@mux/ai 0.54.0 → 0.55.0

package/dist/{transcripts-BCtzC8hZ.d.ts → transcripts-DxSHvTk9.d.ts}

@@ -604,4 +604,4 @@ interface TranscriptFetchStepResult {
 declare function fetchTranscriptDataForUrl(transcriptUrl: string, options?: Pick<TranscriptFetchOptions, "cleanTranscript" | "scope">): Promise<TranscriptFetchStepResult>;
 declare function fetchTranscriptForAsset(asset: MuxAsset, playbackId: string, options?: TranscriptFetchOptions): Promise<TranscriptResult>;
 
-export { fetchTranscriptDataForUrl as $, type AssetTextTrack as A, type ShotsResult as B, type ChunkEmbedding as C, type DeletedShotsResult as D, type Encrypted as E, type SkippedShotsResult as F, type TranscriptEmptyReason as G, type Hotspot as H, type ImageSubmissionMode as I, type TranscriptFetchOptions as J, type TranscriptFetchStepResult as K, LOW_CONFIDENCE_THRESHOLD as L, type MuxAsset as M, type TranscriptResult as N, type WaitForShotsOptions as O, type PlaybackAsset as P, buildTranscriptUrl as Q, buildVttFromCueBlocks as R, type StoragePutObjectInput as S, type TextChunk as T, type UsageMetadata as U, type VTTChunkingConfig as V, type WorkflowCredentials as W, buildVttFromTranslatedCueBlocks as X, concatenateVttSegments as Y, extractTextFromVTT as Z, extractTimestampedTranscript as _, type StoragePresignGetObjectInput as a, fetchTranscriptForAsset as a0, filterVttByScope as a1, findCaptionTrack as a2, getHotspotsForAsset as a3, getHotspotsForPlaybackId as a4, getHotspotsForVideo as a5, getReadyTextTracks as a6, getReliableLanguageCode as a7, getShotsForAsset as a8, parseVTTCues as a9, replaceCueText as aa, requestShotsForAsset as ab, sanitizeUntrustedText as ac, secondsToTimestamp as ad, splitVttPreambleAndCueBlocks as ae, stripVttMetadataBlocks as af, vttTimestampToSeconds as ag, waitForShotsForAsset as ah, type ChunkingStrategy as b, type EncryptedPayload as c, type MuxAIOptions as d, type PlaybackPolicy as e, type ScopedMuxAIOptions as f, type StorageAdapter as g, type TokenChunkingConfig as h, type TokenUsage as i, type ToneType as j, type VideoEmbeddingsResult as k, type WorkflowCredentialsInput as l, type WorkflowMuxClient as m, type WorkflowScope as n, decryptFromWorkflow as o, encryptForWorkflow as p, type VTTCue as q, type CompletedShotsResult as r, type ErroredShotsResult as s, type HotspotOptions as t, type HotspotResponse as u, MAX_CUE_TEXT_CHARS as v, type ParseVTTCuesOptions as w, type PendingShotsResult as x, type Shot as y, type ShotRequestOptions as z };
+export { fetchTranscriptDataForUrl as $, type AssetTextTrack as A, type ShotsResult as B, type ChunkEmbedding as C, type DeletedShotsResult as D, type Encrypted as E, type SkippedShotsResult as F, type TranscriptEmptyReason as G, type Hotspot as H, type ImageSubmissionMode as I, type TranscriptFetchOptions as J, type TranscriptFetchStepResult as K, LOW_CONFIDENCE_THRESHOLD as L, type MuxAsset as M, type TranscriptResult as N, type WaitForShotsOptions as O, type PlaybackAsset as P, buildTranscriptUrl as Q, buildVttFromCueBlocks as R, type StoragePutObjectInput as S, type TokenUsage as T, type UsageMetadata as U, type VTTChunkingConfig as V, type WorkflowCredentials as W, buildVttFromTranslatedCueBlocks as X, concatenateVttSegments as Y, extractTextFromVTT as Z, extractTimestampedTranscript as _, type StoragePresignGetObjectInput as a, fetchTranscriptForAsset as a0, filterVttByScope as a1, findCaptionTrack as a2, getHotspotsForAsset as a3, getHotspotsForPlaybackId as a4, getHotspotsForVideo as a5, getReadyTextTracks as a6, getReliableLanguageCode as a7, getShotsForAsset as a8, parseVTTCues as a9, replaceCueText as aa, requestShotsForAsset as ab, sanitizeUntrustedText as ac, secondsToTimestamp as ad, splitVttPreambleAndCueBlocks as ae, stripVttMetadataBlocks as af, vttTimestampToSeconds as ag, waitForShotsForAsset as ah, type ChunkingStrategy as b, type EncryptedPayload as c, type MuxAIOptions as d, type PlaybackPolicy as e, type ScopedMuxAIOptions as f, type StorageAdapter as g, type TextChunk as h, type TokenChunkingConfig as i, type ToneType as j, type VideoEmbeddingsResult as k, type WorkflowCredentialsInput as l, type WorkflowMuxClient as m, type WorkflowScope as n, decryptFromWorkflow as o, encryptForWorkflow as p, type VTTCue as q, type CompletedShotsResult as r, type ErroredShotsResult as s, type HotspotOptions as t, type HotspotResponse as u, MAX_CUE_TEXT_CHARS as v, type ParseVTTCuesOptions as w, type PendingShotsResult as x, type Shot as y, type ShotRequestOptions as z };

