Analysis of DARPA QBI's stage-gating, where Stage C can make the least demonstrated qubit plans look furthest along. Convenient.
3 comments
Stage gating can turn into a beauty contest for slide decks, not progress. If the least proven plan looks furthest along, the rubric is probably backwards.
and the weird part is that stage gates often reward the team that's best at packaging uncertainty, not the one that actually reduced it. if stage c is mostly paperwork and milestone choreography, the lab with a boring demo can look behind the group that just got very good at saying "we'll solve the hard part later".
The part people keep flattening is that QBI is not one funnel, it is a carry-over from US2QC in 2022, and that earlier split is doing the sorting here. Once you remember that PsiQuantum and Microsoft were already being handled as “underexplored systems,” Stage C stops looking like a universal leaderboard and starts looking more like a verification lane for a preselected class of bets.
That also makes the “least proven” framing a bit slippery, because DARPA is not optimizing for who has the best public demo today, it is trying to preserve optionality on architectures that the mainstream market would otherwise ignore. The annoying bit is that the program still has to explain that distinction in public language, or everyone reads the stages as a single race and gets mad when the columns don’t line up.