI,UAAU,cAAc,aAAa;AAKzC,KAAI,KAAK,wBAAwB,CAAC,IAAI,cACpC,KAAI,MAAM,8BAA8B;CAG1C,IAAI,aAAa;AACjB,MAAK,MAAM,SAAS,QAAQ;EAC1B,MAAM,aAAa,eAAe,YAAY,SAAS,SAAS,iBAAiB,YAAY;AAC7F,MAAI,aAAa,EACf,OAAM,MAAM,YAAY,OAAO;AAEjC,MAAI,QAAQ,QAAS,QAAO;AAC5B,MAAI,IAAI,cAAe,QAAO;AAC9B,MAAI,MAAM,SAAS,KAAK,UAAU,MAAM,CAAC,MAAM;AAC/C,iBAAe;AACf,MAAI,QAAQ,QAAS,QAAO;AAC5B;;AAGF,KAAI,CAAC,IAAI,eAAe;AACtB,MAAI,KAAK,WACP,KAAI,MAAM,SAAS,KAAK,UAAU,KAAK,WAAW,CAAC,MAAM;AAE3D,MAAI,MAAM,mBAAmB;AAC7B,MAAI,KAAK;;AAEX,QAAO;;;;;;AAOT,MAAM,qBAA6C;CACjD,8BAA8B;CAC9B,4BAA4B;CAC5B,kCAAkC;CAClC,gCAAgC;CAChC,8BAA8B;CAC9B,4BAA4B;CAC7B;AAOD,SAAgB,mBACd,KACA,QACA,MACA,SACM;CACN,MAAM,UAAkC,EAAE,gBAAgB,oBAAoB;AAC9E,KAAI,WAAW,KAAK;AAClB,UAAQ,iBAAiB,OAAO,SAAS,cAAc,EAAE;AACzD,SAAO,OAAO,SAAS,mBAAmB;;AAE5C,KAAI,UAAU,QAAQ,QAAQ;AAC9B,KAAI,IAAI,KAAK"}
+{"version":3,"file":"sse-writer.js","names":[],"sources":["../src/sse-writer.ts"],"sourcesContent":["import type * as http from \"node:http\";\nimport type { SSEChunk, StreamingProfile, RecordedTimings } from \"./types.js\";\nimport { httpSink, runFrameLoop, type FrameCodec } from \"./response/frame-loop.js\";\n\nexport function delay(ms: number, signal?: AbortSignal): Promise<void> {\n  if (ms <= 0 || signal?.aborted) return Promise.resolve();\n  return new Promise((resolve) => {\n    const onAbort = (): void => {\n      clearTimeout(timer);\n      resolve();\n    };\n    // Remove the abort listener when the timer fires, so a long stream sharing\n    // one lifecycle signal does not keep one listener per delayed frame.\n    const timer = setTimeout(() => {\n      signal?.removeEventListener(\"abort\", onAbort);\n      resolve();\n    }, ms);\n    signal?.addEventListener(\"abort\", onAbort, { once: true });\n  });\n}\n\nexport interface StreamOptions {\n  latency?: number;\n  streamingProfile?: StreamingProfile;\n  recordedTimings?: RecordedTimings;\n  replaySpeed?: number;\n  signal?: AbortSignal;\n  onChunkSent?: () => void;\n  /** When set, emitted as the final chunk before [DONE] (OpenAI stream_options.include_usage). */\n  usageChunk?: SSEChunk;\n  /**\n   * When true, a single OpenRouter `: OPENROUTER PROCESSING` SSE comment\n   * (keepalive) line is emitted FIRST — before the first data chunk — matching\n   * real OpenRouter bytes (or-capture/chat-stream.sse). A faithful client skips\n   * `:`-prefixed lines; opt-in, default off (so strict OpenAI parsers aren't\n   * surprised). Used only by the OpenRouter chat surface.\n   */\n  openRouterProcessing?: boolean;\n}\n\n/** OpenRouter's SSE keepalive comment line (colon-prefixed, per SSE syntax). */\nconst OPENROUTER_PROCESSING_COMMENT = \": OPENROUTER PROCESSING\\n\\n\";\n\nexport function calculateDelay(\n  chunkIndex: number,\n  profile?: StreamingProfile,\n  fallbackLatency?: number,\n  recordedTimings?: RecordedTimings,\n  replaySpeed?: number,\n): number {\n  const speed = replaySpeed ?? 1.0;\n  let delayMs: number;\n\n  if (profile) {\n    // StreamingProfile has highest precedence\n    let fromProfile = true;\n    if (chunkIndex === 0 && profile.ttft !== undefined) {\n      delayMs = profile.ttft;\n    } else if (profile.tps !== undefined && profile.tps > 0) {\n      delayMs = 1000 / profile.tps;\n    } else {\n      delayMs = fallbackLatency ?? 0;\n      fromProfile = false;\n    }\n    // Jitter only applies when the delay came from ttft/tps, not fallback\n    if (fromProfile && profile.jitter && profile.jitter > 0) {\n      delayMs *= 1 + (Math.random() * 2 - 1) * profile.jitter;\n      if (delayMs < 0) delayMs = 0;\n    }\n  } else if (recordedTimings) {\n    // Recorded timings (second precedence)\n    if (chunkIndex === 0) {\n      delayMs = recordedTimings.ttftMs;\n    } else {\n      const idx = chunkIndex - 1;\n      if (idx < recordedTimings.interChunkDelaysMs.length) {\n        delayMs = recordedTimings.interChunkDelaysMs[idx];\n      } else {\n        // Excess chunks: derive average from recorded inter-chunk delays\n        const totalInterChunk = recordedTimings.interChunkDelaysMs.reduce((a, b) => a + b, 0);\n        delayMs =\n          recordedTimings.interChunkDelaysMs.length > 0\n            ? totalInterChunk / recordedTimings.interChunkDelaysMs.length\n            : 0;\n      }\n    }\n  } else {\n    delayMs = fallbackLatency ?? 0;\n  }\n\n  delayMs = Math.max(0, delayMs);\n  return speed > 0 ? delayMs / speed : delayMs;\n}\n\n// Internal (not part of the public API): the total wait before a non-streaming\n// (`stream: false`) JSON response is written. Sums calculateDelay over the\n// content chunk count that `contentLength` at `chunkSize` would produce, so a\n// fixture with an explicit `streamingProfile` or `recordedTimings` pays roughly\n// the same ttft/tps cost as its streamed content chunks. It is an\n// approximation, not an exact match: a streamed reply also pays delays for its\n// role and finish chunks. There is deliberately no fallback-latency parameter:\n// bare `latency` / `--latency` is an inter-SSE-chunk gap and never delays a\n// non-streaming response. With neither a profile nor recorded timings, this\n// returns 0.\n// Kept as a line comment, not JSDoc: the bundled .d.ts drops this non-public\n// declaration but keeps a JSDoc block, which would then attach to writeSSEStream.\nexport function calculateTotalDelay(\n  contentLength: number,\n  chunkSize: number,\n  profile?: StreamingProfile,\n  recordedTimings?: RecordedTimings,\n  replaySpeed?: number,\n): number {\n  const chunkCount = Math.max(1, Math.ceil(contentLength / chunkSize));\n  let total = 0;\n  for (let chunkIndex = 0; chunkIndex < chunkCount; chunkIndex++) {\n    total += calculateDelay(chunkIndex, profile, undefined, recordedTimings, replaySpeed);\n  }\n  return total;\n}\n\nexport async function writeSSEStream(\n  res: http.ServerResponse,\n  chunks: SSEChunk[],\n  optionsOrLatency?: number | StreamOptions,\n): Promise<boolean> {\n  const opts: StreamOptions =\n    typeof optionsOrLatency === \"number\" ? { latency: optionsOrLatency } : (optionsOrLatency ?? {});\n  // No lifecycle: an external caller gets close-awareness only through its own\n  // `signal` (a gone client still returns true).\n  // Base contract (public API): with no frames, an aborted signal does not\n  // stop the writer; it still finishes the stream (any trailer, then end())\n  // and returns true. The frame loop returns false there, so pass it no\n  // signal when there are no frames (it reads the signal only per frame and\n  // after the loop).\n  return runFrameLoop(\n    httpSink(res),\n    chunks,\n    sseCodec({\n      done: true,\n      usageChunk: opts.usageChunk,\n      openRouterProcessing: opts.openRouterProcessing,\n    }),\n    {\n      latency: opts.latency,\n      streamingProfile: opts.streamingProfile,\n      recordedTimings: opts.recordedTimings,\n      replaySpeed: opts.replaySpeed,\n    },\n    { signal: chunks.length > 0 ? opts.signal : undefined, onChunkSent: opts.onChunkSent },\n  );\n}\n\nconst SSE_HEADERS: Record<string, string> = {\n  \"Content-Type\": \"text/event-stream\",\n  \"Cache-Control\": \"no-cache\",\n  Connection: \"keep-alive\",\n};\n\n// Internal (not part of the public API): data-only SSE (`data: <json>\\n\\n`).