Distinguish an object, a pipeline stage, a metric, and an outcome.
Mark observable and unobservable components of a generative-search event.
Frame a bounded GEO question without implying access to a closed-system algorithm.
PrerequisitesRead the course evidence-ceiling legend.Bring one answer surface or use the supplied synthetic fixture.
Builds on
Read the course evidence-ceiling legend. · Bring one answer surface or use the supplied synthetic fixture.
Investigates
What is the actual object of optimization?
Feeds
L01 system card v1 and source inventory.
02
Before class
Arrive with a prepared artifact
Reading route
Core PAPER-16/PAPER-19; Extend PAPER-06/PAPER-15.
Viewing route
Open the complete text-first lecture packageEight-minute course orientation: question, evidence boundary, artifact sequence. The current equivalent is notes, slide script, worked case, and no-video transcript; no recording is claimed.
Readiness check
Define retrieval and citation in one sentence each; identify one claim that a screenshot cannot support.
Bring
One completed scope card naming system, surface, locale, account state, query, time, and unknowns.
03
Explanation
Concepts, assumptions, and boundary
The object comes before the tactic
GEO is treated here as the study and responsible intervention of source visibility in generative search systems—not as a bag of prompt or copywriting tricks. A usable research object must name a source, query distribution, system surface, response, observation window, and measurable event. Discovery, retrieval, context selection, generation, attribution, and downstream action are different stages. A source can be crawlable yet never retrieved, retrieved yet not selected, selected yet not cited, or cited without changing the generated answer. Conflating these events makes a result impossible to diagnose or reproduce.
Closed systems require epistemic restraint
A screenshot can establish what a named surface displayed at a named time under recorded conditions. It does not disclose the platform’s ranking function, training data, hidden candidate set, or causal reason for the output. We therefore separate observation, interpretation, mechanism hypothesis, intervention, and outcome in every artifact. Open retrieval sandboxes can expose intermediate variables; commercial systems usually cannot. The course uses open systems to learn mechanisms and bounded field observations to study externally visible behavior, while keeping those evidence classes distinct.
Figure design. A left-to-right chain—discover, retrieve, select, generate, attribute, outcome—with a measured denominator under every arrow and a red hatched band behind hidden closed-platform stages. Every arrow is conditional; closed stages may remain unobserved.Long description
Six rounded nodes are connected by conditional arrows. Each arrow is annotated with a possible failure and its observable denominator. A background band marks stages that may remain unobservable in a closed platform; the caption explicitly rejects algorithm disclosure.
05
Interactive check
One action, one feedback state
Action
Classify twelve statements as observation, inference, hypothesis, or guarantee.
Feedback
Each answer reveals the missing system, surface, denominator, date, or hidden-stage caveat.
Accessible alternative
A printable answer table contains the same statements and rationales.