package/dist/workflows/index.d.ts

@@ -1,7 +1,7 @@
-export { J as ASK_QUESTIONS_DEFAULT_MAX_ANSWER_OPTION_LENGTH, K as ASK_QUESTIONS_DEFAULT_MAX_FREE_FORM_ANSWER_LENGTH, N as ASK_QUESTIONS_MAX_QUESTIONS_PER_CALL, O as ASK_QUESTIONS_MAX_QUESTION_LENGTH, P as AskQuestionsOptions, Q as AskQuestionsResult, U as AskQuestionsType, V as AudioTranslationOptions, W as AudioTranslationResult, X as AutoCensorProfanityOptions, Y as BurnedInCaptionsAnalysis, Z as BurnedInCaptionsOptions, _ as BurnedInCaptionsPromptOverrides, $ as BurnedInCaptionsPromptSections, a0 as BurnedInCaptionsResult, a1 as CaptionReplacement, a2 as CensorMode, a3 as Chapter, a4 as ChapterSystemPromptSections, a5 as ChaptersOptions, a6 as ChaptersPromptOverrides, a7 as ChaptersPromptSections, a8 as ChaptersResult, a9 as ChaptersType, aa as DEFAULT_TRANSCRIPT_WINDOWING, ab as EditCaptionsOptions, ac as EditCaptionsResult, ad as EmbeddingsOptions, ae as EmbeddingsResult, af as EngagementInsightsOptions, ag as EngagementInsightsResult, ah as GENERATE_TEXT_CALLS_TO_ACTION, ai as GENERATE_TEXT_CHANNELS, aj as GENERATE_TEXT_CHANNEL_CEILINGS, ak as GENERATE_TEXT_FORMATS, al as GENERATE_TEXT_LENGTH_BOUNDS, am as GENERATE_TEXT_MAX_ARTIFACTS, an as GENERATE_TEXT_MAX_AUDIENCE_CHARS, ao as GENERATE_TEXT_MAX_BRAND_TERMS, ap as GENERATE_TEXT_MAX_BRAND_TERMS_TOTAL_CHARS, aq as GENERATE_TEXT_MAX_BRAND_TERM_CHARS, ar as GENERATE_TEXT_MAX_CONCURRENT_GENERATIONS, as as GENERATE_TEXT_MAX_INSTRUCTIONS_CHARS, at as GENERATE_TEXT_MAX_SHOT_FRAMES, au as GENERATE_TEXT_MAX_VARIANTS, av as GENERATE_TEXT_VOICES, aw as GOOGLE_VISION_LIKELIHOOD_TO_SCORE, ax as GenerateTextArtifact, ay as GenerateTextCallToAction, az as GenerateTextChannel, aA as GenerateTextFormat, aB as GenerateTextKeyedItem, aC as GenerateTextLengthLimit, aD as GenerateTextOptions, aE as GenerateTextResult, aF as GenerateTextVariant, aG as GenerateTextVoice, aH as GeneratedTextArtifact, aI as GeneratedTextVariant, aJ as HIVE_SEXUAL_CATEGORIES, aK as HIVE_VIOLENCE_CATEGORIES, aL as HiveModerationOutput, aM as HiveModerationSource, aN as ModerationOptions, aO as ModerationProvider, aP as ModerationResult, aQ as ModerationSurfaceStatus, aR as MomentInsight, aS as OverallInsight, aT as ProfanityDetectionPayload, aU as Question, aV as QuestionAnswer, aW as QuestionAnswerType, aX as RepairAttempt, aY as RepairCall, aZ as RepairOptions, a_ as RepairPlan, a$ as RepairResult, b0 as RepairSpan, b1 as ReplacementRecord, b2 as SpeakerReplacement, b3 as StaticRenditionCleanupOutcome, b4 as SummarizationOptions, b5 as SummarizationPromptOverrides, b6 as SummarizationPromptSections, b7 as SummaryAndTagsResult, b8 as SummaryType, b9 as TextLengthLimit, ba as TextRepairModelOptions, bb as ThumbnailModerationScore, bc as ThumbnailModerationSkipReason, bd as ThumbnailModerationStatus, be as TranscriptModerationScore, bf as TranscriptModerationSkipReason, bg as TranscriptModerationStatus, bh as TranslationChunkingOptions, bi as TranslationOptions, bj as TranslationPayload, bk as TranslationResult, bl as aggregateTokenUsage, bm as applyOverrideLists, bn as applyReplacements, bo as applySpeakerReplacements, bp as applyTextRepair, bq as askQuestions, br as buildReplacementRegex, bs as buildTranscriptWindows, bt as burnedInCaptionsSchema, bu as censorVttContent, bv as chapterSchema, bw as chaptersSchema, bx as createReplacer, by as createTextRepairGenerator, bz as editCaptions, bA as extractTranscriptSegmentsForHotspots, bB as findGenerateTextLengthViolation, bC as generateChapters, bD as generateEmbeddings, bE as generateEngagementInsights, bF as generateText, bG as generateVideoEmbeddings, bH as getModerationScores, bI as getSummaryAndTags, bJ as hasBurnedInCaptions, bK as mapShotsToHotspots, bL as measureGenerateTextLength, bM as normalizeTranslatedVtt, bN as planTextRepair, bO as profanityDetectionSchema, bP as questionAnswerSchema, bQ as repairJsonSchema, bR as repairText, bS as resolveGenerateTextLengthLimit, bT as resolveGenerateTextLengthLimits, bU as resolveGenerateTextOptions, bV as runTextRepairLoop, bW as selectGenerateTextShotFrames, bX as shouldSplitChunkTranslationError, bY as summarySchema, bZ as transformCueText, b_ as translateAudio, b$ as translateCaptions, c0 as translationSchema, c1 as validateNeverTranslateTerms, c2 as verifyNeverTranslateTerms } from '../index-CmTC1w4V.js';