\n// `done: true` writes the optional usage chunk then `data: [DONE]` after the\n// last frame (OpenAI chat); `done: false` writes no trailer.\n// `openRouterProcessing` writes the OpenRouter keepalive comment first. Line\n// comments, not JSDoc: see calculateTotalDelay.\nexport function sseCodec<F = SSEChunk>(opts: {\n  done: boolean;\n  usageChunk?: SSEChunk;\n  openRouterProcessing?: boolean;\n}): FrameCodec<F> {\n  return {\n    headers: SSE_HEADERS,\n    ...(opts.openRouterProcessing ? { prelude: () => OPENROUTER_PROCESSING_COMMENT } : {}),\n    encode: (frame) => `data: ${JSON.stringify(frame)}\\n\\n`,\n    trailer: () => {\n      if (!opts.done) return [];\n      const parts: string[] = [];\n      if (opts.usageChunk) parts.push(`data: ${JSON.stringify(opts.usageChunk)}\\n\\n`);\n      parts.push(\"data: [DONE]\\n\\n\");\n      return parts;\n    },\n  };\n}\n\n// Internal: typed SSE (`event: <type>\\ndata: <json>\\n\\n`), used by Anthropic.\n// `terminalSentinel` writes `data: [DONE]` after the last frame, only when the\n// stream was not aborted.\nexport function typedSseCodec<F extends { type: string }>(opts?: {\n  terminalSentinel?: boolean;\n}): FrameCodec<F> {\n  return {\n    headers: SSE_HEADERS,\n    encode: (frame) => `event: ${frame.type}\\ndata: ${JSON.stringify(frame)}\\n\\n`,\n    trailer: () => (opts?.terminalSentinel ? [\"data: [DONE]\\n\\n\"] : []),\n  };\n}\n\n/**\n * Default rate-limit response headers matching OpenAI's format.\n * Values are static — aimock doesn't track actual request counts.\n */\nconst RATE_LIMIT_HEADERS: Record<string, string> = {\n  \"x-ratelimit-limit-requests\": \"60\",\n  \"x-ratelimit-limit-tokens\": \"150000\",\n  \"x-ratelimit-remaining-requests\": \"0\",\n  \"x-ratelimit-remaining-tokens\": \"0\",\n  \"x-ratelimit-reset-requests\": \"1s\",\n  \"x-ratelimit-reset-tokens\": \"6m0s\",\n};\n\nexport interface ErrorResponseOptions {\n  /** Override the Retry-After header value (seconds). Default: 1. Only applied on 429. */\n  retryAfter?: number;\n}\n\nexport function writeErrorResponse(\n  res: http.ServerResponse,\n  status: number,\n  body: string,\n  options?: ErrorResponseOptions,\n): void {\n  const headers: Record<string, string> = { \"Content-Type\": \"application/json\" };\n  if (status === 429) {\n    headers[\"Retry-After\"] = String(options?.retryAfter ?? 1);\n    Object.assign(headers, RATE_LIMIT_HEADERS);\n  }\n  res.writeHead(status, headers);\n  res.end(body);\n}\n"],"mappings":";;;AAIA,SAAgB,MAAM,IAAY,QAAqC;AACrE,KAAI,MAAM,KAAK,QAAQ,QAAS,QAAO,QAAQ,SAAS;AACxD,QAAO,IAAI,SAAS,YAAY;EAC9B,MAAM,gBAAsB;AAC1B,gBAAa,MAAM;AACnB,YAAS;;EAIX,MAAM,QAAQ,iBAAiB;AAC7B,WAAQ,oBAAoB,SAAS,QAAQ;AAC7C,YAAS;KACR,GAAG;AACN,UAAQ,iBAAiB,SAAS,SAAS,EAAE,MAAM,MAAM,CAAC;GAC1D;;;AAuBJ,MAAM,gCAAgC;AAEtC,SAAgB,eACd,YACA,SACA,iBACA,iBACA,aACQ;CACR,MAAM,QAAQ,eAAe;CAC7B,IAAI;AAEJ,KAAI,SAAS;EAEX,IAAI,cAAc;AAClB,MAAI,eAAe,KAAK,QAAQ,SAAS,OACvC,WAAU,QAAQ;WACT,QAAQ,QAAQ,UAAa,QAAQ,MAAM,EACpD,WAAU,MAAO,QAAQ;OACpB;AACL,aAAU,mBAAmB;AAC7B,iBAAc;;AAGhB,MAAI,eAAe,QAAQ,UAAU,QAAQ,SAAS,GAAG;AACvD,cAAW,KAAK,KAAK,QAAQ,GAAG,IAAI,KAAK,QAAQ;AACjD,OAAI,UAAU,EAAG,WAAU;;YAEpB,gBAET,KAAI,eAAe,EACjB,WAAU,gBAAgB;MACrB;EACL,MAAM,MAAM,aAAa;AACzB,MAAI,MAAM,gBAAgB,mBAAmB,OAC3C,WAAU,gBAAgB,mBAAmB;OACxC;GAEL,MAAM,kBAAkB,gBAAgB,mBAAmB,QAAQ,GAAG,MAAM,IAAI,GAAG,EAAE;AACrF,aACE,gBAAgB,mBAAmB,SAAS,IACxC,kBAAkB,gBAAgB,mBAAmB,SACrD;;;KAIV,WAAU,mBAAmB;AAG/B,WAAU,KAAK,IAAI,GAAG,QAAQ;AAC9B,QAAO,QAAQ,IAAI,UAAU,QAAQ;;AAevC,SAAgB,oBACd,eACA,WACA,SACA,iBACA,aACQ;CACR,MAAM,aAAa,KAAK,IAAI,GAAG,KAAK,KAAK,gBAAgB,UAAU,CAAC;CACpE,IAAI,QAAQ;AACZ,MAAK,IAAI,aAAa,GAAG,aAAa,YAAY,aAChD,UAAS,eAAe,YAAY,SAAS,QAAW,iBAAiB,YAAY;AAEvF,QAAO;;AAGT,eAAsB,eACpB,KACA,QACA,kBACkB;CAClB,MAAM,OACJ,OAAO,qBAAqB,WAAW,EAAE,SAAS,kBAAkB,GAAI,oBAAoB,EAAE;AAQhG,QAAO,aACL,SAAS,IAAI,EACb,QACA,SAAS;EACP,MAAM;EACN,YAAY,KAAK;EACjB,sBAAsB,KAAK;EAC5B,CAAC,EACF;EACE,SAAS,KAAK;EACd,kBAAkB,KAAK;EACvB,iBAAiB,KAAK;EACtB,aAAa,KAAK;EACnB,EACD;EAAE,QAAQ,OAAO,SAAS,IAAI,KAAK,SAAS;EAAW,aAAa,KAAK;EAAa,CACvF;;AAGH,MAAM,cAAsC;CAC1C,gBAAgB;CAChB,iBAAiB;CACjB,YAAY;CACb;AAOD,SAAgB,SAAuB,MAIrB;AAChB,QAAO;EACL,SAAS;EACT,GAAI,KAAK,uBAAuB,EAAE,eAAe,+BAA+B,GAAG,EAAE;EACrF,SAAS,UAAU,SAAS,KAAK,UAAU,MAAM,CAAC;EAClD,eAAe;AACb,OAAI,CAAC,KAAK,KAAM,QAAO,EAAE;GACzB,MAAM,QAAkB,EAAE;AAC1B,OAAI,KAAK,WAAY,OAAM,KAAK,SAAS,KAAK,UAAU,KAAK,WAAW,CAAC,MAAM;AAC/E,SAAM,KAAK,mBAAmB;AAC9B,UAAO;;EAEV;;AAMH,SAAgB,cAA0C,MAExC;AAChB,QAAO;EACL,SAAS;EACT,SAAS,UAAU,UAAU,MAAM,KAAK,UAAU,KAAK,UAAU,MAAM,CAAC;EACxE,eAAgB,MAAM,mBAAmB,CAAC,mBAAmB,GAAG,EAAE;EACnE;;;;;;AAOH,MAAM,qBAA6C;CACjD,8BAA8B;CAC9B,4BAA4B;CAC5B,kCAAkC;CAClC,gCAAgC;CAChC,8BAA8B;CAC9B,4BAA4B;CAC7B;AAOD,SAAgB,mBACd,KACA,QACA,MACA,SACM;CACN,MAAM,UAAkC,EAAE,gBAAgB,oBAAoB;AAC9E,KAAI,WAAW,KAAK;AAClB,UAAQ,iBAAiB,OAAO,SAAS,cAAc,EAAE;AACzD,SAAO,OAAO,SAAS,mBAAmB;;AAE5C,KAAI,UAAU,QAAQ,QAAQ;AAC9B,KAAI,IAAI,KAAK"}

package/dist/transcription.cjs

@@ -5,7 +5,8 @@ const require_sse_writer = require('./sse-writer.cjs');
 const require_chaos = require('./chaos.cjs');
 const require_router = require('./router.cjs');
 const require_recorder = require('./recorder.cjs');
-const require_interruption = require('./interruption.cjs');
+const require_deliver = require('./response/deliver.cjs');
+const require_proxy = require('./response/proxy.cjs');
 
 //#region src/transcription.ts
 /**
@@ -110,7 +111,7 @@ async function handleTranscription(req, res, raw, fixtures, journal, defaults, s
 			return;
 		}
 		if (defaults.record) {
-			const outcome = await require_recorder.proxyAndRecord(req, res, syntheticReq, "openai", req.url ?? defaultPath, fixtures, defaults, raw);
+			const { outcome, clientHungUp } = await require_proxy.withHangUpDetection(req, res, () => require_recorder.proxyAndRecord(req, res, syntheticReq, "openai", req.url ?? defaultPath, fixtures, defaults, raw));
 			if (outcome === "handled_by_hook") return;
 			if (outcome !== "not_configured") {
 				journal.add({
@@ -121,7 +122,8 @@ async function handleTranscription(req, res, raw, fixtures, journal, defaults, s
 					response: {
 						status: res.statusCode ?? 200,
 						fixture: null,
-						source: "proxy"
+						source: "proxy",
+						...require_proxy.proxyInterruption(res, clientHungUp)
 					}
 				});
 				return;
@@ -196,38 +198,22 @@ async function handleTranscription(req, res, raw, fixtures, journal, defaults, s
 			...t.languages !== void 0 ? { languages: t.languages } : {},
 			...t.usage !== void 0 ? { usage: t.usage } : {}
 		};
-		const latency = fixture.latency ?? defaults.latency;
 		const chunkSize = Math.max(1, fixture.chunkSize ?? defaults.chunkSize);
-		const replaySpeed = fixture.replaySpeed ?? defaults.replaySpeed;
-		res.writeHead(200, {
-			"Content-Type": "text/event-stream",
-			"Cache-Control": "no-cache",
-			Connection: "keep-alive"
+		const frames = [];
+		for (let index = 0; index < t.text.length; index += chunkSize) frames.push({
