Skip to contents

Structured debate between two agents

Source: R/presets.R
debate.Rd

Alternating statements in three phases: opening, rounds rebuttals each, and closing. An optional judge then delivers a structured verdict. The phase labels make it easy to analyze argument development over time.

Usage

debate(
  pro,
  con,
  topic,
  rounds = 2L,
  judge = NULL,
  msg_mode = NULL,
  quiet = FALSE,
  ...
)

Arguments

pro, con

Agents arguing for and against.

topic

The motion being debated.

rounds

Number of rebuttal exchanges (default 2).

judge

Optional Agent; if supplied, returns a verdict with a winner, a confidence, and reasoning.

msg_mode

Message construction, "roleflip" (default) or "flat"; NULL uses getOption("LLMRagent.msg_mode"). See conversation().

quiet

Passed through; FALSE prints utterances live.

...

Passed to the agents' underlying LLMR calls.

Value

An object of class agent_debate: a list with transcript (tibble: turn, phase, speaker, text), verdict (list or NULL), and motion. as.data.frame() returns the transcript.

Examples

if (FALSE) { # \dontrun{
cfg <- LLMR::llm_config("groq", "openai/gpt-oss-20b", temperature = 0.7)
d <- debate(
  pro = agent("Pro", cfg, persona = "You argue FOR the motion, rigorously."),
  con = agent("Con", cfg, persona = "You argue AGAINST the motion, rigorously."),
  topic = "Social media does more harm than good to democratic discourse.",
  judge = agent("Judge", cfg, persona = "A strict, impartial debate judge.")
)
d$verdict
} # }