The corpus is currently positioned to surface this business for specific-service and trust-quality LLM queries on roughly 98.0% of its reviews. The remaining 2 reviews do not carry sufficient semantic content for LLM retrieval, though they continue to function for star ratings and social proof.
v1 prompt — known calibration limitations: outcome dimension over-scores on consumption businesses (96% on this corpus vs ~16% expected). See docs/PROMPT_ITERATION.md for prompt-iteration plan.
Percent of reviews mentioning each dimension at least generically. Weakest: occasion
Type: Dessert shop specializing in innovative ice cream and baked goods — Profile confidence: high
The labeler self-reported confidence on each review:
Low-confidence assignments are reviews where the rubric is genuinely ambiguous. They are flagged in the JSON output for optional human review.