Run the browser-side classifier, isolate the normalized prompt intent, then inspect the enriched use case and category. The validation tab keeps the full benchmark set one click away.
Use the classifier for one prompt at a time, then move to the validation set to compare how the same pipeline behaves across the bundled examples.
The SLM returns the prompt intent. Then we enrich the same result with use case and category.
Paste the exact first prompt you want to test. The classifier keeps the intent and strips wording that does not change the business use case.
Normalization Rules
Business-firstPersonal.
All validation prompts from dataset_validation_results.jsonl with SLM prompt intent and our enrichments.
| # | Prompt | Prompt Intent | Usecase | Usecase Category |
|---|---|---|---|---|
| ... | Loading validation prompts... | ... | ... | ... |