BL-SD-034: Session Structure in Event Supply

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

TOC1 closed the cost lever and, in doing so, raised a second question rather than answering it. Trade rate and realised volatility concentrate in the US session at two to three times the trough. If cascade and exhaustion events follow the same concentration, an operator gains a supply-side reason to favour that window even though the cost of trading in it is unchanged. This dossier tests that implication directly, on the same event definition the predecessor hypothesis to BL-SD-033, BL-SD-032, already uses.

2. Economic thesis

Activity is not the same thing as tradeable event supply. A session with more trade rate and more realised volatility has more of everything, but the specific event this programme trades, cascade exhaustion under the BR1 grid, requires a forced-liquidation displacement of a particular shape, not merely more ticks. If exhaustion events cluster with activity, an operator gains a timing lever on the number of opportunities available per unit of session time, independent of anything about their cost. If exhaustion events are smeared evenly across the clock regardless of how much else is happening, that lever does not exist and volume concentration is a fact about the tape, not about the strategy.

There is no counterparty forced to lose in this measurement. It is a question about when a pre-defined event fires, not a trade, and no forward return is computed anywhere in this instrument. That is deliberate, for the same reason it was deliberate in the predecessor: this is the cheapest way to close a lever, because it requires no cost model, no sizing rule, and no P&L, only a count against a clock.

3. Falsifiable claim

Pre-registered before the instrument was run, as an amendment to the TOC1 specification:

Cascade exhaustion events, detected under the pre-registered BR1 grid (alpha 0.75, beta 0.5 percent, flow from trades, unchanged from the predecessor hypothesis), are not distributed proportionally to session clock share. Concentration is measured as the ratio of a session's share of events to its share of the 24-hour clock; a ratio near one for every session falsifies the claim.

The grid is fixed by construction rather than searched. Using the BR1 detection parameters exactly, rather than tuning a grid for this measurement, removes a degree of freedom that would otherwise let this dossier manufacture structure by choosing a favourable definition of event after the fact.

4. Mathematical specification

This is a specification of the measurement rather than of a strategy. No signal fires a trade, no position is sized, and no cost is charged anywhere in this instrument; the fenced block below states what is counted and how concentration is computed, in place of the signal, entry, exit, sizing, and cost terms a strategy dossier would otherwise carry.

Notation per BL-D-011.
 
Event set:
 
  E  =  detect_events( grid = { alpha = 0.75, beta_pct = 0.005,
                                 band_pct = true, flow_from_trades = true } )
 
         reused without modification from the BR1 detector, hl_m2_cascade_events.
 
Bucket assignment, per event e with exhaustion time t_ex(e):
 
  utc_hour(e)      = floor( t_ex(e) / 3,600,000 ) mod 24
  session(e)       = session_of_hour( utc_hour(e) )
                       ASIA 00-08 UTC, EU 07-16 UTC, US 13-21 UTC, OFF the remainder
 
Clock share, derived from the same session mapping rather than hardcoded:
 
  clock_hours(s)   =  count of h in 0..23 with session_of_hour(h) = s
  clock_share(s)   =  clock_hours(s) / 24
 
Observed share:
 
  n(s)             =  count of e in E with session(e) = s
  obs_share(s)     =  n(s) / |E|
 
Concentration index:
 
  conc(s)          =  obs_share(s) / clock_share(s)
 
     conc(s) = 1     event supply proportional to session length
     conc(s) > 1     session supplies more events than its clock share predicts
     conc(s) < 1     session supplies fewer

Causality holds by construction. t_ex(e) is the exhaustion time already assigned by the BR1 detector, which resolves only on data at or before that instant; the hour and session bucketing are deterministic functions of that timestamp with no forward reference. Nothing in this specification touches a future price, a forward return, or a markout. The prohibition is explicit in the pre-registration: computing any forward-return quantity at this stage would have converted a supply-count into a strategy screen without a separate pre-registration to govern it, which R-12 and the TOC1 amendment both bar.

5. Operating envelope

6. Instrument

BL-IS-012, test-driven with 7 of 7 unit tests passing, full suite of 406 tests run clean. The instrument reuses the BR1 detection pipeline, hl_m2_cascade_events.detect_events, without re-derivation, and reuses session_of_hour from the TOC1 instrument so that the session mapping is identical across all three dossiers in this family.

The sample is small by design of the underlying detector rather than by a choice made here: cascade exhaustion under the BR1 grid is a rare event, and this instrument inherits whatever rate the predecessor hypothesis already established. A larger session-level sample requires a longer recording window, not a different measurement.