+export { J as ASK_QUESTIONS_DEFAULT_MAX_ANSWER_OPTION_LENGTH, K as ASK_QUESTIONS_DEFAULT_MAX_FREE_FORM_ANSWER_LENGTH, N as ASK_QUESTIONS_MAX_QUESTIONS_PER_CALL, O as ASK_QUESTIONS_MAX_QUESTION_LENGTH, P as AskQuestionsOptions, Q as AskQuestionsResult, U as AskQuestionsType, V as AudioTranslationOptions, W as AudioTranslationResult, X as AutoCensorProfanityOptions, Y as BurnedInCaptionsAnalysis, Z as BurnedInCaptionsOptions, _ as BurnedInCaptionsPromptOverrides, $ as BurnedInCaptionsPromptSections, a0 as BurnedInCaptionsResult, a1 as CaptionReplacement, a2 as CensorMode, a3 as Chapter, a4 as ChapterSystemPromptSections, a5 as ChaptersOptions, a6 as ChaptersPromptOverrides, a7 as ChaptersPromptSections, a8 as ChaptersResult, a9 as ChaptersType, aa as DEFAULT_TRANSCRIPT_WINDOWING, ab as EditCaptionsOptions, ac as EditCaptionsResult, ad as EmbeddingsOptions, ae as EmbeddingsResult, af as EngagementInsightsOptions, ag as EngagementInsightsResult, ah as GENERATE_TEXT_CALLS_TO_ACTION, ai as GENERATE_TEXT_CHANNELS, aj as GENERATE_TEXT_CHANNEL_CEILINGS, ak as GENERATE_TEXT_FORMATS, al as GENERATE_TEXT_LENGTH_BOUNDS, am as GENERATE_TEXT_MAX_ARTIFACTS, an as GENERATE_TEXT_MAX_AUDIENCE_CHARS, ao as GENERATE_TEXT_MAX_BRAND_TERMS, ap as GENERATE_TEXT_MAX_BRAND_TERMS_TOTAL_CHARS, aq as GENERATE_TEXT_MAX_BRAND_TERM_CHARS, ar as GENERATE_TEXT_MAX_CONCURRENT_GENERATIONS, as as GENERATE_TEXT_MAX_INSTRUCTIONS_CHARS, at as GENERATE_TEXT_MAX_SHOT_FRAMES, au as GENERATE_TEXT_MAX_VARIANTS, av as GENERATE_TEXT_VOICES, aw as GOOGLE_VISION_LIKELIHOOD_TO_SCORE, ax as GenerateTextArtifact, ay as GenerateTextCallToAction, az as GenerateTextChannel, aA as GenerateTextFormat, aB as GenerateTextKeyedItem, aC as GenerateTextLengthLimit, aD as GenerateTextOptions, aE as GenerateTextResult, aF as GenerateTextVariant, aG as GenerateTextVoice, aH as GeneratedTextArtifact, aI as GeneratedTextVariant, aJ as HIVE_SEXUAL_CATEGORIES, aK as HIVE_VIOLENCE_CATEGORIES, aL as HiveModerationOutput, aM as HiveModerationSource, aN as ModerationOptions, aO as ModerationProvider, aP as ModerationResult, aQ as ModerationSurfaceStatus, aR as MomentInsight, aS as OverallInsight, aT as ProfanityDetectionPayload, aU as Question, aV as QuestionAnswer, aW as QuestionAnswerType, aX as RepairAttempt, aY as RepairCall, aZ as RepairOptions, a_ as RepairPlan, a$ as RepairResult, b0 as RepairSpan, b1 as ReplacementRecord, b2 as SpeakerReplacement, b3 as StaticRenditionCleanupOutcome, b4 as SummarizationOptions, b5 as SummarizationPromptOverrides, b6 as SummarizationPromptSections, b7 as SummaryAndTagsResult, b8 as SummaryType, b9 as TextLengthLimit, ba as TextRepairModelOptions, bb as ThumbnailModerationScore, bc as ThumbnailModerationSkipReason, bd as ThumbnailModerationStatus, be as TranscriptModerationScore, bf as TranscriptModerationSkipReason, bg as TranscriptModerationStatus, bh as TranslationChunkingOptions, bi as TranslationOptions, bj as TranslationPayload, bk as TranslationResult, bl as aggregateTokenUsage, bm as applyOverrideLists, bn as applyReplacements, bo as applySpeakerReplacements, bp as applyTextRepair, bq as askQuestions, br as buildReplacementRegex, bs as buildTranscriptWindows, bt as burnedInCaptionsSchema, bu as censorVttContent, bv as chapterSchema, bw as chaptersSchema, bx as createReplacer, by as createTextRepairGenerator, bz as editCaptions, bA as extractTranscriptSegmentsForHotspots, bB as findGenerateTextLengthViolation, bC as generateChapters, bD as generateEmbeddings, bE as generateEngagementInsights, bF as generateText, bG as generateVideoEmbeddings, bH as getModerationScores, bI as getSummaryAndTags, bJ as hasBurnedInCaptions, bK as mapShotsToHotspots, bL as measureGenerateTextLength, bM as normalizeTranslatedVtt, bN as planTextRepair, bO as profanityDetectionSchema, bP as questionAnswerSchema, bQ as repairJsonSchema, bR as repairText, bS as resolveGenerateTextLengthLimit, bT as resolveGenerateTextLengthLimits, bU as resolveGenerateTextOptions, bV as runTextRepairLoop, bW as selectGenerateTextShotFrames, bX as shouldSplitChunkTranslationError, bY as summarySchema, bZ as transformCueText, b_ as translateAudio, b$ as translateCaptions, c0 as translationSchema, c1 as validateNeverTranslateTerms, c2 as verifyNeverTranslateTerms } from '../index-BBk0AcY1.js';
