BL-P2-001: The Six Doors
BIJECTIVE LABS
BL-P2-001: The Six Doors
Status: RELEASED
Classification: PUBLIC
Revision: 1
Date: 2026-09-07
1. The framing
A strategy makes money in only a small number of ways. Either the signal is large enough to clear the cost, or the cost is small enough to be cleared, or you are paid rather than charged for transacting, or the risk premium you harvest exceeds what you pay for it, or you use the information defensively instead of offensively.
Programme II enumerated those routes as six doors and measured each one to a conclusion, rather than proposing strategies and discovering the constraint afterwards.
The enumeration is the useful part. It is finite, it is checkable, and closing a door closes every strategy that would have gone through it.
2. Door 1: enlarge the directional signal
The route. If the round trip is 11 basis points, find a signal worth more than 11.
What was tested. Positioning drift. The claim was that a large accumulation of open interest in one direction is followed by directional movement, so a sufficiently extreme reading carries enough drift to clear the cost.
The measurement [E-0020]. At a 72-hour horizon with capped-notional sizing, net +0.94 basis points, CI95 [-1.36, +3.33], n=18,129. At 24 hours, +0.08 with [-1.24, +1.36].
Why it closed. Not the interval, which merely fails to reach significance. The mirror test.
If accumulation predicts direction, then release should predict the opposite direction. Forward return was positive after accumulation at +1.4 basis points and positive after release at +6.7. The conditioning variable does not separate outcomes; it sorts on something correlated with return in both states.
A signal that cannot fail its own mirror has not been shown to carry direction.
Axis: idea. Classification: UNIVERSAL.
3. Door 2: reduce cost by quoting first
The route. If crossing costs 11 basis points and quoting costs 4, quote.
What was tested. An information map across horizons, measuring how much signal survives when entry is delayed to allow a passive fill.
The measurement. The skip-one-bar variant returned gross +5.02 basis points with CI95 [+0.10, +10.82]. The effect is real.
Across three sequential subperiods it returned +8.9, then -1.0, then -3.7.
Why it closed. On decay, not on magnitude. A gross effect that is strongly positive early, flat in the middle, and negative late is not an effect with a cost problem. It is an effect that stopped existing during the sample, and quoting rather than crossing does not restore it.
Axis: idea. Classification: UNIVERSAL.
4. Door 3: get paid for providing liquidity
This door was closed twice, and the first closure was wrong. That sequence is the most instructive thing in this document.
The first closure. A spread census on seven instruments returned a median quoted spread of 1.31 basis points, below any plausible maker round trip. The door was closed on the grounds that the universe does not quote wide enough to pay for making.
Seven instruments is not a census. The closure was revisited.
The re-measurement [E-0051]. All 676 perpetual instruments on the venue, three passes:
| Statistic | Value |
|---|---|
| Median quoted spread | 5.26 bps |
| Mean quoted spread | 8.08 bps |
| Instruments with two ticks of room or more | 62.1 pct |
| Instruments at one tick or less | 37.9 pct |
The median is four times the first estimate. On the fee schedule alone, most of the universe looks profitable to quote, and the first closure had been reached on a sample too small to support it.
One caveat belongs on this measurement and is carried in the artifact rather than in the original narrative: the maker fee was marked unverified at the time of measurement and reported as a grid. The census establishes what the universe quotes, not what a specific operator nets against a confirmed schedule.
The second closure [E-0002]. The door closed on the gate after the spread, which is adverse selection.
A resting order is filled preferentially by counterparties who know something. Measuring post-fill price movement across 9,352,722 trades on the control instrument gives a markout at 60 seconds of -0.53 basis points against a half-spread proxy of 0.14. The adverse selection exceeds the spread being earned, and it does so within ten seconds on five of six candidate instruments.
The single instrument that appeared to survive was rejected on concentration: 72.5 percent of its result came from one day.
Axis: adverse selection. Classification: ENVELOPE-BOUND, because the binding
term is queue position and reaction speed rather than a property of the market.
What the sequence teaches. The first closure reached the right conclusion by the wrong route, and nobody would have known if the door had not been reopened. A verdict that happens to be correct on inadequate evidence is not a verdict, and the programme treated it as one for several days.
