humarb Research 08Methodological and descriptive empirical preprint · Prepublication

Instrument Equivalence Across Perpetual-Futures Venues

Evidence Levels, Unit Normalization, and Reference Prices

Download paper Read the abstract
Perpetual futuresInstrument equivalenceContract unitsReference pricesPre-IPO marketsMarket microstructure
Abstract157 words

Equivalence is an evidence decision, not a string match.

Perpetual futures with similar names need not represent the same economic claim. We propose an equivalence framework that separates seven evidence levels, normalizes contract units, and classifies pairs as exactly convertible, reference-basis pairs, or hold-outs. Using official public snapshots, we enumerate 523 Hyperliquid metadata entries across the default and ten named perp DEXs on 20 September 2026, of which 325 were not delisted. For five pre-IPO companies on 14 October 2026, canonical implied equity values show that OKX's native-price ratios of 0.1011 and 0.1031 become residuals of +1.15% and +3.08% after its documented tenfold adjustment. Across 26 eligible exact venue observations, the largest remaining absolute residual is 9.53%, indicating reference-price differences rather than a common unit error. Gate KIMI remains unclassified because its stated and example-implied share counts conflict. A BingX documentation example separately shows a 1,000-fold difference between ordinary and Coin quantity fields. The snapshots do not establish account eligibility, executable size, convergence, or net returns.

阅读中文摘要

不同交易场所中名称相同或相近的永续合约，未必代表同一经济请求权。本文将跨所合约证据划分为七个等级，再将合约分为精确换算、参考价基差配对与暂缓三类。2026 年 9 月 20 日快照覆盖 Hyperliquid 默认及十个命名永续 DEX，共 523 条元数据，其中 325 条未标记退市；这些条目不是唯一标的或可执行套利对。对五家公司 Pre-IPO 永续，本文以原生价格乘假定股数及报价币换算，计算十亿美元计的隐含股权价值。OKX 两组原生价格比在官方十倍重定基后，残差分别为 +1.147811% 与 +3.084383%，说明原始差距主要是单位差。26 条有效精确观测中，最大剩余绝对残差为 9.530398%，反映参考价构造差异。Gate KIMI 的声明股数与示例反推股数冲突，故维持暂缓映射。BingX 官方示例的普通数量与 Coin 数量相差 1,000 倍。稳健性检查支持单位解释和分类方向，但不证明长期持续性、均值回复、收敛、收益或可成交性。本文由 humarb Research 完成，不构成投资建议。

05Reproduced SVG figures
26Eligible exact venue observations
18DOI-checked academic references
00Claimed forecasts or trading returns
Paper contents

I · Research question

A familiar ticker is not sufficient evidence of the same claim.

Cross-venue comparisons often align two symbols, subtract native prices, and call the result a basis. That operation suppresses separate questions about listing, lifecycle, account permission, economic identity, unit definitions, reference construction, and executable size. A failure at any stage changes the meaning of the spread.

This paper asks what evidence is required before two perpetual futures can be treated as equivalent. It develops seven evidence levels, a canonical unit for pre-IPO valuations, and three classifications: exact conversion, reference-basis pair, and hold-out. A public order book establishes neither account eligibility nor economic equivalence; a tight price ratio establishes neither documented units nor common settlement.

Research 01 and 02 measure net carry after identity controls. This paper supplies the upstream identity and unit layer; it does not repeat their funding, cost, or return estimates.

Read Research 01 Read Research 02

II · Data and method

Normalize implied value before interpreting a residual.

The historical evidence comprises a 20 September 2026 Hyperliquid coverage snapshot, exact and range-preserved pre-IPO observations from 14 October 2026, and BingX's official SHIB-USDT documentation example. Delisted lines, conflicts, unavailable interfaces, and range-only summaries remain visible rather than being silently deleted or disaggregated.

Canonical implied equity value · USD billions
Vi = Pi Ni109 Xi

For two native pricing units without an additional order-contract multiplier, the reference-share hedge ratio is qB = qA(NB/NA). Thus one Binance OPENAI unit corresponds to ten OKX units after the rebase, before residual basis. Actual order sizing must separately check contract multipliers, minimum quantities, rounding, and permissions.

I

Evidence before comparison

Documents, metadata, active status, two-sided books, account permission, economic equivalence, and executable size are recorded as distinct levels with timestamps.

No promotion
II

Unit before residual

Documented share counts determine canonical value. USDT, USDC, and USD1 follow the source convention of one USD for descriptive comparison; stablecoin basis remains a limitation.

Value neutral
III

Classification before interpretation

Shared references enter exact-conversion review; independent references become reference-basis pairs; conflicts, unavailable evidence, and negative controls remain hold-outs.

