A near miss is data, and it is almost always read as reassurance

One event, two opposite conclusions

Something almost failed and then did not. The reassuring reading is that the defences worked, the margin held, and the episode demonstrates the system is sound. The alarming reading is that a chain of events reached the last barrier, that the barrier held partly by luck, and that the same chain will recur.

Both readings are consistent with what was observed, and nothing in the event itself decides between them. What decides is usually which reading is cheaper, and the reassuring one costs nothing while the alarming one costs an investigation, a change and an admission.

What the evidence suggests people actually do

Research examining how people interpret near misses has found a consistent tendency: individuals who learn that a hazardous situation resolved without harm subsequently rate the risk as lower and choose less protective options than those who never learned of the episode at all. The near miss, which is information about how close the system came to failing, functions as evidence that it does not fail.

The studies behind this are a mixture of scenario experiments and analyses of organisational records, with the usual limitations attached to both. The direction of the effect is reasonably consistent across them; precise magnitudes are not, and scenario work in particular may not capture how people respond when consequences are real. Treat the pattern as well-motivated rather than as a precisely measured quantity.

Why the reassuring reading is structurally favoured

A near miss produces no damage, so it generates no record unless somebody chooses to create one, and creating one requires effort that will be visible while the harm avoided will not. Every incentive in a busy organisation points at moving on.

The outcome also supplies the wrong feedback. The system behaved acceptably, which trains everybody involved that the current arrangement is adequate, and that training happens automatically. This is the mechanism sometimes described as the gradual normalisation of a departure from the standard: a shortcut is taken, nothing bad happens, the shortcut becomes ordinary, and the margin it consumed is silently removed from the design.

Reading a near miss correctly

The productive question isn’t whether anything went wrong but how many independent things had to hold for nothing to go wrong. If the answer is several, the system has depth and the event is genuinely reassuring. If the answer is one, the event is a report that the last barrier is now the only barrier, which is the most valuable information any system ever produces about itself.

A second question is what stopped it. A designed defence that operated as intended is evidence of a working design. A person who happened to notice, an unusual quiet period, a delay that happened to intervene — these are not defences, they’re coincidences, and a system whose safety record rests on them has no safety record at all.

The counting problem

Near misses are far more numerous than failures, which makes them statistically valuable and practically difficult. They are the only source of information about the frequency of the precursor conditions, and by definition the actual failures are too rare to learn from. Any system that only studies the events with consequences is studying the smallest and least informative part of its own history.

The difficulty is that near misses are only counted where somebody has an interest in counting them and no penalty for doing so. Where reporting one is treated as an admission, the recorded count falls while the underlying rate does not, and the improving figures become evidence of nothing except that people stopped writing things down.

The personal version

The same reasoning applies to individual conduct, where it’s easier to observe and harder to admit. The journey that nearly went wrong, the deadline that was met by an accident of timing, the backup that was never tested and turned out not to be needed — each is an occasion where the margin was consumed and no consequence arrived to mark it.

The habit worth building is small: when something nearly went badly, write down what would have had to be different for it to have gone badly, and how many of those things there were. That count is the actual state of your margin, and it is not available from the outcome, which said only that this time it was enough.