Break One Arrow

Several connected changes could explain a recurring problem. Test one specific link while keeping the surrounding conditions as comparable as practical, so the result can meaningfully update your causal picture.

One link. One bounded change. A contrasting prediction.

Choose a consequential uncertain arrow in a candidate dynamic. Specify how a safe change would perturb that link and what the favored account and strongest rival predict. Keep other conditions stable where practical, observe through the relevant delay, and record remaining confounds. A small test narrows uncertainty; it does not automatically prove causation.

Try this when…

We changed everything and cannot tell what mattered.

This arrow is plausible but untested.

One cheap observation could weaken our explanation.

Can we preserve the rest while changing this connection?

What you can use it for

Workflow dynamics

Test whether a particular handoff or interruption changes later completion.

Practice feedback

Inspect one return from information into the next attempt.

Shared routines

Perturb a low-stakes trigger while tracking substitution and real outcomes.