Are we celebrating speed or fewer mistakes?
In a hypothetical tool-packing experiment, we propose separating similar parts into designated trays instead of keeping them in a shared container. The question is whether this arrangement reduces selection of the wrong part. Once work starts, packing might become faster while errors remain unchanged. Renaming the project a 'packing speed experiment' at that point would conceal the difference between the original question and what attracted our attention later.
We do not need a large manual to begin. We need a plan an independent reader can understand before results appear: the question, comparison procedure, measurements, stopping point and exclusions with their reasons. Boundaries and organization remain relevant, but here they take the form of decisions that can be revisited against a preserved record.
A plan card before opening the boxes
We specify the primary outcome as boxes containing an incorrect part out of all inspected boxes, defining both an error and the inspection procedure. Packing time becomes a secondary outcome, so it cannot quietly replace the primary one later. We identify who arranges trays and who inspects boxes, and address differences in participant familiarity and tool types between the two methods.
We choose follow-up duration and size according to the precision required, recording the stopping rule before examining the direction of results. Exclusions also need advance definitions. A box damaged independently before inspection might be excluded, with its count and reason retained. A box containing the very packing error under study cannot be excluded because it 'spoiled the experiment': that error is part of the subject.
Four positions along the timeline
The diagram shows a fixed plan, observation and report, with a branch from observation to a documented amendment. The amendment also leads to the report, appearing beside the original plan. The arrow from plan to observation represents time order and intended procedure. It does not mean that having a plan guarantees sound measurement or a successful intervention.
Suppose a supply interruption forces us to change tray types. We record when this happened and what changed. We might begin a new phase under an amended plan and distinguish the phases in analysis where needed. We do not erase the earlier description or treat cases collected before the amendment as though they arose under the new conditions.
What does preregistration do?
In research practice, specifying a plan and submitting it to a registry before the study is called preregistration. Its role here is to make planned work distinguishable from additions made after seeing data. It does not establish that the question matters, the measurement is valid or the comparison is sound. A declared plan can still be weak and benefit from methodological review before implementation.
We could instead plan exploratory work from the outset: observing where workers hesitate between similar parts and developing a new question from those observations. That can be useful when described accurately. The difficulty arises when an exploratory observation is presented as an earlier prediction that survived independent examination, even though it helped generate that prediction.
A necessary amendment versus a redefined success
Imagine inspection reveals faulty instructions for classifying parts. A reasonable response might be to pause measurement, repair the instrument and determine how the problem affected existing records. Documentation makes that change understandable. Continuing with defective instructions simply because they were written beforehand does not serve the question; neither does concealing the correction.
A different change is driven only by the result: repeatedly extending observation until a desired difference appears, or replacing the primary outcome because another looks better. Such decisions must be disclosed. Editing the title does not restore the study's earlier status. We can present the new analysis clearly and seek an appropriate subsequent test.
An ending that preserves decision history
The final report gives visible space to the original question and result, amendments and reasons, and new questions. If errors did not fall under the specified criterion, that result remains visible even if speed improved. If information is insufficient for judgment, we report uncertainty instead of turning it into a definite victory or defeat.
The plan provides an organizing boundary, while amendments provide flexibility within a reviewable record. Together they allow learning without moving the finish line after crossing it. The experiment succeeds in giving an honest answer to its question, including an answer that leads us to change the procedure or abandon it.



