BL-SD-004: Order-Flow Recalibrated to 2 Sigma

Status: RELEASED

Classification: PUBLIC

Revision: 1

Date: 2026-09-07

Venue: Bybit perpetuals Programme: I Verdict: FALSIFIED Axis of death: Cost Transferability: ENVELOPE-BOUND

1. Identification

The Kill Ledger records this hypothesis under the title "order-flow recalibrated to 2 sigma." The source decision record specifies the actual threshold used as 2.5 sigma, up from the 2.0 sigma of the predecessor. Both figures are given here rather than silently reconciled.

2. Economic thesis

Unchanged from BL-SD-003: order-flow imbalance over the top five book levels predicts short-horizon directional continuation. This variant's distinct claim is that tightening the entry threshold and shrinking the exit brackets to match the distribution actually observed, rather than an aspirational wider target, would produce a positive expectation at a smaller magnitude per trade, taken more selectively.

Who is forced to lose. Unchanged: the passive counterparty on the side the imbalance is about to sweep.

3. Falsifiable claim

Pre-registered as the shared winner-promotion criteria for all three Parallel Shadow Portfolio variants:

Each variant produces at least 50 shadow trades in 24 hours, an EV per trade at or above 0 basis points after a 7 basis point blended fee assumption, an aggregate p_tp at or above 0.40, no per-symbol p_sl above 0.30, and a positive mean realised move.

4. Mathematical specification

Notation per BL-D-011.
 
Signal:
 
  s_t  =  ofi_l5_norm(t)
 
Entry:
 
  fire at t when  |s_t| >= 2.5  AND  is_warmed_up(t) = true
  cooldown: 5,000 ms per symbol since last exit
  direction  d_t = sign(s_t)
  order type: Market IOC (taker)
 
Exit, path-dependent, first touch wins:
 
  tp_price   =  entry_mid + d_t * entry_mid * 2 / 10000     (PostOnly, maker)
  sl_price   =  entry_mid - d_t * entry_mid * 4 / 10000     (Market IOC, taker)
  timeout: 10,000 ms, Market IOC reduce-only
 
Sizing:
 
  notional_usd = 5.0
  symbols = {HYPE, TON, SUI, TAO, ENA}     3 dropped from the 8-symbol
                                            universe: INJ and NEAR for
                                            elevated p_sl, LINK for
                                            under-firing
 
Cost, branch by outcome, per source assumption:
 
  c_rt(tp_hit)   =  6.0 (taker) + 1.0 (maker) = 7.0 bps
  c_rt(sl_hit)   =  6.0 (taker) + 6.0 (taker) = 12.0 bps
  c_rt(timeout)  =  6.0 (taker) + 6.0 (taker) = 12.0 bps
 
Net return per event:
 
  r_net(t) = d_t * (exit_mid - entry_mid) / entry_mid * 1e4  -  c_rt(outcome)

Causality holds by construction, identical to BL-SD-003: this variant changes only the threshold and bracket constants, not the entry, exit, or outcome-check logic.

5. Operating envelope

6. Instrument

The directional-alpha strategy module, reused unmodified from BL-SD-003 with a configuration-only change to thresholds and universe; zero lines of new Rust were required. Aggregation via the same shadow analysis tooling.

7. Results

Every criterion but the one that mattered most passed. The EV threshold failed by a wide margin.

8. Verdict, axis of death, and transferability

Verdict: FALSIFIED. Axis: cost. Transferability: ENVELOPE-BOUND.

Three of four pre-registered gates passed cleanly: sample size, hit rate, and stop-loss discipline were all within bounds, and direction was positive and consistent across all five symbols in the reduced universe. The EV gate failed because the mean realised move, +0.64 basis points, sits an order of magnitude below the 7.0 to 12.0 basis point cost the exit paths actually incur.

Applying the test in BL-D-002 section 4: at a sufficiently low round-trip cost, +0.64 basis points is a positive expectation. The classification is envelope-bound. The source decision record states this plainly: direction works, magnitude is structurally smaller than the fee floor.

9. Cost and capacity analysis

Mean realised move (gross)       +0.64 bps
EV per trade (fee-adjusted)      -6.36 bps
Projected daily return           -58.4 pct on $100 notional basis

Capacity was not the binding constraint; at 18,367 trades in 24 hours across 5 symbols the strategy generated ample throughput. The projected daily loss scales with that same throughput, which is the structural problem: a small negative EV multiplied by a high trade count compounds quickly.

10. Envelope sensitivity and reopening conditions

Binding parameter: round-trip cost against a sub-basis-point gross edge.

Gross edge, mean realised          +0.64 bps
Reversal condition                 c_rt < 0.64 bps

No Bybit channel reaches that level. Declared maker-to-maker at 4.0 basis points (BL-FM-002 section 3.1) leaves a net of -3.36 basis points on the take-profit path alone, before accounting for the stop-loss and timeout paths, which remain double-taker under any tier. The mechanism is envelope-bound by the strict test, but the gap to any reachable channel is wide.

This finding is grouped in the Kill Ledger with BL-SD-003. Both report a gross edge of roughly 0.6 to 1.0 basis points against an 11.0 basis point taker-to-taker floor: the same order-flow signal, at the same structural magnitude, tested at two different bracket configurations one day apart. A reader should read the two dossiers as a single finding confirmed twice rather than as two independent discoveries.

11. Reference implementation

The directional-alpha strategy module, shared with BL-SD-003 and published as 05_INSTRUMENTS/strategies/rust/src/directional_alpha.rs. No code defects were found; this measurement changed configuration only. That claim was checked rather than assumed during publication: the entry threshold, bracket geometry, cooldown, timeout and symbol universe differ between the two measurements, and no line of entry, exit or outcome logic does.

12. Evidence register

This measurement is recorded in narrative decision documents; registration as a standalone document-evidence artifact is scheduled per BL-SD-000 section 8.

13. Lessons extracted

Tightening a threshold from 2.0 to 2.5 sigma and shrinking the exit brackets to match the observed distribution improved every gate except the one that determines profitability. That is the specific, reusable lesson: selectivity improves hit rate and reduces drawdown variance, but it does not, on its own, change the ratio between the signal's structural magnitude and the venue's fee floor. The fix that eventually worked was not tighter selectivity within this bracket family; it was abandoning the short-horizon bracket entirely in favour of wider targets, which is the pivot documented in BL-SD-007 and BL-SD-008.

14. Cross-references

BL-SD-003 for the predecessor bracket configuration and the shared fee-floor finding. BL-SD-005 and BL-SD-006 for the two other Parallel Shadow Portfolio variants tested the same day. BL-D-002 section 4 for the envelope classification test. Ledger row PS-A.

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