D563 - Runs rank by answered imports, through one key

Status: decided Date: 2026-09-04

A run records unanswered, the distinct imports it called with nothing behind them, and ranks by answered (imports minus unanswered) directly below imports. One ranking_key orders the record, the frontier and the table; a record that predates the measurement holds None and ranks as though nothing was answered.

Why: a percentage of calls follows whatever the guest loops on, so it hides work that removes whole functions from the work list. A count of functions is stable against a hot loop and is the work list itself. A guest that reaches further calls new imports, some unimplemented, so ranking answered above imports would report progress as regression. Separate copies of the ordering drift, and a table that disagrees with the record is the more convincing and the wrong one.

Rejected:

• More precision in standing: makes a loop-dominated number more exact, not more useful.
• Some(0) for old records: a stale record claims a perfect score and refuses honest runs.
• One tuple per consumer: they drifted as soon as a rung was added.