Replication Materials for "Architecture-Dependent Behavioral Diagnosticity: Focal Leverage and Event-Path Responses in a Liquidation Cascade"

This repository contains the replication materials for the manuscript: “Architecture-Dependent Behavioral Diagnosticity: Focal Leverage and Event-Path Responses in a Liquidation Cascade.” The study examines whether the predictive meaning of the same observable leverage choice depends on the institutional architecture that maps the choice into collateral exposure. Using the complete eligible population of Hyperliquid wallets and positions reconstructed around the October 10–11, 2025 liquidation cascade, the analysis distinguishes cross-margin and isolated-margin architectures and evaluates focal leverage choice, acute competing responses, and post-liquidation re-entry. The repository documents the final unified analysis pipeline and contains: the frozen unified analysis specification; the final sample-flow, model, and multiple-testing registries; Stage 1 code and aggregate results for wallet construction, liquidation reconstruction, acute-response classification, first re-entry, and ordered subsequent outcomes; Stage 2 code and aggregate results for wallet-level and exact-position focal-diagnosticity models; Stage 2.2 robustness and proportional-hazards diagnostic outputs; Stage 2.3 minute-60 multinomial and late-survivor model outputs; validation reports, run logs, input checksums, and SHA-256 file checksums; source files for the manuscript tables and figures; a public data dictionary, replication-file manifest, and raw-data access instructions. The repository supports the reported sample construction, main tables, figures, robustness analyses, and Online Appendix results. The final common master sample contains 54,551 wallets after applying the unified position-validity rules. The acute-response analysis contains 54,332 eligible wallets. Raw wallet-level archives and wallet addresses are not redistributed in this deposit. The underlying Hyperliquid archives are publicly accessible through requester-pays object storage and official node-fill archives. Detailed access and reconstruction instructions are provided in CURRENT_JOEP_raw_data_access_instructions_2026-07-26.md. The materials report conditional predictive associations and do not identify causal effects or traders’ underlying psychological motives. Wallets may represent individuals, institutions, or automated strategies. Version 2.0.0 replaces the earlier, narrower replication release with the frozen unified analysis used for the integrated manuscript. This version harmonizes the sample, variable definitions, outcome reconstruction, statistical models, multiple-testing families, tables, figures, and validation records. For file-level descriptions and the recommended reproduction sequence, see: README_JOEP_REPLICATION_2026-07-26.md CURRENT_JOEP_replication_file_manifest_2026-07-26.csv CURRENT_JOEP_public_data_dictionary_2026-07-26.csv SHA256SUMS.txt Author: Kyoungyoung JangDepartment of Consumer Science, Inha UniversityORCID: 0009-0003-9743-8657