7. Results

Expected event counts under a strict clock-proportional null, for reference against the observed 11, 12, 13, 4: 11.7, 10.0, 13.3, 5.0. The observed counts sit inside the range a null this thin would produce by chance; no session's concentration index departs from 1 by more than the noise floor at n=40 supports.

The result stands in direct tension with what the predecessor's activity finding would predict. BL-SD-033 found trade rate and realised volatility peaking in the US session at two to three times the trough. If tradeable event supply tracked activity, the US concentration index here would read well above 1. It reads at 0.98, indistinguishable from proportional.

8. Verdict, axis of death, and transferability

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

The idea under test, that cascade exhaustion events cluster with session activity and so hand an operator a supply-side timing lever, fails on its own terms. Concentration indices of 0.80 to 1.20 across four sessions, none outside the range n=40 would produce under a proportional null, is a flat result. Event supply tracks clock time, not activity.

This corrects a hypothesis this programme raised against itself. TOC1 stated plainly that volume and volatility concentrating in the US session made it plausible that tradeable events would concentrate there too, and named that as the live half of the concentration question TOC2 was written to settle. It settled against the prior. Activity concentration and event-supply concentration are not the same measurement and one cannot be inferred from the other; both had to be checked, and checking them separately is the finding worth keeping.

The verdict is UNIVERSAL rather than envelope-bound because there is no parameter of the operator's cost schedule, capital, or execution channel anywhere in this measurement to substitute against. The finding is a fact about when this venue's forced-liquidation flow occurs, not about the terms under which any operator trades it.

9. Cost and capacity analysis

Not applicable to a characterisation. No cost was modelled and no capacity question arises from a pure event count.

10. Envelope sensitivity and reopening conditions

No binding parameter exists to name, because the concentration indices show no departure from proportional to substitute a parameter against.

Reversal condition:  none computable.

A concentration index inside 0.80 to 1.20 on n=40 leaves no room for an envelope-substitution argument the way a cost finding would. There is no operator parameter, taker fee, latency band, or capital scale that turns a proportional event distribution into a concentrated one; the number that would need to move is the event count itself, which is a property of the market's forced-liquidation flow, not of this operator's terms.

What would reopen it: materially more data. At n=40, four sessions each carry roughly ten events, and a true concentration index in the 1.3 to 1.5 range could plausibly still read as 0.80 to 1.20 by chance at that sample size. A rerun at several times this sample, still on the fixed BR1 grid, would either confirm the flat reading with tighter intervals or surface a departure this measurement was underpowered to see. That rerun was not performed here and is not scheduled; it is named as the honest condition under which this verdict would be revisited.

What would not reopen it: re-deriving the event grid to search for a definition of exhaustion that produces session structure. The grid is fixed to BR1 specifically to remove that degree of freedom, and searching it after seeing this result would be the p-hacking R-12 exists to prevent.

Combined with BL-SD-033: two of the three timing levers named in the TOC1 amendment are now closed. Cost does not vary by session; event supply does not either. Only reversion quality, tested in BL-SD-035, remains.

11. Reference implementation

BL-IS-012 is the specification, and the shared session machinery is published as 05_INSTRUMENTS/strategies/python/src/session_structure.py with 52 tests. The BR1 detector this row depends on remains scheduled; the session partition it feeds is what is out.

The distinction matters for a row whose finding is about event supply. What the published module establishes is that every hour of the day receives exactly one session label under a rule fixed before any counting, which is the property an event-supply comparison across sessions rests on. Whether enough events arrive in each session is then a question about the detector's output rather than about the partition, and the detector is the half still scheduled.

12. Evidence register

This measurement is recorded in the narrative report shared with BL-SD-033 and BL-SD-035, backed by the instrument's own reproducible output, and is registered as D-026 in BL-A-006.

13. Lessons extracted

Activity and opportunity are not interchangeable, and a programme that infers one from the other without measuring both separately will build a timing lever that does not exist. TOC1 produced a plausible-sounding implication; this dossier is the discipline of testing that implication rather than carrying it forward as an assumption, and it happened to be wrong. The predecessor's own honesty about the implication being untested, not yet measured, is what made this dossier possible to write cleanly: the claim was named as a hypothesis before it was checked, not asserted as a finding it never earned.

14. Cross-references

BL-SD-033 for the cost measurement this dossier follows and for the activity-concentration fact whose implication this dossier tests. BL-SD-035 for the companion measurement on reversion quality by session, the third and final lever in this family. BL-SD-032 for the cost-bound hypothesis this family of measurements was written to serve. Ledger row TOC2.

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