Every one of the 1,654 stances carries a year, so the corpus can be read as an agenda in motion. Each row is an era; each column is the kind of question being asked, as a share of all stances in that era. Read down a column, not across a row. Capability spikes to 20% when generative models arrive and falls back to 8% once the answer stops being interesting. Practice and workflow questions climb from 14% to 52% and never give ground. The profession's own questions about expertise and formation fall from nearly half the agenda to 7%.
Era is the corpus's one ordinal axis, and the eras are very unevenly sized — 81 stances before 2015 against 567 in 2026 — so these read as periods with distinct preoccupations rather than as a trend line. The sociology-of-professions era gives way to the Frey–Osborne automation debate, then to reliability, then to business models.
Change in each cluster's share of the agenda between the generative shock (2023–24) and now. A cooling cluster is losing share of a growing conversation, not necessarily shrinking — the Capability Frontier holds 39 stances then and 41 now, still the largest cluster in the corpus, but a smaller slice of a much bigger argument. Four clusters have dropped out of the record altogether since 2022; two arrive for the first time in 2026.
The honest boundary of the artefact. Two of these are gaps in the corpus itself, surfaced by probing for topics that turned out not to exist; the rest are properties of the index that would mislead if read as properties of the world.