CitePack runs one Decision Loop for one offering, in one buying category, for buyers in one country. It asks ChatGPT, Claude, Gemini and Perplexity the buying questions for that decision, through their official APIs, and keeps every answer. One loop is 120 answers: 60 on Day 0 and 60 when the same decision is measured again. Five buying questions are drafted for the decision and approved by the customer before anything is asked: one for each of five buying situations (category consideration, job suitability, mandatory eligibility, balancing priorities and change of solution). Each question can carry up to two buyer specifics. Once approved, the questions are sealed and go to every engine exactly as written. Each engine is asked three times per question, with web search on. Each runs one fixed model, chosen to match the model its free app gives buyers, and every answer records the model that wrote it. The buyer's country is set on every engine and written into every question, so the 60 observations of a map are all about one market. The Decision Map reads those answers: the options put forward, the company's place in each answer where the order is clear, the other names that appear, the words used about the company checked against its own pages, and the cited pages kept apart from the pages found by search. Being mentioned, being offered and being cited are recorded separately. The Work Plan follows the searches, the companies looked up by name and the pages read. Each item names its question, answers, evidence, kind of work, effort, owner and the check for the next measurement. The Action Pack holds the material for each item. Measure Again repeats the same sealed questions on Day 28, or on any day the customer chooses. The Before/After Snapshot reports each engine as Same or Different, with the count of changed answers, and flags any model change. It reports movement or none. It gives no score and makes no promise about the answers.