No forced pair

III · Main results

Unit conversion explains scale; it does not erase reference basis.

01 / 05
Figure 1

Seven evidence levels for cross-venue comparison. No lower level implies a higher one. Source: methodological construction; public snapshots do not establish account permission or executable size.

Figure 2

Native OKX/Binance ratios before and after the documented factor of ten. Units: percent. Source: official public marks, 2026-10-14 08:33–08:42 UTC.

Figure 3

Canonical-value residuals against stated anchors, in percent. Gate KIMI, pseudo-anchors, delisted lines, and range-only summaries are excluded. Source: exact public marks, 2026-10-14.

Figure 4

Gate KIMI normalization under stated and worked-example share counts. Units: implied equity value in USD billions. Source: Gate announcement 100758 and 2026-10-14 marks.

Figure 5

Ordinary and Coin quantities at five official SHIB-USDT bid levels, logarithmic quantity axis. Source: BingX official API documentation; embedded example, not a live quote.

Scroll horizontally or use the arrows. Select a figure to inspect its details.

IV · Evaluation and robustness

Sensitivity changes residual size, not the evidence rules.

Four validation families were executed: early-versus-late windows, venue and unit-evidence grouping, company hold-outs and share-count sensitivity, and alternative median anchors with log residuals. All variants are disclosed; none was selected to optimize a return.

Same-day window check

All six rounded early ratios lie on the same side of one as their exact later counterparts. This is repetition within one day, not long-run persistence or convergence.

Venue and company groups

Documented-unit observations: 18, median absolute residual 0.886410%. Price-consensus observations: eight, 0.754531%. Removing POLYMARKET reduces the maximum from 9.530398% to Gate OURA's 5.050092%.

Share-count sensitivity

Gate KIMI produces +15.966637% at ten billion shares and −0.381780% at 8.59024827 billion. The best-fitting factor is not selected; the conflict remains a hold-out.

Alternative anchors and log form

OKX median-anchor residuals are +0.844275% and +0.922550%. Levels change, while unit interpretation, reference classification, and hold-outs remain intact.

No predictive or return estimand is manufactured.

There is no forecast for an out-of-sample test, no model family for a Reality Check, and no realized execution for cost-adjusted P&L. Assigning p-values to two same-day windows would overstate the evidence.

V · Evidence boundary

What the paper does not claim.

  • Not 325 unique or executable assets. The Hyperliquid totals are metadata and non-delisted entries, not unique underlyings, approved pairs, or account-eligible markets.
  • Not account permission or execution. No account was connected and no order was placed. A two-sided public book is not a fill guarantee.
  • Not convergence or long-run persistence. Two same-day pre-IPO windows cannot establish stationarity, mean reversion, half-life, or a common terminal payoff.
  • Not statistical significance or investment returns. No confidence interval, annualized income, arbitrage feasibility, or cost-adjusted return is reported.
  • Not universal unit semantics. Price-consensus factors are not documentation; BingX's official example is not generalized to every symbol or API version.
  • Not full raw-response authentication. Calculations reproduce from preserved observations, but the task package did not include every historical response byte.

USDT, USDC, and USD1 follow the source convention of one USD for descriptive normalization. Quote-currency, collateral, stablecoin, tax, legal, and jurisdictional risks require separate review.

VI · Limitations and conclusion

Unresolved evidence stays unresolved.

Historical raw responses are incomplete, several venues were inaccessible during recheck, anchors and marks are not synchronized executable quotes, and official specifications can change after a snapshot. Bybit coverage is unknown rather than empty. Gate KIMI's documentary conflict requires a venue clarification or a later formal notice, not a better-fitting estimate.

The next research step is a versioned multi-day panel with raw-response hashes, synchronized books, delisted and unavailable states, and complete quote and collateral metadata. Only then should basis dynamics, rolling out-of-sample tests, dependence-aware inference, multiple-testing control, and complete execution-cost accounting be attempted.

Verify identity first, normalize units second, classify reference differences third, and leave unsupported equivalence unresolved.

VII · Downloads and citation

Cite this preprint accurately.

Cite the concept DOI 10.5281/zenodo.23931254; the version DOI 10.5281/zenodo.23931255 identifies v1.

Concept DOI10.5281/zenodo.23931254Version DOI · 10.5281/zenodo.23931255  |  Record · zenodo.org/records/23931255  |  Status · preprint, not peer reviewed
English PDF 中文摘要 BibTeX metadata.json LaTeX source Reproduction package

Liu, H. (2026). Instrument equivalence across perpetual-futures venues: Evidence levels, unit normalization, and reference prices [Prepublication author manuscript]. humarb Research. https://humarb.ai/research/instrument-equivalence/

Figure