Which of two competing edges wins: higher priority, then lower sequence, then edge id.
The id is a tiebreak, not a preference — and without it this ordering is not total. Priority
and sequence both default to 0 and Submit persists those defaults, so a hand-authored or
LLM-authored spec routinely produces a genuine tie; dependencies load with no ORDER BY, so the
winner was decided by row order, which can differ between polls of the same graph. The worst
interleaving is not a wrong branch but NO branch: poll 1 picks X→B and skips C; poll 2's row
order flips, picks Y→C — already Skipped — and skips B. Both branches Skipped, and the graph
settles Complete having executed neither.
Also used to ask whether an unevaluable edge could have beaten the winner, so the two questions
cannot disagree about what "beats" means.
Which of two competing edges wins: higher priority, then lower sequence, then edge id.
The id is a tiebreak, not a preference — and without it this ordering is not total. Priority and sequence both default to 0 and
Submitpersists those defaults, so a hand-authored or LLM-authored spec routinely produces a genuine tie; dependencies load with noORDER BY, so the winner was decided by row order, which can differ between polls of the same graph. The worst interleaving is not a wrong branch but NO branch: poll 1 picksX→Band skips C; poll 2's row order flips, picksY→C— already Skipped — and skips B. Both branches Skipped, and the graph settles Complete having executed neither.Also used to ask whether an unevaluable edge could have beaten the winner, so the two questions cannot disagree about what "beats" means.