 export { D as DEFAULT_DESCRIPTION_LENGTH, a as DEFAULT_SUMMARY_KEYWORD_LIMIT, b as DEFAULT_TITLE_LENGTH } from '../summarization-BFmhXeAu.js';
 import 'ai';
-import '../transcripts-BCtzC8hZ.js';
+import '../transcripts-DxSHvTk9.js';
 import '@mux/ts';
 import 'zod';
 import '@ai-sdk/anthropic';

package/dist/workflows/index.js

@@ -5,6 +5,10 @@ import { generateText, jsonSchema, NoObjectGeneratedError as NoObjectGeneratedEr
 import { APICallError as APICallError2, NoObjectGeneratedError, RetryError } from "ai";
 import { z as z2 } from "zod";
 
+// node_modules/@workflow/serde/dist/index.js
+var WORKFLOW_SERIALIZE = /* @__PURE__ */ Symbol.for("workflow-serialize");
+var WORKFLOW_DESERIALIZE = /* @__PURE__ */ Symbol.for("workflow-deserialize");
+
 // src/lib/prompt-fragments.ts
 import dedent from "dedent";
 
@@ -526,7 +530,7 @@ function rethrowWithTokenUsage(error, collectedUsage) {
 }
 
 // src/lib/mux-ai-error.ts
-var MuxAiError = class extends Error {
+var MuxAiError = class _MuxAiError extends Error {
   constructor(message, opts) {
     super(message);
     /** Structural brand for duck-type identification. */
@@ -542,6 +546,32 @@ var MuxAiError = class extends Error {
   static is(value) {
     return typeof value === "object" && value !== null && "__robots_error" in value && value.__robots_error === true;
   }
+  static [WORKFLOW_SERIALIZE](instance) {
+    const incomplete = instance;
+    return {
+      message: instance.publicMessage,
+      publicType: instance.publicType,
+      retryable: instance.retryable,
+      stack: instance.stack,
+      usage: getErrorTokenUsage(instance),
+      finishReason: incomplete.finishReason,
+      rawFinishReason: incomplete.rawFinishReason,
+      diagnostics: instance.diagnostics
+    };
+  }
+  static [WORKFLOW_DESERIALIZE](value) {
+    const error = new _MuxAiError(value.message, { type: value.publicType, retryable: value.retryable });
+    Object.assign(error, {
+      usage: value.usage,
+      diagnostics: value.diagnostics,
+      ...value.finishReason !== void 0 && { finishReason: value.finishReason },
+      ...value.rawFinishReason !== void 0 && { rawFinishReason: value.rawFinishReason }
+    });
+    if (value.stack !== void 0) {
+      error.stack = value.stack;
+    }
+    return error;
+  }
 };
 function extractErrorDetail(error) {
   if (error instanceof Error && error.message) {
@@ -615,7 +645,8 @@ async function withRetry(fn, {
   maxRetries = DEFAULT_RETRY_OPTIONS.maxRetries,
   baseDelay = DEFAULT_RETRY_OPTIONS.baseDelay,
   maxDelay = DEFAULT_RETRY_OPTIONS.maxDelay,
-  shouldRetry = defaultShouldRetry
+  shouldRetry = defaultShouldRetry,
+  onRetry
 } = {}) {
   let lastError;
   for (let attempt = 0; attempt <= maxRetries; attempt++) {
@@ -628,6 +659,7 @@ async function withRetry(fn, {
       if (isLastAttempt || !shouldRetry(retryError, attempt + 1)) {
         throw error;
       }
+      onRetry?.(error, attempt + 1);
       const delay = calculateDelay(attempt + 1, baseDelay, maxDelay);
       console.warn(
         `Attempt ${attempt + 1} failed: ${retryError.message}. Retrying in ${Math.round(delay)}ms...`
@@ -721,10 +753,54 @@ async function withContentPolicyErrorHandling(operation) {
   }
 }
 async function withContentPolicyAwareRetry(operation, retryOptions) {
-  return withRetry(
-    () => withContentPolicyErrorHandling(operation),
-    retryOptions
-  );
+  const retriedUsage = [];
+  try {
+    const response = await withRetry(
+      () => withContentPolicyErrorHandling(operation),
+      {
+        ...retryOptions,
