Keep only the rows that are not what the register already says

outcome_at is deliberately outside the default comparison: it changes every run by construction, so including it would make every row differ and the filter would filter nothing. Without the filter every quiet night rewrites every row, and the alert on the outcome field becomes background noise — which is the thing people stop reading, and the alert exists to be read.

Usage

round_changed(
  register,
  payload,
  fields = c("outcome", "outcome_detail", "applied_as")
)

Arguments

register

The register as read, before the round rewrote anything in it. Handed the frame the round has already mutated, this compares a value against itself and reports that nothing ever changes.

payload

The rows this round produced, from outcome_payload() or from identity_payload().

fields

The columns a change can show in. The default is the three an outcome can differ in; the identity body passes its own two.

Value

payload, reduced to the rows that changed something.

Details

One function with a fields argument rather than a copy per family, for the reason register_field_import() is one transport for two doors: two copies of the same comparison are a place to fix a defect once and leave it standing in the other. What differs between the families is which columns can carry a change, and that is exactly what the argument says.