Magic estimation
The whole technique is one bet: discussing the stories where you already agree is a waste.
Argue only where the team disagrees. That's the whole technique.
Magic estimation works the way Planning Poker would if you stripped out the conversations nobody needed to have. The team silently places every story into a size bucket; the only stories that get a real estimation conversation are the ones where placements diverged. Stories where the team agreed in silence get the silent consensus as their number. That bet — that agreement-in-silence is real agreement — is the source of both the speed and the failure mode.
When the bet pays off, it's the fastest sizing technique that produces real numbers. A 50-story backlog gets sized in 30 minutes; Planning Poker would take three sessions to do the same work. When the bet doesn't pay off, the team has shipped a unanimous "5" on a story that was secretly a mix of 3s and 8s, and nobody knows until the sprint runs out.
Where the bet pays off
Long backlogs with stories the team mostly already understands. Roadmap-level passes where rough sizes are enough. Re-anchoring after a reference-story change. In each case the team's shared model is strong enough that silent agreement is probably real agreement, and the time saved is bigger than the precision lost.
Where the bet fails
Stories about to enter a sprint. Teams that don't yet have a shared model — new teams, teams with new members, teams that haven't refined together. Anything with high scope variance. In each case silence hides the disagreement that Planning Poker's reveal-and-discuss exists to surface, and the technique buys speed with the wrong currency.
The shape of a session
Buckets laid out (1, 2, 3, 5, 8, 13); backlog given to every team member; 20 minutes of silent placement; 20 minutes on the disagreements. Stories that ended up in different buckets get a real Planning Poker round. Stories that didn't keep their silent consensus.
Don't run magic estimation on stories you're about to commit to. The silence is exactly the wrong tool for the moment.
See affinity estimation for the closest cousin technique; other estimation techniques for how both fit alongside Planning Poker.