+        onRetry: (error, attempt) => {
+          const usage = getErrorTokenUsage(error);
+          if (usage) {
+            retriedUsage.push(usage);
+          }
+          retryOptions?.onRetry?.(error, attempt);
+        }
+      }
+    );
+    return includeRetriedUsage(response, retriedUsage);
+  } catch (error) {
+    rethrowWithTokenUsage(error, retriedUsage);
+  }
+}
+function includeRetriedUsage(response, retriedUsage) {
+  if (!retriedUsage.length || typeof response !== "object" || response === null) {
+    return response;
+  }
+  const result = response;
+  const merge = (usage) => {
+    const combined = aggregateTokenUsage([...retriedUsage, getErrorTokenUsage({ usage }) ?? {}]);
+    return {
+      ...usage,
+      ...combined,
+      ...combined.cachedInputTokens !== void 0 || combined.cacheWriteTokens !== void 0 ? {
+        inputTokenDetails: {
+          ...usage?.inputTokenDetails,
+          ...combined.cachedInputTokens !== void 0 && { cacheReadTokens: combined.cachedInputTokens },
+          ...combined.cacheWriteTokens !== void 0 && { cacheWriteTokens: combined.cacheWriteTokens }
+        }
+      } : {},
+      ...combined.reasoningTokens !== void 0 ? {
+        outputTokenDetails: { ...usage?.outputTokenDetails, reasoningTokens: combined.reasoningTokens }
+      } : {}
+    };
+  };
+  const descriptors = Object.getOwnPropertyDescriptors(response);
+  descriptors.usage = { value: merge(result.usage), enumerable: true, configurable: true };
+  if ("totalUsage" in result) {
+    descriptors.totalUsage = { value: merge(result.totalUsage), enumerable: true, configurable: true };
+  }
+  return Object.create(Object.getPrototypeOf(response), descriptors);
 }
 function rethrowContentPolicyError(error) {
   const block = extractContentPolicyBlock(error);
@@ -860,6 +936,35 @@ function findBlockedCategory(ratings) {
   return ratings?.find((rating) => rating.blocked === true)?.category ?? void 0;
 }
 
+// src/lib/model-call-diagnostics.ts
+import { APICallError as APICallError3 } from "ai";
+function recordModelCallFailure(error, diagnostics, observedUsage = getErrorTokenUsage(error)) {
+  if (typeof error !== "object" || error === null) {
+    return;
+  }
+  const failure = error;
+  if (failure.diagnostics) {
+    return;
+  }
+  failure.diagnostics = { ...diagnostics };
+  console.warn("[@mux/ai] Model call failed.", {
+    ...diagnostics,
+    finishReason: failure.finishReason,
+    statusCode: failure.statusCode,
+    errorName: error instanceof Error ? error.name : void 0,
+    requestId: APICallError3.isInstance(error) ? error.responseHeaders?.["x-request-id"] ?? error.responseHeaders?.["request-id"] : void 0,
+    usage: observedUsage
+  });
+}
+async function withModelCallDiagnostics(diagnostics, operation) {
+  try {
+    return await operation();
+  } catch (error) {
+    recordModelCallFailure(error, diagnostics);
+    throw error;
+  }
+}
+
 // src/lib/text-repair-capabilities.ts
 var OPENAI_REPAIR_CAPABILITIES = {
   "gpt-5.1": { supportsReasoningNone: true },
@@ -1021,13 +1126,14 @@ function createTextRepairGenerator(options) {
   return async (plan) => {
     let usage;
     const started = Date.now();
+    const maxOutputTokens = options.maxOutputTokens ?? Math.min(32768, Math.max(1024, plan.spans.reduce((total, span) => total + Math.min(...span.budgets.map((limit) => limit.value * (limit.unit === "words" ? 4 : 2))), 0) + 128));
     try {
       const response = await withContentPolicyErrorHandling(() => generateText({
         model: options.model,
         providerOptions: options.providerOptions ?? getTextRepairProviderOptions(options.model),
         maxRetries: 0,
         abortSignal: AbortSignal.timeout(options.timeoutMs ?? 3e4),
-        maxOutputTokens: options.maxOutputTokens ?? Math.min(32768, Math.max(1024, plan.spans.reduce((total, span) => total + Math.min(...span.budgets.map((limit) => limit.value * (limit.unit === "words" ? 4 : 2))), 0) + 128)),
+        maxOutputTokens,
         output: Output.object({ schema: jsonSchema(repairJsonSchema(plan)), name: "paragraph_replacements" }),
         system: promptDedent`
           ${NON_DISCLOSURE_CONSTRAINT}
