BL-SD-028: Momentum Follow After Cascade Exhaustion

Status: RELEASED

Classification: PUBLIC

Revision: 1

Date: 2026-09-08

Venue: Hyperliquid perpetuals Programme: I Verdict: FALSIFIED Axis of death: Idea Transferability: UNIVERSAL

1. Identification

This dossier tests the mirror image of the primary cascade-exhaustion hypothesis. BL-SD-022 found that price continues away from the oracle after exhaustion rather than reverting, which raised the direct question this dossier answers: if the price is not reverting, is it continuing profitably enough to be followed instead. This was pre-registered as the final avenue on this specific event class before the family was closed.

2. Economic thesis

The multi-confirmation cascade hypothesis measured a negative recovery fraction after liquidation-cascade exhaustion: price does not revert toward the oracle, it moves further away. A separate reading of that same fact treats it as evidence of momentum rather than failed reversion. If forced liquidation flow displaces price and the displacement continues rather than closes, a taker entering in the cascade's own direction, either at the onset of the cascade or at the moment exhaustion is detected, might ride that continuation profitably.

Who is forced to lose. The design names the same structural loser as the reversion hypothesis it grew out of, from the opposite direction: a late counterparty on the other side of the continuation, one who is either still absorbing the cascade's forced flow or fading it believing in reversion. If momentum is real, that counterparty is paying for taking the wrong side of a continuing move; if it is not, there is no forced loser and no edge, and this dossier exists to determine which is true rather than to assume it.

3. Falsifiable claim

Pre-registered before any return figure was computed, and explicitly framed as the last avenue on this event class:

Following the cascade direction, entered either at onset or at exhaustion, produces a net positive return at the taker round trip, and beats a random-entry-in-regime null. If every entry point fails at the primary cost, the cascade-exhaustion event class is closed entirely.

4. Mathematical specification

Notation per BL-D-011.
 
Event set, reused without modification from the cascade-exhaustion detector (BL-SD-022):
 
  PRIMARY grid      alpha = 0.75, beta = 0.5%
  HIGH-POWER grid   alpha = 0.6,  beta = 0.25%
  cascade direction d, as detected at exhaustion time t_ex
 
Two entry points, tested as separate arms:
 
  onset entry       t_entry = t0        (ride the cascade itself)
  exhaustion entry  t_entry = t_ex      (ride the continuation after exhaustion)
 
Momentum return, direction d, at horizon H:
 
  mom_ret(t_entry, H) = d * ( mid(t_entry + H) - mid(t_entry) ) / mid(t_entry) * 1e4
 
  Positive mom_ret means price continued in the cascade's own direction.
 
Net return:
 
  net_ret = mom_ret - c_rt
            c_rt primary = 9.0 bps, taker-to-taker round trip (a chasing entry
            crosses on both legs, the conservative assumption)
            c_rt sensitivity = 6.0 bps, taker-to-maker
 
Horizon primary H* = 120 s; secondary {30, 60, 300} s reported as robustness only.

Causality holds by construction: both entry points, onset and exhaustion, are defined from the same event-detection pipeline used and audited in BL-SD-022, which establishes that t0 and t_ex are resolved from data available at or before those times. mom_ret looks forward only from the entry point to the horizon, which is a post-hoc outcome measurement, not an input to either entry decision.

5. Operating envelope

6. Instrument

The momentum-close comparison instrument, reusing the test-driven pure functions from the reversion-edge instrument, reversion_ret, net_ret, the block bootstrap, and the null p-value calculation, with new glue code specific to the two-entry-point comparison. The glue itself was not separately unit-tested, which is recorded here as a downgrade on the instrument rather than smoothed over.

The high-power grid, alpha 0.6, was not run. The primary grid's gross return was negative by a wide enough margin that the pre-registration's own logic treated the result as conclusive rather than marginal, and running a second grid to confirm a result that was not close to the threshold was judged not to add information proportionate to the cost of running it.

7. Results

Both entries fail against the random-entry-in-regime null. Neither entry point approaches the 9.0 basis point cost threshold at the gross level, which means the net figures are not close calls decided by cost assumptions: the direction itself loses money before any fee is applied.

8. Verdict, axis of death, and transferability

Verdict: FALSIFIED. Axis: idea. Transferability: UNIVERSAL.

