The Reversibility Test: How Much Decision Does This Decision Need?

Not every decision deserves a spreadsheet. Not every decision deserves a shrug. The useful question is how much uncertainty you must reduce before acting—and how much of the result you can repair after you learn.

The reversibility test is a practical way to size the decision process. It does not sort life into two clean boxes. Reversibility has dimensions: money can be refunded while trust cannot; software can be rolled back while copied data cannot be recalled; a trial week can end while an injury persists. The protocol makes those dimensions visible before confidence takes over.

Why reversibility changes the right amount of analysis

When a choice is cheap to try, quick to observe and easy to undo, delay can cost more than a small failure. A one-week calendar experiment may teach more than another hour of speculation. When a choice exposes health, livelihood, sensitive data, another person's rights or a large irreversible commitment, speed is not automatically a virtue.

NASA's systems-engineering guidance makes a related point in a higher-stakes setting: not all decisions require the same analytical rigor, and the value of reducing uncertainty depends partly on whether more information could change the choice and whether the schedule can absorb the delay. This guide adapts that general discipline to ordinary decisions. It is not NASA's method and should not substitute for professional, legal, medical, financial or safety review.

Step 1: classify six kinds of reversibility

Write the decision in one sentence, then score each dimension as easy, costly or not meaningfully reversible. Do not average the scores. One severe dimension can dominate the whole choice.

1. Material: Can money, property or time be recovered?
2. Operational: Can the old workflow, tool or arrangement be restored?
3. Informational: Can exposed data, a public statement or a copied file be recalled?
4. Relational: Can trust, consent and reputation be repaired?
5. Bodily: Can effects on health, safety or physical capacity be undone?
6. Path: Does choosing this remove future options or create dependence?

A subscription with a cancel button may be materially reversible but informationally sticky. A job offer may be operationally reversible but relationally and financially costly. A public launch can be technically rolled back after screenshots and expectations have escaped.

Step 2: mark the blast radius

List who bears the downside if the decision is wrong. Start with people who did not choose it. A test that affects only your own morning routine is different from an experiment that changes a team's workload, a child's privacy or a customer's access.

Use three rings: self, known others and unknown others. For each ring, write the worst plausible consequence rather than the most dramatic imaginable one. If another person carries meaningful downside, reversibility requires their informed participation—not merely your ability to stop later.

Step 3: price the delay

Analysis has a cost. Waiting can preserve options, but it can also allow a small problem to compound, miss a deadline or consume attention that belongs elsewhere. Write what one day, one week and one month of delay would cost. Include opportunity, stress and information that will arrive naturally.

Then ask whether research will reduce decision-relevant uncertainty. More reading is useful only if a plausible result would change the action. If every answer leads to the same safe trial, start the trial. If one unresolved fact could expose someone to serious harm, find that fact first.

Step 4: design the recovery before acting

A decision is not reversible because you feel willing to change your mind. It is reversible because a recovery path exists and you have preserved the resources to use it.

• Back up data before migration and test that the backup opens.
• Keep the old process available through a defined overlap period.
• Limit a pilot by users, money, geography or time.
• Name the person authorized to stop it.
• Reserve the budget, time and access required to restore the prior state.
• Write how affected people will be informed and repaired if the trial causes harm.

“We can always switch back” is not a recovery plan. It ignores lost data, contract terms, retraining, social cost and the tendency to defend a choice after investing in it.

Step 5: choose an observable signal and a stop rule

Define what the experiment must teach. Use one outcome that matters and one guardrail that must not deteriorate. For a notification change, the outcome might be uninterrupted work blocks; the guardrail might be response time for family messages. For a new team tool, the outcome might be fewer handoffs; the guardrail might be accessibility or error recovery.

Set a review date before beginning. Add a stop rule tied to evidence: “Restore the previous process if two required records are lost,” not “stop if it feels bad.” Feelings matter, especially in personal decisions, but a vague exit rule is easy to renegotiate when sunk costs and public commitment accumulate.

Step 6: match the process to the class

Worked example: moving a personal email domain

The monthly fee may be refundable, but the real decision has other dimensions. Messages can be exported, yet DNS changes can interrupt delivery. A custom domain preserves address portability, but contacts may still depend on the old provider's formats. Other household members can lose access. Recovery codes can be misplaced.

A reversible version looks different from an instant switch: inventory accounts; verify domain control; export mail and contacts; create the new addresses; test authentication records; run a short overlap; send test messages in both directions; keep the old account until delivery and recovery are proven. The choice of provider matters. The sequence makes the choice less brittle.

Failure modes the test is meant to catch

• Binary labels: calling a choice reversible because one component can be undone.
• Invisible stakeholders: treating other people's exposure as your experiment.
• Research theater: collecting information that cannot change the decision.
• Rollback fiction: promising reversal without preserved data, access or budget.
• Sunk-cost defense: continuing mainly because stopping would admit the first choice was wrong.
• Permanent pilots: starting a trial without a review owner or date.

Research on escalation of commitment does not identify one universal cure, and laboratory effects do not map perfectly onto every life decision. Still, the pattern is credible enough to design around: negative evidence can arrive after money, effort and identity have become attached to the original choice. A prewritten review makes that evidence harder to explain away.

The one-minute reversibility card

1. Decision and deadline:
2. Hardest thing to recover:
3. Who bears downside without choosing it:
4. Cost of waiting one week:
5. Smallest safe trial:
6. Backup or recovery path:
7. Success signal and guardrail:
8. Stop rule, owner and review date:

Use the card to produce an action, not a feeling of control. If the risk is low, the action may be “try it today.” If the risk is high, the action may be “do not proceed until a qualified person verifies this condition.” Both are decisions.

Sources and scope

• NASA, Systems Engineering Handbook, section 6.8 on decision analysis and matching rigor to uncertainty and consequence.
• NASA, Systems Engineering Handbook Appendix, including analysis-of-alternatives terminology.
• Feldman and Wong, A New Inaction-Effect Perspective on the Sunk-Cost Fallacy.
• NIST, Risk Management Framework, a domain-specific example of framing, assessment, response and ongoing monitoring.

The six-dimension test and card are original editorial synthesis. The institutional sources support the general principles of proportional analysis, explicit alternatives, uncertainty, recovery and review; they do not endorse this personal protocol.