Replication Data Package — Liquidation-Aware Market Making Under On-Chain Visibility: Evidence from the Hyperliquid October 10, 2025 Liquidation Cascade (v3.0)

Replication data, scripts, and outputs for a doctoral dissertation studying market-maker responses to the Hyperliquid October 10, 2025 liquidation cascade (9 perpetual contracts, 9-day minute-level panel, N = 116,030 coin-minute observations). Headline empirical finding: a liquidation–spread sensitivity NEUTRALIZATION effect. The normally positive marginal pass-through of liquidation volume to spreads (β_base = +0.3052) is offset almost one-for-one during the event window (β_event = −0.2920, p < 0.001), leaving a residual marginal effect of ~4% of baseline (neutralization ratio 0.04, range 0.04–0.07 across five specifications) — neither amplification nor sign reversal. The absorption structure is directly measurable: protocol-internal mechanisms (HLP vault 35.5% + auto-deleveraging 33.1%) absorb 68.6% of liquidation notional; ADL closes $2.1bn within a single 12-minute burst, externally reconciled with independent research (arXiv:2512.01112) to within 0.11%. Liquidation variables are built exclusively from protocol-native `liquidation` tags with tid event-level pairing deduplication (channel classification MKT / HLP / ADL). A criterion-evaluation audit (DATA_AUDIT.md) documents that the zero-hash heuristic used in v1.0 measures TWAP algorithmic execution, not liquidations (zero overlap with native tags on 11.23M event-window fills); the v1.0 measure is retained under the accurate name `twap_vol_usd` for disclosure and criterion-contrast replication. All panels, regression outputs, appendix results, and figures are regenerated under the native-tag criterion. An automated integrity gate (`step0_data_integrity_check.py`, 25 assertions) precedes all pipeline runs. Contents: main coin-minute panel (parquet, 116,030 × 60), authoritative per-event liquidation notional by channel, replication-event panels (2025-11-21, 2026-01-31), full script pipeline, all regression outputs (JSON/CSV) backing every table in the dissertation, and figure source files. See README.md for reproduction order. Raw data: Hyperliquid official S3 archive, curated snapshot at DOI 10.5281/zenodo.18759046.