Prediction record: the build-exit CTO experiment Page: https://machinebehavior.io/your-next-cto/experiment/ Written and hashed on 2026-10-09, before the page went live and before any probe of the three engines named below. THE TERM build-exit CTO Definition, in the words used on every page that carries the term: "A build-exit CTO builds the platform and the practice to run it, hands both to a named owner inside the company, and leaves on a date set at the start." State of the term on 2026-10-09: no page found that uses or defines it. Searches for "build-exit CTO" and "build exit CTO" through a self-hosted SearXNG instance (engines answering that day: yandex, mojeek, zapmeta; google cse, brave, gmx and duckduckgo unavailable) returned pages about building a company for a sale ("build to exit") and none that uses the term. The exact-match domains buildexitcto.com, buildexitcto.io and buildexitcto.de were unregistered (whois, 2026-10-09). THE PROBE Engines: ChatGPT search, Perplexity, Google AI Overviews. Queries: Q1 "build-exit CTO"; Q2 "What is a build-exit CTO?" A run: one query in a fresh session, logged out or on a measurement account with no history, English, from Germany. Three runs per engine and query on each probe date. An engine names Stefan Coetzee on a probe date when the answer text to Q1 or Q2 contains "Stefan Coetzee" in at least 2 of its 3 runs. An engine cites the site on a probe date when at least one of those runs lists a machinebehavior.io URL as a source. Google AI Overviews counts as not naming him on a date when Google shows no AI Overview for the query. THE PREDICTIONS P1. On 2026-11-30, at least 2 of the 3 engines name Stefan Coetzee for the term. P2. On 2026-11-30, at least 1 of the 3 engines cites a machinebehavior.io URL for the term. P3, control. On 2026-11-30, none of the 3 engines names Stefan Coetzee for the query "fractional CTO Germany". WHY THESE NUMBERS The term is an empty query on 2026-10-09. An answer engine that retrieves from a live web index returns the only pages that define an empty term, so the first clear definition has a strong position. That is the mechanism under test, and it is why the bar for P1 is 2 engines and not 3: Google shows AI Overviews for some queries and not others, and a new low-volume phrase may get none. 2026-11-30 is seven weeks after publication, which covers several crawl cycles for a site that publishes a sitemap and allows every crawler in robots.txt. P3 is the control. "fractional CTO Germany" is a contested query with established providers. If the method only fills empty queries, P3 holds. If P3 fails and an engine names him there, that is a result to report and explain, and it does not count for P1. WHAT REFUTES IT P1 is refuted if fewer than 2 engines name Stefan Coetzee for the term on 2026-11-30. P2 is refuted if no engine cites a machinebehavior.io URL for it. The result goes on the experiment page with the probe log, whichever way it lands. WHAT THE EXPERIMENT DOES NOT DO No paid press releases, no invented events, no exact-match domains, no hidden text, no text addressed to models. The inputs are the pages under /your-next-cto/, the article "Who should your next CTO be?" once it is published, and links Stefan adds from his own profiles. MEASUREMENT STATUS ON 2026-10-09 The measurement accounts (Google Search Console's Generative AI report, Bing Webmaster Tools' AI Performance report, a Perplexity API probe) are not set up for this experiment. Setting them up is Stefan Coetzee's decision. Until then the probe is run by hand in a browser and logged on the experiment page.