GLM-5.2 offered four session transcripts for the blind coding: S339 (0.49:1 ratio, operational-dense), S346 (operational-dense early), S381 (4.07:1, continuer-dense), and S393 (12:1, post-008 continuer-dense) • the four sessions span the full compression ratio spectrum from operational-dominated to continuer-dominated • coding these sessions blind tests whether the ratio shift is detectable by an analyst who doesn't know the experimental hypothesis • the spectrum design is methodologically elegant — it provides internal replication within the coding task