The value of an imperfect record

AA visitor to a digital archive sees rows of neat images and may imagine an equally orderly past. Yet the notebooks behind one coastal weather collection were anything but uniform. Some observers recorded the sky before breakfast; others waited until the afternoon. Ink ran, pages disappeared and the meaning of a small cross changed between villages. For the fictional Harbour Records Project, these imperfections were not merely obstacles to be removed. They were clues to how knowledge had been made. Its historians wanted to preserve the relationship between an observation and the circumstances in which somebody had written it down. A perfectly tidy spreadsheet could sever that relationship while appearing to improve the evidence.

BThe project's first plan was simple: recruit volunteers to copy handwritten temperature readings into a searchable table. Every page was assigned to two volunteers working independently. If their entries differed, a staff member examined the image. Agreement made obvious transcription mistakes less likely, but it could not guarantee that a reading meant what the team initially thought. The number twelve, for instance, could be copied perfectly even if the observer had used a different temperature scale. The researchers therefore added instrument descriptions and notes about local conventions. These details were slow to collect, but without them comparisons between places would have been deceptively precise. Accurate copying and accurate interpretation turned out to be different tasks.

CA greater surprise came from gaps. Early software treated every empty box as a missing observation. A historian noticed, however, that one observer left the rainfall column blank after recording a dry day in the margin. For this observer a blank usually meant no rain, whereas another observer wrote a zero for the same condition and used a blank when absent. The team built separate rules for each notebook rather than forcing one convention on the entire collection. Some gaps remained unresolved. They were retained as uncertainty instead of being filled with the average of neighbouring days, because doing that would disguise precisely the unusual weather the project hoped to investigate.

DThese decisions changed how volunteers were trained. The original tutorial rewarded speed and the number of completed pages. The revised version asked contributors to explain ambiguous entries and allowed them to flag uncertainty without losing credit. Some experienced volunteers initially disliked the change, feeling that their expertise was being questioned. In response, staff published examples in which a confident first reading had been overturned by a note elsewhere in the book. Volunteers then began to discuss competing interpretations in a shared annotation space. The purpose was not to arrive at a unanimous judgement on every mark. It was to leave a visible account of why a particular interpretation had been adopted.

EThe public website reflected the same approach. Each table entry linked back to a page image, and a coloured marker indicated whether the interpretation was secure or disputed. Users could download a simplified dataset, but it was accompanied by a confidence field rather than presented as a record of uniformly reliable measurements. This irritated a software company that wanted a single clean number for each day. For historians, however, the refusal to provide that apparent certainty was a strength. A comparison based on doubtful entries could still be worth exploring, provided that its dependence on those entries remained visible. Uncertainty changed what could responsibly be concluded; it did not automatically make a source useless. A teacher using the collection illustrated the issue by giving students two versions of the same small table. One omitted all the warnings. Students working from that version wrote firmer conclusions, although they had received less information. The exercise did not measure the archive's educational impact systematically. It simply made visible why a shorter dataset should not automatically be mistaken for a clearer account of the past.

FThe collection was never intended to replace modern weather stations. Its value lay in extending the range of questions that researchers could ask about earlier communities and environments. Nor did the project establish that every local observer was equally careful. Instead, it supplied tools for investigating differences in their practices. The broader lesson concerns digitisation itself. Converting an old record into data is not simply a transfer from a fragile container to a durable one. It is a sequence of choices about which distinctions survive. An archive becomes more useful when those choices can be inspected and challenged, even if the resulting interface is less reassuringly neat than its designers first imagined.