Run the STORM pipeline — tempest

This is a scripted workflow that:

discovers perspectives and research questions,
runs a multi-perspective research loop (search/fetch + expert synthesis),
creates an outline,
writes a cited report in Markdown.

Usage

tempest_run(
  topic,
  config = tempest_config(),
  retriever = NULL,
  knowledge = NULL,
  n_experts = 3,
  experts = NULL,
  research_strategy = c("key_questions", "conversation"),
  max_rounds = 3,
  max_questions_per_perspective = 3,
  parallel_writing = FALSE,
  steps = c("perspectives", "research", "outline", "write", "polish"),
  output_dir = NULL,
  resume = FALSE,
  run_id = NULL,
  progress = NULL,
  verbose = TRUE
)

Arguments

topic: Research topic or question.
config: A TempestConfig.
retriever: Optional TempestRetriever or compatible retriever with a ResearchWorkspace at $workspace and search(query, k) and fetch(url) methods. search() returns a data frame with non-empty title and HTTP/HTTPS url columns; snippet is optional. Tempest derives or verifies source_id. Create the workspace with tempest_research_workspace() and source IDs with tempest_source_id(). If NULL, a retriever is created from config.
knowledge: Optional accepted organizational knowledge from tempest_artifact_knowledge(). It supplies accepted evidence records without selecting or authorizing executable programs. The live Graft view and knowledge object are not serialized; materialized evidence and artifact selections are retained in the bundle.
n_experts: Number of expert profiles to generate when experts is NULL (default 3).
experts: Optional list of active profiles created by tempest_expert(). When supplied, STORM uses this selected team and does not generate experts.
research_strategy: Either "key_questions" (default, faster) or "conversation" (more thorough but slower). Key questions uses predefined questions; conversation dynamically generates follow-up questions.
max_rounds: Maximum rounds per perspective for the "conversation" strategy (default 3).
max_questions_per_perspective: Maximum questions per perspective for the "key_questions" strategy (default 3).
parallel_writing: If TRUE, write report sections in parallel using the mirai package. Failed parallel sections are retried sequentially.
steps: Character vector controlling which steps to run. Defaults to all.
output_dir: Optional directory for persisted STORM run artifacts. When supplied, a current schema-10 product bundle with schema-5 STORM state is written under a topic-specific subdirectory.
resume: If TRUE and output_dir contains a previous run, load saved current-format artifacts and skip stages recorded as complete. Older, future, missing, extra, or mismatched product shapes are rejected rather than migrated.
run_id: Optional run directory name. Defaults to a slug of topic.
progress: Optional function called with one plain named progress record as STORM workflow stages start, finish, fail, persist artifacts, or make final artifacts available. See the package vignette for the eleven stable fields.
verbose: If TRUE, prints progress messages.

Value

A validated completed Tempest research product. Read it with tempest_report(), tempest_sources(), tempest_claims(), tempest_claim_supports(), and tempest_trajectory_review().

Examples

if (FALSE) { # \dontrun{
cfg <- tempest_config()
result <- tempest_run("History of jazz", config = cfg)
cat(tempest_report(result))
} # }