AI as second coder - never the author | DATASPHERES AI - Dataspheres AI

Supports Module 4 objectives; aligned to CLO 4. AI-assisted coding proposes segments against the study codebook at corpus scale. Three conditions make the ...

Supports Module 4 objectives; aligned to CLO 4. AI-assisted coding proposes segments against the study codebook at corpus scale. Three conditions make the practice methodologically defensible: Every application stores the exact quote, character offsets, memo, and confidence value, and is individually reviewable. Machine-proposed codes receive intercoder review — accepted, refined, or rejected — with the same standard applied to any new coder. Applications from low-fidelity sources carry capped confidence values and a memo prefix identifying the source quality. Intercoder reliability from the sample study - two blind coders vs the original, Cohen’s kappa per dimension. The sample study practices this rule on itself: 53 segments were double-coded by two blind coders (kappa 0.885 theme / 0.909 stance), the disagreements caught the original coder drifting, and the codebook was amended . Full analysis: Intercoder Reliability .