VIX-family maths,
every step open.

VolX adopts the Cboe VIX-family methodology — variance swap replication via a strip of out-of-the-money options — formalised in Cboe's 2003 white paper. Same maths family as the equity VIX, Deribit DVOL, and Volmex BVIV.

What VolX computes per 60-second tick:

• → select two expiries bracketing 30 days (near + next)
• → solve the forward via put-call parity at the ATM strike
• → fit the IV surface as a natural cubic spline in log-moneyness
• → apply the Carr-Madan variance integral
• → interpolate in total-variance space to the exact 30-day point

σ²_T  =  (2 e^{rT} / T) · ∫ Q(K) / K² dK  −  (F / K₀ − 1)² / T

The variance integral runs independently per venue: one strip for Deribit, one for OKX, one for Bybit. Each venue produces its own σ²₃₀d number reflecting only its own order book.

A venue's strip is rejected if it has fewer than 5 strikes with both call and put legs quoted > 0.000000001 USD. Rejected venues are excluded from that 60-second tick and contribute to the confidence score collapse.

The three venue values are combined by taking the median, not the mean. A single bad venue cannot drag the median around the way it could the mean.

BVOL  =  100 · √median(σ²_dervt,  σ²_okx,  σ²_bybit)

For two surviving venues the median collapses to the simple mean of the two. For one surviving venue the value passes through unchanged. The published value is then 100 · √σ²₃₀d — annualised vol percent.

If a venue's σ²₃₀d deviates from the cross-venue median by more than 5% for 5 consecutive ticks (5 minutes), the engine drops that venue from the blend until its quotes return to consensus.

Availability rollback: if the drop would leave the active set empty (e.g. all three venues simultaneously diverging), the policy reverts — every venue is kept active and the confidence score absorbs the degraded state. The system never publishes a null for a transient quorum collapse.

Why three venues, not two: Volmex BVIV blends two venues (Deribit + OKX). With only two sources no real outlier policy is possible — drop one and you collapse to a single-venue index. VolX's third venue (Bybit) is specifically what makes the 5%/5-tick drop policy viable.

Every published index tick carries a [0.0, 1.0] confidence value computed from three multiplied factors:

confidence  =  venue_factor  ×  freshness_factor  ×  strike_factor

Multiplicative so two simultaneous degradations compound rather than mask each other. A perfect snapshot (3/3 venues, all quotes fresh, ≥ 8 strikes) yields confidence = 1.0. Downstream consumers can filter, gate, or weight by the score.

Two layers of filtering protect the index from contaminated input.

Normalizer layer (per quote)

• · Drop if last tick > 5 s old
• · Drop if ask ≤ bid
• · Drop if (ask − bid) / mid > 0.30
• · Drop if mid < intrinsic value (1e-9 tolerance)

Engine layer (per snapshot)

• · Drop strike if mark_iv ≤ 0.001 or non-finite
• · Drop strike if either call or put leg is missing for that (strike, expiry)
• · Reject the expiry if fewer than 5 strikes survive

DVOL is the closest published reference: same VIX-family methodology, but on a single-venue (Deribit-only) input universe. Restricting VolX to Deribit-only input and running the engine should reproduce DVOL within numerical noise — the canonical correctness check.

Acceptance bar per the PRD: |VolX − DVOL| / DVOL ≤ 2% sustained over 30 consecutive days. Until that window passes, the index is shipped but not yet validated.