@semiont/jobs 0.6.9 → 0.6.11

package/dist/worker-main.js

@@ -9928,7 +9928,8 @@ Example format:
   /**
    * Build a prompt for detecting structural tags
    *
-   * @param content - The full text content to analyze (NOT truncated for structural analysis)
+   * @param content - The text content to analyze — a chunk sized by the
+   *   caller from derived provider limits; NEVER re-truncated here
    * @param category - The specific category to detect
    * @param schemaName - Human-readable schema name
    * @param schemaDescription - Schema description
@@ -10356,7 +10357,7 @@ ${piece}
   }
   if (items.length * YIELD_COLLAPSE_BAND < counted) {
     throw new YieldCollapseError(
-      `Extraction found ${items.length} entities where a count call reports ~${counted} mentions (band \xD7${YIELD_COLLAPSE_BAND}) on a ${piece.length}-char chunk \u2014 silent yield collapse (F7): deterministic \u2014 a same-size retry returns the identical under-report.`,
+      `Extraction found ${items.length} entities where a count call reports ~${counted} mentions (band \xD7${YIELD_COLLAPSE_BAND}) on a ${piece.length}-char chunk \u2014 silent yield collapse: deterministic \u2014 a same-size retry returns the identical under-report.`,
       [...items],
       { found: items.length, counted, pieceChars: piece.length }
     );
@@ -10997,9 +10998,8 @@ async function processCommentJob(content, inferenceClient, params, buildAnnotati
       found += comments.length;
       const fresh = dedupe(comments.map(
         (c) => (
-          // Match the pre-#651 CommentAnnotationWorker: include format and
-          // language on the body TextualBody. Optional in the schema, but
-          // consumers that do language-aware rendering rely on them.
+          // Format and language go on the body TextualBody: optional in the
+          // schema, but consumers that do language-aware rendering rely on them.
           buildAnnotation("commenting", c, [
             { type: "TextualBody", value: c.comment, purpose: "commenting", format: "text/plain", language: bodyLanguage }
           ])
@@ -11040,8 +11040,7 @@ async function processAssessmentJob(content, inferenceClient, params, buildAnnot
       const fresh = dedupe(assessments.map(
         (a) => (
           // Single-object body with purpose aligned to motivation, matching the
-          // pre-#651 AssessmentAnnotationWorker's shape and the majority of
-          // persisted assessments. Do not switch to an array or to
+          // majority of persisted assessments. Do not switch to an array or to
           // purpose='describing' — that loses the "this is an assessment, not
           // a description" signal and breaks existing readers that access
           // `body.value` directly on the object.
@@ -11109,11 +11108,11 @@ async function processReferenceJob(content, inferenceClient, params, buildAnnota
       params.includeDescriptiveReferences ?? false,
       logger2,
       params.sourceLanguage,
-      // Liveness heartbeat (DETECTION-HEARTBEAT): fires at chunk boundaries and
-      // every ~15 s while a call is in flight, so a long single-chunk call is
-      // not silent. It repeats the current position rather than inventing an
-      // advance — the stall watchdog, janitor and client timeout need a signal,
-      // not a monotone.
+      // Liveness heartbeat: fires at chunk boundaries and every ~15 s while a
+      // call is in flight, so a long single-chunk call is not silent. It
+      // repeats the current position rather than inventing an advance — the
+      // stall watchdog, janitor and client timeout need a signal, not a
+      // monotone.
       () => emitTypeProgress(entityTypeName),
       (verdict) => {
         underReported = {
@@ -11270,7 +11269,7 @@ async function processGenerationJob(inferenceClient, params, onProgress, logger2
       `Unsupported outputMediaType for generation: ${outputMediaType}. Generation produces ${GENERATABLE_MEDIA_TYPES.join(" or ")}.`
     );
   }
-  const title = params.title ?? "Untitled";
+  const title = params.title;
   const entityTypes = (params.entityTypes ?? []).map(String);
   if (outputMediaType === "application/pdf") {
     onProgress(5, { code: "generating-resource" });
@@ -11548,10 +11547,9 @@ function startWorkerProcess(config) {
         // the question never arose, never that durability was ruled out.
