Skip to contents

Run progressive sequential coding on all entries

Source: R/09_coding.R
run_progressive_coding.Rd

Processes entries strictly one at a time. For each entry, the AI reads the text and codes applicable segments using existing codes or creating new ones. Entries with no applicable content are skipped.

Usage

run_progressive_coding(
  data,
  provider,
  config = list(),
  learning_context = NULL,
  research_focus = "",
  checkpoint = NULL,
  concepts = NULL,
  resume_state = NULL,
  audit_log = NULL,
  response_cache = NULL,
  fabrication_log = NULL,
  framework_spec = NULL,
  live_tracker = NULL
)

Arguments

data

Tibble with std_text, std_id columns

provider

AIProvider object

config

Coding config section

learning_context

LearningContext object (or NULL)

research_focus

Research focus string

checkpoint

CheckpointManager (or NULL)

concepts

Character vector of core research concepts (or NULL)

resume_state

ProgressiveCodingState from a previous partial run (or NULL)

audit_log

An AuditLog object (from init_audit_log) for recording each coding decision (entry skipped, code assigned, new code created), or NULL to disable audit logging for this step.

response_cache

An optional ResponseCache object (from init_response_cache). When provided, raw API responses for each per-entry coding ai_complete() call are written to the cache and a reference is recorded in the audit log. Pass NULL to skip raw-response capture.

fabrication_log

An optional FabricationLog object (from init_fabrication_log). T0.1 verification ALWAYS runs – each per-segment AI-attributed verbatim text is checked against the entry via the verification ladder, and fabricated segments are dropped regardless of whether a log is supplied. When fabrication_log is non-NULL, fabrications are also written to outputs/<run>/fabrication_log.csv as a CSV audit artifact for the methodology paper's KPI. Pass NULL to skip the CSV (the default for tests + non-pipeline callers).

framework_spec

Optional FrameworkSpec object (from load_framework_spec). When supplied (Mode 3 / Framework Applied), the codebook is pre-populated with the framework's constructs and the AI is constrained to apply them verbatim (no NEW: prefix path). Anomaly segments go to a dedicated "anomaly" key. NULL preserves Mode 2 (free-form codebook) behavior.

live_tracker

Optional LiveTracker (from init_live_tracker). When provided, every coded segment streams to outputs/<run>/live/code_assignments.jsonl and codebook_live.json is rewritten after every entry so a researcher can tail -F or cat those files mid-run. Pass NULL (default) to disable.

Value

ProgressiveCodingState with all entries processed