Run provocateur questioning across themes (Mode 1 entry point)

Orchestrates the five (or selected) provocation categories per theme, assembling a ResearcherReflectionLog that captures every provocation issued. The log is the AI's contribution to a Mode 1 analysis – the THEMES + CODES are the researcher's authorship, kept on the same log object as researcher_authored_codes / researcher_authored_themes.

Usage

run_provocateur_questioning(
  data,
  theme_set,
  provider,
  config = list(),
  categories = .VALID_PROVOCATION_CATEGORIES,
  audit_log = NULL,
  response_cache = NULL,
  fabrication_log = NULL,
  resume_log = NULL
)

Arguments

data

Tibble with std_id + std_text (standardized corpus).

theme_set

ThemeSet object. Each theme drives one round of provocations. The researcher must have authored these themes; the provocateur does NOT name themes.

provider

AIProvider object.

config

Optional config list (used for logging/context).

categories

Character vector of category names to run (default: all five).

audit_log

Optional AuditLog from init_audit_log.

response_cache

Optional ResponseCache.

fabrication_log

Optional FabricationLog.

resume_log

Optional ResearcherReflectionLog to append to (resume semantics).

Value

A ResearcherReflectionLog with provocations populated.

Details

Per AC7 (universal Tier-0): every provocation that cites verbatim evidence runs through verify_quote; fabricated provocations are dropped silently and recorded to the audit + fabrication logs.