Headline Finding

Carbon's turnover assumption isn't a guess. The simulator's Base default of 0.75 matches Ostium's 90-day median (0.755) — the closest RWA-native comparable — and sits just below Variational's 1.03 (RFQ-based RWA perp). The cross-peer median of 1.39 is pulled up by reward-incentivized CLOBs (Lighter 2.75, dYdX v4 2.55) that don't represent organic RWA flow.

No simulator default needs material adjustment. The model is well-calibrated.

Methodology

Six protocols selected to span the relevant comparison axes for Carbon (CFD-style perp DEX, RWA focus, RFQ + bilateral hedging architecture):

• Hyperliquid, Lighter, dYdX v4 — high-volume crypto perp DEXes (active-trader benchmark)
• Variational — RFQ + OLP architecture (RFQ-based RWA comparable)
• Ostium — RWA-native, trades commodities/FX (Carbon's market-fit twin)
• GMX v2 — LP-pool reference (vault-economics comparable)

Three primary metrics requested: Open Interest (daily series), Daily Volume (series), Turnover Ratio (volume/OI computed daily). Liquidations attempted as stretch — DeFiLlama doesn't expose perp liquidation history, so that metric is unavailable from this dataset.

Data quality guardrails applied: known incentive programs flagged (HL, Lighter, dYdX v4, GMX v2 had active rewards during the window); only total volume available (not organic) — treat all turnover ratios as upper bounds on organic flow.

Open Interest — 90-Day Daily Series

Hyperliquid dominates absolute OI ($6.83B 90-day median, left axis). Lighter and Variational cluster around $700M (right axis). Ostium runs at ~$225M but declining through the window. dYdX v4 and GMX v2 sit near the floor.

Turnover Ratio — The Headline Number

Daily volume divided by daily OI. The single most-important calibration number for the simulator.

Calibration to Carbon's Defaults

The right reference for Carbon is the RWA-native cluster: Variational + Ostium midpoint = 0.89. Carbon's stage-decay defaults sit in or just below this band:

Why we're conservative on Base. Pre-launch product — the model should under-promise. Once we have 90 days of our own data we'll calibrate up if observed turnover supports it.

Why we ignore the cross-peer median (1.39). It's contaminated by Lighter and dYdX v4 — both CLOB venues with reward-incentivized scalper cohorts that don't represent organic RWA flow. Anchoring to the wrong reference would inflate our model.

Caveats and limitations

• No organic-volume distinction. DeFiLlama does not separate organic from total volume for perp derivatives. All volume figures (and therefore all turnover ratios) are upper bounds on organic flow.
• Reward-program contamination in 4 of 6 protocols. dYdX v4 (DYDX trading rewards), GMX v2 (ARB incentives), Lighter (points program suspected). Hyperliquid 2026 reward status not confirmed by DeFiLlama. This pushes turnover ratios up — read all numbers as ceiling rather than floor.
• Variational and Ostium are relatively young. Their turnover may not yet reflect mature-market behavior. Best-available anchors but numbers may evolve.
• Liquidations data gap. DeFiLlama doesn't expose perp liquidation history. Our 3.5%/mo liquidation-rate assumption remains uncalibrated. Direct on-chain log queries per protocol is the next step (queued).
• 90 days is a short window. Single-event volatility (e.g., the Feb 5 spike on Hyperliquid where one day showed 4.75× turnover) can move medians materially. Quarterly refresh is the mitigation.

Citation

Carbon Labs, "90-Day Perp DEX Benchmark," 2026-04-25. Source: DeFiLlama AI Terminal. Comparable protocols: Hyperliquid, Lighter, dYdX v4, Variational, Ostium, GMX v2. Window: 90 days through 2026-04-25, daily granularity.

Refresh cadence: Quarterly, or sooner if a new comparable launches (e.g., another RWA-native perp DEX) or major reward programs at Lighter/dYdX/HL end.

Conventions

Core Equations

Parameters — Full Reference

Same content as the info-icon tooltips on the Simulator tab, laid out in one place.

Preset Scenarios

Key Assumptions & Caveats

• Volume counts one leg, not both. Daily volume = closes (= opens in balanced book). Applying spread markup once matches the CFD industry "round-trip spread" convention.
• Overnight markup is annualized. The blended 4% already accounts for the mix of 252 trading days (stocks) and 365 days (FX carry). Don't re-annualize.
• Liquidation capture assumes P(gap|liq) ≈ 30%. This is an estimate — calibrate with 30 days of operating data. If most liquidations are intraday, capture drops toward 0.
• Commission is a placeholder pending VitaVerse fee schedule confirmation (§13 open items in operating rules).
• Tail reserve and insurance default to 0 — they're soft costs analyzed separately. Toggle sliders up to model them.
• Demand side is independent of vault size. Util + turnover describe the absolute book activity. Shrinking vault concentrates the same absolute demand into higher effective utilization — see "Bear deallocation" strategy.
• Front-end trading fee is separate. Carbon charges an additional 1–2 bps at the front end (goes to treasury, not CLP). Users see total cost = solver spread + FE fee.

Methodology & sources

• Solver revenue model — full mathematical derivation including spread revenue, overnight rate revenue, liquidation capture, and cost stack. Available on request.
• Solver operating rules — capital allocation (30/50/20 split), circuit breakers, per-symbol parameter table. Available on request.
• RWA risk analysis — gap-risk distribution, volatility regimes, leverage calibration per asset class. Available on request.
• Empirical validation — turnover ratio benchmarked against 6 comparable on-chain perp DEXes (Hyperliquid, Lighter, dYdX v4, Variational, Ostium, GMX v2) over 90 days. See the Validation tab above for the full analysis, charts, and calibration vs Carbon defaults.