Introduction

The Bijective Labs Research Record

An institutional record of proprietary alpha research conducted by Bijective Labs across two derivatives venues between May and September 2026.

This is not a strategy catalogue. Its subject is the apparatus: the method by which a hypothesis is proposed, priced, measured, and killed before it can consume capital. Winners are cheap to publish and expensive to trust. A complete and audited record of what was actually measured, with the instruments attached, is the rarer document, and it is the one held here.

Every quantitative research group accumulates this knowledge eventually. The usual price is time or capital. This record is an attempt to make it available at neither.

What the record contains

Asset	Volume
Pre-registrations and decision records	58
Verdict reports	43
Machine-readable evidence artifacts	59
Research instruments	105 scripts, 20,647 lines
Strategy families carried to a documented verdict	~30
Trading and infrastructure source	28,802 lines Python, 15,713 lines Rust
Data substrate	8.20 GB across two venues
Programme span	2026-05-04 to 2026-09-07, 179 commits

Three strategy families remain open: two event-limited leads that reached a measurement but never the power to settle it, and one that was never characterised at all. Every other family is closed.

The claim this record makes

A disciplined falsification programme closed thirty hypotheses in four months without committing capital to any of them.

On the single occasion when the system touched a live venue, an 18-hour passive-quoting run produced 823 triggers and zero fills. Capital exposure across the whole programme was nil, because the funnel is built to reach a verdict before capital is at risk rather than after.

There is no projection in this record, no simulated track record, and no implied forward return. Bijective Labs publishes a method and the measurements it produced. It does not represent that any strategy is profitable.

How to read a verdict

Every measurement here was taken under a declared operating envelope: a fee tier, an execution channel, a latency band, a measurement cadence, a capital scale, and a defined set of venue access. That envelope is a design decision, chosen and held fixed so results stayed comparable across four months of work. It is stated in the operating assumptions that accompany each programme and verdict, and it should be read before interpreting any result in this record.

Because of it, every verdict carries a transferability classification:

UNIVERSAL. The mechanism itself was falsified. The result carries to anyone who tests it, on any infrastructure. A signal whose forward return is positive on both sides of its own conditioning variable is not expensive, it is wrong.

ENVELOPE-BOUND. The mechanism measured real, and the binding term was a parameter of the envelope rather than a property of the market. A signal worth 8 basis points that does not clear an 11 basis point round trip is closed for an operator paying eleven and open for an operator paying three.

Where a verdict is envelope-bound, the dossier names the binding parameter and states the value at which the verdict reverses. Readers running faster infrastructure, a better fee tier, or wider venue access are expected to reach different conclusions on those families, and the record is built to tell them exactly which ones.

Reading paths

If you are	Start at	You are looking for
An allocator or institutional client	The Kill Ledger, then Volume VI	Evidence of process discipline and intellectual honesty
A quantitative researcher	Volume I, then the Volume IV dossiers	Transferable method and reusable instruments
An engineer	Volume V, then Volume VIII	The measurement plant, and how it was verified
Operating a different envelope	The Kill Ledger, filtered to ENVELOPE-BOUND	Which mechanisms are worth re-testing on your own infrastructure

Nobody is expected to read this record from front to back.

Volume map

Path	Volume	Contents
`01_DOCTRINE/`	I	The method: falsification chain, pre-registration, evaluation standard, rules register
`02_PROGRAM_V1/`	II	First programme, hunting inefficiency
`03_PROGRAM_V2/`	III	Second programme, layered probability
`04_DOSSIERS/`	IV	The Kill Ledger and one dossier per strategy family
`05_INSTRUMENTS/`	V	The measurement apparatus and reference implementations
`06_INCIDENTS/`	VI	The incident book
`07_ECONOMICS/`	VII	Cost, capacity, capital
`08_INFRASTRUCTURE/`	VIII	The measurement plant
`09_APPENDICES/`		Evidence register, notation, statistics, data dictionary
`_evidence/`		Machine-readable artifacts, redacted

Evidence

No number in a released document stands without an evidence identifier of the form [E-nnnn], resolving through the register in 09_APPENDICES/ to a machine-readable artifact under _evidence/. Nothing is transcribed from memory or from a summary. Where a figure was later corrected, both figures appear together with the correction and its cause.

Retractions are published rather than buried. This record contains two verdicts that were withdrawn after audit and one instrument that marked itself superseded after it was found to double-count a cost term. Those documents carry the status WITHDRAWN or SUPERSEDED and remain in place. They are among the strongest evidence here.