Rossum.ai recently introduced a new field type: Reasoning Fields (inline LLM fields). Here is all you need to know about them from a Solution Architect:

1. Reasoning Fields are designed to output "single value" predictions. It's better to have many specialized fields than one that solves many problems simultaneously. Parsing address? Create one Reasoning Field per address chunk (as opposed to having one fields that resolves all the chunks at once).
2. Reasoning Fields remember previous predictions and manual changes (rather aggressively!). Same inputs are likely to return the same output even when the prompt or the underlying model changes. Currently, the best way to overwrite this is by introducing a "cache buster" field.
3. Reasoning Fields are LLM-based. Naturally, like any other LLM, they are not a good fit for tasks requiring accuracy or reproducibility. While reproducibility is solved by the caching mentioned above, accuracy is still a problem. Favor Formula Fields for deterministic outputs where possible (such as mathematical calculations).

Data matching (Master Data Hub; MDH) results are by default returned into an "enum" field in Rossum (also known as "Options" field). Enum field allows to select only one of the values returned (it behaves as a regular HTML select element with options). What if you need to select multiple values, however?