Why
The PoC contrasts free-form prompting with schema-constrained decoding on a extraction task and measures the parse-failure rate each produces.
How it works
Not yet built.
Where it lands in Jayverse
- Number: use schema-constrained decoding for any LLM-generated reading or indicator. If Number ever parses model output into structured readings, measure the parse-failure rate of free-form prompting against schema-constrained generation before shipping either.
- Personas: constrain any LLM-generated persona metadata to a schema. Persona NFT attributes or descriptions produced by a model should be schema-guided rather than free-form JSON, so a malformed field never reaches the mint step.
- Auditor: require a parse-failure-rate number for any LLM step in the pipeline. Wherever a Jayverse service depends on an LLM producing structured output (Number, Personas, or elsewhere), the Auditor row should record the measured failure rate rather than assuming the model's JSON always parses.
Key expressions
| Expression | 뜻 · 쓰이는 자리 |
|---|---|
| flaky | 불안정한(가끔 실패하는) · 신뢰할 수 없는 기존 방식을 묘사할 때. "turning a flaky text-to-JSON step" |
| hope for | 보장 없이 되기를 바라다 · 강제하지 않고 운에 맡기는 방식을 대조할 때. "forcing valid JSON instead of hoping for it" |
| contrast X with Y | X를 Y와 대조하다 · 두 접근 방식을 비교할 때. "contrasts free-form prompting with schema-constrained decoding" |
| reliable interface | 신뢰할 수 있는 인터페이스 · 애플리케이션이 의지할 수 있는 안정적 출력. "a reliable interface an application can depend on" |
| parse-failure rate | 파싱 실패율 · 출력이 형식에 안 맞아 실패하는 비율을 측정할 때. "measures the parse-failure rate each produces" |
| constrained decoding | 제약 디코딩 · 모델이 정해진 스키마를 벗어나지 않게 강제로 생성시키는 기법. "Constrained decoding and schema-guided generation make a model emit output" |