Skip to contents

Assembles, from the objects themselves, the paragraph and tables that LLM annotation studies should report and rarely do: the instrument (codebook name, version, hash), the protocol (model, parameters, prompt hash), the validation result with its confidence interval, per-category performance, and the gold set's complete test-split ledger – every evaluation that ever touched the sealed split, not just the flattering one.

Usage

coding_report(validation, gold, protocol)

Arguments

validation

A validate_protocol() result.

gold

The gold_set() used (for the ledger and split sizes).

protocol

The locked protocol() (for instrument identifiers).

Value

A character vector of report lines (class coding_report), with a print method. Paste into the appendix and edit; this is a draft whose numbers are right, not finished prose.