@@ -1055,6 +1161,12 @@ function createTextRepairGenerator(options) {
       return { replacements: output, usage, elapsedMs: Date.now() - started };
     } catch (error) {
       const cause = NoObjectGeneratedError2.isInstance(error) && error.finishReason && error.finishReason !== "stop" ? new IncompleteGenerationError(error.finishReason, void 0, getErrorTokenUsage(error)) : error;
+      recordModelCallFailure(cause, {
+        phase: "length_repair",
+        provider: typeof options.model === "object" ? options.model.provider : "gateway",
+        modelId: typeof options.model === "object" ? options.model.modelId : options.model,
+        maxOutputTokens
+      }, usage ?? getErrorTokenUsage(cause));
       return { usage: usage ?? getErrorTokenUsage(cause), elapsedMs: Date.now() - started, error: cause instanceof Error ? cause.message : String(cause), errorCause: cause };
     }
   };
@@ -6348,7 +6460,7 @@ async function generateEngagementInsightsInternal(assetId, options, collectedUsa
 }
 
 // src/workflows/generate-text.ts
-import { APICallError as APICallError3, generateText as generateTextWithModel, Output as Output7 } from "ai";
+import { APICallError as APICallError4, generateText as generateTextWithModel, Output as Output7 } from "ai";
 import { z as z8 } from "zod";
 var GENERATE_TEXT_MAX_VARIANTS = 5;
 var GENERATE_TEXT_MAX_ARTIFACTS = 5;
@@ -6471,7 +6583,7 @@ function resolveGenerateTextOptions(options) {
   }
   return { ...options, variants };
 }
-var LENGTH_TARGET_RATIO = 0.9;
+var LENGTH_TARGET_RATIO = 0.85;
 function resolveLengthTarget(limit) {
   return Math.max(1, Math.floor(limit.value * LENGTH_TARGET_RATIO));
 }
@@ -6760,122 +6872,124 @@ function readUsage(response) {
 }
 async function extractBriefWithModel(args) {
   "use step";
-  const model = await createLanguageModelFromConfig(args.provider, args.modelId, args.credentials);
-  const response = await withContentPolicyAwareRetry(() => generateTextWithModel({
-    model,
-    maxRetries: 0,
-    output: Output7.object({
-      name: "editorial_brief",
-      description: "Source-grounded editorial brief used to write every requested artifact.",
-      schema: briefSchema
-    }),
-    system: args.systemPrompt,
-    messages: [
-      {
-        role: "user",
-        content: [
-          { type: "text", text: args.userPrompt },
-          ...args.imageUrls.map((url) => ({ type: "image", image: url }))
-        ]
-      }
-    ]
-  }));
-  const output = getGeneratedOutputWithContentPolicyHandling(response);
-  if (!output) {
-    throw new Error("Editorial brief output missing");
-  }
-  return {
-    brief: clampBriefArrays(briefSchema.parse(output)),
-    usage: readUsage(response),
-    unexpectedKeys: detectUnexpectedKeysFromRawText(response.text, briefSchema.keyof().options)
-  };
-}
-var artifactSchema = z8.object({ content: z8.string() });
-async function generateArtifactWithModel(args) {
-  "use step";
-  const model = await createLanguageModelFromConfig(args.provider, args.modelId, args.credentials);
-  const attempt = async (messages) => {
+  return withModelCallDiagnostics({ phase: "brief", provider: args.provider, modelId: args.modelId }, async () => {
+    const model = await createLanguageModelFromConfig(args.provider, args.modelId, args.credentials);
     const response = await withContentPolicyAwareRetry(() => generateTextWithModel({
       model,
       maxRetries: 0,
       output: Output7.object({
-        name: "generated_text",
-        description: "One finished artifact written from the editorial brief.",
-        schema: artifactSchema
+        name: "editorial_brief",
+        description: "Source-grounded editorial brief used to write every requested artifact.",
+        schema: briefSchema
       }),
       system: args.systemPrompt,
-      messages
+      messages: [
+        {
+          role: "user",
+          content: [
+            { type: "text", text: args.userPrompt },
+            ...args.imageUrls.map((url) => ({ type: "image", image: url }))
+          ]
+        }
+      ]
     }));
     const output = getGeneratedOutputWithContentPolicyHandling(response);
     if (!output) {