+			type: "transcript.text.delta",
+			delta: t.text.slice(index, index + chunkSize)
+		});
+		await require_deliver.deliverStream(res, journalEntry, fixture, {
+			...require_sse_writer.sseCodec({ done: false }),
+			writeHeadImmediately: true,
+			trailer: () => [`data: ${JSON.stringify(done)}\n\n`, "data: [DONE]\n\n"]
+		}, frames, {
+			latency: fixture.latency ?? defaults.latency,
+			streamingProfile: fixture.streamingProfile,
+			recordedTimings: fixture.recordedTimings,
+			replaySpeed: fixture.replaySpeed ?? defaults.replaySpeed
 		});
-		let chunkIndex = 0;
-		const interruption = require_interruption.createInterruptionSignal(fixture);
-		for (let index = 0; index < t.text.length; index += chunkSize) {
-			const chunkDelay = require_sse_writer.calculateDelay(chunkIndex, fixture.streamingProfile, latency, fixture.recordedTimings, replaySpeed);
-			if (chunkDelay > 0) await require_sse_writer.delay(chunkDelay, interruption?.signal);
-			if (interruption?.signal.aborted) break;
-			res.write(`data: ${JSON.stringify({
-				type: "transcript.text.delta",
-				delta: t.text.slice(index, index + chunkSize)
-			})}\n\n`);
-			interruption?.tick();
-			chunkIndex++;
-		}
-		if (interruption?.signal.aborted) {
-			journalEntry.response.interrupted = true;
-			journalEntry.response.interruptReason = interruption.reason();
-			interruption.cleanup();
-			res.destroy();
-			return;
-		}
-		res.write(`data: ${JSON.stringify(done)}\n\n`);
-		res.write("data: [DONE]\n\n");
-		res.end();
-		interruption?.cleanup();
 		return;
 	}
 	if (responseFormat === "verbose_json" || t.words != null || t.segments != null) {

package/dist/transcription.cjs.map

@@ -1 +1 @@
-{"version":3,"file":"transcription.cjs","names":["getContext","getTestId","matchFixtureDiagnostic","applyChaosAsync","flattenHeaders","resolveRequestDecisions","strictNoMatchMessage","strictNoMatchLogLine","strictOverrideField","renderError","proxyAndRecord","resolveResponse","isErrorResponse","isTranscriptionResponse","createInterruptionSignal","calculateDelay","delay"],"sources":["../src/transcription.ts"],"sourcesContent":["import { renderError } from \"./response/errors.js\";\nimport { resolveRequestDecisions, strictOverrideField } from \"./response/policy.js\";\nimport type * as http from \"node:http\";\nimport type { ChatCompletionRequest, Fixture, HandlerDefaults } from \"./types.js\";\nimport {\n  isTranscriptionResponse,\n  isErrorResponse,\n  flattenHeaders,\n  getTestId,\n  resolveResponse,\n  getContext,\n  strictNoMatchMessage,\n  strictNoMatchLogLine,\n} from \"./helpers.js\";\nimport { matchFixtureDiagnostic } from \"./router.js\";\nimport { calculateDelay, delay, writeErrorResponse } from \"./sse-writer.js\";\nimport type { Journal } from \"./journal.js\";\nimport { applyChaosAsync } from \"./chaos.js\";\nimport { proxyAndRecord } from \"./recorder.js\";\nimport { createInterruptionSignal } from \"./interruption.js\";\n\n/**\n * Extract the multipart boundary string from a Content-Type header.\n *\n * RFC 2046 §5.1.1 defines the boundary parameter as either a bare token\n * (`boundary=abc123`) or a quoted-string (`boundary=\"abc123\"`), and its\n * `bchars` production INCLUDES a space — which is exactly the case the quoted\n * form exists for, and the one a capture that stops at whitespace can never\n * read. So the quotes are consumed BY the match rather than stripped after it:\n * a post-strip runs on an already-truncated value and leaves `boundary=\"a b c\"`\n * as the useless `\"a`, and the delimiter then never matches, so every form\n * field silently falls back to its default (`whisper-1`, `gpt-image-1`, …).\n *\n * Only DQUOTE quotes. An apostrophe is a legal RFC 2045 token character and a\n * legal RFC 2046 `bchar`, so `boundary='abc'` is a BARE token whose value\n * literally includes the quotes, and stripping them would break a working\n * request.\n *\n * The `(?:^|[;\\s])` prefix keeps a decoy parameter such as `myboundary=` from\n * matching, and `\\s*=\\s*` tolerates the LWSP some clients put around the `=`.\n */\nexport function extractBoundary(contentType: string | undefined): string | undefined {\n  if (!contentType) return undefined;\n  const match = contentType.match(/(?:^|[;\\s])boundary\\s*=\\s*(?:\"([^\"]*)\"|([^\\s;]+))/i);\n  if (!match) return undefined;\n  // Group 1 is the quoted form (possibly the empty string), group 2 the bare\n  // token; an empty boundary is meaningless, so it reads as absent.\n  const boundary = match[1] !== undefined ? match[1] : match[2];\n  return boundary || undefined;\n}\n\n/**\n * Extract a text field from multipart form data using boundary-based parsing.\n * Splits the body by the multipart boundary so each part is isolated, then\n * checks each part's Content-Disposition header for the target field name.\n * This avoids false matches from binary audio data that might contain\n * header-like byte sequences.\n */\nexport function extractFormField(\n  raw: string,\n  fieldName: string,\n  boundary: string | undefined,\n): string | undefined {\n  if (!boundary) {\n    // Fallback: no boundary available, use simple regex (best-effort)\n    console.warn(\"extractFormField: no multipart boundary found, using best-effort regex fallback\");\n    const escaped = fieldName.replace(/[.*+?^${}()|[\\]\\\\]/g, \"\\\\$&\");\n    const pattern = new RegExp(\n      `Content-Disposition:\\\\s*form-data;(?:[^\\\\r\\\\n]*;)?\\\\s*name=\"${escaped}\"[^\\\\r\\\\n]*\\\\r\\\\n\\\\r\\\\n([^\\\\r\\\\n]*)`,\n      \"i\",\n    );\n    const match = raw.match(pattern);\n    return match?.[1];\n  }\n\n  // Split by boundary delimiter — each chunk is one part\n  const delimiter = `--${boundary}`;\n  const parts = raw.split(delimiter);\n\n  for (const part of parts) {\n    // Skip the preamble (before first boundary) and epilogue (after closing boundary)\n    if (!part || part.trimStart().startsWith(\"--\")) continue;\n\n    // Split part into headers and body at the first blank line (\\r\\n\\r\\n)\n    const headerEnd = part.indexOf(\"\\r\\n\\r\\n\");\n    if (headerEnd === -1) continue;\n\n    const headers = part.slice(0, headerEnd);\n    const body = part.slice(headerEnd + 4);\n\n    // Check if this part's Content-Disposition names the target field\n    const cdMatch = headers.match(\n      /Content-Disposition:\\s*form-data;(?:[^\\r\\n]*;)?