BL-SD-038: Fragility-Conditioned Volatility Risk Premium

Status: RELEASED

Classification: PUBLIC

Revision: 1

Date: 2026-09-08

Venue: Deribit options, Bybit perpetual hedge Programme: II Verdict: CLOSED Axis of death: Capital, venue Transferability: ENVELOPE-BOUND

1. Identification

H-V1 is recorded here with a procedural irregularity stated plainly rather than smoothed over: the independent audit of 2026-07-30 notes that H-V1 drew against the programme's twelve-hypothesis budget without the dedicated conditioning instrument its own pre-registration specified ever being built or run. What closed the door instead was a capital-scale finding produced by the Layer-0 and Layer-5 mechanism checks that preceded pre-registration. Both facts are on the record, and the second is why the verdict stands regardless of the first.

2. Economic thesis

An option seller is paid because they carry risk the buyer wants to shed. In a market where implied volatility runs persistently above realised volatility, the average premium collected is positive by construction: buyers pay for insurance, and someone has to be willing to underwrite it. That premium is not itself an edge. It is compensation for tail risk, and it is paid out precisely when volatility expands and the underwriter needs it most.

H-V1's claim is narrower and falsifiable in a way passive harvesting is not: that the programme's own fragility predictor, already shown to separate calm from turbulent regimes at high statistical confidence, is sharp enough to avoid the specific episodes where the premium turns negative, converting a passive risk premium into something that clears the mandate's definition of alpha, behaviour not fully priced by the market.

Who is forced to lose. Leveraged position holders who must close as volatility spikes, and insurance buyers who purchase protection only after the risk has already become visible, paying a price that has already moved against them. Why structural. The insurance premium must be positive on average or no one would sell it, and the information behind it survives specifically because it cannot be monetised without a convex instrument; nobody can arbitrage away a signal they have no venue to express.

3. Falsifiable claim

Pre-registered before any conditioning instrument was written:

A delta-hedged short straddle on 30-day at-the-money BTC options, conditioned on a binary fragility gate, produces a higher mean PnL per position than the same construction without conditioning, measured against a pessimistic baseline fixed in advance.

The pessimistic baseline was fixed before any result was seen: 3-day rebalance rather than daily, a 1.5 vol-point implied-volatility haircut for selling at bid rather than mid, 5.0 percent of premium for entry cost rather than the 2.5 percent measured, 30-day at-the-money tenor, BTC only, ETH rejected in advance on grounds given in section 6. The rule exists because three amendments to a previous hypothesis in this programme demonstrated that a favourable assumption will always find a way in if it is not locked before results are seen.

4. Mathematical specification

Notation per BL-D-011.
 
Volatility risk premium, characterisation, no position implied:
 
  VRP[t] = IV_30d[t] - RV_30d_forward[t]
           IV from Deribit DVOL index, RV realised on the programme's own
           kline lake, forward-looking over the same 30-day window IV prices
 
Straddle payoff, delta-hedged, exact path simulation (not the IV-RV
approximation, which is valid only for a hedged position and was corrected
mid-programme, section 6):
 
  PnL_hedged[pos] = sum over rebalance steps j of
                     [ -delta_j * (S_{j+1} - S_j) ]     (hedge leg)
                   + [ premium - max(0, S_T - K) - max(0, K - S_T) ]  (option leg, short)
                   - hedge_fee - entry_cost - iv_haircut_effect
 
  delta_j          Black-Scholes delta at rebalance step j, rebalance
                   frequency a tested parameter (daily baseline, section 6)
  entry_cost       fraction of premium, 2.5 pct measured, 5.0 pct pessimistic
  iv_haircut       vol points subtracted from mid IV to model selling at bid,
                   0 optimistic, 1.0 to 1.5 pessimistic baseline
  hedge_fee        5.5 bps per Bybit perp fill
 
signal:  binary fragility gate, BSS-calibrated risk gate, PASS or FAIL only.
         H-V1 is prohibited from using a continuous gate score (rule set out
         in the pre-registration's data gate).
entry:   sell 30-day ATM BTC straddle on Deribit when gate = PASS
         (conditioned arm); unconditioned arm sells every eligible window
         regardless of gate state, as the comparison population.
exit:    hold to expiry, delta-hedged along the way per the rebalance
         schedule; no early close, no bracket.
sizing:  one contract per eligible window, non-overlapping positions only,
         for inference validity.
cost:    entry_cost + iv_haircut_effect + hedge_fee, per the pessimistic
         baseline in section 6 for the binding verdict, optimistic values
         reported only as sensitivity.
 
