ΣigmaClava

The method

Study, build, publish — and mostly, don’t.

Three stages and one closed loop. The interesting part of the method is not the stages; it is the bar at the exit, and how much of the work never crosses it.

The shape of it

A research house is defined less by what it produces than by what it refuses to produce. Our method is ordinary in its stages and unusual only in its exit condition: the loop is closed, and the default is that work stays inside it.

The diagram below is the whole method. Everything after it is detail about how each stage is actually run.

The research cycle Study, Build and Publish run in sequence, with a return rail carrying what was learned from Publish back to Study, closing the loop. One gated exit leads from Publish to the Library, crossing a bar marked "something true to say". A separate dashed exit leads from Build to OccultOpes, marked "commissioned work", and does not rejoin the cycle. WHAT WE LEARNED Study LIVE WITH IT WHAT WE BUILT Build MAKE THE ARGUMENT WHAT HELD UP Publish ONLY IF IT CLEARS THE BAR SOMETHING TRUE TO SAY Library PUBLIC RECORD COMMISSIONED WORK OccultOpes DELIVERED, NOT PUBLISHED Most work never reaches the bar. That is the design, not a backlog.
Figure 1 The cycle is closed and the exits are narrow. What leaves toward the Library has to clear a stated bar; commissioned work leaves by a different door entirely, and never enters the research record.

The stages

  1. 01

    Study

    Live with the question before answering it.

    The house studies at family scale, in its own home, with lived experience as primary data. This is a deliberate constraint rather than a limitation we are apologising for. A question about what it is like to live with an intelligent system cannot be answered in a session, and it cannot be answered by people who go home at the end of it.

    What this buys is duration and honesty. We see the second month, when novelty is gone and the system is either useful or in the way. We see what a household does when nobody is being observed, because the household is us.

    What it costs is generality, and we will not pretend otherwise. A finding from one home is a hypothesis about homes. We say so, and the standard for publishing rises accordingly.

  2. 02

    Build

    A working system is the argument.

    We research by building. A design that has not been built has not been tested against the part of reality that matters — the part where a person is tired, the network is bad, and the elegant interaction turns out to require a decision nobody wants to make.

    Building is also how a claim becomes falsifiable. A paper can be disagreed with; a system can be used, and using it settles arguments that discussion does not. This is why the unit of publication here is a working thing rather than a description of one.

    It follows that we build fewer things than we could, and finish more of them than is comfortable. An unfinished system is not an argument. It is a sketch of one.

  3. 03

    Publish

    Only when there is something true to say.

    Most of what the cycle produces stays inside it. That is the intended behaviour, not a backlog. A research house that publishes on a schedule is publishing to the schedule, and the work bends to fit.

    What clears the bar goes to the Library, with its method and its limits attached. What does not clear it stays in the house and feeds the next turn of the cycle, which is the more common outcome by a wide margin.

    The bar is deliberately awkward: could we defend this to someone who has read everything else in the field, and who has no reason to be generous? If not, it is not ready, and saying so costs nothing.

The bar

Could we defend this to someone who has read everything else in the field, and who has no reason to be generous?

That is the whole test. It is not a high bar in the sense of requiring a large result — a small, well-bounded, clearly-stated finding passes it easily. It is a high bar in the sense that it is indifferent to how much we would like to publish.

Most work fails it for the same three reasons: the vantage is too narrow to support the claim being made, the effect is real but we cannot yet say what produced it, or the finding is true and already known. None of these are failures of the work. They are the reason the cycle has three stages and not two.

Three basins arranged in a loop with flow circulating between them, and a single thin trail of points leaving the loop at one side.
Plate I The work circulates. The thin trail leaving at the left is everything that has cleared the bar — drawn to scale against what stays in.

What we commit to

We name the vantage
Every finding says where it was seen from — how many households, over how long, and by whom. A result from one home is labelled as a result from one home.
We publish the limits with the claim
The conditions under which we expect a result not to hold are stated in the same document as the result, not left for someone else to discover.
We do not publish to a calendar
There is no quarterly cadence and no minimum volume. A quiet year is a possible outcome of doing this correctly.
We say when we were wrong
A superseded finding stays in the Library with its correction attached. Removing it would make the record tidier and less true.
We separate the research from the commission
Commercial work is delivered through OccultOpes and does not enter the research record. A client cannot buy a finding, and a finding is not a deliverable.

Where this leads

The method exists to serve four questions. Each is a program of inquiry, and each is run this way.

The programs → How intelligence is governed here → What has cleared the bar →