chrismdp / counting-house-goldFernwick Air: building the scored setPublic · Version 3 of 3#3 · chrismdp · Yesterday 12:12am · 0.6ShowingLatest#2 · chrismdp · Yesterday 12:10am · 0.6#1 · chrismdp · 27 Oct 2026, 9:56pm · 0.6ForkWho may see each partWhere each part runsYAMLChecksNo mistakestokenisedFernwick staffScopeops: Fernwick staffthe sampled casesthe sampled casestheir answerstheir answers[agreed][disagreed]the settled answersthe casesTriggerEach week, for the intentbeing scoredLinked to chrismdp/counting-house-live#closedData · untrustedClosed cases, tokenised: whatwas asked, what the clerkanswered, what was sent anddoneCode · bring-inSample a few hundred closedcases, oversampling the riskyand the rarePerson · checkThe first scorer answers therubric's questions, blindPerson · checkThe second scorer answersthe rubric's questions, blindCode · checkOnce both have scored,compare their answersPerson · checkA third person settles eachdisagreementCode · doFile each agreed case in thescored set, and the scorers'agreementUsed by chrismdp/counting-house-judge-test and chrismdp/counting-house-promotionDataThe scored set: real cases,each with its true intent andagreed answers, a slice heldout for examsUsed by chrismdp/counting-house-judge-testDataHow often the scorers agree,per intentFlow only+−Fit