Test statistic (as pre-registered, never computed in this form):
  delta = mean(PnL_hedged | gate = PASS) - mean(PnL_hedged | all positions)

Causality holds by construction for the VRP characterisation: IV at t is an observable market quote at t, and RV is computed only over the forward window the IV price already spans, which is a forward-looking realisation used to grade a forward-looking price, not a leaked input to a decision at t. The hedge simulation uses only price history at or before each rebalance step.

The conditioning statistic delta was never computed. The programme's own audit records this: the R-3 hypothesis budget was charged for H-V1 without a built instrument producing it, and this dossier states that plainly rather than substituting a number that does not exist. What is reported in section 7 is the VRP characterisation and the exact-path hedge simulation that preceded the conditioning test and made the capital question, rather than the conditioning statistic, the one that actually closed the door.

5. Operating envelope

The venue itself is the first-order constraint stated in this table rather than buried in prose: Deribit is not a venue this operator's declared access reaches (BL-FM-002 section 7.1). Every number in section 7 was measured against live Deribit quotes and the programme's own price history, which establishes that the mechanism is real; it does not establish that the mechanism is reachable.

6. Instrument

The VRP characterisation and cost floor ran first, ahead of pre-registration, as a Layer-0 check spending no hypothesis budget. It measured IV against forward realised volatility over the full research era on BTC and ETH, and it measured live Deribit fee and spread quotes rather than assuming a schedule.

The exact-path simulation that followed corrected a material error in the Layer-0 model. The cost-floor report approximated hedged straddle PnL as 0.4 * VRP * sqrt(T), which is valid for a delta-hedged position but was mischaracterised in that report as merely optimistic about gamma. The correct diagnosis, stated in the simulation report, is sharper: IV - RV approximates the hedged payoff, not the unhedged one. An unhedged short straddle held to expiry has payoff premium - abs(S_T - K), which depends only on the terminal price and has no path-dependency at all, so a single large terminal move erases many small wins outright. The correction moved the unhedged t-statistic from an apparent +3.1 under the averaging model to +0.25 under exact path simulation, and moved maximum drawdown from an apparent -5.3 percent to a measured -42.9 to -79.8 percent depending on phase offset. ETH was rejected in advance of the pessimistic-baseline testing on the strength of a separate finding: t = -0.12 at realistic cost, with monotonically worsening subperiods.

7. Results

Volatility risk premium, research era, n = 1,277 days:

Exact-path hedge simulation, BTC, 30-day ATM, three phase offsets, n = 42 or 41 per offset:

Hedged subperiod decomposition, offset 0, the diagnostic table for whether the edge is regime-stable: +22.6, +20.9, +20.1 percent across three subperiods, materially more stable than the unhedged path (+1.8, -9.0, +23.8 percent) across the same subperiods, consistent with the mechanism claim that delta-hedging strips directional risk and leaves a cleaner exposure to realised-versus-implied volatility.

Sensitivity to the assumptions the pre-registration required be tested rather than assumed:

Under the fully combined pessimistic baseline the pre-registration commits to as binding, the edge stays positive but is not statistically significant. It does not reverse sign, so the mechanism itself is not falsified by cost; it simply is not strong enough, at the declared conservative assumptions, to carry a verdict of PASS on its own.

The conditioning statistic itself, delta between gated and ungated PnL, was never computed. No table for it exists because no instrument producing it was built.

8. Verdict, axis of death, and transferability

Verdict: CLOSED. Axis: capital, venue. Transferability: ENVELOPE-BOUND.

The underlying mechanism measured real by every check the programme ran against it. The volatility risk premium on BTC is large, persistent, and independently corroborated by the broader options literature, which the Layer-0 report treats as a mark of credibility rather than a discovery. The delta-hedged harvest of that premium clears live Deribit fees and measured spreads with a wide margin under exact-path simulation, and it is stable across three subperiods in a way the unhedged construction is not.

