Llm
data class LlmRationaleProjector(llmOptions: <Error class: unknown class>, ai: <Error class: unknown class>) : RationaleProjector
LLM-backed RationaleProjector that turns a proposition (or group) into human-readable rationale prose grounded in the source propositions.
Security note (indirect prompt injection). Proposition text and group labels are embedded in the LLM prompt. Since this content typically comes from ingested source documents, it must be treated as untrusted: a crafted document could embed instructions the rationale model may follow. The template wraps proposition text in a labelled data block as a mitigation, but that is not a guarantee. Callers are responsible for sanitizing ingested content upstream and for not granting the rationale output undue authority.
Example usage:
val projector = LlmRationaleProjector
.withLlm(llmOptions)
.withAi(ai)
val artifact = projector.rationale(group)
println(artifact.text)Content copied to clipboard