-      throw new Error("Generated text output missing");
+      throw new Error("Editorial brief output missing");
     }
     return {
-      content: artifactSchema.parse(output).content.trim(),
+      brief: clampBriefArrays(briefSchema.parse(output)),
       usage: readUsage(response),
-      unexpectedKeys: detectUnexpectedKeysFromRawText(response.text, artifactSchema.keyof().options)
+      unexpectedKeys: detectUnexpectedKeysFromRawText(response.text, briefSchema.keyof().options)
     };
-  };
-  const judge = (content) => content ? findGenerateTextLengthViolation(content, args.limits) : "empty";
-  const initialMessages = [{ role: "user", content: args.userPrompt }];
-  const first = await attempt(initialMessages);
-  const firstViolation = judge(first.content);
-  const firstEscaped = hasEscapedLineBreaks(first.content);
-  if (!firstViolation && !firstEscaped) {
-    return { content: first.content, usages: [first.usage], unexpectedKeys: first.unexpectedKeys };
-  }
-  let candidate = first;
-  const usages = [first.usage];
-  const unexpectedKeys = [...first.unexpectedKeys];
-  if (firstViolation === "empty" || firstEscaped) {
-    const feedback = [
-      firstViolation && firstViolation !== "empty" ? describeLengthRevision(firstViolation) : void 0,
-      firstEscaped ? ESCAPED_TEXT_REVISION : void 0
-    ].filter((line) => Boolean(line)).join("\n");
-    const revisionMessages = firstViolation === "empty" ? [{ role: "user", content: `${args.userPrompt}
+  });
+}
+var artifactSchema = z8.object({ content: z8.string() });
+async function generateArtifactWithModel(args) {
+  "use step";
+  const diagnostics = { phase: "draft", provider: args.provider, modelId: args.modelId };
+  return withModelCallDiagnostics(diagnostics, async () => {
+    const model = await createLanguageModelFromConfig(args.provider, args.modelId, args.credentials);
+    const attempt = async (messages) => {
+      const response = await withContentPolicyAwareRetry(() => generateTextWithModel({
+        model,
+        maxRetries: 0,
+        output: Output7.object({
+          name: "generated_text",
+          description: "One finished artifact written from the editorial brief.",
+          schema: artifactSchema
+        }),
+        system: args.systemPrompt,
+        messages
+      }));
+      const output = getGeneratedOutputWithContentPolicyHandling(response);
+      if (!output) {
+        throw new Error("Generated text output missing");
+      }
+      return {
+        content: artifactSchema.parse(output).content.trim(),
+        usage: readUsage(response),
+        unexpectedKeys: detectUnexpectedKeysFromRawText(response.text, artifactSchema.keyof().options)
+      };
+    };
+    const judge = (content) => content ? findGenerateTextLengthViolation(content, args.limits) : "empty";
+    const initialMessages = [{ role: "user", content: args.userPrompt }];
+    const first = await attempt(initialMessages);
+    const firstViolation = judge(first.content);
+    const firstEscaped = hasEscapedLineBreaks(first.content);
+    if (!firstViolation && !firstEscaped) {
+      return { content: first.content, usages: [first.usage], unexpectedKeys: first.unexpectedKeys };
+    }
+    let candidate = first;
+    const usages = [first.usage];
+    const unexpectedKeys = [...first.unexpectedKeys];
+    if (firstViolation === "empty" || firstEscaped) {
+      const feedback = [
+        firstViolation && firstViolation !== "empty" ? describeLengthRevision(firstViolation) : void 0,
+        firstEscaped ? ESCAPED_TEXT_REVISION : void 0
+      ].filter((line) => Boolean(line)).join("\n");
+      const revisionMessages = firstViolation === "empty" ? [{ role: "user", content: `${args.userPrompt}
 
 ${renderSection({ tag: "revision_request", content: "The previous draft was empty. Write the complete artifact." })}` }] : [...initialMessages, { role: "assistant", content: JSON.stringify({ content: first.content }) }, { role: "user", content: renderSection({ tag: "revision_request", content: feedback }) }];
-    candidate = await attempt(revisionMessages).catch((error) => rethrowWithTokenUsage(error, usages));