What closes it is not the mechanism. It is that Deribit sits outside this operator's declared venue access (BL-FM-002 section 7.1), that a single BTC option contract carries a minimum notional of 6,345 USD with an unbounded loss profile, and that safely running the construction requires diversification across eight or more concurrent positions in time, which the programme's own estimate places at 50,000 to 100,000 USD of working capital against a declared operating scale of 1,000 to 2,000 USD. The mechanism is not in question. The capital and the venue are, and both are declared parameters of this operator's envelope rather than properties of the market, which is what places the verdict in the ENVELOPE-BOUND class under the test in BL-D-002 section 4.

The procedural irregularity in section 1 is recorded here rather than omitted: the pre-registered conditioning instrument was never built, and the verdict rests on the mechanism and capital findings that preceded and, functionally, superseded it. A later operator who reopens this hypothesis with adequate capital and venue access should treat the conditioning question as still genuinely open, not as answered by this dossier.

9. Cost and capacity analysis

Deribit fee                       3 bps underlying, capped 12.5 pct premium
Entry cost, measured              2.5 pct of premium
Entry cost, pessimistic baseline  5.0 pct of premium
Bybit hedge fee                   5.5 bps per fill
Hedge cost realised, simulation   0.206 pct spot per position
Minimum notional per contract     6,345 USD

Capacity was measured indirectly through the capital requirement rather than through a throughput study: the construction needs 8 or more concurrent, time-diversified positions to bring maximum drawdown from an unhedged tail-risk profile down to a managed one, which at the measured minimum notional sets the floor at 50,000 to 100,000 USD. Below that scale the position count needed for diversification does not fit, and the tail-risk profile is not one this record treats as acceptable at any premium.

10. Envelope sensitivity and reopening conditions

Binding parameters: capital scale and venue access, jointly.

Our capital           : 1,000 to 2,000 USD working
Required capital       : 50,000 to 100,000 USD, for adequate time
                          diversification against a 6,345 USD minimum
                          per-contract notional and an unbounded loss profile
Our venue access        : Bybit perpetuals, Hyperliquid perpetuals
Required venue access   : Deribit or an equivalent options venue, not
                          currently reachable
Reversal condition      : capital at or above roughly 50,000 USD AND
                          options venue access, simultaneously

Neither parameter alone reverses the verdict. Capital without venue access has nowhere to execute the option leg; venue access without capital cannot diversify the tail risk to an acceptable drawdown. Both are declared parameters of this operator's envelope rather than properties of the market, and an operator whose envelope already clears both should treat the mechanism measurement in section 7 as the starting point for their own verdict rather than as something to re-derive from zero.

What would reopen it: capital and venue access crossing the thresholds above, at which point the unresolved question is exactly the one the pre- registration asked and never answered: does conditioning on the fragility gate raise the harvested premium above the unconditioned baseline, net of the pessimistic costs already measured here. That instrument does not exist yet and would need to be built before any capital is committed.

What would not reopen it: a cheaper Deribit fee tier or a better hedge execution alone. Neither touches the binding constraint, which is capital scale for tail diversification, not transaction cost.

11. Reference implementation

The exact-path simulation is published as 05_INSTRUMENTS/strategies/python/src/volatility_premium.py, with 19 tests, on the option-pricing primitives in 05_INSTRUMENTS/strategies/python/src/option_pricing.py, with 10. The conditioning instrument specified in the pre-registration was never written and has no implementation to publish; this is stated in section 4 and repeated here rather than implied by its absence, and the published module records the same absence instead of supplying a substitute.

12. Evidence register

13. Lessons extracted

The theory-level correction in section 6, that IV - RV approximates a hedged payoff and not an unhedged one, is worth carrying forward on its own: an averaging model can overstate an unhedged tail-risk construction by more than an order of magnitude on both the significance statistic and the drawdown, precisely because averaging cannot show what a single terminal move does to a position with no path-dependency. This is now treated as a standing caution against approximating any convex payoff with a linear risk-premium model.

The quota-accounting lapse noted in section 1 is recorded as a process lesson rather than an outcome one. A hypothesis's budget line should reflect an instrument that was actually run. Where a capital or venue question closes a door before the pre-registered instrument is built, the honest record is to say so, not to retrofit a completed test that never happened.

14. Cross-references

BL-SD-037 for the crowding-to-volatility finding that motivated this family. BL-FM-002 sections 6 and 7 for the capital and venue parameters this verdict is bound by. BL-D-002 section 4 for the ENVELOPE-BOUND classification test. BL-SD-000 ledger row H-V1 and the envelope-bound table in section 5, row H-V1/D5a.

Copyright (c) 2026 Bijective Labs. All rights reserved.