Measuring Optimization Depth: A Pre-Registered Instrument for the Stated-Actual Optimization Gap in Organizational Self-Description (PRISM-O)

Organizations improve themselves at different depths — economic outputs, coordination structure, activity chains, delivered value — and a long literature holds both that deeper interventions are the durable ones and that organizational talk decouples from organizational action. Neither claim has a general-purpose measurement instrument. This paper introduces PRISM-O, a pre-registered instrument in which large language model operators classify every improvement intervention in a frozen two-channel panel of public artifacts — strategy prose versus reported concrete interventions — onto a four-rung optimization-depth ladder under a fixed mechanical rubric, reporting the stated-actual optimization gap only beyond twice the cross-family operator noise floor. Constructed-ground-truth validation is a ceiling result (macro accuracy 1.000, identity masking inert). On a pinned 40-organization filing panel, the floors speak: 4 of 35 organizations resolve, so by pre-registered rule the decoupling hypotheses demote to descriptive reporting — pooled depth distributions separate in the predicted stated-deeper direction (p < .001) but the permutation falsification bar is not cleared. Nearly half of segmented improvement spans specify nothing codable, more in strategy prose than in enacted reports. The instrument, rubric, bank, logs, and estimator publish for reuse; the instrument measures the public record, never internal ground truth. Includes zharnikov-2026bd-measuring-optimization-depth.yaml (Paper Spec v0.1.0) – a machine-readable specification of the paper's claims, assumptions, and dependencies. The paper's full machine-first bundle (the SPINE claim/dependency graph and the ONTOLOGY term module) lives in the public repository; see https://github.com/spectralbranding/paper-spec for the standard. This PDF is generated programmatically from that machine-first source under a research-as-repository model.