BL-D-002: The Operating Envelope as a Research Object

Status: RELEASED

Classification: PUBLIC

Revision: 1

Date: 2026-09-07

This article defines the method. The values Bijective Labs held its own parameters at are declared in BL-FM-002.

1. A verdict is a joint statement

A quantitative verdict is never a statement about the market alone. It is a statement about the market as seen through a particular apparatus, by an operator paying particular costs, reacting at a particular speed, sampling at a particular cadence, holding a particular amount of capital, and able to reach a particular set of instruments.

Most research writing suppresses the second half of that sentence. The result is a literature of findings that cannot be reused, because a reader has no way to tell which conclusions describe the market and which describe somebody else's fee schedule.

The envelope is the second half, written down.

2. The seven parameters

An envelope is defined by seven declared quantities. Five belong to the operator and two to the measurement.

The last two are properties of the research apparatus rather than of the trading operation, and they are the ones most often left undeclared. A verdict computed on a substrate that cannot resolve the premise is not a verdict, and a verdict computed on correlated instruments treated as independent is not a verdict either.

3. Declaring one

Three requirements.

Fix it before the programme, not per hypothesis. An envelope that drifts between tests destroys comparability across the programme. If it must change, the change is dated and every prior verdict is marked as taken under the earlier values.

State the reasoning, not just the value. A cadence of 93 seconds means nothing on its own. A cadence of 93 seconds chosen because the rate budget forces a trade between wallet coverage and refresh speed, and because the research question at that stage was about breadth rather than freshness, tells a reader what would change if they made the other choice.

Publish it where a verdict cannot be read without it. In this record the envelope sits in the front matter, ahead of every volume.

4. Classifying a verdict against the envelope

Every settled verdict takes one of two labels.

UNIVERSAL. The mechanism itself failed. No substitution of any envelope parameter rescues it. The result carries to any operator.

ENVELOPE-BOUND. The mechanism held and a declared parameter blocked it. A reader with a different value on that parameter is testing a different question.

The test to distinguish them is a single question, asked honestly: if every envelope parameter were arbitrarily favourable, would this hypothesis produce a positive expectation?

If yes, it is envelope-bound. If no, it is universal.

A signal whose forward return is positive on both sides of its own conditioning variable stays wrong at zero cost, at zero latency, with unlimited capital. That is universal. A signal producing 3.0 basis points of drift against a 9.0 basis point round trip becomes positive the moment the round trip drops below 3.0. That is envelope-bound, and saying otherwise would be a false report about the market.

BL-D-005 maps axes of death to these labels.

5. The reversal point

An envelope-bound verdict is incomplete until it states the value at which it reverses. This is the part that converts a record of our failures into something another operator can act on.

The computation is usually trivial and should be shown rather than asserted.

Measured gross edge          +3.0 bps at a 30 s horizon, n = 486, p_perm = 0
Round trip, our envelope      9.0 bps  taker to taker, tier 0
Net, our envelope            -6.0 bps  CI95 [-6.5, -5.3]
 
Reversal condition            c_rt < 3.0 bps
Reachable at                  maker to maker, same venue, same tier

Three properties make a reversal point trustworthy.

It names one parameter. If two must move together, say so and give both.

It is arithmetic on measured quantities, not an extrapolation. The gross edge above was measured; only the cost was substituted.

It states reachability. A reversal at a round trip below 3.0 basis points is useful because such a channel exists on that venue. A reversal requiring a negative round trip is not a reopening condition, it is a restatement of the kill.

Where the archive does not contain enough to compute a reversal point, the dossier says which measurement is missing. That absence is a finding, and reporting it is more useful than a guess.

6. The failure mode

An undeclared envelope does not produce a wrong verdict. It produces a verdict that cannot be used, and worse, one that gets used anyway.

The specific damage is contamination. A team that closes three hypotheses in a class, all of them on cost, without recording cost as the axis, will come to believe the class is empty. They will stop proposing hypotheses in it. The belief will outlive the fee schedule that produced it, and no one will notice, because the reasoning was never written down.

Seven of the thirty-seven settled verdicts in this record died on cost. Without the envelope, that record would read as evidence about the market. With it, the record reads as evidence about a fee tier, and it says exactly which fee tier.

7. What the envelope is not

It is not an excuse, and the register is careful not to let it become one.

A parameter is only envelope-bound if it was declared before the measurement. Discovering after a disappointing result that some parameter could have been better, and reclassifying the verdict on that basis, is the same fitting error that pre-registration exists to prevent, arriving through a different door.

An envelope-bound verdict is also still a verdict. The hypothesis is closed for this operator, on these parameters, and no capital is committed to it. The label records who else might reach a different answer. It does not soften the finding.

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