Predictions

What We Think Happens Next.


How to read these

Market predictions with the mechanism behind them, the evidence for and against, and the condition that would prove each one wrong. Each carries instruments read on a cadence. None carries a resolution date, because the date evidence arrives is not something anyone can predict.

A prediction ends one of four ways: confirmed when the falsifier could no longer fire, transformed when the market moves along a different path and it keeps its history, contradicted by direct evidence, or retired after twelve months of silence.

How a signal becomes a prediction → the method, the gates and the lifecycle

The live set

The billing unit stopped growing

Every vendor charging per employee is selling against a number that is no longer rising, and the market has already priced it.

Rating A Live 3 checks 8 counted signals

The gate is the trap

AI cannot be priced until correctness is proved, and correctness is not being measured, so the value is being taken on the cost line instead.

Rating A Live 4 checks 6 counted signals

Permissions, not capability

What an agent is allowed to touch decides where it can be deployed. The market resolved the question in four days.

Rating B Live 2 checks 9 counted signals

Payroll stops being the decision

The evaluation is collapsing into somebody else's, and the moment a growing business used to switch is being deliberately removed.

Rating B Live 4 checks 14 counted signals

The frontline runs on the shift record

Compensable time is a payroll term. The system that decides who works which shift is upstream of gross-to-net, whether or not it calls itself payroll.

Rating B Live 3 checks 2 counted signals

Compliance by web edit

UK payroll rules now change mostly through guidance, specifications and enforcement, and none of those arrives on the calendar a vendor's release plan is built around.

Rating B Live 1 checks 3 counted signals