Data science is the unglamorous craft of turning recorded facts into defensible decisions. It is older than its name. A nineteenth-century actuary pricing a life policy, a public-health officer counting cholera cases on a Soho street and a modern e-commerce team A/B-testing a checkout button are all doing the same job: gathering observations, asking a sharp question and reporting the answer with its uncertainty attached. What is new is the volume of recorded life — every click, every transaction, every sensor — and the cheap arithmetic to sift it. The job has not changed; the haystack has.

Every honest piece of analysis follows the same loop. One: a question, written down before the data is opened. Two: a dataset, with its provenance noted. Three: cleaning — fixing missing values, mis-typed entries and obvious errors, which routinely consumes most of the time. Four: analysis — the calculation or model that addresses the question. Five: communication — a chart, a number or a paragraph that a decision-maker can act on. Skip step one and you will find something interesting; skip step three and it will be wrong; skip step five and no one will care.

Byte: cleaning is the job

Surveys of working data scientists consistently report that roughly 60–80% of their time is spent finding, fixing and reshaping data — not modelling. If your mental image of the job is staring at equations, adjust it to staring at a slightly-wrong spreadsheet.

1. 1.Which step is most often skipped by amateurs and most regretted later?

   Writing the question down before opening the dataPicking the colour palette of the final chartChoosing whether to use Python or RBuying a faster laptop