         ...error instanceof CommitDurabilityError ? { durability: error.durability } : {},
         // Whether this failure is the END, answered by the same predicate
-        // the queue applies at failJob (JOB-RESTART-SAFETY P5). Without
-        // it a client cannot tell a recovering run from a dead one: it
-        // sees job:fail either way and would end its stream on a job the
-        // queue is about to re-run.
+        // the queue applies at failJob. Without it a client cannot tell a
+        // recovering run from a dead one: it sees job:fail either way and
+        // would end its stream on a job the queue is about to re-run.
         willRetry: willRetryAfter(job, failureClass)
       }).catch(() => {
       });
@@ -11676,8 +11674,9 @@ async function handleJobInner(adapter, config, job, completedUnitsByJob, signal,
       // The durability write, per chunk, awaited; folds into the terminal
       // durability evidence like every commit, and carries the unit's cursor.
       commitChunk,
-      // …and the other direction: where an earlier attempt left each unit
-      // (CHUNK-GRAIN-RESUME P3). Empty on a first attempt.
+      // …and the other direction: where an earlier attempt left each unit, so
+      // a partway unit resumes at its offset instead of the top. Empty on a
+      // first attempt.
       job.unitCursors
     );
     await emitEvent(session, "job:complete", {
@@ -11695,8 +11694,8 @@ async function handleJobInner(adapter, config, job, completedUnitsByJob, signal,
       // The durability write, per chunk, awaited; folds into the terminal
       // durability evidence like every commit, and carries the unit's cursor.
       commitChunk,
-      // …and the other direction: where an earlier attempt left each unit
-      // (CHUNK-GRAIN-RESUME P3). Empty on a first attempt.
+      // …and the other direction: where an earlier attempt left each unit.
+      // Empty on a first attempt.
       job.unitCursors
     );
     await emitEvent(session, "job:complete", {
@@ -11714,8 +11713,8 @@ async function handleJobInner(adapter, config, job, completedUnitsByJob, signal,
       // The durability write, per chunk, awaited; folds into the terminal
       // durability evidence like every commit, and carries the unit's cursor.
       commitChunk,
-      // …and the other direction: where an earlier attempt left each unit
-      // (CHUNK-GRAIN-RESUME P3). Empty on a first attempt.
+      // …and the other direction: where an earlier attempt left each unit.
+      // Empty on a first attempt.
       job.unitCursors
     );
     await emitEvent(session, "job:complete", {
@@ -11753,8 +11752,8 @@ async function handleJobInner(adapter, config, job, completedUnitsByJob, signal,
       // The durability write, per chunk, awaited; folds into the terminal
       // durability evidence like every commit, and carries the unit's cursor.
       commitChunk,
-      // …and the other direction: where an earlier attempt left each unit
-      // (CHUNK-GRAIN-RESUME P3). Empty on a first attempt.
+      // …and the other direction: where an earlier attempt left each unit.
+      // Empty on a first attempt.
       job.unitCursors
     );
     if (signal?.aborted) {
@@ -11780,8 +11779,8 @@ async function handleJobInner(adapter, config, job, completedUnitsByJob, signal,
       // The durability write, per chunk, awaited; folds into the terminal
       // durability evidence like every commit, and carries the unit's cursor.
       commitChunk,
-      // …and the other direction: where an earlier attempt left each unit
-      // (CHUNK-GRAIN-RESUME P3). Empty on a first attempt.
+      // …and the other direction: where an earlier attempt left each unit.
