Liquidation OI Reset & Rebuild

Descriptive filters. Completed UTC hours in the preceding 72 completed hours, separately by exact market ID. At least $100K total recorded long plus short liquidation notional. Rank the largest ten overall by notional, then earliest hour, then contract. These are observed-heavy hours, not inferred clusters or all-market cascades.

Snapshot formulas. U = OI in USD / mark price on the same recorded market snapshot, a contract-equivalent base-asset exposure measure. For hour H, U_before uses H−5m; U_after uses H+55m; U_1h uses H+115m; U_4h uses H+295m. These buckets immediately precede H, hour end, +1h and +4h boundaries. Event-hour change = 100 × (U_after / U_before − 1). Only a reset of at least 1% (U_after ≤ 0.99 × U_before) qualifies: rebuilt fraction = 100 × (U_follow − U_after) / (U_before − U_after), separately at +1h and +4h. Negative values mean further contraction; above 100% means above pre-event level.

Availability. Future boundaries are Pending, even if a bucket already exists. Mature missing or invalid exact anchors are Unavailable. OI and marks must be finite and positive; no nearby bucket, price fallback or other venue fills a gap. Incomplete event-hour minute-bar recording invalidates comparisons. With no qualifying reset, no rebuild fraction is calculated.

Recorded observations only. Snapshot bucket times identify engine snapshots, not each field’s original exchange publication time. They do not prove intrabucket synchronization or source freshness. Liquidation amounts are recorded public-feed observations, not complete market totals. Rebuilt OI includes both sides; it does not establish that liquidated positions reopened, leverage itself returned, or price must reverse. This is retrospective context, not a predictive cascade-risk score.