The week we could not grade

The Scoreboard · Ins7ghts Daily, Friday 23 October 2026

Thirteen calls, graded. One right, one that was never a real call, eleven that cannot be settled yet, and no misses. Zero wrong is not a good week. It usually means the horizons are set too far out to be checked, and that is what happened here: most of the thirteen run to the end of Q1 2027 or beyond, so they will sit on this board for months before anybody can say.

What was right. On 16 October we wrote that governance turns into a product feature, and that when the control plane belongs to a vendor your governance posture becomes a function of your platform contract. On 22 October ModelOp released ModelOp Center 3.5, an enterprise AI command center that lists low-code, no-code and pro-code agents across Microsoft and Amazon and tracks what they spend. Different vendor, same move, six days after the call. Right.

What was not a real call. Also on 16 October, on the Pentagon's $50 billion for autonomous weapons production, we said the autonomy accountability specs it sets would become the reference standard in adjacent regulated industries inside two years. Nobody publishes a register of reference standards. No event can prove or disprove that sentence, which makes it a nice-sounding paragraph rather than a call, and it comes off the board.

Still open, and the one I am least comfortable about. On 17 October we said the next data-centre or GPU-cloud refinancing would get priced in public, and that the spread it cleared at would reset what every unlisted peer could tell its lenders. What has happened since points the other way: the money keeps arriving privately, and privately is where nobody has to publish a price. There is no settling event, so it stays open. It is not ageing well.