Build the study manifest for a run

One object tying together the design, personas, tools, workflow, served model identifiers, generation parameters, and package versions, with a manifest_hash that changes when any of those changes. It does not hash the transcript or replies (those are outcomes, not identity), so two runs of the same design with different sampled replies share a manifest_hash but differ in their per-call records.

Usage

agent_manifest(run)

Arguments

run

An object accepted by as_agent_run() (a chat agent, a conversation, a preset, a pipeline, an experiment, a think_harder() result, or an agent_run).

Value

An object of class agent_manifest (a list); see Details. Print shows the short hash, kind, models, and headline counts.

See also

hash_persona(), hash_tool_spec(), hash_workflow(), archive_agent_study(), LLMR::llm_hash()