BL-P2-000: Programme II, Overview

BIJECTIVE LABS

BIJECTIVE LABS . SEP 7, 2026

BL-P2-000: Programme II, Overview

Status: RELEASED

Classification: PUBLIC

Revision: 1

Date: 2026-09-07

Period: 2026-07-26 to 2026-08-03 Venue: Bybit perpetuals, multi-year substrate Hypothesis budget: 12 declared, 8 used Decision documents: 10

1. Why a second programme

The first programme searched for inefficiencies: narrow, specific dislocations that could be captured if you were quick enough and cheap enough. It closed 22 strategy families and found that at this capital scale it was neither.

The second programme changed the question rather than the effort applied to it.

Does this configuration have a higher probability of being right than wrong, under conditions large enough to matter, rather than in a narrow edge?

The distinction is economic. A narrow edge is only worth having if you can transact cheaply enough to keep it, and the first programme established that the cost structure at this capital scale consumes edges of the size available. So the second programme looked instead for configurations whose expectation is positive after all costs, verified out of sample, and reachable at the capital actually held.

The revised definition of alpha, adopted 2026-07-28, states it directly:

A configuration of condition, signal, and execution whose net expectation is positive after all our costs, verified out of sample, and which fits our capital.

That definition admits risk-premium harvesting, which the earlier definition excluded, and it is why Programme II tested a volatility premium that Programme I would never have proposed.

2. What it is not

The layered structure this programme uses is not a stack of retail indicators with a vote at the end.

Every layer produces a probabilistic assessment with a stated basis, and every decision at every layer must be attributable to a specific piece of information. The design constraint is decision accountability rather than signal count. A layer that cannot say what it observed and why that observation changes the distribution does not enter the specification.

Nothing in the programme outputs a certainty. The first programme's most expensive errors came from treating point estimates as facts, and the second was built to make that harder.

3. Framework Alpha

The eight-layer falsification chain in BL-D-001 is this programme's contribution to the method, and it was formalised at its opening. Three cross-cutting rules came with it.

A five-angle audit before every verdict. Calculation, cost, data, sample, mechanism. Run by default rather than on request.

Every kill carries an axis of death. Idea, data, implementation, cost, condition, or method, so that a falsification does not spread to axes that were never tested.

Dead is not the same as underpowered. A hypothesis is killed only with the statistical power to kill it. Without that power the label is unproven, and the branch stays open.

The third rule is the one that shaped the record most, and BL-D-004 section 5 carries it forward.

4. The budget

The programme capped itself at twelve tested hypotheses in total, declared before any work began.

The cap is a multiple-testing constraint made structural. A programme that can test without limit will eventually find something and will have no way to say what the finding is worth. Eight of the twelve were used, and the programme closed with four remaining rather than spending them for the sake of it.

Alongside the cap, the substrate was partitioned and sealed:

Era	Range	Status
Held-out A	Substrate floor to 2021-12-31	Sealed, unopened
Research	2022-01-01 to 2025-06-30	All development
Held-out B	2025-07-01 onward	Opened once

5. What was measured

Six economic mechanisms, referred to throughout the programme as doors, were each measured to a conclusion. BL-P2-001 carries them in full.

Door	Mechanism	Outcome
1	Enlarge the directional signal	Closed; the conditioning variable does not separate outcomes
2	Reduce cost by quoting first	Closed; the effect decays to nothing across subperiods
3	Get paid for providing liquidity	Closed on adverse selection, after a first closure that was wrong
4	Reach a better fee tier	Closed; requires volume the capital scale cannot generate
5a	Harvest the volatility premium	Closed on capital and venue access; the premium is real
5b	Buy volatility when conditioned	Falsified; the market already prices the conditioning signal
6	Use the signal defensively	Validated and in production

The programme's economic conclusion, recorded 2026-07-28 and unchanged since: no mechanism satisfies measured edge, cost clearance, and capital fit simultaneously.

Door 6 is the exception that proves the framing. The volatility signal is real and validated out of sample twice. It cannot be sold, because that requires an options desk. It can be used to decline trades, and that is what it does.

6. The eight hypotheses

Code	Verdict	Axis
H-P1	FALSIFIED	Idea; crowding predicts volatility expansion, not direction
H-V1	CLOSED	Capital and venue
H-R1	CLOSED	Cost
H-R2	FALSIFIED	Idea; the stop cuts the reversion itself
H-P2	FALSIFIED	Idea; mirror test failed
H-X1	WITHDRAWN	Procedure
H-X2	NOT PROVEN	Power
H-X4	FALSIFIED	Mechanism inverted out of sample

The last one is the only hypothesis in the entire archive to die on that axis. It survived research-era validation, survived conditioning, beat a permutation null by 5.83 sigma, and then reversed sign on sealed held-out data.

Every gate before the seal had passed. Only the seal caught it, and it could only catch it once.

7. How the programme closed

Not with a verdict but with an audit.

On 2026-08-03, while building an unrelated eligibility layer, a discrepancy surfaced that traced to the rebalance hour, a parameter nineteen pre-registrations had used without recording it as a choice. Re-running across all twenty-four phases confirmed the programme's most important kill and revised its most promising number downward by roughly sixty percent.

BL-IN-006 is the case study. The closing assessment in the source is the right summary of the whole programme:

This audit produces certainty, not opportunity.

8. What it produced

Eleven numbered rules, R-01 through R-11 in the programme's own numbering and carried into BL-D-009. A substrate of 12,970,563 rows. An instrument set with tests written before each measurement. And a knowledge map that says which classes of edge on this venue are closed, which are closed only for this operator, and which were never opened.

BL-P2-003 is that map. It is the most directly reusable document in this volume.

9. Cross-references

BL-P2-001 the six doors. BL-P2-002 the gates and the sealed era. BL-P2-003 the knowledge map. BL-P2-004 the decision ledger. BL-IN-006 the closing audit.