\\s*name=\"([^\"]+)\"/i,\n    );\n    if (cdMatch && cdMatch[1] === fieldName) {\n      // Return the body value, trimming trailing \\r\\n from the part boundary\n      return body.replace(/\\r\\n$/, \"\");\n    }\n  }\n  return undefined;\n}\n\nexport async function handleTranscription(\n  req: http.IncomingMessage,\n  res: http.ServerResponse,\n  raw: string,\n  fixtures: Fixture[],\n  journal: Journal,\n  defaults: HandlerDefaults,\n  setCorsHeaders: (res: http.ServerResponse) => void,\n  endpointType: \"transcription\" | \"translation\" = \"transcription\",\n): Promise<void> {\n  setCorsHeaders(res);\n  const defaultPath =\n    endpointType === \"translation\" ? \"/v1/audio/translations\" : \"/v1/audio/transcriptions\";\n  const path = req.url ?? defaultPath;\n  const method = req.method ?? \"POST\";\n\n  const contentType = Array.isArray(req.headers[\"content-type\"])\n    ? req.headers[\"content-type\"][0]\n    : req.headers[\"content-type\"];\n  const boundary = extractBoundary(contentType);\n\n  const model = extractFormField(raw, \"model\", boundary) ?? \"whisper-1\";\n  const responseFormat = extractFormField(raw, \"response_format\", boundary) ?? \"json\";\n  const stream = extractFormField(raw, \"stream\", boundary) === \"true\";\n\n  const syntheticReq: ChatCompletionRequest = {\n    model,\n    messages: [],\n    _endpointType: endpointType,\n    _context: getContext(req),\n  };\n\n  const testId = getTestId(req);\n  const { fixture, skippedBySequenceOrTurn } = matchFixtureDiagnostic(\n    fixtures,\n    syntheticReq,\n    journal.getFixtureMatchCountsForTest(testId),\n    defaults.requestTransform,\n  );\n\n  if (fixture) {\n    journal.incrementFixtureMatchCount(fixture, fixtures, testId);\n    defaults.logger.debug(`Fixture matched: ${JSON.stringify(fixture.match).slice(0, 120)}`);\n  } else {\n    defaults.logger.debug(`No fixture matched for request`);\n  }\n\n  if (\n    await applyChaosAsync(\n      res,\n      fixture,\n      defaults.chaos,\n      req.headers,\n      req.url,\n      journal,\n      { method, path, headers: flattenHeaders(req.headers), body: syntheticReq },\n      fixture ? \"fixture\" : \"proxy\",\n      defaults.registry,\n      defaults.logger,\n    )\n  )\n    return;\n\n  if (!fixture) {\n    const effectiveStrict = resolveRequestDecisions(defaults.strict, req.headers).fixtureMiss;\n    if (effectiveStrict) {\n      const strictMessage = strictNoMatchMessage(skippedBySequenceOrTurn);\n      defaults.logger.error(strictNoMatchLogLine(method, path, skippedBySequenceOrTurn));\n      journal.add({\n        method,\n        path,\n        headers: flattenHeaders(req.headers),\n        body: syntheticReq,\n        response: {\n          status: 503,\n          fixture: null,\n          ...strictOverrideField(defaults.strict, req.headers),\n        },\n      });\n      writeErrorResponse(\n        res,\n        503,\n        renderError(\"openai\", \"strict-miss\", {\n          message: strictMessage,\n          type: \"invalid_request_error\",\n          code: \"no_fixture_match\",\n        }).body,\n      );\n      return;\n    }\n    if (defaults.record) {\n      const outcome = await proxyAndRecord(\n        req,\n        res,\n        syntheticReq,\n        \"openai\",\n        req.url ?? defaultPath,\n        fixtures,\n        defaults,\n        raw,\n      );\n      if (outcome === \"handled_by_hook\") return;\n      if (outcome !== \"not_configured\") {\n        journal.add({\n          method,\n          path,\n          headers: flattenHeaders(req.headers),\n          body: syntheticReq,\n          response: { status: res.statusCode ?? 200, fixture: null, source: \"proxy\" },\n        });\n        return;\n      }\n    }\n\n    journal.add({\n      method,\n      path,\n      headers: flattenHeaders(req.headers),\n      body: syntheticReq,\n      response: {\n        status: 404,\n        fixture: null,\n        ...strictOverrideField(defaults.strict, req.headers),\n      },\n    });\n    writeErrorResponse(\n      res,\n      404,\n      renderError(\"openai\", \"fixture-miss\", {\n        message: \"No fixture matched\",\n        type: \"invalid_request_error\",\n        code: \"no_fixture_match\",\n      }).body,\n    );\n    return;\n  }\n\n  const response = await resolveResponse(fixture, syntheticReq);\n\n  if (isErrorResponse(response)) {\n    const status = response.status ?? 500;\n    journal.add({\n      method,\n      path,\n      headers: flattenHeaders(req.headers),\n      body: syntheticReq,\n      response: { status, fixture },\n    });\n    writeErrorResponse(res, status, renderError(\"openai\", \"fixture\", response.error).body, {\n      retryAfter: response.retryAfter,\n    });\n    return;\n  }\n\n  if (!isTranscriptionResponse(response)) {\n    journal.add({\n      method,\n      path,\n      headers: flattenHeaders(req.headers),\n      body: syntheticReq,\n      response: { status: 500, fixture },\n    });\n    writeErrorResponse(\n      res,\n      500,\n      renderError(\"openai\", \"unknown-response\", {\n        message: \"Fixture response is not a transcription type\",\n        type: \"server_error\",\n      }).body,\n    );\n    return;\n  }\n\n  const journalEntry = journal.add({\n    method,\n    path,\n    headers: flattenHeaders(req.headers),\n    body: syntheticReq,\n    response: { status: 200, fixture },\n  });\n\n  const t = response.transcription;\n\n  // Only the transcription endpoint streams modern models. Whisper-1 ignores\n  // `stream=true`, and translations continue to return their JSON response.\n  if (endpointType === \"transcription\" && stream && model !== \"whisper-1\") {\n    const done = {\n      type: \"transcript.text.done\",\n      text: t.text,\n      ...(t.languages !== undefined ? { languages: t.languages } : {}),\n      ...(t.usage !== undefined ? { usage: t.usage } : {}),\n    };\n    const latency = fixture.latency ?? defaults.latency;\n    const chunkSize = Math.max(1, fixture.chunkSize ?? defaults.chunkSize);\n    const replaySpeed = fixture.replaySpeed ?? defaults.replaySpeed;\n    res.writeHead(200, {\n      \"Content-Type\": \"text/event-stream\",\n      \"Cache-Control\": \"no-cache\",\n      Connection: \"keep-alive\",\n    });\n    let chunkIndex = 0;\n    const interruption = createInterruptionSignal(fixture);\n    for (let index = 0; index < t.text.length; index += chunkSize) {\n      const chunkDelay = calculateDelay(\n        chunkIndex,\n        fixture.streamingProfile,\n        latency,\n        fixture.recordedTimings,\n        replaySpeed,\n      );\n      if (chunkDelay > 0) await delay(chunkDelay, interruption?.signal);\n      if (interruption?.signal.aborted) break;\n      res.write(\n        `data: ${JSON.stringify({ type: \"transcript.text.delta\", delta: t.text.slice(index, index + chunkSize) })}\\n\\n`,\n      );\n      interruption?.tick();\n      chunkIndex++;\n    }\n    if (interruption?.signal.aborted) {\n      journalEntry.response.interrupted = true;\n      journalEntry.response.interruptReason = interruption.reason();\n      interruption.cleanup();\n      res.destroy();\n      return;\n    }\n    res.write(`data: ${JSON.stringify(done)}\\n\\n`);\n    // The live `gpt-transcribe&stream=true` stream ends with the `[DONE]`\n    // sentinel after `transcript.text.done`. Omitting it leaves clients that\n    // loop until the sentinel waiting on a stream that never terminates.\n    res.write(\"data: [DONE]\\n\\n\");\n    res.end();\n    interruption?.cleanup();\n    return;\n  }\n\n  const useVerbose = responseFormat === \"verbose_json\" || t.words != null || t.segments != null;\n\n  if (useVerbose) {\n    const verboseBody: Record<string, unknown> = {\n      task: endpointType === \"translation\" ? \"translate\" : \"transcribe\",\n      language: t.language ?? \"english\",\n      duration: t.duration ?? 