BL-SD-001: Order-Flow Taker Entries, the Original Configuration

Status: RELEASED

Classification: PUBLIC

Revision: 1

Date: 2026-09-07

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

1. Identification

This is the strategy the programme was running before any research discipline existed to test it against. It is entry zero in the record for that reason: the closure of this line is what motivated the pre-registration and cost-first disciplines that everything after it follows.

2. Economic thesis

Order-flow imbalance, aggregated over the top five book levels and normalised to a rolling standard deviation, predicts short-horizon price direction: a one-sided sweep of resting liquidity signals that informed size is working the book, and price should continue in the sweep direction long enough to cross a Market IOC round trip. Entries fired when the normalised imbalance exceeded a fixed threshold, taken immediately as a taker fill.

Who is forced to lose. The passive side of the sweep, the same structural counterparty every order-flow thesis in this record names. The thesis did not fail because the counterparty was wrong. It failed because the venue's taker fee schedule left nothing behind after paying for entering and exiting the trade.

3. Falsifiable claim

No falsifiable claim was pre-registered. The strategy was run live in a demo account, observed for 25 hours, and the fee arithmetic that closed it was computed after the fact from the observed win rate rather than checked against a stated threshold in advance. That absence is recorded here rather than smoothed over, because it is precisely the gap BL-D-003 was later written to close.

The breakeven identity itself functioned as the kill criterion once computed, even though it was not written down before the data existed.

4. Mathematical specification

Notation per BL-D-011.
 
Signal, order-flow imbalance over the top 5 book levels, Welford-normalised:
 
  s_t  =  ofi_l5_norm(t)      z-score of level-5 order-flow imbalance
                               at or before t
 
Entry:
 
  fire at t when  |s_t| >= theta      theta = 1.0 (pre-recalibration value
                                       in force during the 356-fill window)
  direction  d_t = sign(s_t)
  order type: Market IOC (taker), age filter: signal rejected if older than
              500 ms at submission
 
Exit:
 
  Not specified by any coded rule. The source documents this explicitly as
  Bug Layer 2: a configured max_hold_ms of 2000 existed in the parameter file
  but was never read by any execution path. Positions accumulated until a
  later opposite-direction signal closed them, or remained open. The realised
  win/loss figures below are therefore descriptive statistics of whatever
  fills happened to close a position, not the outcome of a designed exit.
 
Sizing:
 
  Base order size per fill; the source does not state a fixed notional for
  the 356-fill window specifically.
 
Cost, as assumed by the source at measurement time:
 
  c_rt = 2 * 6.0 bps = 12.0 bps      taker fee, both legs, non-VIP
 
Net return per event (breakeven identity used to close the line):
 
  WR_breakeven = (avg_loss + c_rt) / (avg_win + avg_loss)

Causality holds for the entry: s_t is computed from book state at or before t, and the age filter discards stale signals. Causality for the exit cannot be assessed, because no exit rule existed to check. This is stated as a defect rather than reconstructed after the fact.

5. Operating envelope

6. Instrument

No dedicated backtest or shadow-analysis tool measured this line; it was read directly from demo account fill records and reconciled by hand into a win rate, an average win, and an average loss. That is weaker evidence than an in-repository instrument with test coverage, and the verdict is downgraded accordingly in how much weight it carries as a measurement, though not in its conclusion, because the conclusion is arithmetic rather than statistical.

A second instrument defect compounds the first. The position tracker itself was found, in the same investigation, to treat every sell as a reset to flat rather than as a proper short (recorded in the dated position-tracking-fix report, 2026-05-24), and the absence of any coded exit rule (section 4) means the realised win and loss figures reflect whatever fills happened to close exposure, not a consistent trading rule. The win rate and fee arithmetic are accepted here because they close the line on cost grounds regardless of these defects; a reader should not treat the underlying 356-fill sample as a clean dataset for any purpose beyond that closure.

7. Results

No confidence interval is reported because none was computed at the time; the closure did not require one. Breakeven win rate above 100 percent is a structural result, true for any sample the strategy could have produced under the stated cost and payoff asymmetry.

8. Verdict, axis of death, and transferability

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

The breakeven win rate the strategy needed, 122.2 percent, exceeds the maximum possible win rate of 100 percent. No sample size and no amount of selectivity changes that arithmetic while the payoff asymmetry and the cost stay fixed. This is a closure by construction rather than by statistical power.

Applying the test in BL-D-002 section 4: at zero round-trip cost, the required win rate falls to avg_loss / (avg_win + avg_loss) = 10 / 18 = 55.6 percent, and the observed 64 percent clears that bar. The mechanism is therefore envelope-bound, not universal: a sufficiently cheap round trip converts this into a working strategy against the same signal and the same observed win rate.

9. Cost and capacity analysis

Fee, taker to taker, as assumed          12.0 bps
Avg win                                   8.0 bps
Avg loss                                 10.0 bps
Breakeven win rate required             122.2 pct
Observed win rate                        64.0 pct

Capacity was not assessed; the line was closed on unit economics before frequency or size became a relevant question.

10. Envelope sensitivity and reopening conditions

Binding parameter: round-trip cost.

Breakeven WR(c_rt) = (10 + c_rt) / 18
 
Solve for the cost at which observed WR (64 pct) equals breakeven WR:
  0.64 = (10 + c_rt) / 18
  c_rt = 0.64 * 18 - 10 = 1.52 bps
 
Reversal condition:  c_rt < 1.52 bps

This is the honest reading and it is a hard one. The cheapest channel this venue's declared schedule offers is maker to maker at 4.0 bps round trip (BL-FM-002 section 3.1). At that cost the breakeven win rate is (10 + 4) / 18 = 77.8 percent, which the observed 64 percent does not clear. Taker to taker at the declared 11.0 bps and maker to taker at 7.5 bps fail by a wider margin still.

No channel Bybit publishes at any fee tier reaches 1.52 basis points. The mechanism is envelope-bound by the strict test in section 8, because zero cost would rescue it, but no reachable envelope on this venue does. A reader holding this signal and this observed win rate should treat it as closed in practice, not merely closed at the current tier.

11. Reference implementation

None. This line predates the Rust strategy framework; the OFI computation it used was later carried into the directional-alpha strategy module and its siblings, which are the reference implementations for BL-SD-003 through BL-SD-006.

12. Evidence register

This measurement is recorded in narrative decision documents rather than as a second independent machine-readable artifact; the single evidence row above is the one resolved artifact per BL-SD-000 section 8.

13. Lessons extracted

This is the founding lesson of the entire record, and it produced two rules rather than one. BL-D-008, the cost-first doctrine, exists because this closure took 25 hours and hundreds of fills to reach when the breakeven arithmetic could have been checked before a single order was sent. BL-D-003, the pre-registration standard, exists because this line had no falsifiable claim to check the result against; the win rate was interpreted after the fact rather than measured against a threshold fixed in advance.

The instrument defects found alongside the closure, an asymmetric position tracker and an unread exit-timeout parameter, are the earliest recorded instance of the failure mode BL-D-006 was later written to prevent: a verdict is only as good as the pipeline that produced the numbers behind it.

14. Cross-references

BL-SD-003 through BL-SD-006 for the order-flow-derived strategies built after the pivot away from this line. BL-D-002 section 4 for the envelope classification test applied in section 8. BL-D-003 and BL-D-008 for the two disciplines this closure motivated. Ledger row T0.

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