Finding direction
How much evidence do I need before taking a reversible next step?
For a low-stakes reversible next step, you need enough evidence to show that the downside is bounded, the action can be stopped or undone, the key uncertainty is worth testing, and you know what observation will inform the next decision. You do not need proof that the direction will work. Raise the evidence bar as cost, irreversibility, impact on other people, or safety risk rises. If consent, legal, health, financial, or material consequences are involved, pause for qualified input instead of treating a small experiment as permission to proceed.
Match the evidence to the downside
There is no universal number of facts, hours of research, or confidence percentage that makes a reversible next step ready. Current UK government analysis guidance uses a proportionality principle: the effort and assurance should reflect the decision's scale, complexity, risk, impact, and uncertainty. That guidance is written for public analysis, not personal choices, but the practical interpretation is useful: demand more evidence when a mistake would be costly, difficult to undo, unsafe, or consequential for other people.
For a genuinely low-stakes step, the useful threshold is often modest. You should understand the obvious downside, have a credible stop or recovery path, and know why the action could reduce one decision-relevant uncertainty. Evidence that only makes you feel generally busier is not enough; evidence that changes what you will do next can be.
Use the STEP threshold
STEP is a Forever Free Compass framework, not an external research finding. Write one sentence for each field before acting.
- Stakes bounded: name the maximum time, money, exposure, obligation, and effect on other people, plus how you can stop or recover.
- Test one uncertainty: identify the assumption whose answer would materially change your next decision.
- Evidence rule written: state what you will observe, save, or ask, and what different results would mean.
- Permission preserved: confirm consent, access, duties, safety requirements, and any need for qualified guidance.
| STEP field | Not ready | Ready for a low-stakes test |
|---|---|---|
| Stakes bounded | I can probably back out | I will spend at most two hours and can stop without an ongoing obligation |
| Test one uncertainty | See whether this path works | Learn whether I want to repeat the core activity after trying it twice |
| Evidence rule written | Notice how it feels | Save both outputs and write a short note after each session |
| Permission preserved | It is only an experiment | No one else is committed, required access is approved, and no safety rule is bypassed |
Seek enough insight, not definitive proof
The 2026 Test and Learn guidance describes early prototypes as quick, low-cost ways to expose practical issues. At very small scale, it says the aim is enough insight for the next iteration, not definitive or generalisable evidence. It also recommends selecting uncertain or risky assumptions, collecting evidence systematically, and setting decision points and criteria before testing.
Applied to a personal direction, this means a reversible action may begin before the whole path is proven. The action should still be real enough to produce relevant information. Reading one more broad article may add little; completing a small sample of the actual activity, checking a constraint, or asking an informed person a precise question may produce evidence closer to the decision.
Run a simple sensitivity check
The Green Book recommends making assumptions explicit and testing how changes in important assumptions affect an appraisal. For a small personal step, use a plain-language version: ask what would have to be different for this action to stop being worth its downside.
If the step is only sensible when several optimistic assumptions all hold, gather more evidence or shrink the step. If it remains acceptable even when the time doubles, the response is lukewarm, or the first attempt fails, the evidence threshold may already be sufficient for learning. This is an interpretation for low-stakes planning, not a formal economic appraisal.
Raise the bar or pause when the stakes rise
The AQuA Book links greater assurance to greater complexity and business risk, and it says the total elimination of risk is not possible. The goal is an acceptable, transparent level of remaining risk, with attention concentrated on the areas of greatest consequence.
Do not use the STEP threshold alone for health treatment, legal rights, regulated financial choices, immediate safety, confidentiality, employment duties, or commitments that materially affect another person. Those situations may be hard to reverse even when the first action looks small. Obtain the needed consent or qualified input, follow applicable rules, and redesign the action if the recovery path is unclear.
Decide what the evidence changes
After the action, compare what happened with the evidence rule you wrote. Record what you observed, which assumption changed, what remains unknown, and whether the next commitment is still proportionate.
Stop when the downside was larger than expected or permission cannot be preserved. Adapt when the question matters but the test did not reach it. Repeat when another bounded observation could reduce the same uncertainty. Expand only when the evidence supports a larger commitment and the new stakes pass a fresh STEP threshold.
Original contribution
The Forever Free Compass take
Forever Free Compass uses the STEP threshold as an original decision aid: Stakes bounded, Test one uncertainty, Evidence rule written, Permission preserved. It asks for enough support to justify a specific reversible action, not enough evidence to prove an entire direction. STEP is not a validated assessment, a statistical confidence rule, or permission to ignore professional standards. Its evidence threshold rises with cost, irreversibility, effects on other people, safety exposure, and the difficulty of recovery.
Sources
- Test and Learn — HM Treasury and Evaluation Task Force. Proportional methods; prioritising critical uncertainties; quick, low-cost prototypes; small-scale evidence; predefined decision points; enough insight for iteration rather than definitive proof.
- Government functional standard GovS 010: Analysis — UK Government Analysis Function. Proportionate analysis and quality assurance; identifying uncertainty; matching analytical detail to the decision; avoiding unwarranted confidence.
- The Green Book (2026) — HM Treasury. Proportional evidence effort; explicit assumptions and uncertainty; research and pilots; sensitivity analysis and switching values.
- The AQuA Book — UK Government Analysis Function and partner analytical bodies. Fit-for-purpose evidence; assurance proportional to complexity and risk; explicit assumptions; risk prioritisation; limits of eliminating risk.
This is educational reflection, not medical or mental-health care. If distress is persistent, severe, or involves immediate safety, contact a qualified professional or local emergency service.