0,\n      text: t.text,\n    };\n    if (t.words && t.words.length > 0) {\n      verboseBody.words = t.words;\n    }\n    if (t.segments && t.segments.length > 0) {\n      verboseBody.segments = t.segments;\n    }\n    res.writeHead(200, { \"Content-Type\": \"application/json\" });\n    res.end(JSON.stringify(verboseBody));\n  } else {\n    res.writeHead(200, { \"Content-Type\": \"application/json\" });\n    res.end(\n      JSON.stringify({\n        text: t.text,\n        ...(t.languages !== undefined ? { languages: t.languages } : {}),\n        ...(t.usage !== undefined ? { usage: t.usage } : {}),\n      }),\n    );\n  }\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AAyCA,SAAgB,gBAAgB,aAAqD;AACnF,KAAI,CAAC,YAAa,QAAO;CACzB,MAAM,QAAQ,YAAY,MAAM,qDAAqD;AACrF,KAAI,CAAC,MAAO,QAAO;AAInB,SADiB,MAAM,OAAO,SAAY,MAAM,KAAK,MAAM,OACxC;;;;;;;;;AAUrB,SAAgB,iBACd,KACA,WACA,UACoB;AACpB,KAAI,CAAC,UAAU;AAEb,UAAQ,KAAK,kFAAkF;EAC/F,MAAM,UAAU,UAAU,QAAQ,uBAAuB,OAAO;EAChE,MAAM,UAAU,IAAI,OAClB,+DAA+D,QAAQ,sCACvE,IACD;AAED,SADc,IAAI,MAAM,QAAQ,GACjB;;CAIjB,MAAM,YAAY,KAAK;CACvB,MAAM,QAAQ,IAAI,MAAM,UAAU;AAElC,MAAK,MAAM,QAAQ,OAAO;AAExB,MAAI,CAAC,QAAQ,KAAK,WAAW,CAAC,WAAW,KAAK,CAAE;EAGhD,MAAM,YAAY,KAAK,QAAQ,WAAW;AAC1C,MAAI,cAAc,GAAI;EAEtB,MAAM,UAAU,KAAK,MAAM,GAAG,UAAU;EACxC,MAAM,OAAO,KAAK,MAAM,YAAY,EAAE;EAGtC,MAAM,UAAU,QAAQ,MACtB,oEACD;AACD,MAAI,WAAW,QAAQ,OAAO,UAE5B,QAAO,KAAK,QAAQ,SAAS,GAAG;;;AAMtC,eAAsB,oBACpB,KACA,KACA,KACA,UACA,SACA,UACA,gBACA,eAAgD,iBACjC;AACf,gBAAe,IAAI;CACnB,MAAM,cACJ,iBAAiB,gBAAgB,2BAA2B;CAC9D,MAAM,OAAO,IAAI,OAAO;CACxB,MAAM,SAAS,IAAI,UAAU;CAK7B,MAAM,WAAW,gBAHG,MAAM,QAAQ,IAAI,QAAQ,gBAAgB,GAC1D,IAAI,QAAQ,gBAAgB,KAC5B,IAAI,QAAQ,gBAC6B;CAE7C,MAAM,QAAQ,iBAAiB,KAAK,SAAS,SAAS,IAAI;CAC1D,MAAM,iBAAiB,iBAAiB,KAAK,mBAAmB,SAAS,IAAI;CAC7E,MAAM,SAAS,iBAAiB,KAAK,UAAU,SAAS,KAAK;CAE7D,MAAM,eAAsC;EAC1C;EACA,UAAU,EAAE;EACZ,eAAe;EACf,UAAUA,2BAAW,IAAI;EAC1B;CAED,MAAM,SAASC,0BAAU,IAAI;CAC7B,MAAM,EAAE,SAAS,4BAA4BC,sCAC3C,UACA,cACA,QAAQ,6BAA6B,OAAO,EAC5C,SAAS,iBACV;AAED,KAAI,SAAS;AACX,UAAQ,2BAA2B,SAAS,UAAU,OAAO;AAC7D,WAAS,OAAO,MAAM,oBAAoB,KAAK,UAAU,QAAQ,MAAM,CAAC,MAAM,GAAG,IAAI,GAAG;OAExF,UAAS,OAAO,MAAM,iCAAiC;AAGzD,KACE,MAAMC,8BACJ,KACA,SACA,SAAS,OACT,IAAI,SACJ,IAAI,KACJ,SACA;EAAE;EAAQ;EAAM,SAASC,+BAAe,IAAI,QAAQ;EAAE,MAAM;EAAc,EAC1E,UAAU,YAAY,SACtB,SAAS,UACT,SAAS,OACV,CAED;AAEF,KAAI,CAAC,SAAS;AAEZ,MADwBC,uCAAwB,SAAS,QAAQ,IAAI,QAAQ,CAAC,aACzD;GACnB,MAAM,gBAAgBC,qCAAqB,wBAAwB;AACnE,YAAS,OAAO,MAAMC,qCAAqB,QAAQ,MAAM,wBAAwB,CAAC;AAClF,WAAQ,IAAI;IACV;IACA;IACA,SAASH,+BAAe,IAAI,QAAQ;IACpC,MAAM;IACN,UAAU;KACR,QAAQ;KACR,SAAS;KACT,GAAGI,mCAAoB,SAAS,QAAQ,IAAI,QAAQ;KACrD;IACF,CAAC;AACF,yCACE,KACA,KACAC,2BAAY,UAAU,eAAe;IACnC,SAAS;IACT,MAAM;IACN,MAAM;IACP,CAAC,CAAC,KACJ;AACD;;AAEF,MAAI,SAAS,QAAQ;GACnB,MAAM,UAAU,MAAMC,gCACpB,KACA,KACA,cACA,UACA,IAAI,OAAO,aACX,UACA,UACA,IACD;AACD,OAAI,YAAY,kBAAmB;AACnC,OAAI,YAAY,kBAAkB;AAChC,YAAQ,IAAI;KACV;KACA;KACA,SAASN,+BAAe,IAAI,QAAQ;KACpC,MAAM;KACN,UAAU;MAAE,QAAQ,IAAI,cAAc;MAAK,SAAS;MAAM,QAAQ;MAAS;KAC5E,CAAC;AACF;;;AAIJ,UAAQ,IAAI;GACV;GACA;GACA,SAASA,+BAAe,IAAI,QAAQ;GACpC,MAAM;GACN,UAAU;IACR,QAAQ;IACR,SAAS;IACT,GAAGI,mCAAoB,SAAS,QAAQ,IAAI,QAAQ;IACrD;GACF,CAAC;AACF,wCACE,KACA,KACAC,2BAAY,UAAU,gBAAgB;GACpC,SAAS;GACT,MAAM;GACN,MAAM;GACP,CAAC,CAAC,KACJ;AACD;;CAGF,MAAM,WAAW,MAAME,gCAAgB,SAAS,aAAa;AAE7D,KAAIC,gCAAgB,SAAS,EAAE;EAC7B,MAAM,SAAS,SAAS,UAAU;AAClC,UAAQ,IAAI;GACV;GACA;GACA,SAASR,+BAAe,IAAI,QAAQ;GACpC,MAAM;GACN,UAAU;IAAE;IAAQ;IAAS;GAC9B,CAAC;AACF,wCAAmB,KAAK,QAAQK,2BAAY,UAAU,WAAW,SAAS,MAAM,CAAC,MAAM,EACrF,YAAY,SAAS,YACtB,CAAC;AACF;;AAGF,KAAI,CAACI,wCAAwB,SAAS,EAAE;AACtC,UAAQ,IAAI;GACV;GACA;GACA,SAAST,+BAAe,IAAI,QAAQ;GACpC,MAAM;GACN,UAAU;IAAE,QAAQ;IAAK;IAAS;GACnC,CAAC;AACF,wCACE,KACA,KACAK,2BAAY,UAAU,oBAAoB;GACxC,SAAS;GACT,MAAM;GACP,CAAC,CAAC,KACJ;AACD;;CAGF,MAAM,eAAe,QAAQ,IAAI;EAC/B;EACA;EACA,SAASL,+BAAe,IAAI,QAAQ;EACpC,MAAM;EACN,UAAU;GAAE,QAAQ;GAAK;GAAS;EACnC,CAAC;CAEF,MAAM,IAAI,SAAS;AAInB,KAAI,iBAAiB,mBAAmB,UAAU,UAAU,aAAa;EACvE,MAAM,OAAO;GACX,MAAM;GACN,MAAM,EAAE;GACR,GAAI,EAAE,cAAc,SAAY,EAAE,WAAW,EAAE,WAAW,GAAG,EAAE;GAC/D,GAAI,EAAE,UAAU,SAAY,EAAE,OAAO,EAAE,OAAO,GAAG,EAAE;GACpD;EACD,MAAM,UAAU,QAAQ,WAAW,SAAS;EAC5C,MAAM,YAAY,KAAK,IAAI,GAAG,QAAQ,aAAa,SAAS,UAAU;EACtE,MAAM,cAAc,QAAQ,eAAe,SAAS;AACpD,MAAI,UAAU,KAAK;GACjB,gBAAgB;GAChB,iBAAiB;GACjB,YAAY;GACb,CAAC;EACF,IAAI,aAAa;EACjB,MAAM,eAAeU,8CAAyB,QAAQ;AACtD,OAAK,IAAI,QAAQ,GAAG,QAAQ,EAAE,KAAK,QAAQ,SAAS,WAAW;GAC7D,MAAM,aAAaC,kCACjB,YACA,QAAQ,kBACR,SACA,QAAQ,iBACR,YACD;AACD,OAAI,aAAa,EAAG,OAAMC,yBAAM,YAAY,cAAc,OAAO;AACjE,OAAI,cAAc,OAAO,QAAS;AAClC,OAAI,MACF,SAAS,KAAK,UAAU;IAAE,MAAM;IAAyB,OAAO,EAAE,KAAK,MAAM,OAAO,QAAQ,UAAU;IAAE,CAAC,CAAC,MAC3G;AACD,iBAAc,MAAM;AACpB;;AAEF,MAAI,cAAc,OAAO,SAAS;AAChC,gBAAa,SAAS,cAAc;AACpC,gBAAa,SAAS,kBAAkB,aAAa,QAAQ;AAC7D,gBAAa,SAAS;AACtB,OAAI,SAAS;AACb;;AAEF,MAAI,MAAM,SAAS,KAAK,UAAU,KAAK,CAAC,MAAM;AAI9C,MAAI,MAAM,mBAAmB;AAC7B,MAAI,KAAK;AACT,gBAAc,SAAS;AACvB;;AAKF,KAFmB,mBAAmB,kBAAkB,EAAE,SAAS,QAAQ,EAAE,YAAY,MAEzE;EACd,MAAM,cAAuC;GAC3C,MAAM,iBAAiB,gBAAgB,cAAc;GACrD,UAAU,EAAE,YAAY;GACxB,UAAU,EAAE,YAAY;GACxB,MAAM,EAAE;GACT;AACD,MAAI,EAAE,SAAS,EAAE,MAAM,SAAS,EAC9B,aAAY,QAAQ,EAAE;AAExB,MAAI,EAAE,YAAY,EAAE,SAAS,SAAS,EACpC,aAAY,WAAW,EAAE;AAE3B,MAAI,UAAU,KAAK,EAAE,gBAAgB,oBAAoB,CAAC;AAC1D,MAAI,IAAI,KAAK,UAAU,YAAY,CAAC;QAC/B;AACL,MAAI,UAAU,KAAK,EAAE,gBAAgB,oBAAoB,CAAC;AAC1D,MAAI,IACF,KAAK,UAAU;GACb,MAAM,EAAE;GACR,GAAI,EAAE,cAAc,SAAY,EAAE,WAAW,EAAE,WAAW,GAAG,EAAE;GAC/D,GAAI,EAAE,UAAU,SAAY,EAAE,OAAO,EAAE,OAAO,GAAG,EAAE;GACpD,CAAC,CACH"}
+{"version":3,"file":"transcription.cjs","names":["getContext","getTestId","matchFixtureDiagnostic","applyChaosAsync","flattenHeaders","resolveRequestDecisions","strictNoMatchMessage","strictNoMatchLogLine","strictOverrideField","renderError","withHangUpDetection","proxyAndRecord","proxyInterruption","resolveResponse","isErrorResponse","isTranscriptionResponse","deliverStream","sseCodec"],"sources":["../src/transcription.ts"],"sourcesContent":["import { renderError } from \"./response/errors.js\";\nimport { resolveRequestDecisions, strictOverrideField } from \"./response/policy.js\";\nimport type * as http from \"node:http\";\nimport type { ChatCompletionRequest, Fixture, HandlerDefaults } from \"./types.js\";\nimport {\n  isTranscriptionResponse,\n  isErrorResponse,\n  flattenHeaders,\n  getTestId,\n  resolveResponse,\n  getContext,\n  strictNoMatchMessage,\n  strictNoMatchLogLine,\n} from \"./helpers.js\";\nimport { matchFixtureDiagnostic } from \"./router.js\";\nimport { sseCodec, writeErrorResponse } from \"./sse-writer.js\";\nimport type { Journal } from \"./journal.js\";\nimport { applyChaosAsync } from \"./chaos.js\";\nimport { proxyAndRecord } from \"./recorder.js\";\nimport { proxyInterruption, withHangUpDetection } from \"./response/proxy.js\";\nimport { deliverStream } from \"./response/deliver.js\";\nimport type { FrameCodec } from \"./response/frame-loop.js\";\n\n/** One `transcript.text.delta` frame of a streamed transcription. */\ninterface TranscriptDelta {\n  type: \"transcript.text.delta\";\n  delta: string;\n}\n\n/**\n * Extract the multipart boundary string from a Content-Type header.