+      // Empty on a first attempt.
       job.unitCursors
     );
     await emitEvent(session, "job:complete", {
@@ -11801,7 +11800,6 @@ async function handleJobInner(adapter, config, job, completedUnitsByJob, signal,
       onProgress,
       config.logger
     );
-    const genParams = job.params;
     const genReferenceId = referenceIdOf(job);
     const storageUri = job.params.storageUri;
     const expectedExtension = capabilitiesOf(genResult.format)?.extension;
@@ -11820,9 +11818,9 @@ async function handleJobInner(adapter, config, job, completedUnitsByJob, signal,
       storageUri,
       sourceResourceId: resourceId,
       ...genReferenceId ? { sourceAnnotationId: genReferenceId } : {},
-      ...genParams.prompt ? { generationPrompt: genParams.prompt } : {},
-      ...genParams.language ? { language: genParams.language } : {},
-      ...genParams.entityTypes && genParams.entityTypes.length > 0 ? { entityTypes: genParams.entityTypes } : {},
+      ...job.params.prompt ? { generationPrompt: job.params.prompt } : {},
+      ...job.params.language ? { language: job.params.language } : {},
+      ...job.params.entityTypes && job.params.entityTypes.length > 0 ? { entityTypes: job.params.entityTypes } : {},
       generator,
       // The resource cites the job it fulfils; who asked for it is derived
       // downstream from that job's events, never stated here.
@@ -11944,26 +11942,25 @@ var WORKER_AWAITED_OPERATIONS = [
   "job:claim",
   "browse:resource-requested",
   // Canonical geometry for a geometry-bearing detection: the consult behind
-  // `ConsultAnchoredText` (SMELTER-OWNS-OCR P2). Its omission broke every
-  // PDF detection job at the transport probe
-  // (.plans/WORKER-ANCHORED-TEXT-CHANNEL.md); the census below now fails the
+  // `ConsultAnchoredText`, answered by the Smelter. Without it every PDF
+  // detection job fails at the transport probe; the census below fails the
   // BUILD when this list and the declared awaits drift.
   "browse:anchored-text-requested",
-  // Durability acknowledgement for a unit's annotations (JOB-RESTART-SAFETY
-  // P6). The worker AWAITS this one — a unit may not advance until its
-  // annotations are in the event log — so its replies must be in the narrow
-  // channel set or every commit fails fast with `bus.unsubscribed`.
+  // Durability acknowledgement for a unit's annotations. The worker AWAITS
+  // this one — a unit may not advance until its annotations are in the event
+  // log — so its replies must be in the narrow channel set or every commit
+  // fails fast with `bus.unsubscribed`.
   "mark:commit",
-  // The durability probe for a commit whose acknowledgement never routed
-  // (COMMIT-ACK-FALSE-FAILURE F1). SINGULAR by design: the annotation LIST
-  // channel is the multi-MB fan-out this narrowing exists to keep out, and a
-  // rare error path is no reason to let it back in.
+  // The durability probe for a commit whose acknowledgement never routed.
+  // SINGULAR by design: the annotation LIST channel is the multi-MB fan-out
+  // this narrowing exists to keep out, and a rare error path is no reason to
+  // let it in.
   "browse:annotation-requested"
 ];
 var WORKER_CONSUMED_BROADCASTS = [
   // The queue announcement the claim adapter races for.
   "job:queued",
-  // Cooperative cancellation of the ACTIVE job (JOB-RESTART-SAFETY P4).
+  // Cooperative cancellation of the ACTIVE job: it stops at a unit boundary.
   "job:cancel-requested"
 ];
 var WORKER_ANSWERED_OPERATIONS = [
@@ -12085,8 +12082,9 @@ async function startAgentWorker(opts) {
     generator,
     // Byte reads for decode-path media only. A geometry-bearing type's text
     // never comes from bytes here — it is CONSULTED from the Smelter's
-    // canonical anchored text over the bus (SMELTER-OWNS-OCR P2), and this
-    // worker cannot derive even by mistake (READ-VS-EXTRACT P2).
+    // canonical anchored text over the bus, and this worker cannot derive
+    // even by mistake: deriving needs the anchored-text store, which only
+    // the Smelter holds.
     contentReads: contentReads2,
     logger: logger2
   });