01 / ABSTRACT
Comparability is a measurement assumption, not a property of the field name.
Cross-venue perpetual-futures analysis often treats a funding rate as a directly comparable scalar. We test that premise using official documentation and read-only public interfaces for 21 centralized and on-chain venues. We construct a dated data dictionary covering funding formulas, settlement schedules, caps, field semantics, clock and rate units, open-interest and turnover bases, fees, liquidation backstops, and pre-trade rules.
Fifteen venues share a clamped interest-minus-premium core, but three use distinct formula structures and three require separate treatment. Instrument-level intervals invalidate an exchange-wide eight-hour assumption: Gate’s 1,070 rows on 3 November 2026 contain 420 four-hour contracts, while GRVT and Variational also return multiple intervals. A new sample of 1,840 public history rows shows median absolute timestamp offsets from the nearest hour of 36 ms at Hyperliquid, 937 ms at Extended, and 3,680 ms at Pacifica; all 100 edgeX records are exact.
Deterministic examples show 100-fold unit errors, twofold annualization errors, and fee differences exceeding one typical funding payment. The results provide a reproducible normalization layer for funding-rate and execution-cost measurement, not evidence of trading profitability.
中文摘要
跨场所永续合约研究常把资金费率视为可以直接相减的标量，但公开字段是否描述同一经济对象，取决于公式、结算周期、语义阶段、数值单位、事件时间和交易状态。本文依据官方文档与公开只读接口，对10家中心化交易所和11家链上或混合场所进行机制普查，形成包含29个字段的日期化数据字典。
合约级证据否定了全所统一8小时的假设。对Hyperliquid、Pacifica、Extended和edgeX共1,840条公开历史记录的分析显示，时间戳记录精度具有场所差异。确定性示例进一步表明，单位、周期和截断位置误读可造成超过单次典型资金费支付的误差。本文提出可复现的输入审计层，不评估交易盈利能力，也不构成投资建议。
下载一页中文摘要 PDF02 / PRINCIPAL FINDINGS
The same label can conceal different economic objects.
Mechanism before arithmetic
A current forecast, last-settled rate, trailing realized rate, and current accrual are not interchangeable phases. Outer funding caps and inner premium clamps must also remain separate.
Intervals are instrument-level
Gate’s repeated snapshots contain eight-hour, four-hour, and one-hour rows. GRVT and Variational independently confirm mixed schedules; an exchange-wide default is not a safe normalization rule.
Deterministic errors do not average away
Reading percentage points as decimals creates a 100-fold error. Reading a four-hour rate as eight-hour halves the daily estimate. Higher-frequency collection cannot repair a wrong unit or phase.
The estimand is public record-timestamp offset—not internal settlement latency, cash availability, execution quality, or realized P&L.
03 / EVIDENCE & COVERAGE
A dated evidence layer with explicit failure semantics.
Baseline materials were collected on 19–20 October 2026. The 3 November refresh saved complete raw JSON for 11 venues and public web samples for Bybit, MEXC, and Deribit. Seven venues were not refreshed because of network errors, access restrictions, timeouts, connection resets, or non-JSON responses. Refresh failure is not evidence that a rule is absent.
| Evidence layer | Coverage | Interpretation |
|---|---|---|
| Mechanism dictionary | 21 venues · 29 fields | Dated public rules, not permanent parameters |
| Complete refresh JSON | 11 venues | Raw responses plus UTC manifests |
| Additional web samples | 3 venues | Targeted public corroboration |
| Settlement histories | 4 venues · 1,840 rows | Record timestamp analysis only |
| Refresh failures | 7 venues | Reported explicitly; never treated as missing rules |
All rates, limits, fees, and statuses must be rechecked against venue pages on the day of use. Reproduction scripts make only low-frequency read-only requests; users must review current venue terms.
04 / FIGURES
Normalization, interval heterogeneity, and error magnitudes.
Select any figure to open the vector original. Illustrative error amounts are deterministic examples, not return forecasts.
05 / NORMALIZATION DICTIONARY
Six fields required before two rates can be compared.
The normalization tuple is Mvit = (Svit, Uvit, hvit, τvit, Qvit, Zvit). It records semantic phase, numeric unit, interval, event time, quantity basis, and trading state for each venue, instrument, and timestamp.
- SSemantic phase
Current forecast, last settled, trailing realized, or current accrual.
- UUnits and formula basis
Decimal versus percentage points; rate formula, inner clamp, outer cap, and sign convention.
- hSettlement interval
Instrument-level hours, including cap acceleration and continuous-accrual exceptions.
- τEvent and record time
Seconds, milliseconds, microseconds, or nanoseconds, with source timestamp meaning preserved.
- QQuantity and cash-flow basis
Position multiplier, settlement-price basis, open-interest and turnover conventions.
- ZTradability state
Market status, trading hours, RFQ or indicative books, transfer state, and pre-trade constraints.
The reproduction package contains the complete 21-row venue dictionary in CSV and JSON, all 29 field definitions, raw public responses, UTC manifests, analysis outputs, and scripts.
06 / ROBUSTNESS & LIMITATIONS
Robust to alternative coding; bounded by public evidence.
Robustness checks
- Repeated-date Gate instrument distributions.
- Venue-group comparisons and alternative Variational taxonomy coding.
- Formula, cap-position, interval, fee, and unit sensitivity calculations.
- Alternative timestamp summaries and explicit venue coverage.
Evidence boundaries
- No trading-profitability or guaranteed-return claim.
- No statistical-significance or forecasting claim.
- No account-level fees, private endpoints, orders, or executable books.
- Public snapshots can change and do not establish permanent venue rules.
No historical executable orders or account-level fees were observed. Deterministic examples are not return forecasts.
07 / HUMARB RESEARCH SERIES
The antecedent input audit for net-carry research.
Research 01 and 02 address executable or framed net carry once inputs are comparable. Research 03 asks whether venue-native fields represent comparable inputs at all.
08 / DOWNLOADS & CITATION
Paper, source, evidence, and reproducible analysis.
@misc{liu2026crossvenue,
author = {Liu, Hang},
title = {{A Cross-Venue Data Dictionary for Perpetual Futures: Funding-Rate Mechanisms, Pre-Trade Rules, and Measurement Error Across 21 Exchanges}},
year = {2026},
month = nov,
publisher = {humarb Research},
howpublished = {humarb Research 03, Methodological preprint},
doi = {10.5281/zenodo.23916662},
url = {https://doi.org/10.5281/zenodo.23916662}
}