\n *\n * RFC 2046 §5.1.1 defines the boundary parameter as either a bare token\n * (`boundary=abc123`) or a quoted-string (`boundary=\"abc123\"`), and its\n * `bchars` production INCLUDES a space — which is exactly the case the quoted\n * form exists for, and the one a capture that stops at whitespace can never\n * read. So the quotes are consumed BY the match rather than stripped after it:\n * a post-strip runs on an already-truncated value and leaves `boundary=\"a b c\"`\n * as the useless `\"a`, and the delimiter then never matches, so every form\n * field silently falls back to its default (`whisper-1`, `gpt-image-1`, …).\n *\n * Only DQUOTE quotes. An apostrophe is a legal RFC 2045 token character and a\n * legal RFC 2046 `bchar`, so `boundary='abc'` is a BARE token whose value\n * literally includes the quotes, and stripping them would break a working\n * request.\n *\n * The `(?:^|[;\\s])` prefix keeps a decoy parameter such as `myboundary=` from\n * matching, and `\\s*=\\s*` tolerates the LWSP some clients put around the `=`.\n */\nexport function extractBoundary(contentType: string | undefined): string | undefined {\n  if (!contentType) return undefined;\n  const match = contentType.match(/(?:^|[;\\s])boundary\\s*=\\s*(?:\"([^\"]*)\"|([^\\s;]+))/i);\n  if (!match) return undefined;\n  // Group 1 is the quoted form (possibly the empty string), group 2 the bare\n  // token; an empty boundary is meaningless, so it reads as absent.\n  const boundary = match[1] !== undefined ? match[1] : match[2];\n  return boundary || undefined;\n}\n\n/**\n * Extract a text field from multipart form data using boundary-based parsing.\n * Splits the body by the multipart boundary so each part is isolated, then\n * checks each part's Content-Disposition header for the target field name.\n * This avoids false matches from binary audio data that might contain\n * header-like byte sequences.\n */\nexport function extractFormField(\n  raw: string,\n  fieldName: string,\n  boundary: string | undefined,\n): string | undefined {\n  if (!boundary) {\n    // Fallback: no boundary available, use simple regex (best-effort)\n    console.warn(\"extractFormField: no multipart boundary found, using best-effort regex fallback\");\n    const escaped = fieldName.replace(/[.*+?^${}()|[\\]\\\\]/g, \"\\\\$&\");\n    const pattern = new RegExp(\n      `Content-Disposition:\\\\s*form-data;(?:[^\\\\r\\\\n]*;)?\\\\s*name=\"${escaped}\"[^\\\\r\\\\n]*\\\\r\\\\n\\\\r\\\\n([^\\\\r\\\\n]*)`,\n      \"i\",\n    );\n    const match = raw.match(pattern);\n    return match?.[1];\n  }\n\n  // Split by boundary delimiter — each chunk is one part\n  const delimiter = `--${boundary}`;\n  const parts = raw.split(delimiter);\n\n  for (const part of parts) {\n    // Skip the preamble (before first boundary) and epilogue (after closing boundary)\n    if (!part || part.trimStart().startsWith(\"--\")) continue;\n\n    // Split part into headers and body at the first blank line (\\r\\n\\r\\n)\n    const headerEnd = part.indexOf(\"\\r\\n\\r\\n\");\n    if (headerEnd === -1) continue;\n\n    const headers = part.slice(0, headerEnd);\n    const body = part.slice(headerEnd + 4);\n\n    // Check if this part's Content-Disposition names the target field\n    const cdMatch = headers.match(\n      /Content-Disposition:\\s*form-data;(?:[^\\r\\n]*;)?\\s*name=\"([^\"]+)\"/i,\n    );\n    if (cdMatch && cdMatch[1] === fieldName) {\n      // Return the body value, trimming trailing \\r\\n from the part boundary\n      return body.replace(/\\r\\n$/, \"\");\n    }\n  }\n  return undefined;\n}\n\nexport async function handleTranscription(\n  req: http.IncomingMessage,\n  res: http.ServerResponse,\n  raw: string,\n  fixtures: Fixture[],\n  journal: Journal,\n  defaults: HandlerDefaults,\n  setCorsHeaders: (res: http.ServerResponse) => void,\n  endpointType: \"transcription\" | \"translation\" = \"transcription\",\n): Promise<void> {\n  setCorsHeaders(res);\n  const defaultPath =\n    endpointType === \"translation\" ? \"/v1/audio/translations\" : \"/v1/audio/transcriptions\";\n  const path = req.url ?? defaultPath;\n  const method = req.method ?? \"POST\";\n\n  const contentType = Array.isArray(req.headers[\"content-type\"])\n    ? req.headers[\"content-type\"][0]\n    : req.headers[\"content-type\"];\n  const boundary = extractBoundary(contentType);\n\n  const model = extractFormField(raw, \"model\", boundary) ?? \"whisper-1\";\n  const responseFormat = extractFormField(raw, \"response_format\", boundary) ?? \"json\";\n  const stream = extractFormField(raw, \"stream\", boundary) === \"true\";\n\n  const syntheticReq: ChatCompletionRequest = {\n    model,\n    messages: [],\n    _endpointType: endpointType,\n    _context: getContext(req),\n  };\n\n  const testId = getTestId(req);\n  const { fixture, skippedBySequenceOrTurn } = matchFixtureDiagnostic(\n    fixtures,\n    syntheticReq,\n    journal.getFixtureMatchCountsForTest(testId),\n    defaults.requestTransform,\n  );\n\n  if (fixture) {\n    journal.incrementFixtureMatchCount(fixture, fixtures, testId);\n    defaults.logger.debug(`Fixture matched: ${JSON.stringify(fixture.match).slice(0, 120)}`);\n  } else {\n    defaults.logger.debug(`No fixture matched for request`);\n  }\n\n  if (\n    await applyChaosAsync(\n      res,\n      fixture,\n      defaults.chaos,\n      req.headers,\n      req.url,\n      journal,\n      { method, path, headers: flattenHeaders(req.headers), body: syntheticReq },\n      fixture ? \"fixture\" : \"proxy\",\n      defaults.registry,\n      defaults.logger,\n    )\n  )\n    return;\n\n  if (!fixture) {\n    const effectiveStrict = resolveRequestDecisions(defaults.strict, req.headers).fixtureMiss;\n    if (effectiveStrict) {\n      const strictMessage = strictNoMatchMessage(skippedBySequenceOrTurn);\n      defaults.logger.error(strictNoMatchLogLine(method, path, skippedBySequenceOrTurn));\n      journal.add({\n        method,\n        path,\n        headers: flattenHeaders(req.headers),\n        body: syntheticReq,\n        response: {\n          status: 503,\n          fixture: null,\n          ...strictOverrideField(defaults.strict, req.headers),\n        },\n      });\n      writeErrorResponse(\n        res,\n        503,\n        renderError(\"openai\", \"strict-miss\", {\n          message: strictMessage,\n          type: \"invalid_request_error\",\n          code: \"no_fixture_match\",\n        }).body,\n      );\n      return;\n    }\n    if (defaults.record) {\n      const { outcome, clientHungUp } = await withHangUpDetection(req, res, () =>\n        proxyAndRecord(\n          req,\n          res,\n          syntheticReq,\n          \"openai\",\n          req.url ?? defaultPath,\n          fixtures,\n          defaults,\n          raw,\n        ),\n      );\n      if (outcome === \"handled_by_hook\") return;\n      if (outcome !