BL-SD-007: Taker Wider Edge on a Static Universe

Status: RELEASED

Classification: PUBLIC

Revision: 1

Date: 2026-09-07

Venue: Bybit perpetuals Programme: I Verdict: NOT PROVEN Blocking axis: Sample Transferability: -

1. Identification

This is the pivot away from short-horizon brackets entirely. Every hypothesis from BL-SD-003 through BL-SD-006 had shown a real order-flow signal too small to clear Bybit's fee floor at horizons under a minute. T1 tested the opposite lever: hold longer, target wider, trade less often.

2. Economic thesis

A historical backtest over 78 million microstructure rows identified specific signal, symbol, hour-of-day, and horizon combinations where price reached a 100 basis point favourable move within a bounded horizon at a win rate above 55 percent, wide enough to clear the venue's taker round trip with margin even at the worst-case fee tier. The thesis carries this forward as a live shadow strategy: fire the T0-validated signal types on the historical 8-symbol Tier 1.5 universe, restricted to the hours of day the backtest found favourable, and hold toward a 100 basis point target with a 50 basis point stop.

Who is forced to lose. Unchanged from the order-flow family: the passive counterparty on the side being swept. The distinguishing claim here is magnitude and horizon, not a new structural mechanism.

3. Falsifiable claim

Pre-registered before the shadow window began, per the roadmap's Phase T1 acceptance gate:

Shadow EV per trade is positive, aggregate p_tp is at or above 0.50, no per-symbol p_sl exceeds 0.30, and at least 20 shadow trades accumulate within the observation window.

4. Mathematical specification

Notation per BL-D-011.
 
Signal: T0-validated triggers, evaluated per symbol in the static universe
(the same construction later generalised in BL-SD-008):
 
  ofi_3sigma:    |ofi_l5_norm(t)| >= 3.0
  vpin_extreme:  vpin(t) >= 0.85  AND  |ofi_l5_norm(t)| >= 1.0
  spread_spike:  spread_bps(t) >= 2.0  AND  |ofi_l5_norm(t)| >= 0.5
 
Entry, fire at t when a trigger fires AND both additional gates hold:
 
  symbol in {HYPE, TON, SUI, TAO, ENA, LINK, NEAR, INJ}      (static universe)
  hour_of_day_utc(t) in the T0-validated active-hour set      (source does
                        not preserve the exact hour set actually run;
                        the T0 backtest's top combo used UTC hour 14)
 
  direction  d_t = sign(ofi_l5_norm(t))
  order type: Market IOC (taker)
 
Exit, path-dependent, first touch wins:
 
  tp_price  =  entry_mid + d_t * entry_mid * 100 / 10000
  sl_price  =  entry_mid - d_t * entry_mid * 50  / 10000
  timeout: 4 hours, per the T0-validated horizon
 
Sizing:
 
  notional_usd = 5.0     per entry
 
Cost:
 
  c_rt = 2 * fee_taker    exact per-trade blended assumption for this
                            specific 26-hour window is not separately
                            preserved in the source; the T0/T1 roadmap
                            assumes a worst-case 12.0 bps taker-to-taker
                            floor for the acceptance gate

Causality holds for the signal triggers themselves, each read from state at or before t. The hour-of-day gate is deterministic and causal by construction. The exact per-symbol cooldown value actually configured for this specific 26-hour run is not preserved in the source beyond the roadmap's general specification of "cooldown per symbol"; that gap is stated here rather than reconstructed.

5. Operating envelope

6. Instrument

The taker wider-edge strategy implementation, later generalised for BL-SD-008, run in shadow mode against the static Tier 1.5 universe with the hour-of-day filter active. The instrument itself carries test coverage; the finding here is about event supply from the market, not about the instrument's correctness.

7. Results

Five trades in 26 hours is below the pre-registered power floor by a wide margin; no downstream metric in the acceptance gate can be evaluated meaningfully at this sample size.

8. Verdict, blocking axis, and transferability

Verdict: NOT PROVEN. Axis: sample. Transferability: dash, per BL-SD-000 section 2.

This is not a falsification. The diagnosis, reached the same day the shadow window closed, is that the static 8-symbol Tier 1.5 universe is too efficient a pool for a wider-edge strategy targeting moves of 100 basis points or more: Universe Discovery 2.0's subsequent 570-symbol, 7-day analysis found that three of the eight Tier 1.5 symbols did not pass a movement-frequency filter at all, and the median altcoin in the broader market produced roughly 20 qualifying moves per week against the Tier 1.5 set's 23. The event supply the hour-filtered static universe could offer was never going to reach the pre-registered floor of 20 trades in 26 hours.

Per BL-D-005 section 2, Power is explicitly not a death axis: it produces NOT PROVEN and the hypothesis stays open rather than closed. Nothing here says the T0-validated signal or the 100/50 basis point bracket does not work; it says the pool it was tested against did not supply enough events to find out.

9. Cost and capacity analysis

Neither cost nor capacity could be assessed. Five events is insufficient to estimate a per-trade EV with any confidence, and the finding that closes this line is about event frequency rather than about unit economics.

10. Envelope sensitivity and reopening conditions

Binding parameter: universe breadth, not round-trip cost.

Reversal condition:  not computable on cost; the constraint here is
                      event supply, not price.

This is a case where the reversal arithmetic that section 10 asks for does not apply in its usual form, and the honest statement is that a different measurement, not a cheaper fee, is what reopens this line. Universe Discovery 2.0 measured the fix directly: the top 20 candidate symbols by a movement-spec composite score fired 30 to 37 qualifying moves per 7 days against the Tier 1.5 set's typical 20 to 28, a 1.4 to 1.8 times improvement in frequency and a 2.5 times improvement in typical magnitude. BL-SD-008 is that reopening, executed within 24 hours of this result.

What would reopen it on the static universe as specified: a materially longer observation window, since the underlying signal-and-bracket logic was never shown to fail, only to be under-sampled. That path was not taken; the universe was widened instead.

11. Reference implementation

The taker wider-edge strategy implementation, shared with BL-SD-008 and published as 05_INSTRUMENTS/strategies/rust/src/taker_wider_edge.rs, with 15 tests.

The sharing claim was checked rather than assumed during publication. The strategy core takes its instrument universe as a parameter and holds no selection logic of its own, so the difference between this hypothesis and BL-SD-008 is entirely which symbol list was supplied to it. The rolling universe machinery that produced the second list sits upstream of the strategy and is not part of this core. That is also why the finding here is about event supply rather than about the signal: the same signal code, on a static list, simply did not fire often enough to evaluate.

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

Widening the bracket and lengthening the horizon fixed the magnitude problem diagnosed across BL-SD-003 through BL-SD-006, but it introduced a new constraint the programme had not yet measured: an efficient, well-monitored universe of eight symbols does not produce 100-basis-point dislocations often enough to sample. The fix was not a better signal; it was a bigger pool. That reframing, from signal quality to universe breadth, is the direct ancestor of the rolling-universe design in BL-SD-008.

14. Cross-references

BL-SD-003 through BL-SD-006 for the short-horizon order-flow family this strategy pivoted away from. BL-SD-008 for the rolling-universe successor that resolved the sample constraint found here. BL-D-005 section 2 for the power-axis and NOT PROVEN distinction. Ledger row T1.

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