The purgatory promotion regression searched up to 99,999 nonces for an event ID below a random baseline. A sufficiently low baseline could exhaust the search and fail without a product defect.
Sign two candidates with distinct nonces and sort their IDs to assign winner and loser deterministically. Preserve their shared timestamp, declared Git state and losing-first submission order. Production promotion behavior is unchanged.
Validation: the same-second purgatory integration test passes; workspace all-target Clippy and formatting pass.
Assisted-by: GPT-6