Skip to contents

Record one (entry, code, segment) assignment to the live tracker

Source: R/live_tracking.R
live_record_assignment.Rd

Appends a JSONL line to code_assignments.jsonl. Safe to call with tracker = NULL (no-op).

Usage

live_record_assignment(
  tracker,
  entry_id,
  code_key,
  code_name,
  segment,
  is_new_code = FALSE,
  entry_index = NA_integer_
)

Arguments

tracker

A LiveTracker or NULL

entry_id

Character entry id (std_id)

code_key

Codebook key (sanitized id)

code_name

Human-readable code name

segment

List with text, start_char, end_char

is_new_code

Logical; whether this assignment created the code

entry_index

Integer position of the entry in the input (for ordering)

Value

The (possibly updated) tracker, invisibly.

Details

schema relationship to the audit log. code_assignment events live in TWO places by design:

  • outputs/<run>/live/code_assignments.jsonl (this writer): append-only EVENT log. One JSONL line per (entry_id, code_key, segment) triple. Carries text + offsets + verification_status inline so a researcher can tail -F the file mid-run.

  • outputs/<run>/ai_decisions.jsonl (R/audit_log.R): append-only DECISION log. One JSONL line per audit event; step = "coding" + decision_type = "code_assignment" records the same logical event. Carries methodology stamp + AI metadata (model, request_id, methodology_mode) but NOT segment text or offsets.

The two are joined by (entry_id, code_key). The split is intentional: the audit log is the methodology-of-record (provenance, AC9 stamping, replay) while the live tracker is the researcher's real-time view (text + offsets + verification status for tail -F). An earlier docstring was silent on this; downstream consumers had to infer the join from the field names. Now documented.