The corrected score still needs the story of its correction
I explain why a results service can hold the right current value and still fail to account for what it supplied earlier, using an update decision I had to reverse.
Read the essayFile sectionThe leadership journal
Where the most convenient data explanation breaks down
A better timing measure is not proof of a faster feed
A revised latency account changed the evidence and the population being measured. I explain why its lower median could not be presented as an operational improvement.
Read the essayTwo matching annotations can carry one shared mistake
A high agreement figure concealed errors both annotators made. I examine why consistency and correctness need different evidence in a sports-data service.
Read the essay