The Core Laws
Settle the plan before the work
Steering a plan costs a sentence. Steering a finished piece of work costs the afternoon.
Executed work is heavy. Once something exists, a document, a booking, an analysis, changing course means unwinding it, and its mere existence argues for keeping it. The plan is the last cheap moment: at plan time a wrong assumption costs one sentence to fix. Discovered in the finished artifact, it costs the afternoon, plus the temptation to ship the flawed version because it's already there.
But a plan is only as good as the reality it touched. The plausible approach fails on the constraint nobody mentioned: the record that was deleted, the account that was deactivated, the rule that lives in a neighboring system. Walk the plan against the actual terrain it will land on, and interrogate it rather than admire it. A plan you approved without reading is not a plan; it's a delay.
How to practice it
Plan first on anything multi-step.
Ask for the approach before the artifact. Read it. The sentence you fix there is the afternoon you don't lose later.
Ask what would make it wrong.
Ask which assumption, if false, sinks the plan. Then check that assumption against the real system, not the model's memory of it.
Check the edges it touches.
Plans fail where they meet other people's rules: the deleted, the deactivated, the renamed, the department you forgot. Walk those seams deliberately.
Don't rubber-stamp.
Approving an unread plan converts the cheap moment into ceremony. If you won't interrogate it, you've just chosen to debug it instead.
Deep scripture: Codeuteronomy II — Design is settled before code is written