== \"not_configured\") {\n        journal.add({\n          method,\n          path,\n          headers: flattenHeaders(req.headers),\n          body: syntheticReq,\n          response: {\n            status: res.statusCode ?? 200,\n            fixture: null,\n            source: \"proxy\",\n            ...proxyInterruption(res, clientHungUp),\n          },\n        });\n        return;\n      }\n    }\n\n    journal.add({\n      method,\n      path,\n      headers: flattenHeaders(req.headers),\n      body: syntheticReq,\n      response: {\n        status: 404,\n        fixture: null,\n        ...strictOverrideField(defaults.strict, req.headers),\n      },\n    });\n    writeErrorResponse(\n      res,\n      404,\n      renderError(\"openai\", \"fixture-miss\", {\n        message: \"No fixture matched\",\n        type: \"invalid_request_error\",\n        code: \"no_fixture_match\",\n      }).body,\n    );\n    return;\n  }\n\n  const response = await resolveResponse(fixture, syntheticReq);\n\n  if (isErrorResponse(response)) {\n    const status = response.status ?? 500;\n    journal.add({\n      method,\n      path,\n      headers: flattenHeaders(req.headers),\n      body: syntheticReq,\n      response: { status, fixture },\n    });\n    writeErrorResponse(res, status, renderError(\"openai\", \"fixture\", response.error).body, {\n      retryAfter: response.retryAfter,\n    });\n    return;\n  }\n\n  if (!isTranscriptionResponse(response)) {\n    journal.add({\n      method,\n      path,\n      headers: flattenHeaders(req.headers),\n      body: syntheticReq,\n      response: { status: 500, fixture },\n    });\n    writeErrorResponse(\n      res,\n      500,\n      renderError(\"openai\", \"unknown-response\", {\n        message: \"Fixture response is not a transcription type\",\n        type: \"server_error\",\n      }).body,\n    );\n    return;\n  }\n\n  const journalEntry = journal.add({\n    method,\n    path,\n    headers: flattenHeaders(req.headers),\n    body: syntheticReq,\n    response: { status: 200, fixture },\n  });\n\n  const t = response.transcription;\n\n  // Only the transcription endpoint streams modern models. Whisper-1 ignores\n  // `stream=true`, and translations continue to return their JSON response.\n  if (endpointType === \"transcription\" && stream && model !== \"whisper-1\") {\n    const done = {\n      type: \"transcript.text.done\",\n      text: t.text,\n      ...(t.languages !== undefined ? { languages: t.languages } : {}),\n      ...(t.usage !== undefined ? { usage: t.usage } : {}),\n    };\n    const chunkSize = Math.max(1, fixture.chunkSize ?? defaults.chunkSize);\n    const frames: TranscriptDelta[] = [];\n    for (let index = 0; index < t.text.length; index += chunkSize) {\n      frames.push({ type: \"transcript.text.delta\", delta: t.text.slice(index, index + chunkSize) });\n    }\n    // The live `gpt-transcribe&stream=true` stream ends with the `[DONE]`\n    // sentinel after `transcript.text.done`. Omitting it leaves clients that\n    // loop until the sentinel waiting on a stream that never terminates.\n    const codec: FrameCodec<TranscriptDelta> = {\n      ...sseCodec<TranscriptDelta>({ done: false }),\n      writeHeadImmediately: true,\n      trailer: () => [`data: ${JSON.stringify(done)}\\n\\n`, \"data: [DONE]\\n\\n\"],\n    };\n    await deliverStream(res, journalEntry, fixture, codec, frames, {\n      latency: fixture.latency ?? defaults.latency,\n      streamingProfile: fixture.streamingProfile,\n      recordedTimings: fixture.recordedTimings,\n      replaySpeed: fixture.replaySpeed ?? defaults.replaySpeed,\n    });\n    return;\n  }\n\n  const useVerbose = responseFormat === \"verbose_json\" || t.words != null || t.segments != null;\n\n  if (useVerbose) {\n    const verboseBody: Record<string, unknown> = {\n      task: endpointType === \"translation\" ? \"translate\" : \"transcribe\",\n      language: t.language ?? \"english\",\n      duration: t.duration ?? 0,\n      text: t.text,\n    };\n    if (t.words && t.words.length > 0) {\n      verboseBody.words = t.words;\n    }\n    if (t.segments && t.segments.length > 0) {\n      verboseBody.segments = t.segments;\n    }\n    res.writeHead(200, { \"Content-Type\": \"application/json\" });\n    res.end(JSON.stringify(verboseBody));\n  } else {\n    res.writeHead(200, { \"Content-Type\": \"application/json\" });\n    res.end(\n      JSON.stringify({\n        text: t.text,\n        ...(t.languages !== undefined ? { languages: t.languages } : {}),\n        ...(t.usage !== undefined ? { usage: t.usage } : {}),\n      }),\n    );\n  }\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AAiDA,SAAgB,gBAAgB,aAAqD;AACnF,KAAI,CAAC,YAAa,QAAO;CACzB,MAAM,QAAQ,YAAY,MAAM,qDAAqD;AACrF,KAAI,CAAC,MAAO,QAAO;AAInB,SADiB,MAAM,OAAO,SAAY,MAAM,KAAK,MAAM,OACxC;;;;;;;;;AAUrB,SAAgB,iBACd,KACA,WACA,UACoB;AACpB,KAAI,CAAC,UAAU;AAEb,UAAQ,KAAK,kFAAkF;EAC/F,MAAM,UAAU,UAAU,QAAQ,uBAAuB,OAAO;EAChE,MAAM,UAAU,IAAI,OAClB,+DAA+D,QAAQ,sCACvE,IACD;AAED,SADc,IAAI,MAAM,QAAQ,GACjB;;CAIjB,MAAM,YAAY,KAAK;CACvB,MAAM,QAAQ,IAAI,MAAM,UAAU;AAElC,MAAK,MAAM,QAAQ,OAAO;AAExB,MAAI,CAAC,QAAQ,KAAK,WAAW,CAAC,WAAW,KAAK,CAAE;EAGhD,MAAM,YAAY,KAAK,QAAQ,WAAW;AAC1C,MAAI,cAAc,GAAI;EAEtB,MAAM,UAAU,KAAK,MAAM,GAAG,UAAU;EACxC,MAAM,OAAO,KAAK,MAAM,YAAY,EAAE;EAGtC,MAAM,UAAU,QAAQ,MACtB,oEACD;AACD,MAAI,WAAW,QAAQ,OAAO,UAE5B,QAAO,KAAK,QAAQ,SAAS,GAAG;;;AAMtC,eAAsB,oBACpB,KACA,KACA,KACA,UACA,SACA,UACA,gBACA,eAAgD,iBACjC;AACf,gBAAe,IAAI;CACnB,MAAM,cACJ,iBAAiB,gBAAgB,2BAA2B;CAC9D,MAAM,OAAO,IAAI,OAAO;CACxB,MAAM,SAAS,IAAI,UAAU;CAK7B,MAAM,WAAW,gBAHG,MAAM,QAAQ,IAAI,QAAQ,gBAAgB,GAC1D,IAAI,QAAQ,gBAAgB,KAC5B,IAAI,QAAQ,gBAC6B;CAE7C,MAAM,QAAQ,iBAAiB,KAAK,SAAS,SAAS,IAAI;CAC1D,MAAM,iBAAiB,iBAAiB,KAAK,mBAAmB,SAAS,IAAI;CAC7E,MAAM,SAAS,iBAAiB,KAAK,UAAU,SAAS,KAAK;CAE7D,MAAM,eAAsC;EAC1C;EACA,UAAU,EAAE;EACZ,eAAe;EACf,UAAUA,2BAAW,IAAI;EAC1B;CAED,MAAM,SAASC,0BAAU,IAAI;CAC7B,MAAM,EAAE,SAAS,4BAA4BC,sCAC3C,UACA,cACA,QAAQ,6BAA6B,OAAO,EAC5C,SAAS,iBACV;AAED,KAAI,SAAS;AACX,UAAQ,2BAA2B,SAAS,UAAU,OAAO;AAC7D,WAAS,OAAO,MAAM,oBAAoB,KAAK,UAAU,QAAQ,MAAM,CAAC,MAAM,GAAG,IAAI,GAAG;OAExF,UAAS,OAAO,MAAM,iCAAiC;AAGzD,KACE,MAAMC,8BACJ,KACA,SACA,SAAS,OACT,IAAI,SACJ,IAAI,KACJ,SACA;EAAE;EAAQ;EAAM,SAASC,+BAAe,IAAI,QAAQ;EAAE,MAAM;EAAc,EAC1E,UAAU,YAAY,SACtB,SAAS,UACT,SAAS,OACV,CAED;AAEF,KAAI,CAAC,SAAS;AAEZ,MADwBC,uCAAwB,SAAS,QAAQ,IAAI,QAAQ,CAAC,aACzD;GACnB,MAAM,gBAAgBC,qCAAqB,wBAAwB;AACnE,YAAS,OAAO,MAAMC,qCAAqB,QAAQ,MAAM,wBAAwB,CAAC;AAClF,WAAQ,IAAI;IACV;IACA;IACA,SAASH,+BAAe,IAAI,QAAQ;IACpC,MAAM;IACN,UAAU;KACR,QAAQ;KACR,SAAS;KACT,GAAGI,mCAAoB,SAAS,QAAQ,IAAI,QAAQ;KACrD;IACF,CAAC;AACF,yCACE,KACA,KACAC,2BAAY,UAAU,eAAe;IACnC,SAAS;IACT,MAAM;IACN,MAAM;IACP,CAAC,CAAC,KACJ;AACD;;AAEF,MAAI,SAAS,QAAQ;GACnB,MAAM,EAAE,SAAS,iBAAiB,MAAMC,kCAAoB,KAAK,WAC/DC,gCACE,KACA,KACA,cACA,UACA,IAAI,OAAO,aACX,UACA,UACA,IACD,CACF;AACD,OAAI,YAAY,kBAAmB;AACnC,OAAI,YAAY,kBAAkB;AAChC,YAAQ,IAAI;KACV;KACA;KACA,SAASP,+BAAe,IAAI,QAAQ;KACpC,MAAM;KACN,UAAU;MACR,QAAQ,IAAI,cAAc;MAC1B,SAAS;MACT,QAAQ;MACR,GAAGQ,gCAAkB,KAAK,aAAa;MACxC;KACF,CAAC;AACF;;;AAIJ,UAAQ,IAAI;GACV;GACA;GACA,SAASR,+BAAe,IAAI,QAAQ;GACpC,MAAM;GACN,UAAU;IACR,QAAQ;IACR,SAAS;IACT,GAAGI,mCAAoB,SAAS,QAAQ,IAAI,QAAQ;IACrD;GACF,CAAC;AACF,wCACE,KACA,KACAC,2BAAY,UAAU,gBAAgB;GACpC,SAAS;GACT,MAAM;GACN,MAAM;GACP,CAAC,CAAC,KACJ;AACD;;CAGF,MAAM,WAAW,MAAMI,gCAAgB,SAAS,aAAa;AAE7D,KAAIC,gCAAgB,SAAS,EAAE;EAC7B,MAAM,SAAS,SAAS,UAAU;AAClC,UAAQ,IAAI;GACV;GACA;GACA,SAASV,+BAAe,IAAI,QAAQ;GACpC,MAAM;GACN,UAAU;IAAE;IAAQ;IAAS;GAC9B,CAAC;AACF,wCAAmB,KAAK,QAAQK,2BAAY,UAAU,WAAW,SAAS,MAAM,CAAC,MAAM,EACrF,YAAY,SAAS,YACtB,CAAC;AACF;;AAGF,KAAI,CAACM,wCAAwB,SAAS,EAAE;AACtC,UAAQ,IAAI;GACV;GACA;GACA,SAASX,+BAAe,IAAI,QAAQ;GACpC,MAAM;GACN,UAAU;IAAE,QAAQ;IAAK;IAAS;GACnC,CAAC;AACF,wCACE,KACA,KACAK,2BAAY,UAAU,oBAAoB;GACxC,SAAS;GACT,MAAM;GACP,CAAC,CAAC,KACJ;AACD;;CAGF,MAAM,eAAe,QAAQ,IAAI;EAC/B;EACA;EACA,SAASL,+BAAe,IAAI,QAAQ;EACpC,MAAM;EACN,UAAU;GAAE,QAAQ;GAAK;GAAS;EACnC,CAAC;CAEF,MAAM,IAAI,SAAS;AAInB,KAAI,iBAAiB,mBAAmB,UAAU,UAAU,aAAa;EACvE,MAAM,OAAO;GACX,MAAM;GACN,MAAM,EAAE;GACR,GAAI,EAAE,cAAc,SAAY,EAAE,WAAW,EAAE,WAAW,GAAG,EAAE;GAC/D,GAAI,EAAE,UAAU,SAAY,EAAE,OAAO,EAAE,OAAO,GAAG,EAAE;GACpD;EACD,MAAM,YAAY,KAAK,IAAI,GAAG,QAAQ,aAAa,SAAS,UAAU;EACtE,MAAM,SAA4B,EAAE;AACpC,OAAK,IAAI,QAAQ,GAAG,QAAQ,EAAE,KAAK,QAAQ,SAAS,UAClD,QAAO,KAAK;GAAE,MAAM;GAAyB,OAAO,EAAE,KAAK,MAAM,OAAO,QAAQ,UAAU;GAAE,CAAC;AAU/F,QAAMY,8BAAc,KAAK,cAAc,SALI;GACzC,GAAGC,4BAA0B,EAAE,MAAM,OAAO,CAAC;GAC7C,sBAAsB;GACtB,eAAe,CAAC,SAAS,KAAK,UAAU,KAAK,CAAC,OAAO,mBAAmB;GACzE,EACsD,QAAQ;GAC7D,SAAS,QAAQ,WAAW,SAAS;GACrC,kBAAkB,QAAQ;GAC1B,iBAAiB,QAAQ;GACzB,aAAa,QAAQ,eAAe,SAAS;GAC9C,CAAC;AACF;;AAKF,KAFmB,mBAAmB,kBAAkB,EAAE,SAAS,QAAQ,EAAE,YAAY,MAEzE;EACd,MAAM,cAAuC;GAC3C,MAAM,iBAAiB,gBAAgB,cAAc;GACrD,UAAU,EAAE,YAAY;GACxB,UAAU,EAAE,YAAY;GACxB,MAAM,EAAE;GACT;AACD,MAAI,EAAE,SAAS,EAAE,MAAM,SAAS,EAC9B,aAAY,QAAQ,EAAE;AAExB,MAAI,EAAE,YAAY,EAAE,SAAS,SAAS,EACpC,aAAY,WAAW,EAAE;AAE3B,MAAI,UAAU,KAAK,EAAE,gBAAgB,oBAAoB,CAAC;AAC1D,MAAI,IAAI,KAAK,UAAU,YAAY,CAAC;QAC/B;AACL,MAAI,UAAU,KAAK,EAAE,gBAAgB,oBAAoB,CAAC;AAC1D,MAAI,IACF,KAAK,UAAU;GACb,MAAM,EAAE;GACR,GAAI,EAAE,cAAc,SAAY,EAAE,WAAW,EAAE,WAAW,GAAG,EAAE;GAC/D,GAAI,EAAE,UAAU,SAAY,EAAE,OAAO,EAAE,OAAO,GAAG,EAAE;GACpD,CAAC,CACH"}

package/dist/transcription.d.cts.map

@@ -1 +1 @@
-{"version":3,"file":"transcription.d.cts","names":[],"sources":["../src/transcription.ts"],"sourcesContent":[],"mappings":";;;;;;iBAsGsB,mBAAA,MACf,MAAA,CAAK,sBACL,MAAA,CAAK,uCAEA,oBACD,mBACC,uCACY,MAAA,CAAK,0EAE1B"}
+{"version":3,"file":"transcription.d.cts","names":[],"sources":["../src/transcription.ts"],"sourcesContent":[],"mappings":";;;;;;iBA8GsB,mBAAA,MACf,MAAA,CAAK,sBACL,MAAA,CAAK,uCAEA,oBACD,mBACC,uCACY,MAAA,CAAK,0EAE1B"}

package/dist/transcription.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"transcription.d.ts","names":[],"sources":["../src/transcription.ts"],"sourcesContent":[],"mappings":";;;;;;iBAsGsB,mBAAA,MACf,MAAA,CAAK,sBACL,MAAA,CAAK,uCAEA,oBACD,mBACC,uCACY,MAAA,CAAK,0EAE1B"}
+{"version":3,"file":"transcription.d.ts","names":[],"sources":["../src/transcription.ts"],"sourcesContent":[],"mappings":";;;;;;iBA8GsB,mBAAA,MACf,MAAA,CAAK,sBACL,MAAA,CAAK,uCAEA,oBACD,mBACC,uCACY,MAAA,CAAK,0EAE1B"}

package/dist/transcription.js

@@ -1,11 +1,12 @@
 import { renderError } from "./response/errors.js";
 import { resolveRequestDecisions, strictOverrideField } from "./response/policy.js";
 import { flattenHeaders, getContext, getTestId, isErrorResponse, isTranscriptionResponse, resolveResponse, strictNoMatchLogLine, strictNoMatchMessage } from "./helpers.js";
-import { calculateDelay, delay, writeErrorResponse } from "./sse-writer.js";
+import { sseCodec, writeErrorResponse } from "./sse-writer.js";
 import { applyChaosAsync } from "./chaos.js";
 import { matchFixtureDiagnostic } from "./router.js";
 import { proxyAndRecord } from "./recorder.js";
-import { createInterruptionSignal } from "./interruption.js";
+import { deliverStream } from "./response/deliver.js";
+import { proxyInterruption, withHangUpDetection } from "./response/proxy.js";
 
 //#region src/transcription.ts
 /**
@@ -110,7 +111,7 @@ async function handleTranscription(req, res, raw, fixtures, journal, defaults, s
 			return;
 		}
 		if (defaults.record) {
-			const outcome = await proxyAndRecord(req, res, syntheticReq, "openai", req.url ?? defaultPath, fixtures, defaults, raw);
+			const { outcome, clientHungUp } = await withHangUpDetection(req, res, () => proxyAndRecord(req, res, syntheticReq, "openai", req.url ?? defaultPath, fixtures, defaults, raw));
 			if (outcome === "handled_by_hook") return;
 			if (outcome !== "not_configured") {
 				journal.add({
@@ -121,7 +122,8 @@ async function handleTranscription(req, res, raw, fixtures, journal, defaults, s
 					response: {
 						status: res.statusCode ?? 200,
 						fixture: null,
-						source: "proxy"
+						source: "proxy",
+						...proxyInterruption(res, clientHungUp)
 					}
 				});
 				return;
@@ -196,38 +198,22 @@ async function handleTranscription(req, res, raw, fixtures, journal, defaults, s
 			...t.languages !== void 0 ? { languages: t.languages } : {},
 			...t.usage !== void 0 ? { usage: t.usage } : {}
 		};
-		const latency = fixture.latency ?? defaults.latency;
 		const chunkSize = Math.max(1, fixture.chunkSize ?? defaults.chunkSize);
-		const replaySpeed = fixture.replaySpeed ?? defaults.replaySpeed;
-		res.writeHead(200, {
-			"Content-Type": "text/event-stream",
-			"Cache-Control": "no-cache",
-			Connection: "keep-alive"
+		const frames = [];
+		for (let index = 0; index < t.text.length; index += chunkSize) frames.push({
+			type: "transcript.text.delta",
+			delta: t.text.slice(index, index + chunkSize)
+		});
+		await deliverStream(res, journalEntry, fixture, {
+			...sseCodec({ done: false }),
+			writeHeadImmediately: true,
+			trailer: () => [`data: ${JSON.stringify(done)}\n\n`, "data: [DONE]\n\n"]
+		}, frames, {
+			latency: fixture.latency ?? defaults.latency,
+			streamingProfile: fixture.streamingProfile,
+			recordedTimings: fixture.recordedTimings,
+			replaySpeed: fixture.replaySpeed ?? defaults.replaySpeed
 		});
-		let chunkIndex = 0;
-		const interruption = createInterruptionSignal(fixture);
-		for (let index = 0; index < t.text.length; index += chunkSize) {
-			const chunkDelay = calculateDelay(chunkIndex, fixture.streamingProfile, latency, fixture.recordedTimings, replaySpeed);
-			if (chunkDelay > 0) await delay(chunkDelay, interruption?.signal);
-			if (interruption?.signal.aborted) break;
-			res.write(`data: ${JSON.stringify({
-				type: "transcript.text.delta",
-				delta: t.text.slice(index, index + chunkSize)
-			})}\n\n`);
-			interruption?.tick();
-			chunkIndex++;
-		}
-		if (interruption?.signal.aborted) {
-			journalEntry.response.interrupted = true;
-			journalEntry.response.interruptReason = interruption.reason();
-			interruption.cleanup();
-			res.destroy();
-			return;
-		}
-		res.write(`data: ${JSON.stringify(done)}\n\n`);
-		res.write("data: [DONE]\n\n");
-		res.end();
-		interruption?.cleanup();
 		return;
 	}
 	if (responseFormat === "verbose_json" || t.words != null || t.segments != null) {