Seán Ernest Murray ⚡️

8am-10pm, 7 days a week This was 8 months of setting up Fuse’s support function in 2022. No back office tooling, no exception handling, pre GPT-3.5. After 10pm every day, I’d clear 100s of exceptions with python scripts: switch date changes, reissues, cancellations, industry data reports.... Many, many lessons. Here is 3:

My favourite thing to come from the Guinness research lab was the Student t-distribution Now a common Bayesian prior, it was originally invented to compare yield differences across varieties of barley. Farmers would test n=4-10 plots per variety and falsely flag one as better. William Gosset found the normal distribution gave false positives here due to low sample sizes. So he created a generalised "fat tail" distribution (which approaches normal as n->∞) to capture small sample uncertainty. He confirmed his findings by manually drawing small samples from 3,000 criminal height records (from Scotland Yard). Funny thing is - Guinness forbade publications to avoid leaking trade secrets. So Gosset, a brewer amongst academics, published it under the wry pseudonym "Student". Student t is now a go-to choice for loose priors, modelling small-sample or fat-tailed data (asset returns, clinical trials, A/B tests). Some might say Guinness were hiding their light under a bushel :p

the @Muse prompt (alongside being quite good) serving some pearls to live by