A company must calculate an invoice tax using a published rate table and exact rounding rules. It needs the same correct amount for the same inputs and has no prediction or language-generation requirement. Which approach is most appropriate?

1. Train a clustering model to group invoices and use the cluster identifier as the tax amount.
2. Train a regression model to approximate the calculation even though the complete rules are already available.
3. Use deterministic application logic implementing the approved rate table and rounding rules.
4. Ask a generative model to estimate the tax from an informal description of the invoice.