Producing a plain-language account of why the two sides differ and what the likely cause is, given both entries, the applicable rate and prior resolutions for this pair.
Language generation over structured inputs, with the causal claim framed as likely rather than determined
LLM API