BL-D-008: The Cost-First Doctrine

Status: RELEASED

Classification: PUBLIC

Revision: 1

Date: 2026-09-07

1. The pre-check

Before a hypothesis is written, before data is pulled, before anyone opens a notebook, write down two numbers.

c_floor    the round-trip cost of the cheapest execution channel available
m_plaus    the plausible magnitude of the signal being proposed

If m_plaus < c_floor, stop. There is nothing to measure.

This is layer 0 of the falsification chain and it is the cheapest test in the entire programme. It takes minutes. It has closed hypotheses that would otherwise have consumed weeks.

2. Why it sits before everything else

The instinct in quantitative research is to establish whether an effect exists and worry about monetising it afterwards. That ordering is correct for academic work and wrong for a trading programme, because it inverts the cost of being wrong.

Establishing that an effect exists is expensive. Establishing that it is too small to trade is arithmetic.

The programme learned this on a hypothesis that did everything right. Aggregated informed-flow bursts produced a directional drift of +3.0 basis points at a 30-second horizon, on 486 events, with a sign-permutation p-value of zero. The signal is real, well powered, and mechanistically sensible. The taker round trip on that venue is 9.0 basis points. Net, the result is -6.0 with a confidence interval entirely below zero.

The measurement was correct and the answer had been available on day one for the price of comparing two numbers.

3. Estimating the plausible magnitude before you have measured it

The obvious objection to layer 0 is that you cannot know the signal magnitude before testing. In practice you can bound it, and a bound is enough.

Three sources of bound, in order of preference:

A comparable measurement already in the record. The archive now contains measured drift magnitudes for a range of flow, identity, and positioning signals on both venues. A new hypothesis in a class already measured inherits an order of magnitude.

The dispersion of the underlying quantity. A signal cannot systematically extract more than the variable it keys on moves. Cross-sectional funding dispersion on this substrate is roughly 15 basis points per settlement across the tradeable universe, which bounds any funding-keyed strategy before a single backtest.

The mechanism's own scale. If the claim is that a participant is forced to transact at a worse price, the size of the concession is bounded by what forces them. Liquidation mechanics, settlement timing, and margin thresholds all have observable magnitudes.

Where no bound can be established, that is itself a layer-0 result: the hypothesis is not yet specific enough to test.

4. The cost floor is an envelope parameter, not a market fact

c_floor belongs to the operator. Fee tier, execution channel, venue, and capital scale all move it, and none of them are properties of the market.

A hypothesis closed at 11.0 is not closed at 3.0. This is why every cost-axis verdict in the record is classified ENVELOPE-BOUND and carries a reversal point, and why published cost instruments take the fee schedule as an input rather than a constant.

Seven of the thirty-seven settled verdicts in this record died on cost. All seven are live questions for an operator with a better schedule.

5. Quoted cost is not realised cost

The schedule above is the floor, not the expectation. Measured execution costs on this substrate ran considerably higher.

The gap between 11.0 quoted and 15.02 measured is spread, slippage against real depth, and the fills that do not happen at the price the model assumed. A layer-0 check run against the quoted schedule is optimistic by roughly a third.

One measured result is worth carrying into any cost model, because it contradicts the first lever most operators reach for. Extending the passive wait budget makes total cost worse rather than better. The orders that fail to fill are not a random subset; they are the names moving away, and their drift compounds while the order rests. Measured drift on unfilled names rose from +11.7 basis points at a 60-second budget to +102.2 basis points at one hour.

Patience is not a cost lever on this substrate. That finding closed a door the programme had expected to walk through.

6. Passive execution has a second cost

Quoting rather than crossing looks like the obvious answer to a cost problem, and the arithmetic of the fee schedule supports it. The arithmetic is incomplete.

A resting order is filled preferentially by counterparties who know something. A census of 676 perpetual symbols found a median quoted spread of 5.26 basis points against a maker round trip well below it, with 62.1 percent of the universe carrying two ticks of room or more [E-0051]. On the fee schedule alone, most of the universe looks profitable to quote.

Markout says otherwise. On the control symbol, post-fill price movement at 60 seconds runs to -0.53 basis points against a half-spread proxy of 0.14 over 9,352,722 trades [E-0002]. The adverse selection exceeds the spread being earned.

This is why Adverse selection is a separate axis in BL-D-005 rather than a subcategory of cost. It is a cost that scales with how attractive your quote is, which means it cannot be escaped by quoting better.

7. What the doctrine buys

Layer 0 does not find edges. It ends conversations.

Its value is measured in the hypotheses that never became projects. That number is not recorded here, because a hypothesis stopped at layer 0 does not consume a pre-registration and does not enter the Kill Ledger. The visible evidence is indirect: the second programme tested eight hypotheses against a budget of twelve and closed six economic mechanisms, in roughly five weeks, on a substrate the first programme had spent three months on.

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