5. Door 4: reach a better fee tier
The route. Trade enough volume to qualify for a schedule that makes doors 1 and 2 passable.
Why it closed. Qualification requires 5,000,000 USD of thirty-day volume. At the declared capital scale and maximum sustainable throughput, the distance is roughly 333 times.
This is the shortest entry in the programme and it took minutes to settle. It is included because enumerating a door and closing it on arithmetic is exactly what the enumeration is for.
Axis: capital. Classification: ENVELOPE-BOUND, trivially.
6. Door 5a: harvest the volatility premium
The route. Stop predicting direction. Sell insurance and collect the premium buyers pay for it.
The measurement [E-0007, E-0008].
| Statistic | Value |
|---|---|
| Variance risk premium, BTC | +8.10 volatility points, t=22.0, n=1,277 days |
| Days positive | 75.3 pct |
| Variance risk premium, ETH | +3.98 volatility points, t=7.7 |
| Hedged overlay simulation | +1.51 pct mean, t=2.43, n=42 |
| Unhedged overlay simulation | +0.40 pct mean, t=0.25 |
A t-statistic of 22 is not a marginal finding. The premium is one of the most robust measurements in this archive.
Why it closed. Margin of 6,345 USD per contract with an unbounded loss profile, against a declared working capital of 1,000 to 2,000. Simulated drawdown ranges from -10.5 percent under optimistic assumptions to -79.8 percent unhedged, and -27.4 percent on the pessimistic hedged combination. Running it with managed risk requires 50,000 to 100,000 USD.
Axis: capital and venue access. Classification: ENVELOPE-BOUND.
This is the clearest envelope-bound result in the record. The mechanism is not in question. The operator was the wrong size for it.
7. Door 5b: buy volatility when conditioned
The route. If selling volatility is out of reach, invert it. Use the fragility signal to identify when volatility is underpriced and buy it.
The measurement. The premium in the highest fragility quintile against the lowest: +0.63 volatility points at t=0.54 on BTC, +2.39 at t=1.51 on ETH.
The sign is wrong for the hypothesis. The fragility gate increases the premium rather than decreasing it, meaning that when our signal says fragile, options are already more expensive, not less.
Why it closed. The market prices the conditioning signal. Whatever the fragility measure detects, the options market has detected it too and has already adjusted.
Axis: idea. Classification: UNIVERSAL. This is a genuine falsification and it
is the more interesting of the two volatility results, because it says something
about the signal rather than about the operator.
8. Door 6: use the signal defensively
The route. If the volatility signal cannot be sold, use it to decline.
The measurement [E-0012].
| Statistic | Value |
|---|---|
| Lift over base rate at a 20 pct block rate | 1.49x |
| Volatility expansions avoided | 28.4 pct |
| Brier skill, in sample | +0.01251 |
| Brier skill, out of sample | +0.01053 |
| Sample | 2,356,200 |
| Calibration error, out of sample | 0.02260 |
The in-sample to out-of-sample gap is small, which is the property that matters for a component intended to run in production.
The skill score sits below the programme's own threshold of 0.02 for a
calibrated probability output. Rather than tune until it cleared, the component
was kept in binary mode: it emits pass or block, not a confidence. R-02
requires exactly that.
The lift figure carries a correction. It was first reported as 1.52x and revised to 1.49x after the expansion label was made strictly causal. Both figures are in the record.
Axis: none. This door is open and the component is in production. It is held at
RESTRICTED classification.
9. The verdict on all six
Recorded 2026-07-28 and unchanged: no mechanism satisfies measured edge, cost clearance, and capital fit simultaneously.
That sentence has three clauses and the doors fail different ones. Doors 1, 2 and 5b fail the first. Doors 3 and 4 fail the second. Door 5a fails the third with the first two intact.
Only door 6 satisfies all three, and it satisfies them by not trading.
For a reader whose envelope differs, doors 3, 4, 5a and the cost-bound half of this analysis are open questions rather than settled ones. Doors 1, 2 and 5b are settled for everyone.
Copyright (c) 2026 Bijective Labs. All rights reserved.