Vary a persona along named dimensions

Turn one persona_frame() into a designed set of personas. Two modes:

Usage

persona_variants(p, vary, n = NULL, .config = NULL)

# S3 method for class 'persona_set'
print(x, ...)

Arguments

p

A persona_frame() (the base).

vary

A named list of the dimensions to vary. Enumerated mode reads the level vectors (e.g. list(age = c("28", "52"), risk = c("cautious", "tolerant"))); generative mode reads only the names.

n

Number of briefs to generate (generative mode only). Ignored when enumerating.

.config

Optional generative LLMR::llm_config(). When NULL (default), the set is enumerated offline.

x

A persona_set.

...

Ignored.

Value

An object of class persona_set: a tibble with a persona list column of persona_frame() objects, an id column (each frame's hash), a variant_of column (the base hash), and one column per varied attribute.

Details

• Enumerated (default, .config = NULL): vary is a named list of level vectors and the result is their full Cartesian product. Each combination is rendered by appending the varied attributes to the base brief in plain, factual language (no stereotyping copula), so the design is legible and the base text is never rewritten.

• Generative (.config is a generative LLMR::llm_config() and n is given): one structured call asks the model for n individuated briefs that vary names(vary), under a hard-coded anti-essentialism instruction. Use this when you want fluent, non-templated briefs; audit the result with persona_audit().

Varying a demographic attribute does not license writing a stereotype. The enumerated renderer states attributes flatly; the generative prompt forbids caricature. Neither mode certifies the output: it produces briefs to be inspected, not a population to be trusted.

See also

persona_frame(), persona_audit()

Examples

base <- persona_frame("A first-time voter in a swing district.",
                      source = "synthetic")
set <- persona_variants(base, vary = list(age = c("19", "24"),
                                          leaning = c("undecided", "left")))
set
set$persona[[1]]$attributes
if (FALSE) { # \dontrun{
cfg <- LLMR::llm_config("openai", "gpt-4o-mini")
gen <- persona_variants(base, vary = list(age = NA, occupation = NA),
                        n = 5, .config = cfg)
persona_audit(gen)
} # }