Community · E2 · artifact verified

Read customer-inquiry emotion in six questions

A Japanese-language experiment scoring 100 labeled customer inquiries with the official TypeSafe SDK: one request per inquiry answers six questions at once - sentiment, emotion, anger intensity, urgency, churn risk, and sarcasm.

01 · Role in the system

What Jev does here

Each inquiry is sent as fitted state with six typed questions in a single system_one call through the official SDK's typed builders: sentiment and emotion as Choices (four-way sentiment; gratitude, joy, anger, irritation, disappointment, anxiety, or none), anger as a 0-3 Score, and urgency, churn risk, and sarcasm as Nouls. The experiment set carries intended emotion labels for all 100 Japanese inquiries, and the repository's analyze.py runs the batch and writes per-inquiry results with confidences to JSONL for comparison against the labels - a compact template for wiring the same six-question probe into a support desk.

02 · Control boundary

Where Jev sits

Official SDK typed builders (Choice, Score, Noul) batched into one system_one call per inquiry; results written to JSONL for label comparison.

Code owns the loop, permissions, thresholds, validation, and side effects. Jev owns only the bounded judgments described above.

03 · Known limits

What this evidence does not prove

  • A 100-inquiry experiment with author-supplied emotion labels; agreement analysis is in the repository's results rather than a published benchmark.
  • Japanese-language inquiries only; the question schema is bilingual but the corpus is not.
  • No UI or integration - the deliverable is the analyze.py template and its results.

04 · Attribution

Public sources

This is a Community record: the project was published by a third-party community author.