fix: remove confidence from rawWriteInterpretationSchema to fix OpenAI structured output 400

z.record() produces additionalProperties in JSON Schema, which OpenAI's structured
outputs rejects — it requires every key in required to appear in properties.
Confidence was never populated by the LLM in practice; unresolvedFields handles
ambiguity instead. Remove confidence from the raw schema, default it to {} in
the normalizer, and default the ?? lookup to 1 (fully trusted) so existing
commit-policy thresholds are unaffected.

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

Author: MaxLinCode

packages/core/src/index.ts

@@ -1066,7 +1066,6 @@ export const rawWriteInterpretationSchema = z.object({
     scheduleFields: rawScheduleFieldsSchema.nullable(),
     taskFields: rawTaskFieldsSchema.nullable(),
   }),
-  confidence: z.record(z.string(), z.number().min(0).max(1)),
   unresolvedFields: z.array(z.string().min(1)),
 });
 

packages/core/src/write-commit.ts

@@ -161,7 +161,7 @@ function commitScheduleFields(input: {
       continue;
     }
 
-    const fieldConfidence = interpretation.confidence[fieldPath] ?? 0;
+    const fieldConfidence = interpretation.confidence[fieldPath] ?? 1;
     if (fieldConfidence < CONFIDENCE_THRESHOLD) {
       needsClarification.push(fieldPath);
       continue;
@@ -217,7 +217,7 @@ function commitTaskFields(input: {
       continue;
     }
 
-    const fieldConfidence = interpretation.confidence[fieldPath] ?? 0;
+    const fieldConfidence = interpretation.confidence[fieldPath] ?? 1;
     if (fieldConfidence < CONFIDENCE_THRESHOLD) {
       needsClarification.push(fieldPath);
       continue;

packages/core/src/write-interpretation.ts

@@ -74,7 +74,7 @@ export function normalizeRawWriteInterpretation(
     taskName: raw.taskName,
     fields: resolvedFields,
     sourceText,
-    confidence: raw.confidence,
+    confidence: {},
     unresolvedFields: raw.unresolvedFields,
   };
 }