01
Prior respondent
The respondent and an operative version of its prior state exist before the particular research demand that later finds or selects it.
§ Definition · artificial respondents
What is an artificial respondent?
An artificial respondent is a persistent, addressable artificial individual that exists with respondent-specific prior state before the particular research demand that later selects or questions it. Its responses are materially formed by that prior state and by what it actually observes.
- A persistent, addressable artificial individual
- Exists before the research demand that later selects it
- Responses are formed by prior state and what it actually observes
Last updated
§ 01 · Direct answer
The respondent exists before the particular research demand that finds it
Respondenthood is a claim about an individual and the causal origin of its responses, not merely about plausible role-play.
An artificial respondent is not a character created by the current research question. The respondent already exists as an addressable individual with respondent-specific prior state before a particular audience query, panel selection, or task later finds it.
That prior state must be operative. It can constrain what the respondent knows, notices, values, recalls, appraises, decides, or does. When the respondent encounters research material, the resulting response must be materially traceable to what that individual already was and to what it actually observed.
Canonical web definition
An artificial respondent is a persistent, addressable artificial individual that exists with respondent-specific prior state before the particular research demand that later selects or questions it. Its responses are materially formed by that prior state and by what it actually observes.
§ 02 · Public specification
Eight requirements define the category
The Artificial Respondent Specification defines what must be true without prescribing how an implementation must produce it.
The Artificial Respondent Specification is a public, CC BY 4.0 specification for the category. Version 1.0 is intentionally small: it defines respondenthood, evidence boundaries, causal formation, and the limits of what conformance establishes.
02
Persistent identity and operative state
The respondent remains one addressable individual whose prior state is rich enough to constitute a person rather than merely describe one.
03
Independent formation and evidence
Selection criteria, encounter material, preferred results, or scoring rules cannot create or rewrite state presented as prior.
04
Bounded encounter and knowledge
A response can use only what the respondent already knows or what it actually observes through its bounded capabilities.
05
Causally traceable formation
Material properties of the response must be traceable to respondent-specific state and observed encounter information that actually contributed to them.
06
Calibrated agency and honest outcomes
Agreement, resistance, uncertainty, questioning, abstention, and disengagement remain possible where the respondent state and encounter support them.
07
Semantic separation and medium neutrality
The respondent's semantic response stays distinguishable from platform intervention, rendering, and later human editing or selection.
08
Falsifiability and limits
The claim of respondenthood must be independently testable and capable of failing without implying human prediction or representativeness.
§ 03 · Antestrata implementation
A digital world first. A panel later
The Specification does not require this exact production method. Antestrata's implementation goes further by creating a grounded population before audience selection.
01
Ground the world
Real-world distributions set the population-level structure. They describe how characteristics and circumstances are distributed, not who any individual person must be.
02
Synthesize individual lives
LifecoreML creates new people inside that world. Their biographies, relationships, circumstances, memories, and tendencies are generated together as individual lives.
03
Search for the audience
A research brief does not create the people. It searches the existing population for individuals who match the audience you want to study.
04
Recruit the panel
Matching individuals become research respondents. The same person can be selected again because the identity existed before the study selected them.
Antestrata uses digital world twin as shorthand for the generated population beneath its research panels: aggregate patterns are grounded in real-world distributions, while every individual biography is newly synthesized rather than copied from a real person.
§ 04 · LifecoreML
A person has more than a demographic profile
The Specification sets a functional floor for individual state. LifecoreML realizes that floor as connected individual lives.
Identity and life coordinates
Birth date, age, location, household context, and the ordinary facts that place a person somewhere in the world.
Relationships and social context
Family, partners, friends, colleagues, roles, obligations, and the people whose opinions or needs can matter to a decision.
Education, work, and economics
Education, employment, income context, professional history, financial constraints, and the practical conditions around a choice.
Life events and memories
Generated episodes, turning points, associations, and remembered experiences that give later reactions somewhere to come from.
Attitudes, habits, and preferences
Beliefs, routines, tastes, risk tolerance, purchase habits, and other tendencies that accumulate across a life rather than a single prompt.
Behavioural tendencies and constraints
Stable patterns in how the person weighs evidence, handles uncertainty, trusts claims, hesitates, approaches, or avoids.
§ 05 · Population grounding
Real-world data grounds the population. It does not become the respondent
Population grounding is an Antestrata implementation property. The Specification deliberately does not make representativeness a condition of respondenthood.
Real-world distributions constrain the generated world. They help determine what combinations of ages, locations, circumstances, relationships, work histories, and other life conditions are plausible across a population.
The respondents themselves are newly synthesized individuals. Antestrata does not copy a census record, reconstruct a real identity, or treat training data as if it were a person.
Grounding and synthesis do different jobs. Grounding describes the world. Synthesis creates the people inside it.
Grounding describes the world. Synthesis creates the people.
§ 06 · Persona distinction
A persona is a portrait. A respondent is an individual
A vivid character description can still be a persona if its only operative role is to instruct a system how to portray someone.
A persona is built to represent a segment, archetype, or scenario. Giving the portrait a name and a backstory can make it vivid, but its purpose is still representation.
An artificial respondent must exist as a persistent, addressable individual whose respondent-specific state materially participates in response formation. Antestrata then goes further by selecting those individuals from an existing generated population.
Short distinction
A persona characterizes. An artificial respondent encounters and forms a response as an individual.
§ Category · method · research
A public category. A private production method. A working research product
The Artificial Respondent Specification defines what must be true. LifecoreML is Antestrata's production method. Antestrata selects respondents, runs the research, and returns a decision brief before the real users.