Both pre-registered entry points fail, and both fail at the gross level, which settles the question this dossier exists to answer without any dependence on the cost model. Following the cascade's own direction after a liquidation-cascade exhaustion event is not profitable, whether entered at the moment the cascade begins or at the moment exhaustion is detected.

This finding, read alongside BL-SD-022, resolves an ambiguity that the reversion hypothesis's own negative recovery fraction had left open. BL-SD-022 measured that price does not revert toward the oracle after exhaustion; it moves further away, a fact that is consistent with two different readings: either the reversion thesis is simply wrong about direction and a momentum thesis is right, or there is no tradeable directional edge at the point of detection at all, in either direction. This dossier settles it: both readings are tested directly and both lose gross. A negative recovery fraction against the oracle is not the same finding as a positive tradeable continuation, and the direct test rejects the second claim even though the first remains true.

The precise statement is that at the point a cascade-exhaustion event is detected, the market is locally efficient with respect to both directions this family tested: fading the dislocation loses, and following it loses. No envelope substitution changes a mechanism that loses at the gross level in both directions it was tested in, which places this squarely in the UNIVERSAL class by the same test applied throughout this record.

9. Cost and capacity analysis

Gross, onset entry           -1.59 bps
Gross, exhaustion entry      -5.84 bps
Round trip, taker-to-taker    9.00 bps

No capacity measurement was taken. The event class this family draws on produces roughly 1.5 to 6 events per day depending on grid, adequate for the sample this dossier reports, but the mechanism failed before frequency became a relevant question at either entry point.

10. Envelope sensitivity and reopening conditions

No binding parameter exists to name, because both tested entries lose at the gross level, before any cost is applied.

Reversal condition:  none computable.

Setting c_rt to zero and adding the 9.0 basis point cost back to both net figures still leaves both entries losing money, -1.59 and -5.84 basis points respectively. No fee schedule, maker rebate, or execution channel converts a negative gross move into a profitable trade in either direction this family tested.

What would reopen it: a demonstration, on data this programme has not seen, that a different horizon or a different definition of the cascade's own direction produces a positive gross return in one of the two directions. That is a testable claim and it was not tested here, because doing so on the same events this verdict was drawn from would be horizon shopping, which the design's own integrity rules and R-12 both prohibit.

What would not reopen it: running the high-power grid on this same event definition. The primary grid's gross figures were not marginal, and the pre-registration's own decision not to run the second grid reflects that the result was conclusive rather than underpowered.

11. Reference implementation

The cascade-exhaustion detector this row reuses is published as 05_INSTRUMENTS/strategies/python/src/cascade_events.py with 27 tests, shared with BL-SD-022. The momentum-close comparison glue specific to this dossier is not published, and that is a decision rather than an omission: section 6 records that it lacked dedicated unit coverage, and publishing untested glue would carry that gap forward under the appearance of having closed it. It should be re-verified against its documented behaviour before any future reuse.

What is published is the half both rows share. BL-SD-022 bet against the cascade and this row bet with it, on the same detector, which is why the detector is the part a reader needs in order to check either verdict.

12. Evidence register

This measurement is recorded in the pre-registration ledger rather than as a separately registered evidence artifact, and is registered as D-019 in BL-A-006.

13. Lessons extracted

This hypothesis closed the cascade-exhaustion event class on Hyperliquid, and the record it leaves is precise about what that means. The programme's internal handoff notes an amendment, dated 2026-07-14, clarifying that this falsification, together with the fade hypothesis in BL-SD-022 and the identity-aware maker hypothesis in BL-SD-027, closes those three specific approaches on this venue rather than closing the venue itself to further research. Subsequent Hyperliquid work, including the whale-cartography and session-structure families elsewhere in this record, was built from the lessons these three falsifications produced rather than as variants of any of them.

The narrower methodological lesson is about reading a null result correctly. A negative recovery fraction was read, provisionally, as suggestive of a momentum opportunity by the team that reported it in BL-SD-022. This dossier is the direct test of that suggestion, and it corrects it: a price that fails to revert is not the same fact as a price that continues profitably, and the difference between those two readings is exactly the gap between an observation and a pre-registered test of that observation.

14. Cross-references

BL-SD-022 for the reversion hypothesis whose recovery-fraction finding motivated this dossier directly. BL-SD-027 for the third falsified Hyperliquid avenue named in the same closing amendment. BL-D-002 section 4 for the reversal-point test applied in section 10. Ledger row M3B.

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