-    usages.push(candidate.usage);
-    unexpectedKeys.push(...candidate.unexpectedKeys);
-  }
-  const violation = judge(candidate.content);
-  if (!violation || violation === "empty") {
-    return { content: candidate.content, usages, unexpectedKeys, violation };
-  }
-  const repaired = await repairText({
-    content: candidate.content,
-    limits: args.limits,
-    protectedTerms: args.protectedTerms,
-    model
-  });
-  usages.push(...repaired.attempts.flatMap((item) => item.call.usage ? [item.call.usage] : []));
-  if (repaired.status !== "valid") {
-    const cause = repaired.attempts.at(-1)?.call.errorCause;
-    let failure;
-    if (isIncompleteGenerationError(cause)) {
-      failure = new IncompleteGenerationError(cause.finishReason, cause.rawFinishReason);
-    } else if (MuxAiError.is(cause)) {
-      failure = new MuxAiError(cause.publicMessage, { type: cause.publicType, retryable: cause.retryable });
-    } else {
-      const apiError = APICallError3.isInstance(cause) ? cause : void 0;
-      console.warn("[@mux/ai] Bounded text repair failed.", {
-        provider: args.provider,
-        modelId: args.modelId,
-        reason: repaired.reason,
-        errorName: cause instanceof Error ? cause.name : void 0,
-        statusCode: apiError?.statusCode,
-        requestId: apiError?.responseHeaders?.["x-request-id"] ?? apiError?.responseHeaders?.["request-id"]
-      });
-      failure = new MuxAiError("Unable to complete text generation.", {
-        type: "processing_error",
-        retryable: apiError?.isRetryable ?? true,
-        cause
-      });
+      diagnostics.phase = "format_repair";
+      candidate = await attempt(revisionMessages).catch((error) => rethrowWithTokenUsage(error, usages));
+      usages.push(candidate.usage);
+      unexpectedKeys.push(...candidate.unexpectedKeys);
+    }
+    const violation = judge(candidate.content);
+    if (!violation || violation === "empty") {
+      return { content: candidate.content, usages, unexpectedKeys, violation };
+    }
+    diagnostics.phase = "length_repair";
+    const repaired = await repairText({
+      content: candidate.content,
+      limits: args.limits,
+      protectedTerms: args.protectedTerms,
+      model
+    });
+    usages.push(...repaired.attempts.flatMap((item) => item.call.usage ? [item.call.usage] : []));
+    if (repaired.status !== "valid") {
+      const cause = repaired.attempts.at(-1)?.call.errorCause;
+      let failure;
+      if (isIncompleteGenerationError(cause)) {
+        failure = new IncompleteGenerationError(cause.finishReason, cause.rawFinishReason);
+      } else if (MuxAiError.is(cause)) {
+        failure = new MuxAiError(cause.publicMessage, { type: cause.publicType, retryable: cause.retryable });
+      } else {
+        const apiError = APICallError4.isInstance(cause) ? cause : void 0;
+        failure = new MuxAiError("Unable to complete text generation.", {
+          type: "processing_error",
+          retryable: apiError?.isRetryable ?? true,
+          cause
+        });
+      }
+      if (typeof cause === "object" && cause !== null && "diagnostics" in cause) {
+        failure.diagnostics = cause.diagnostics;
+      }
+      rethrowWithTokenUsage(failure, usages);
     }
-    rethrowWithTokenUsage(failure, usages);
-  }
-  return { content: repaired.content, usages, unexpectedKeys };
+    return { content: repaired.content, usages, unexpectedKeys };
+  });
 }
 async function generateText7(assetId, options) {
   "use workflow";
@@ -9354,7 +9468,7 @@ async function translateAudio(assetId, toLanguageCode, options = {}) {
 
 // src/workflows/translate-captions.ts
 import {
-  APICallError as APICallError4,
+  APICallError as APICallError5,
   generateText as generateText9,
   NoObjectGeneratedError as NoObjectGeneratedError3,
   NoOutputGeneratedError as NoOutputGeneratedError2,
@@ -9509,7 +9623,7 @@ function isProviderServiceError(error) {
   if (RetryError2.isInstance(error)) {
     return isProviderServiceError(error.lastError);
   }
-  if (APICallError4.isInstance(error)) {
+  if (APICallError5.isInstance(error)) {
     return true;
   }
   if (error instanceof Error && "cause" in error) {