An outage doesn't change the price of redundancy
The price of redundancy after an outage is the price the week before it. What the failure moved is the burden of proof, and the burden reverts.
Rare knowledge doesn't feel rare
Nobody can hand you a list of what only they know. Knowledge transfer before an engineer leaves starts in your own systems: they already record who to ask.
An accepted cost is a decision nobody is making
A number that never moves reads as normal in every review it passes through. Reopening one starts with a question about who went looking for the cause.
More architecture won't move the objection
Technical objections survive better architecture when a stakeholder measures a new system with a requirement learned somewhere else.
A data migration is scoped by what the system can't do yet
A migration estimate prices the copy. Three capabilities the system had before the plan existed decide the rest, and whatever it lacks is the project.
Your safe migration window closes before the deadline
Migration safety spends time: rehearsal, phased rollout, observation, rollback. Once the schedule can't afford them, the safe migration window has closed.
Architecture bends around what you won't change
Designing around components you can't control means picking which side holds the workaround. The one you refuse to change shapes everything else.
Cloud cost optimization is an influence problem
Cloud cost optimization becomes an influence problem past the configuration layer, where the savings live in code another team owns and answers for.
A constraint you can fail overrules your architecture
An architecture can clear every performance spec and still fail a line that has nothing to do with speed: a price ceiling, a margin it quietly destroys.
Infrastructure facts don't belong in service code
A fact about infrastructure stored in service code is a second source of truth, and two copies drift. Generate the view from the source instead.
Migrate one variable at a time, or you won't know what broke
Hold every variable but one and a failure has one suspect. Bundle two and the suspects grow like the subsets: three places to look, then seven.
The irreversible decisions go first
Most risks are prices you can pay. The one with no affordable way back is different. Settle those decisions first, while you still have a choice.