Short pieces on lessons from the projects, each linked to the case study it came from.
8th CPC
When the date cannot move, choose what absorbs the risk
On most programmes, when something slips, the date moves. On the 8th CPC the date was set in statute and could not.
That left a different question. If the date is fixed, which variable is allowed to take the strain? I chose coordination. Running the modules as parallel streams cost more of my time and more coordination overhead, and in return no single module could sink the date.
Naming the variable early is what makes the rest of the plan honest. A plan that pretends everything is fixed is a plan that has not decided where it will break.
The question is not whether the model is good enough. It is whether the people accountable for the output can yet tell.
On GHCI the process was new, so the auditors had never operated it. An AI suggestion they could not evaluate would have been a suggestion nobody should trust. So Phase 1 launched without AI, and Phase 2 will be accepted against a golden dataset of cases already answered by hand.
It cost a phase of efficiency. It bought an acceptance test that means something.
A unit losing money at twice its required headcount looks like a staffing problem. It usually is not.
Cutting first removes capacity and keeps the dependency: the work still has to be done, and only a few people can do it. Standardise the knowledge so it stops living in individuals, move the work to the right tier, and only then resize.
The order matters. Reverse it and the risk moves to whoever is left.