BL-SD-049: Defensive Regime Gate
BIJECTIVE LABS
BL-SD-049: Defensive Regime Gate
Status: RELEASED
Classification: RESTRICTED
Revision: 1
Date: 2026-09-08
Venue: Bybit, production surface Programme: II Verdict: VALIDATED Axis of death: None. In production. Transferability: RESTRICTED
1. Identification
| Field | Value |
|---|---|
| Internal code | D6 |
| Family | Defensive regime gate, production risk overlay |
| Venue | Bybit |
| Programme | II |
| Pre-registered | Not applicable; a risk-management calibration rather than an alpha hypothesis, consuming zero R-3 quota |
| Verdict issued | 2026-07-28, corrected 2026-07-30 |
| Predecessor | The fragility predictor characterised in the information map this gate is built from, shared with BL-SD-048 |
This is the one row in the entire kill ledger that reached production. Per
BL-FM-004 section 4, production risk gate parameters are held at
RESTRICTED classification: what is closed is published in full, and this
component is not closed, it is running.
2. Economic thesis
Not every use of a validated signal is an offensive bet. Where a signal cannot
be sold directly, as BL-SD-047 and BL-SD-048 establish for the volatility
premium, it can still be used to decline: reduce exposure ahead of the
conditions it flags as likely to turn adverse, without ever taking a position
that requires the signal to be right about direction.
Who is forced to lose, if the mechanism works as stated. Nobody, by design. A defensive gate does not extract a payment from a counterparty; it avoids a cost the operator would otherwise have incurred. The gate's value is measured in exposure avoided, not in profit captured, and that distinction is what keeps it inside the programme's own definition of a risk-management overlay rather than a strategy requiring a forced-loser thesis.
3. Falsifiable claim
Withheld under BL-FM-004 section 4. This component is in production, and
its specific trigger thresholds, block-rate policy, and signal weighting are
held at RESTRICTED classification along with every other production risk
gate parameter in the record.
4. Mathematical specification
Withheld under BL-FM-004 section 4.
5. Operating envelope
Withheld under BL-FM-004 section 4.
6. Instrument
Withheld under BL-FM-004 section 4.
7. Results
Withheld under BL-FM-004 section 4.
8. Verdict, axis of death, and transferability
Verdict: VALIDATED. Axis: none. Transferability: RESTRICTED.
The gate passes as a binary pass or fail filter and fails its own calibration
test for a continuous, confidence-weighted output. That second result is not
a defect; it is the programme's own pre-registered rule, R-02, doing
exactly what it exists to do. R-02 requires a calibrated probability output
to clear a Brier skill threshold of 0.02 before it may be used to size a
position rather than only to gate one. The measured skill fell short of that
threshold both in sample and out of sample, and the component was held to
binary mode rather than tuned until it cleared. Tuning a risk threshold by
maximising backtest performance would convert a risk-management decision into
a fitted parameter, and that path was declined deliberately.
The property that matters most for a production component is the near identity between the in-sample and out-of-sample calibration result: the gap between them is small, which is evidence the underlying signal survives a time split rather than merely fitting one. That is a materially different finding from a component that looks strong in sample and collapses out of it, and the record elsewhere documents exactly that failure mode in a different row for contrast.
One correction is recorded here because BL-FM-004 section 5 requires it
regardless of classification: an independent audit on 2026-07-30 found the
original expansion label had been defined using information from both halves
of the evaluation split, which is not strictly causal. Correcting the label
to use only the calibration half changed the headline lift figure from 1.52x
to 1.49x at a 20 percent block rate, and the volatility-expansion share
avoided from 28.8 to 28.4 percent. The corrected figures are the only ones
that may be cited; the classification of this row does not change and did not
change under the correction.
9. Cost and capacity analysis
Withheld under BL-FM-004 section 4.
10. Envelope sensitivity and reopening conditions
Withheld under BL-FM-004 section 4.
11. Reference implementation
Withheld under BL-FM-004 section 4.
12. Evidence register
Withheld under BL-FM-004 section 4. Evidence identifiers E-0010 through
E-0012 resolve to this row in the appendix evidence map and carry the
calibration and lift figures cited in section 8; their full content is not
reproduced here.
13. Lessons extracted
Withheld under BL-FM-004 section 4.
14. Cross-references
Withheld under BL-FM-004 section 4. Ledger row D6.
Copyright (c) 2026 Bijective Labs. All rights reserved.