How can I adapt a small experiment without losing track of what changed?

Freeze the version you actually tried

Before adapting, copy five things into a short version note: the learning question, the action, the observation you planned to use, the review point, and the boundaries that must remain intact. Do not overwrite this note. It is the reference for what version one was supposed to test and what it actually did.

The UK government's 2026 Test and Learn guidance warns that adaptation can make teams lose sight of which version generated a result. It recommends defining assumptions and decision criteria before testing and formally documenting later changes. Its setting is public policy, not personal choice; the modest transfer is to preserve a readable before-and-after record.

Use the PATCH change card

PATCH is a Forever Free Compass framework, not an external research finding. Complete it when you decide to alter the experiment, before the next attempt begins.

1. Preserve the prior version: keep its question, planned action, observation, boundaries, and notes unchanged.
2. Alter one defined feature: name the single controllable part that will differ, such as timing, duration, format, sequence, or support.
3. Timestamp the change and reason: record when the new version starts and which observation or practical problem justified it.
4. Capture context: note relevant conditions that shifted independently, including access, workload, another person's availability, or the setting.
5. Hold observations apart: label every note with its version and compare versions before deciding what the change may suggest.

Change one feature for one stated reason

Write the adaptation as a sentence: “For version two, I will change ___ because version one showed ___; the learning question and boundaries remain ___.” If several features must change together, list them all and treat the result as a new bundle. Do not claim that one of those features caused what follows.

Test and Learn describes iteration as feedback informing how an intervention should be adapted, while the Magenta Book says a Theory of Change should keep developing as new evidence appears. These sources support evidence-led revision, not constant tinkering. A change should respond to a named observation or feasibility issue and should update the assumption you are examining.

Record context without turning the log into a diary

Capture context only when it could change the interpretation or the next decision. Useful examples include an interrupted attempt, a changed permission or cost boundary, a new tool, a different setting, or an observation collected at a different time. Keep description separate from explanation: “the attempt moved to the evening” is an observation; “the evening caused the result” is an unsupported causal claim.

CDC's Program Evaluation Framework says context and data needs can change during implementation and may require modifying the design, questions, or purpose. It also says documenting contextual challenges and their effects strengthens transparency. For a personal experiment, one or two decision-relevant context lines are usually more useful than exhaustive notes.

Review versions separately before comparing them

At the review point, first summarize each version on its own: did the planned action happen, what was observed, which boundary or context changed, and what remains unknown? Only then compare them. If the task, measure, timing, and context all changed, the notes describe two different attempts; combining them into one average hides that difference.

The 2026 Magenta Book transparency guidance recommends time-stamping and archiving plans before data collection, and recording any later change with its rationale. It explains that advance plans reduce the risk of adjusting methods after seeing results in ways that bias interpretation. PATCH borrows that transparency principle at a much lighter scale; it does not make a personal note a formal research record.

Keep the conclusion smaller than the change

Choose a modest conclusion: the revised version is feasible enough for one more bounded attempt; the change did not resolve the stated problem; too many conditions changed to compare; or the boundary says stop. An apparent improvement can guide the next step, but it does not establish that the altered feature caused it or that the result will repeat elsewhere.

Use PATCH only for low-stakes, reversible exploration. Do not adapt medication, legal obligations, regulated financial activity, workplace requirements, safety controls, privacy commitments, or another person's involvement through an informal experiment. Preserve consent and required processes, and seek qualified guidance when consequences are material.