Skip to contents

Sends the business ask (and the optional situation) through seven independent, versioned dimension prompts to a large language model, converts the seven raw scores (0 = small space, 10 = large space) to the internal condition scale, and aggregates them with measure_liminal_space().

Usage

assess_liminal_space(
  ask,
  situation = NULL,
  weights = default_weights(),
  chat = liminal_chat(),
  parallel = TRUE,
  max_active = 7L,
  rpm = 500
)

Arguments

ask

A single non-empty character string: the business ask, verbatim, in whatever language it was written.

situation

NULL or a single character string describing the context: data state, history, constraints, known ambiguities.

weights

A named numeric vector with one non-negative weight per dimension, summing to 1. Defaults to default_weights(). Invalid weights abort before any LLM call is made.

chat

An ellmer Chat object used for all seven calls. Defaults to liminal_chat(), which reads its configuration from environment variables. Inject a custom chat to control the client in tests or scripts.

parallel

Logical; run the seven dimension calls concurrently (default) or sequentially.

max_active

Maximum number of simultaneous requests in the concurrent path; at most one per dimension.

rpm

Maximum number of requests per minute in the concurrent path.

Value

A list of class liminal_assessment with elements:

  • liminality — the final score in [0, 1];

  • raw_scores — the seven model scores, integers in [0, 10];

  • scores — the seven condition scores in [0, 1];

  • weights — the weights used;

  • model, base_url — provenance of the client;

  • prompt_versions — the template version used per dimension;

  • ask, situation — the inputs, verbatim.

Details

The seven dimension calls run concurrently by default; concurrency affects timing only, never the scores, because the calls share no state and results are re-assembled in canonical dimension order.

Examples

# \donttest{
if (nzchar(Sys.getenv("PANAMA_API_KEY"))) {
  assessment <- liminal2::assess_liminal_space(
    ask = paste(
      "Give us indicators of project execution built from the",
      "task and timesheet data we already have."
    ),
    situation = paste(
      "Three years of in-house task and timesheet data with",
      "stable definitions."
    )
  )
  print(assessment)
}
# }