Four people on useful failure

A failed survey can still tell us that our sampling window was built around the wrong tide. I write a failure note before redesigning the method, while the original assumptions are still visible. Repeating the survey immediately can erase the evidence of why it failed.

The note distinguishes absence of the species from absence of a fair opportunity to detect it. That distinction prevents negative data from acquiring more confidence than the method deserves. Before moving the recorders, I copy the assumptions behind the failed survey into a separate note: expected tide, calling period, wind and listening range.

The redesigned method is linked back to that note so a later biologist can see what changed and why. Otherwise the improved survey could erase the very evidence that showed the original sampling window was wrong. A second survey is planned only after we have stated what the first one could not have seen.

That prevents a disappointing count from being mistaken either for absence or for proof that the entire method was useless.

A cracked vessel is not automatically a useful experiment; sometimes it is simply careless firing. I keep failures only when I can identify the pressure, moisture or timing they reveal. The broken object must change the next decision if it is to count as knowledge.

Students often want to display every accident as creative discovery, which avoids the discipline of explanation. Useful failure is selective because interpretation, not damage, gives it value. I do not archive every cracked pot.

For the failures I keep, the card identifies clay body, moisture, kiln shelf and the pressure revealed by the fracture. A later ceramicist can distinguish a break that tests an explanation from damage caused after firing or by careless handling. The object enters the teaching collection only when those contrasts make a defensible interpretation possible.

In the workshop, fragments are kept only when firing conditions were documented well enough to support an inference. A broken pot without that context may be interesting, but it cannot carry the explanatory weight of a controlled trial.

Our incident reports begin with the conditions that made an error consequential, not with the individual who triggered it. A harmless action can expose a dangerous dependency if the system offers no recovery path. We test whether another team would have made the same mistake with the same information.

Blame narrows attention; reproducibility widens it to design. The report is complete only when a safer default or clearer signal has been implemented. Our report reconstructs the system state before the incident, the information visible to the operator and the recovery options available afterwards.

Those circumstances show why a routine action became dangerous in this configuration. A later engineer can test whether another user would reach the same point, and whether the revised default removes the dependency rather than merely preventing one person from repeating the error. Our review separates the action that triggered an incident from the conditions that made its consequences likely.

Recommendations are addressed to the system first, and individual responsibility is considered only where the evidence genuinely supports it.

When an activity fails, the loudest explanation is often that learners lacked ability or motivation. I first ask what the task allowed them to notice and what it required them to guess. An exercise can be difficult for the wrong reason and still produce low scores that look impressive.

I compare errors across several tasks before deciding whether a language feature is genuinely insecure. Failure is useful only after the demand of the task has been separated from the skill being measured. I keep the instructions, examples, time allowance and sequence of earlier activities with the pupils’ responses.

A later teacher can then ask whether low performance came from the target language or from information the task forced learners to guess. Comparing the same feature across differently designed activities prevents one impressive-looking failure from becoming an unsupported judgement about ability. I compare performance across tasks that place different demands on memory, language and confidence.

A recurring difficulty then becomes interpretable; a single poor result remains a prompt for investigation, not a diagnosis of the learner.